Escolar Documentos
Profissional Documentos
Cultura Documentos
Structural Dynamics
Edited by
J. DONA
ISBN 0-85334-859-6
EUR 6693 EN
LEGAL NOTICE
Neither the Commission of the European Communities nor any person acting on behalf of
the Commission is responsible for the use which might be made of the following information.
This book, is based on the lecture notes prepared for the advanced course on
structural dynamics held at the Joint Research Centre of the Commission of
the European Communities, lspra Establishment, in October 1978.
Although the subject of structural dynamics has a long history, the last
two decades have seen a remarkable development of numerical methods for
dynamic analysis of engineering structures. The intention in organising the
lspra Course was to provide a review of modern numerical procedures for
structural dynamics and to illustrate their use by application to significant
practical problems.
The first part of the book covers techniques for dynamic response in the
frequency domain as well as methods for transient response by time
integration. The variational methods of structural dynamics are presented
and their finite element implementation discussed in detail. The use of
modal methods for calculating a transient response is described with a
particular emphasis on the problem of reducing a multi-degree of freedom
system to a smaller set of degrees of freedom. A review is then made of
implicit, explicit and combined explicit-implicit operators for transient
response by time integration. This is followed by a presentation of finite
element formulations adapted to implicit and explicit time integrators.
The second part of the book is devoted to a detailed presentation of
practical applications. These cover a wide range of situations, including
dynamic fracture mechanics, anti-missile design, pipe whip accidents and
transient fluid-structure interaction. The latter topic is given particular
attention in view of its growing importance in nuclear reactor safety studies.
Applications are described based on finite difference, finite element and
boundary integral methods.
The present book will prove to be useful to researchers and engineers
engaged in the development and use of numerical methods for dynamic
analysis of structures.
JEAN DONA
Contents
Preface . . . . . . . . . . .
Lisi oj Contributors . . . . . . . . ix
Index 465
List of Contributors
C. ALBERTINI
Commission of the European Communities. Joint Research Centre,
Applied Mechanics Division, J 21020 lspra. Italy.
T. BELYTSCHKO
Department of Civil and Nuclear Engineering, The Technological
Institute, Northwestern University. Evanslon, Illinois 60201, USA.
J. DONA
Commission of the European Communities. Joint Research Centre,
Applied Mechanics Division. 1 21020 lspra, Italy.
P. FASOLI-STELLA
Commission oj the European Communities. Joint Research Centre,
Information Analysis Division, I 21020 lspra. Italy.
K. FULLRD
Central Electricity Generating Board, Research Division, Berkeley
Nuclear Laboratories, Gloucestershire GLI3 9PB, UK.
M. GERADIN
Laboratoire de Techniques Aronautique s et Spatiales, Universit de
Lige, Rue du Val-Benoit 75, B-4000 Lige, Belgium.
R. J. GIBERT
Commissariat l'Energie Atomique, Division des Etudes Mcaniques
et Thermiques, Centre de Saciar. F-91190 Gif-sur-Yvette, France.
ix
X LIST CONTRIBUTORS
J. P. HA LLEUX
Commission of the European Communities, Joint Research Centre,
Applied Mechanics Division, 1 21020 lspra, Italy.
A. V. JONES
Commission of the European Communities, Joint Research Centre,
Information A nalysis Division, I 21020 lspra, Italy.
S. W. KEY
Division 5521, Sandia Laboratories, A lbuquerque, New Mexico87185,
USA.
R. KRIEG
Kernforschungszentrum Karlsruhe GmbH, Institut fr Reaktorent
wicklung, Postjach 3640, D7500 Karlsruhe 1, West Germany.
N. J. KRUTZIK.
Kraftwerk Union A .G.. Berliner Str. 295299, D6050 OJjenbach
(Main), West Germany.
L. LA ZZERI
AMN, Via P. Pesce, 1 16151 Genova Sampierdarena, Italy.
M . MONTA GNA Nl
Commission of the European Communities. Joint Research Centre.
Applied Mechanics Division, 1 21020 lspra, Italy.
D. R. J. OWEN
Department oj Civil Engineering, University of Wales, Singleton Park,
Swansea SA 2 8PP, UK
J. REYNEN
Commission of the European Communities. Joint Research Centre,
Systems A nalysis Division. I 21020 lspra. Italy.
Introduction and General Overview of Structural
Dynamics Problems
J. REYNEN
Commission of the European Communities, Joint Research Centre, lspra,
Italy
1. INTRODUCTION
The advent of digital computers some two decades agoand the development
of numerical methods such as the finite element method (FEM) exploiting
fully the typical features of digital computers (speed and memory) have
caused a revolution in the approach to technical problems, i.e. a shift from
testing towards analysis.
Indeed, until recently the structural engineer has depended on testing to
serve (1) as a proof of design and (2) as a proof of manufacture.
Constant improvement in instrumentation and test equipment increased
the dependence upon testing. However, as the engineer became more
proficient at analysis, he found out that the cost of using analysis for design
certification is far less than that of testing and that the time lapse between
completion of a design and the performance results is far shorter.
Moreover, the engineer is now able to inquire into the behaviour of
situations and locations too remote to be accessible by testing and
instrumentation, of particular importance in the field of postulated
accidents with nuclear reactors.
This triple impact of lower cost, less time and more complete information
has given modern analysis its proper role: analysis is for proof of design;
testing is for proof of manufacture.
In the last decade this revolution has caused a social impact in the sense
that fewer testing personnel are employed and a great demand exists for
digital programmers to make the analysis process more efficient by
developing pre- and post-processors, computer graphics, remote comput-
ing techniques, etc. Testing laboratories have been recycled towards the
more fundamental experiments to define physical properties (constitutive
XI
xii j . REYNEN
{
pure Lagrangian.
updated Lagrangian.
convective coordinates
The choice is mostly dictated by the physics of the problem. Even if
Lagrangian formulations are preferredbecause of the numerical
difficulties in dealing with the non-linearconvective terms and with material
INTRODUCTION AND GENERAL OVERVIEW XVII
3. CONCLUSION
An attempt has been made to highlight the role of modern computerised
analysis in the process of design and of the safety assessment of today's
technical achievements.
XVIII J. REYNEN
1. INTRODUCTION
FIG. 1.
system of particles (Fig. 1). The mass particles m7 are supposed free of
kinematic restraints, and the resulting 3 degrees of freedom system is
assumed to have a potential energy V(r'a) depending only upon the
3N(N l)/2 coordinate differences:
/ = 1.2.3: = ; > y. (1)
For conciseness, let us adopt a matrix formalism. Equation (1) can be
rewritten
r = Lx (2)
where L isa Boolean matrix of dimension 3N( l)/2 3V. If theelastic
restrains of Fig. 1 have spring stiffnesses k^,. alternative forms of the
potential energy of the system are
V(r) = hJSr and V(x) = i v ' Kx (3)
( \') = 0 (6)
M. GERA DIN
c . ' , d cT\ tV
~ - = 0
.Y
,
+ I
"
\) cx
from which result the 3/V Lagrange equations of motion
d (cT\ CV
- + = 0 (8)
\(/ cx
By referring to the explicit forms of 7~and Fin terms of kinematic variables,
we obtain the dynamic equilibrium equations
Mx + Kx = 0 (9)
describing free vibration motion.
[T + p\x - ) - F]dr = 0
=0 (15)
(31)
0 < \ < ; < < .
and by taking variations with respect to ,v, it follows immediately that eqn.
(32) takes the stationary value OJ2 = ? for the corresponding mode .v(/|.
We recall the wellknown orthogonality properties:
x K and x Mx
l) \j) = vA j l) \j) = Vj
The generalised stiffness and mass so introduced. y and ,, are not
independent of each other, since eqn. (32) gives 2 = }'/p. A convenient
choice of norm consists thus of imposing
xJitMx{jj = d,j so that xl)K.\tJ) = 2} (33)
Note that the possible zero frequency modes of eqn. (30). solutions of
Kx = 0, have to be interpreted as either the rigid body modes or the internal
mechanisms of the system, since eqn. (33) shows that they involve no
potential energy.
h = 0 (35)
They are selfstressing modesthat is. internal force distributions in
equilibrium without external forces, capable of existing without inertia
forces.
The selfstressing modes will be denoted by sU), and will be supposed to
form an orthogonal basis
sJt)NstJ) = oj (36)
The other eigenmodes will be denoted by /;(/) and ranged in increasing
order of their eigenvalues
/;
<1> . .2 K)
0 < < \ < < 2
2
VARIATIONAL METHODS OF STRUCTURA L DYNA MICS 9
ALto, r , = 0 1
V (371
hlNh{j) = fou and hJ.Fh^ = J
The eigenmodes of eqn. (34) can be obtained as the stationary points of the
Rayleigh quotient in terms of impulses:
I? N h
= RU,) = (38)
bJfh
Note that for a discrete system of particles, solving the free vibration
eigenvalue problem in the form of either eqn. (30) or eqn. (34) provides the
same nonzero eigensolutions: the choice of either set of degrees of freedom
is essentially a matter of convenience. Once the numerical solution has been
obtained, the transformation from stress parameters to kinematic variables
and its reciprocal are
1 , , 1
v, = " 1 ! 1 /;,,, and /;, = SLxU) (39)
OJ: CO:
v ' Kx
min R(x) = -= ] = 0 i = 1,..., r - 1 (40)
x Mx
Courant's principle states that the rth eigenvalue of problem (30) is the
maximum value that can be given to the minimum of the Rayleigh quotient
(40) by varying the r 1 constraints. This maximum is reached for
v, = Mx,
')
and this minimum is reached for u = ulr). This wellknown result defines an
adequate procedure for obtaining eigenvalues recursively by minimisation
techniques.
The variational principles that have been presented in Section 2 for discrete
systems can easily be extended to the elastodynamics of conservative
systems.
The progression from Hamilton's principle to Toupin's principle is
essentially the same: continuity in space and time are relaxed to provide a
canonical principle operating simultaneously on displacements, velocities
and impulses. Restoring a priori requirements between the three sets of
variables provides either mixed or one-field variational equations that will
be made explicit below.
VARIATIONAL METHODS OF STRUCTURA L DYNA MICS 11
,">J
. '::' . d A ( f f ) ; ;'*.
^ ^ dw() ,
" vd/'
/ w{,)
V
Fie. 2.
the discrete case, that for fixedend values of the displacement field, the
Lagrangian action of a conservative system
and t/, its potential energy, can be split into distinct parts: the strain energy
results from the integration of the strain energy W(Djj) per unit of volume
of reference over the domain V0. The notation W(Du) indicates that the
strain energy density isa function of the displacement derivatives only. The
region I 0 is that occupied by the body in its initial state. If large
displacements are considered, the strain energy density of the hyperelastic
material can be expressed as a function of the Green strain tensor (45) with
which the KirchhoffTrefftz tensor is associated by the constitutive
equations (46).
The second contribution to the potential energy, U2. results from
conservative body loads and surface tractions on the part Sn of the
boundary
+ Xu \pvv + pvu] d V0
A K + <rJ>mu) + Xi Pij = in
ov
(59)
"Mij + trimDmUj) + tj = 0 on Sa
(ii) Varying the stresses j and the surface tractions tj restores the
compatibility in space:
c
\(DliiJ + Dju, + D,umDjum) = 0 in V0
(60)
u. = . on S
(iii) By the variation 7 one restores the dependence (56) between
displacements and velocities.
The functional (57) is particularised to geometrically linear problems
simply by noticing that, for infinitesimal displacements.
\aij(Diuj + Djui + DwmDjU) =s auD,Uj
and
tj = n(aj + <fmDmUj) naj
The only a priori conditions that remain in the canonical principle concern:
(i) The constitutive equations (46).
(ii) The symmetry of the KirchhoffTrefftz tensor atj = . which
implies that rotational equilibrium is satisfied everywhere. A still
more general principle could be derived by relaxing this condition,
as will be briefly mentioned in the next section.
V A R I A T I O N A L METHODS OF S T R U C T U R A L DYNA MICS 15
[,] = F[u. ] dr = 0
&[u,v] = F[u, ] dx = 0
The formal independence of the stress and velocity fields in the associated
variational principle is, however, seriously restricted in practice by the
requirement of satisfying a priori dynamic equilibrium. As one has to
consider infinitesimal displacements, the coupling between the two fields is
reintroduced by
Dau + Xj pij = 0 in F (64)
One way of generalising the functional (63) to finite elastic displacements,
based on the polar decomposition of the Jacobian matrix of the
transformation, was suggested by Fraeijs de Veubeke.' ' The correspond
ing variational principle operates then on three fields: the Piola stress
tensor, the velocities Vj and the matrix of finite rotations associated with the
deformation. The resulting functional will not be made explicit here.
Vj^Dfiu (65)
Hence, in the absence of body forces,
+
dr\cj;) Tp (D'DJ- + D D
' "'0""] = in v
(68)
1 dii
t'T | Su
Scr 1r
" 'Sif
8?[/,/(,)] = 0 8?[,i/(5,)]=0
o prion prion
- '8 s, / -" -
c - Du fc-iX-p'-O
?(/) = 0
1
'
?() = 0
prion prion
Du
W -
t - Du compatibility equations
D\A- X-py-0 equilibrium
Jf[0,u(Sa)] = F[e,u(Sa))dr =0
of the various formulations that have been given, according to the use that
can be made of them. These are summarised in Table I.
The variational principle of widest use, on which the kinematic approach
is based, is of course Hamilton's principle. No limitations are encountered
in its practical implementation. For eigenvalue analysis in the linear range,
a direct consequence of Courant's minimax principle is that the kinematic
approach using purely conforming models guarantees obtaining upper
bounds to the eigenfrequency spectrum, but the price paid for this
behaviour is an excessive cost in the representation of kinetic inertia.
Up to now, the canonical and F[u, ] principles have always been
considered of no practical interest from the point of view of finite element
discretisation. It seems, however, that this assertion ought to be
reconsidered for non-linear analysis. The observation that the com
plementary energy remains a quadratic form of the stresses for
geometrically non-linear problems indicates that a mixed discretisation
could perhaps provide a more adequate approach than a purely kinematic
modelling.
The F[u, v] functional has been regarded independently by Hughes 1 4 and
Geradin, Sanderand Nyssen 16 as a consistent way of lumping masses. The
results obtained with it in eigenvalue analysis are quite promising, as will be
shown in numerical examples. However, the [u, v] approach does not give a
systematic convergence to the eigenspectrum by either upper or lower
bounds. Its main drawback seems to be the practical impossibility of
implementing it together with explicit time integration schemes in non
linear problems.
A few observations can be made in common on the variational principles
based on the complementary energy (Reissner, Toupin and Pian): they
cannot be implemented for geometrically non-linear problems; they do not
allow for an easy handling of other body loads than inertia loads; according
to the Euler equations (68), the actual impulse field contains two sets of
impulse (or stress) modes, just as in the discrete case: an infinite set of self-
stressing modes, solutions of >,CT;J = 0, and particular solutions of the non-
homogeneous equation (64).
The first two observations indicate that the practical application of
complementary energy principles in structural dynamics is limited to
eigenvalue analysis. The last has two consequences:
T
( ) = ( 0 , ,0^0330,20,3^23)
B is thus a matrix of strain functions, and the rigid body modes of the
element are the solutions of Du = 0.
Substituting approximations (74) and (76) into Hamilton's functional
yields the discrete variational equation
with the definitions of the mass and stiffness matrices of the elements
Both matrices are symmetric, and the mass matrix is positive definite. The
positive semidefinite character of the stiffness matrix results from the
existence of rigid body, modes. The generalised loads of the elements,
defined by
QJtdS
. s\
include the external loads applied to the structure and the reactions on the
boundary of the element.
The assembling of the finite element model results from the substitution
of eqn. (75) in eqn. (77): one obtains
l. (81)
According to eqn. (78), the mass matrix constructed on the basis of the
shape functions of the elements is consistent with the stiffness matrix
obtained from the same approximation. In contrast with the discrete system
considered in Section 1, the resulting mass matrixwhich is generally
referred to as a consistent mass matrixis a full matrix. In comparison with
the discrete case, a significant amount of computational effort is thus
required to evaluate the inertia forces in eqn. (82) and to solve the
eigenvalue problem (83).
This observation is the reason for a fairly common practice, which
consists of making the approximation that the mass of the structure is
concentrated at the nodes of the finite element model. Such an
approximation, which is then inconsistent with the repartition of stiffness in
the structure, leads, as in the discrete case, to a mass matrix of diagonal
form. It is then referred to as a lumped mass matrix.
When a lumped mass matrix is used, the guarantee of obtaining upper
bounds to the eigenfrequencies of the structure is lost: in some particular
cases, it is possible to show that the eigenfrequencies will be under-
estimated, 22 but no proof of it can be given in general.
Obtaining cheaper, but still consistent, representation of the kinetic
energy is one of the goals that have been aimed at in the development of
finite element approximations based on other variational equations.
with the definitions of the stiffness, mass and coupling matrices of the
element
K =Yw.
The Euler equations of eqn. (85) with Hamilton's rule q = Ofor = , and
= , are
q: Kq + w g(t) (87a)
M = \ (BlBJ, (91)
C
In this case we would be led back to the classical methods for eigenvalue
analysis and direct time integration in terms of generalised displacements.
However, greater computational efficiency can be obtained if the
parameters r are used explicitly in the computational scheme to express the
inertia forces on the structure.
If free harmonic motion is assumed by setting g(z) = 0.
q = qs\r\ojz and r = rcoscoz
Equations (90) reduce then to the linear eigenvalue problem
Kq coBTr = 0]
(92)
r coBq = 0 I
As the form (92) of the eigenvalue problem is unclassical, its properties and
the way of solving it efficiently will be briefly discussed.
Eigenso/utions. Let us denote by , an eigenfrequency of system (92).
The corresponding eigenmode can be expressed in terms of either
displacement or velocity parameters. We note thus it is (q0). rU)). It is easy to
verify that between two eigenmodes of distinct eigenfrequencies , and Wj
the following orthogonality relations hold:
defined by
Fec= RJ(x)HR(x)dV
Fcs= RT(x)HS(x)dV
(100)
Fss= f ST(x)HS(x)dV
1
A' (D'R)'(D'R)dV (;
vcP
At the element level, the variational derivatives of eqn. (99) together with
bc = 0 at time limits yields the discretised form of the compatibility
equations:
FJ + FJ + Nc - Gq = 0
(102)
Fscc + Fsss - Hq = 0
= C7 ;i04)
Equation (104) is then solved for c and s and put into the form
C = CF-'C1 (106)
K =
K,
we obtain the set of dynamic equations
K
w + KqPP = g (107)
Kpqq+Kqpp = -'1
The final stiffness equation results from the definition of the full sets of
displacements and loads of the element
'q g~
OJc = 1\ i'.. =
j>_ c _0_
and that of the extended mass matrix
"0 0
M.. =
_o N: '
whereby one obtains at the element level
Kw, + M,w, = r, (108)
Expressing equilibrium of the boundary tractions along element interfaces
leads to structural equations of the same form.
The structural eigenvalue problem can be written in the standard form
(K - co2M)w = 0 (109)
The main difference with the kinematic approach lies in the fact that the
mass matrix has a degree of singularity equal to the number of boundary
displacements q, since only the internal parameters c contribute to the
kinetic energy. Hence, the boundary displacements q can be eliminated
statically from the eigenvalue problem,
q= -KMlKqpp
V A R I A T I O N A L M E T H O D S OF S T R U C T U R A L DYNA MICS 31
N o n dimensional
Circular f r e q u e n c y
Cuff2
J -
wm
m : mass/unit of area
D -- bending rigid ty
- skew ancle
The plate bending conforming model (CQ). The element used in the
displacement approach is the purely conforming Fraeijs de Veubeke's
quadrangle represented in Fig. 5. 9 This wellknown element is obtained
from an assemblage of four triangular regions with cubic deflection. The
generalised displacements of the element are the deflection w and the two
slopes w/x and cw/cy at each vertex, and the slopes cw/dn normal to the
external interfaces. The resulting model has 16 degrees of freedom. It is
denoted by C Q in the tables and diagrams.
The plate bending mixed (u, v) model (MXQ). The mixed model, based on
separate approximations of displacements and velocities, is derived from
the purely conforming quadrangle CQ described above. It is denoted by
32 M. GERADIN
FlG. 5.
MXQ. A linear variation of velocities over the whole element has been
adopted to discretise the kinetic energy: account is thus taken of the only
inertia forces associated to rigid body motions. The resulting element
possesses 19 degrees of freedom: 16 generalised displacements and 3
generalised velocities.
As shown in Section 4.2, all the generalised displacements can be
eliminated before solving the eigenvalue problem, from which results a
considerable economy in eigenvalue computation.
The plate bending equilibrium triangle (EQT). This element, represented
in Fig. 6, results from the discretisation of Toupin's principle. The field of
self-stressing modes is obtained by assuming linearly varying moments. The
resulting variables are three concentrated loads at vertices, two bending
moments and one Kirchhoff shear load on each interface. The particular
solution of the equations of dynamic equilibrium is introduced next by
superimposing a cubic moment field that provides a linear variation of
velocities. 1013 The resulting model has 15 degrees of freedom: 12
generalised connecting forces, and 3 integral parameters in terms of which
the inertia forces of the element are represented. A considerable economy in
eigenvalue computation can thus be expected with this EQT model, just as
with MXQ elements, since both are based on a similar approximation of the
kinetic energy.
The numerical results. In the displacement analysis using CQ elements,
the finest grid consists of 6 6 elements and 231 degrees of freedom, from
which 37 are fixed. The results collected in Table II confirm the upper
boundedness guaranteed by the kinematical approach. A significant
deterioration of the finite element solution is observed when the skewness of
the plate is increased.
VARIATIONAL METHODS OF STRUCTURAL DYNAMICS 33
PARTICULAR
SOLUTION
W, - DEFLECTION AT VERTICES
Wlk -- MEAN DEFLECTION
W; and Wnlk- MEAN SLOPES } ON EDGE / -
W - MEAN DEFLECTION
HE AREA
\ ON TH
W- and W - MEAN SLOPES
The equilibrium approach (Tables III and IV) underlines the same
phenomenon. Distinct mesh patterns have been used in the square and skew
cases. In the latter case the finest mesh uses 50 elements and 420 degrees of
freedom. Only 150 of them contribute to the eigenvalue problem.
Note that in most cases the computed frequencies appear to be lower
approximations, except when the discretisation yields to a too rough
representation of inertia forces.
Another important factor, when using triangular elements, is the
skewness of the plate. The best subdivision is obtained as expected, using
two elements per quadrangle, with their common edge along the shorter
diagonal.
Mixed models exhibit not so systematic behaviour as conforming and
equilibrium elements. In most cases they give a convergence by lower
bounds, but such behaviour is not guaranteed.
The analysis with MXQ elements has been pushed up to a 8 8 mesh,
with a total of 352 generalised displacements that are eliminated statically
and 192 velocity parameters in terms of which the eigenvalue problem is
solved. The corresponding results are given in Table V.
All the remarks that can be made about the comparison of the respective
models arise from Fig. 7, which represents, formode 1, the convergence as a
function of three factors: (1) the number of degrees of freedom in terms of
34 M. GERADIN
TABLE 11
CONFORMING APPROACH (CQ)
which the eigenvalue problem is solved; (2) the skewness of the platethe
curves associated with different skew angles have been shifted horizontally;
(3) the type of model used.
The equilibrium models would normally give three distinct convergence
curves according to the mesh subdivision adopted. Only the most
favourable case (subdivision ZV) has been represented in Fig. 7.
TABLE III
EQUILIBRIUM APPROACH (EQT): 0 = 0
1 * 1 2*2 3x3 44
2 subdivisions
MODE
no. m El m
1 3.447 3.453 3 462 3466 3.467 3.469 3
TABLE IV
EQUILIBRIUM A PPROA CH (EQT): 15;30;45 t
3 types of
2x2 2x2
subdivision
^ s ^ ^ 5 ^ ^ S ^ a I ^ a
3.378 3.400 3.502 3.537 3.549 3.567 3 5 6 7 3.565 3.570 3.576 3.574 3 5 7 9
7 374 7733 8096 8.377 8478 8563 8.571 8.607 8651 8632 8 6 4 9 8 667
15 11.82 16.02 21.46 21.23 21.64 21.97 21.84 2 2 . 0 0 22.13 2 2 0 5 22.11 22.16
22.35 2018 24 03 25.60 2558 2598 2 5 9 8 26.12 2624 26.18 26.25 2 6 2 9
39.37 47.97 32.45 31.17 32 54 33.03 3289 3332 33.59 33 41 33.60 33.71
3.374 3562 3.709 3 7 5 8 3.348 3.830 3.857 3.892 3 9 0 6 3907 3907 3.914
7917 8.681 8.884 8.929 9 169 9.257 9 166 9.297 9 3 3 8 9 266 9.338 9365
0 10.95 19.17 22.89 22.07 2405 2 4 5 6 23 91 24.76 2 4 9 7 24.57 24.97 2509
23.11 2108 25 58 25.10 25.33 2 5 3 3 2 5 3 2 25.68 2 5 9 8 2560 2578 25.85
37.92 4515 36.79 3 4 8 8 40.02 40.33 39.14 40.82 41.03 40.34 41.03 41.18
3.344 3 9 4 2 4.084 4.002 4.321 ! 4.370 4 2 3 5 4.406 4 432 4441 4.441 4.450
8302 1025 1043 10.34 11.00 i 11.06 10.82 11.14 11 16 11.00 11.18 11.20
u 2 2 6 9 25.57 26.01 24.67 26.35 26.53
45 11.66 23.35 26.27 2658 25.60 26.72
22.80 2 6 8 8 31.56 25B3 3 0 3 7 3 0 4 0 2 8 5 4 30.84 30.98 2987 31.06 31.17
40.01 4 3 0 1 40.71 35.32 48.09 48.79 4 6 3 7 49.97 50.27 4816 5041 5056
5 0
convergence to mode no 1
non-
dimenstona
i t I
frequency
4 5 J . tuO
/fm
- - QUADRANGLE C Q
TRIANGLE EQT
QUADRANGLE MXQ
40
frequency versus
sxew angle
35
|20Q log
TABLE V
MIXED APPROACH (MXQ)
TABLE VI
SKEW CANTILEVER PLATES: COMPARISON OF RESULTS
analysis of shallow shells, for which quite good results can be obtained using
flat shell elements: within the limits of linear analysis, this approximation is
known to lead to errors on the displacement field and on strain energy
which are of the same order of magnitude as when using shallow shell
elements. 17
Flat shell elements, which are much simpler to derive, exhibit interesting
properties, such as an easier connection to other types of elements such as
beams, and allow for an easier discretisation of shells with strong curvature
discontinuities.
Within a flat shell element, the extensional and bending displacement
fields are fully uncoupled and can thus be analysed separately.
An interesting solution consists of combining an equilibrium plated
bending element, constructed on the basis of a linear approximation of
bending moments, with a quadratic membrane element as shown in Fig. 8.
38 M. GERADIN
conforming equilibrium
membrane bending
triangle triangle
FIG.
/?=24in
L - 12 in
/ = 0 120 in
E- 6 0 3 5 7 lb/in 2 (steel)
P- 0.2817 lb/in 3
=0.3
FIG. 10.
VARIATIONAL METHODS OF STRUCTURA L DYNA MICS 39
TABLE VII
COMPRESSOR BLA DE EIGENFREQUENCIES
|\^ \ \^ \ /
! \
; y /
\/
; \- \.
:^\ \ ^
z M0DE3
y'
MODE 4
\ : /
MODE 5
/
M0DE6
\
\ ; \ ;\~
: y'
y ;/
w \ \, : \.
y
'; / ./,
MODE 7 MODE 8 MODE 9 MODE i o
Frequency : 701 5 Hz Frequency : 6 9 4 - 1 Hz Frequency 7 6 4 5 Hz Frequency : 8 0 1 Hz
Table VII compares the numerical results obtained with the element
described above with those of Olson and Lindberg, 18 and with their
experimental investigation. Olson and Lind berg's element is a shallow shell
element, obtained using a fifth-degree interpolation of displacements. It
uses not only displacements and rotations, but also curvatures as
connecting variables at vertices.
One finds, as in plate bending problems, a tendency for the element to
give lower bounds to the exact frequencies of the structure. This is to be
expected because the eigenmodes, which are represented in Fig. 11, involve
essentially bending deformations.
REFERENCES
1. INTRODUCTION
2. REDUCTION OF MULTI-DEGREE-OF-FREEDOM
MATRICES FOR EIGENVALUE COMPUTATION
The finite element method is now in widespread use for both dynamic and
static analyses. It is a powerful technique but, for dynamics problems, has
43
44 K. FULLRD
Guyan described reduction in the above manner, while Irons gave details
of how the method could be programmed in a similar manner to the front
solution technique used by many finite element programs.
The reduced stiffness matrix is, in fact, simply the inverse of a flexibility
matrix in the same degrees of freedom. Consider an encastre-free beam with
a transverse and rotational degree of freedom at its end. The
displacement-force relationship is given by
EI 12 -61'
(7)
71 -61 Al2
using simple bending theory.
If the rotational term is eliminated, using eqn. (4) we obtain
EJ7-36/2
12 F v = F
4/-
or
3EI
= F = * (8)
IY
3 U 6
Degrees of freedom
FIG. 1. Third-mode frequencies for a cantilever.
selected master
""
FIG. 2. Pipework exampleautomatically selected degrees of freedom.
7777 *"Z
FIG. 3. Fundamental mode of pipework.
fourth mode and none is less than 0-1 % for the fifth mode. The degrees of
freedom automatically chosen for the branch are almost certainly those
which would have been selected, since they are needed for the fundamental
mode of the branch. Bearing in mind the F-direction restraints, again there
is probably little difference between the automatically chosen Y degrees of
freedom and those which might have been selected 'by eye'.
50 K. FULLRD
Generally speaking, Henshell and Ong found that for a large degree of
freedom problem, automatically chosen masters gave more accurate
frequencies than the same number of selected masters. They also said that if
the number of masters was too low, the method was not very satisfactory in
association with the frontal method, because there would be too few
potential choices at any stage of the frontal process and important
freedoms could be eliminated. This would not happen if the whole of the
stiffness and mass matrices were derived before any reduction commenced.
They suggested that the method should not be used if less than about twenty
masters were needed.
The slave displacements should not, strictly speaking, be recovered from
eqn. (2), since they would be static and not dynamic displacements.
Referring to the equations of motion, we can establish a relationship
between slave and master displacements which includes inertia terms:
Then
Mn
M21
Ml2
M22 +
^11
-^2 1
-^12
-^22 -
<, = (
2
+1(-22 + 12){2} (12)
Provided that we are considering values of 2 which are lower than any
eigenvalue 2, of ( 2 , , + ){1}, then eqn. (12) is approximately
TV are integers. N is structure dependent and n is any integer less than TV. A
cooling tower on legs is a typical example of a rotationally periodic
structure. The dynamic characteristics of the complete structure can be
obtained by considering only a substructure, but the matrices are
Hermitian.
[Hermitian matrices have the form (A + iB), where A is real and
symmetric and is real and skew symmetric (BT = B). If C = A 4 iB,
C T = C, where C is the matrix whose elements are the complex con
jugates of those of C. C T or C H is referred to as the Hermitian transposed
matrix. It is a property of Hermitian matrices that their eigenvalues are real
and their eigenvectors are complex.]
It can be readily seen from this form of expressing the response that if OJ
corresponds to a natural frequency of the system, then (A urM) is zero.
In this case {A 'R | is zero and
{Xl)=a>l[C]l{P) (31)
The above discussion is given only to show that with harmonic excitation
there should be no difficulty in obtaining a solution to the structural
response problem. If complex solution techniques are available, then
solving eqn. (27) is simply equivalent to solving a complex stiffness matrix
or impedance matrix:
When the forcing is not harmonic, the above impedance matrix approach
can still be used, with little extra difficulty if the forcing is periodic (via
Fourier components) and by periodic approximation if the forcing is non
periodic. However, as an alternative to this frequency response technique
(so called because the response is related to harmonic excitation) we have
modal methods of solution.
Returning to eqn. (25), the unknown response vector {x} is expressed as
the sum of proportions of the normal mode vectors:
^aa ^ai
(42)
A;., A;; ,
where subscripts a, i refer t o attachment or internal degrees of freedom; [ F J
is the set of forces required to give the appropriate attachment
displacements; and [<J are the constraint mode shapes.
From eqn. (42)
[^]=-Ai71Aia (43)
[Although there must be as many constraint mode shapes as attachment
degrees of freedom, the number of internal degrees of freedom defining the
mode shape can be limited by the use of economisation procedures.]
The original attachment and (reduced set of) internal degrees of freedom
can be related through the constraint and substructure mode shapes to a
smaller number of degrees of freedom:
I 0"
= [T\{q] (44)
0c . q.
where [</>J are the substructure mode shapes; {x.d\ = {qa} are attachment
degrees of freedom; {x-A are original internal degrees of freedom; and {qs}
are modal degrees of freedom.
Matrix [77] is used to transform the substructure mass and stiffness
matrices defined in terms of both attachment and internal degrees of
freedom (sinusoidal excitation):
(-2[[][\ + [Tf[K][T)){q} = IT]T{F} (45)
or
-co2[m) + [k]){q}={f} (45a)
[*] = (46)
58 K. FULLRD
where
kCi = :sc = 0
ks = 1.,
Note that kSi is a diagonal matrix
[m] = (47)
m.
here
Wcc = M a a - 3 ,7 >M ia - MiBK^Kia + A ai Ai7 ',,, ' A ia
"Jss = ,
and 7??ss is also a diagonal matrix: [ss] = [a>2][mss], where [OJ2] is a diagonal
eigenvalue matrix for the substructure.
From the above it can be seen that Acc and mcc are identical with the
reduced stiffness and mass matrices which would have been derived if
attachment degrees of freedom had been the only masters, [k] and [m] can
be regarded as Guyan reduced matrices supplemented by a truncated set of
fixed constraint substructure modes.
The force transformation of eqn. (45) is
(/il JF^K^K^FJ (48)
:/":
UJ * F
where the forces include all externally applied forces and also forces on the
attachment degrees of freedom imposed by adjacent substructures.
When, for example, two substructures are joined, the following matrices
are formed :
"*+* 0 0 "
[A*] = 0 '11 0 (49)
0 0 k[2>
[M*] = (50)
\F*\ (51)
[tit] (53)
*i = {c/} +
and (54)
{x}= {} +
where {U} and {u} are support and structure relative displacements.
Similarly for the other five directions.
Thus, the vectors {/?} and {q} can be expressed in terms of the relative
displacements {U} and {} by
iV}^ |1 \Xo
{V\ {l^o
w< 'MZ0
{>}= { / M + </>2= < {A} > + < {1! 0 > (55)
[B\ {
v i l To
THE MODA L METHOD FOR TRA NSIENT RESPONSE 61
and
OH
{}
0
{ ? } = W i } + {<h = <
!H
{oj
>+ i !0 >
(56)
\i {1}
Lir) J
Substituting eqns. (55) and (56) into eqn. (53) gives
the structure can then be found by use of [A ] {}, where \q\ = {qy\ + \q2\
and \p\ = \Pi\ + \p2\ If the fixing moments and reaction forces are not
needed, the forces and moments may be found by use of [A ] {q J 4 [A 2]{/?j.
As stated previously, true relative motions only occur in the absence of
rotational input motions. In this case the terms [K]{q2\ + [K2]'Kp2\ and
{ 2 } + [C2]{p2} are zero, giving the vector [R] as [M]{q2} +
[M2]{p) + [C 2 ]{,} + [A 2]{/>,}.
If the base mat. on which the structure rests, is stiff enough compared
with its environment to be considered as rigid, then the supports of the
structure all seethe same input and the vectors {pi j , {p1} and {p\ } are zero.
Thus, for linear directions and a rigid base mat {R} = [M]{q2} +
[M2]{p2}. If the elements through which the support motion is trans
mitted have a mass distribution such that their mass matrices are
diagonal, then [/ 2 ], which is not, in general, a square matrix, has all zero
elements.
With the above conditions the familiar seismic equations are formed
from eqn. (58):
[M]{q1} + [C]{ql} + [K]{q1} = W]{q2} (59)
where {I}x is a unit vector for displacements and null for the other
directions of support motion.
THE MODAL METHOD FOR TRANSIENT RESPONSE 63
()
>';..v = = (64)
where
= 2('-)
u X2(tn)X2(tn^)
/ - ,
t = /
and S(tn_), S(tn_) and S(tn_) are known from a previous solution at
',,-
The solution to eqn. (65) gives
S(t') = (C, sint' + C.costfc^1' + A + Bt' (66)
where
C,
4h ,) + S ( t _ 1 ) + - | - - - ^ 6 I 1
2
/
+ ^
2
" +4-0
-b
"7
and = (1 2 ).
2 2
where yn is the participation factor for the z'th mode in the nth direction.
THE MODAL METHOD FOR TRANSIENT RESPONSE 65
1000
- 20
01 0.2 05 1 2 5
Frequency, cps
used to give floor response spectra distinct from the ground response
spectra in that they reflect dynamic characteristics of the intervening
structure. Other artificially generated time histories may have a more
fundamental purpose in that they modify a given free-field time history to
take into account ground conditions different from those where the original
recording was made.
There is another type of artificial accelerogram which is not used to
generate a response spectrum but is created to fit an idealised response
spectrum reasonably closely. The response spectrum is generated from a
large number of distinct single-degree-of-freedom systems, and, strictly
speaking, is a series of points. Joining the points together results in a spiky
plot. The U S N R C have defined, for design purposes, smoothed ground
response spectra which, on a log velocity versus log frequency plot, are a
connected series of straight lines for different levels of damping, based on
the characteristics of a set of response spectra of actual earth tremor
recordings (Fig. 5). In order that floor response spectra may be obtained,
68 K. FULLRD
artifical ground time histories are generated which envelop the design
ground response spectrum.
Whatever the source of the response spectrum, the principle of its use is
simple. For a specific system, we can calculate its natural frequencies. For
each mode, a level of damping is assumed and a spectral value of response is
read off the corresponding response spectrum. This response has to be
scaled by the participation factor for the particular mode and excitation
direction, and the result is a modal response value for the structure,
5/ = i^D.i- This can be compared with the calculation of response in the
previous section. The major difference is that we no longer have a time-
varying response, only a peak value. The effect of excitation duration is lost
and the phase differences between the responding modes are also lost.
The U S N R C spectra are defined for two horizontal and the vertical
directions. The three components are independent and the manner in which
the various modal responses are to be combined reflects this. The most
conservative method of combining the responses would be to simply add
them. What this means is that all the modal response values at a given
location in the system are added together. Ifall the modes are assumed in
phase for this summation, some will have to be out of phase to give a
maximum response at another place in the system. Summation is usually
only carried out for those modes whose natural frequencies are close
together (usually within 10% of the lowest frequency). When there are no
closely spaced modes, the responses are squared and added and a square
root of the total taken (square root sum of squares or SRSS method).
Response here is taken to be stresses, forces or displacements, rotation
effects being ignored. A fuller description of acceptable methods of
combining modal responses is given in the U S NRC Regulatory Guide 1.92,
Revision 1, published in February 1976.
Although the response spectrum method is easy to use, it should not be
forgotten that it is based on the response of single-degree-of-freedom
systems. The method should not be applied to multisupported structures
without a realisation that the relative support motions are being
approximated.
ACKNOWLEDGEMENTS
REFERENCES
1. INTRODUCTION
The stiffness matrix k and the consistent mass matrix w c are wellknown
results and are easily obtained by standard methods. The diagonal mass
matrix is the result of lumping at the nodal points.
For uniform meshing, the equation at any interior node/ is given by eqn.
(4):
c^2
(1 +rD)j-l-fj Duj=jj (4)
+27"=0.
The first expression in eqns. (6) is satisfied by the discrete harmonic solution
in eqn. (7):
S j - C . s i n ^ + Cjcos^ (7)
For a free-free or fixed-fixed boundary condition, = nn, while for the free-
fixed case, = (2A? 1)/2, the same as in the continuous case. The positive
integer is less than or equal to N, the total number of elements making up
the mesh. Since the mode shapes are the same in both the discrete and
continuous cases, a frequency-by-frequency comparison can be made. The
continuous frequency is given by = kcJIN. The result is given by eqn. (8)
and shown in Fig. 2.
. / i ^' 2 "
= sin - 1 4rsin' (8)
2/V 2N\ 2/V,
underlies the bulk of the transient dynamic analyses carried out. The
piecewise quadratic displacement assumption is discussed by Belytschko
and Mullen. ' The piecewise cubic displacement assumption is discussed by
Krieg and Key. 5
In two space dimensions much of the same behaviour is obtained.
Goudreau 2 discusses Laplace's equation which governs the vibration of
0 5
1 0
Mode = n/N ,(2n- \)/N
H=/(0 (9)
Here vv is the lateral displacement and and / are again differentiations in
space and time. The constant cb is given by (Eh2/\2pY 2 , where h is the
thickness of the beam or plate.
T R A N S I E N T RESPONSE BY TIME I N T E G R A T I O N 75
12 m/M = 7 / 8 ^ = = = - . - . - _
'^* ^
.___ m/M = 1/2
^f
? s
",/'
m/M = 1 / 8 , . ' " ' '
,~^'
10
~":r=:^2r~^ ^ /n/Af* 1/8
m/M=\/%~~ ^.
m// ^Q\
=3_
0 9==. ^ "^^ ^ "
x.
x ^ \
x . \ \
x V/77//VM/2 \ \ \
0 8-
~ \
m/M
7{^^
.-"--^
- ^ . \ \
. \
07 \
\ \
.
^ \\
\ \
\
06
\
\
/?// = 7/8 \ s
S,
-
ite difference
Fin
Q4 (consis t en t mass )
Q4 (lumped mass)
\ .
.
0 4
0 02 0 4 06 01 1
/
FIG. 3. The ratio of discrete frequency, , to exact frequency, , versus mode
number, n/N, for selected values of mode number m/M for both a finite difference
and finite element bilinear quadrilateral spatial discretisation of the membrane
equation.2
T h e resulting element stiffness and mass matrices are given in eqn. (11):
. j
12 6 V2
7 'l2 '7 + /2
4 6
+ 2 7 ~T2
12 6 12 6
? I2
? ?
6 6 4
+ r- ~7- /
';
13/ III2 9/ 13/ 2 '
35 210 70 ~42
ill2 I3 13/ 2 P_
210 105 420 140
m,. = (H)
9/ 13/ 2
13/ 11/ 2
70 420 ~35 ~2
2 3
13/ / 11/ 2
"420 140 210 105
TRANSIENT RESPONSE BY TIME INTEGRATION 77
2-Or
\
\
\
\
\
\
\
1 \
\
\
\
1 5
\
\
\
\
^ 10
\ \ \\
\ \ \\
0 5
1 , Diagonal mass a = 1 ^
2, Consistent mass
3, Diagonal mass a = 17 5
0 1 2
Mode - n /N pin-pin, ...
The stiffness matrix k and the consistent mass matrix mc are well-known
results and are easily obtained by standard methods. No rotary inertia is
included. A diagonal mass matrix employing a gradient inertia scaling
parameter is given in eqn. (12): 5
7/2
a/ J /420
(12)
1/2
3
a/ /420_
78 s. w. KEY
Two discrete equations are obtained at each mesh point. The details of this
are covered by Goudreau and Taylor. 4 The discrete solution is given by
eqn. (13):
The mode shapes are once more the same as those in the continuous case, so
that a frequency-by-frequency comparison can be made. For the relation
between /. and mode number for various boundary conditions, Hurty and
Rubinstein 14 (p. 203), can be consulted. The results are contained in Fig. 5.
In shells the membrane and bending response is coupled, so that at best
one has modes which are predominantly membrane or bending in
behaviour with small components of the other. Nevertheless, the wave
equation and the beam equation represent in the limit of a flat plate the
membrane and bending behaviour of shells.
Figures 2, 3 and 5 tend to favour the consistent mass representation as
being more accurate. These are the results that would be obtained in a mode
shape and frequency evaluation. If only the lower modes are being sought,
there is no clear-cut preference except on the basis of computational effort,
and then a diagonal mass is better.
The important point to be made here is that these are the semidiscrete
equations which are to be discretised further when applying a time
integration operator to integrate forward in time. However, it is still the
original partial differential equations to which solutions are sought.
For the discussion here time integration methods are going to be divided
into explicit procedures and implicit procedures. For either the semidiscrete
wave equation, the semidiscrete bending equation or, for that matter, any
discrete model of structural behaviour, we can adopt the notation in eqn.
(14):
Ma + Kd = F(t) (14)
Here d is the vector of nodal degrees of freedom; a is the vector of nodal
accelerations; M is a mass matrix which may or may not be diagonal; is
the linear stiffness matrix, although the restriction to linearity can be
relaxed under appropriate conditions; and F(t) is a vector of nodal 'loads'.
TRANSIENT RESPONSE BY TIME INTEGRA TION 79
We will assume for the moment that we have the solution at time equal to
nAt so that eqn. (15) holds:
Ma" + Kd" = F" (15)
+l
If the time integration procedure being used requires (Kd)" to obtain
i/" + ', then a set of algebraic equations will have to be solved and the scheme
will be called implicit. If, on the other hand, (Kd)"+ ' is not required to
obtain 7" +1 , the scheme will be called explicit. This classification is
consistent with the finite difference literature, where only diagonal mass
matrices are considered.
Explicit Procedures
The most widely used explicit time integration procedure is the central
difference method. As is discussed below, it is the optimal explicit procedure
from a very wide class. First a central difference expression is introduced for
the acceleration a in terms of the velocity and then central difference
expressions are introduced for the velocity in terms of the nodal degree of
freedom. Thus.
yn + ti 2, = t.i,n 2, + \,tM\Kd" F")~)
+ 1/2)
d"^ = d" + /" j
In practice, eqn. (16) is the form used where one old configuration d" and
one old velocity t"" (1 2 | are used to obtain a new velocity i?""*(1;2) and a new
configuration d"+ '. The velocity is carried at the halflime point. It is also
seen that were the stiffness calculation nonlinear owing to either geometric
changes which had occurred or material nonlinearities, there would be no
trouble in computing K(d")d". To take advantage of the speed with which
each time step can be computed M must be diagonal, so that M~ ' in eqn.
(16) is trivial.
The difference method is only conditionally stable, which means that
there is a critical time step A icnt below which A t must be kept. When A t is
above the critical time step, highfrequency modes in the numerical solution
become exponential in time rather than harmonic solutions of the original
semidiscrete system (14) A fter a number of time steps the exponential
solutions dominate.
A stability analysis for the central difference method is straightforward.
First the velocity is eliminated from eqn. ( 16) to give the homogeneous eqn.
(17):
M _ 2d" + dn' ' + A t2M~xKdn =0 (17)
80 S. W. KEY
/2<^ (21)
'-max
This approach to the stability conditions for the central difference method
is discussed by Leech, Hsu and Mack.'
For the linear wave equation with a piecewise linear displacement finite
element mesh of uniform dimension /. the maximum eigenvalue is given by
c2
= 4-. lumped mass
For the linear bending equation with a piecewise cubic displacement finite
element mesh of uniform dimension /, the maximum eigenvalue is given by
2
max = 2 5 2 0 ^ . lumped mass 0 < a < 525
(23)
= 2520 .
4'
consistent mass
The schemes of this form include all explicit two-root and three-root
methods. (The three-root methods contain an extraneous solution about
which little is known in general.) The central difference method is a two-root
method.
It was found that there did not exist a scheme which was unconditionally
stable and of order (?2) when compared with the differential equation (14).
In addition, there exists no time integration operator of the form of eqn.
(24) which is of order (/ 2 ) when compared with the differential equation
(14) which is stable for a time step longer than that of the central difference
method. These are very strong results, and point to the optimal nature of
the central difference method in this class.
Implicit Procedures
There are a great variety of implicit schemes; however, the Newmark
method with the explicit central difference method as a special case is the
most popular method in use today and is examined first. While in practice
the full set of semidiscrete equations (14) is integrated together, the
integration may be uncoupled into modes. The model uncoupling leads to
the second expression of eqns. (6) being integrated mode-by-mode. The
Newmark method applied to eqns. (6), where 2 is an eigenvalue of
M ~ ' of eqn. ( 14) gives
DT" + c2A t2l\<\ b f 1 ) + y ( l E~1) + D]T" = 0 (25)
The superscript is used to denote the value of T(t) at time step (nA t) and
the operator i s defined as usual where Epuk = up + k. The central difference
operator in time is then defined as D = (E 1 )( 1 E'1). The method has
two parameters: and y. A s these two parameters vary, all possible two
root methods are obtained and the parameter is tied closely to the
damping in the scheme. The scheme contains no damping when y = \. It is
for these two reasons that the method has gained such popularity.
Equation (25) has a damped sinusoidal solution, but it is useful to
transform with a onetoone and onto transformation to the complex z
plane as follows:
4 + 22(4 - 2)
13r
1-2
Central /
difference /
\1313 --0 / ^ =1/12
r = l/2
J
I io
/3=1/6
09
/3=1/4\
08
to5
S 10
y = 0 501
= 1/4
- 10;
~-0 51
%" I 102
02 05 10 20 50
FIG. 6. The frequency response for y = 05 and damping for = 025 for the
Newmark time integrator.
The manner in which a"+ l enters these equations for 0 means that an
'estimate' of <7" + ' to use in eqn. (30) for a"4 ' must be obtained. More
84 S. W . KEY
o
^ j ^ r ^ r , =05,y3=-y2=0t
XX. xNv.Farhoomand
09
\ \ NNwHoubolt
08
Wilson/ \ \ \ \
07-
0 6 I 1 1 \ \ \\
FIG. 7. Frequency response and damping for the Houbok, Wilson and
WilsonFarhoomand time integrators along with a generalised threeroot method
of Krieg.
precisely, the first expression in eqns. (31) contains d"+ l on both the right
and left sides with the solution of a nonlinear set of equations required to
find d"+ '. A t every step along the way, because of the nonlinearities, the
solution to eqn. (30) is found only approximately. The resulting error e is
computed in eqn. (32):
e" = F" L(d") Ma" (32)
To discuss the unconditional stability of the Newmark method, energy
considerations need to be introduced. A discrete internal energy is given in
eqn. (33):
t/'=0; U"+i = U" + jAd1(L" + L" + 1 ) (33)
A discrete external work is given in eqn. (34):
W1 = 0; W"+ ' = W + Mi/ T (F" + F,+ ') (34)
TRANSIENT RESPONSE BY TIME INTEGRATION 85
007
0 01 -
0 5
010
/Houbolt /Wilson
0 09 /method / method! =1.4)
008
007
006
:
0 05 -
004
/Nev/ algorithm
003 / (a = - 0 - 3 )
002
0 01 - / 'New algorithm
(a = - 0 0 5 )
0 S^i 1 1
01 0-2 0 3 0-4
//
FIG. 8. D amping ratio versus At for the new Hilber- Hughes-Taylor. Newmark
. Houbolt and Wilson time integrators (/ = (/)).
86 S. W. KEY
04
A wide variety of time integration schemes have been introduced with very
little comment on their relative merits. It was seen that, depending on the
mass representation, the original frequency was distorted either upward
or downward into 3. Depending on the choice of time integrations, the
semidiscrete frequency c was further distorted upward or downward into
. Since it is not practical to carry out calculations at vanishingly small time
steps to minimise these distortions, they are necessarily present when an
analysis is carried out. The goal here is to examine selection of both a spatial
and temporal descretisation based on their combined effect and the class of
applications under consideration.
The first observation is that, depending on the choice of mass
representation, / is either greater than or less than 1. Depending on the
choice of time integration scheme, / is either greater than or less than 1.
However, our goal in a computation is to have / as close to 1 as
practicable.
The second observation is that, depending on the application, one is
interested in all of the frequency content of the model or only a portion at
the low-frequency end of the spectrum. In an impact analysis where stress
waves emanate from the place of impact and propagate throughout the
structure, it is the entire frequency content which is needed to represent the
rapid changes in stress that occur in time as the stress waves propagate.
However, in seismic analyses the frequency content of the acceleration
88 s. w. KEY
pulse is quite low compared with the entire spectrum of the spatial
discretisation. The highest modes in the structure are only minimally
excited and then rapidly damp out, not to participate in the overall
structural response. If one is required to carry out a calculation including
them, needless effort is expended.
While all transient dynamic analyses do not fall into either the impact or
seismic class, it is our experience that these areas are sufficiently important
to base criteria upon them. On the basis of the reasons which are listed the
following recommendations are made:
Impact
Spatial discretisationlumped mass
Temporal discretisationexplicit central difference
All frequencies are important, 0-9 < / < 1 1 . (The lumped mass
depresses the frequencies, while the central difference operator distorts
the frequencies upward, thereby offsetting each other.)
Simple and fast computationally.
No numerical dissipation but artificial viscosity added to control
shock wave formation. Controlled independent of time step.
Conditionally stable, Atcm.dX < 2-0. Accuracy is maximised by
operating as close to the stability limit as possible.
Seismic
Spatial discretisationconsistent mass
Temporal discretisationimplicit Hilber-Hughes-Taylor
Unconditionally stable when applied to linear problems.
Possess numerical dissipation which can be controlled by a parameter
other than the time step. In particular, zero dissipation is possible.
Numerical dissipation does not affect the lower modes too strongly.
All frequencies for which it is required that 0-9 < / < 1 ;
Ao)culK)ff < 2.
(The consistent mass distorts the frequencies upward, while the
Hilber-Hughes-Taylor method depresses the frequencies, thereby
offsetting each other.)
If these two procedures are used for all classes of transient dynamic
calculations, it is useful to have a criterion which indicates which method
will require the least effort. A square two-dimensional mesh is assumed. The
work required to advance the solution one time step using a lumped mass
matrix and the central difference involves a matrix times a vector for each
time step; only the non-zero terms are counted.
TRANSIENT RESPONSE BY TIME INTEGRA TION 89
In the last example the choices of spatial and temporal discretisations are
individually optimal choices but each raises all of the frequencies. The
combination is certainly not optimal.
Thus, a diagonal system is obtained where the elements are explicit and a
TRANSIENT RESPONSE BY TIME INTEGRA TION 91
2 4 6 8 10 12
y
\
9 11
Node number
I - Implicit elements
E - Explicit elements
Equation
number
Shaded region
represents
non-zero entries
Symmetric
Profile of K
symmetric banded set of equations where the elements are implicit. Figure
10 shows a typical circumstance. With a "profile' or 'skyline' solver
considerable computational effort is eliminated.
Of interest are the computational results from this scheme. Consider the
axial impact of a bar 10 units long with a uniform velocity ;0 equal to 1 0.
The impacting end is held fixed. The bar is composed of 20 elements 05
units long, a modulus equal to 100 and a density of 001. The critical
time is 0005, the element barwave transit time. Figure 11 shows the axial
stress history at the impacting end for various combinations of time step,
92 S. W. KEY
- 1-2
oo
12
=0 0
/ = 0 0 0 4 9
0 8 =05
Exact
= 00001
? = 0 25
0 4 -
00
-04-
-08
\'^
-12
00 0 1 02 0-3 04 05
I 2
Va * = 00
/ = 0 00353
0 8
y=0 5
= 025
{ A l l elements have the
04 same material properties)
00
Exact
Explicit
-04
Implicit-Explicit
(Elements 1,11 and 21
are implicit )
1 6 01 0 2 O 3 0 4
Consider the same bar problem, but now 105 units long and 21 elements
in the mesh. Elements 1,11 and 21 are made implicit, the first, middle and
last elements. ForAr = 0003 53, well below the critical time step,, the results
in Fig. 12 are obtained. As should occur, the results are very similar with or
without the implicit designation.
For the final example, let E = IO7 in elements 1 and 21. The critical time
step for elements 1 and 21 diminishes to about 1 /136th of the critical step
for elements 2 to 20. Figure 13 shows the axial stress history in the middle
of the bar. The time step used is governed by the explicit elements.
These results point to a very effective operator. In addition, the technique
for deriving the operator can be applied to other implicit schemes.
94 S. W. KEY
I X 5 25
u
1 2- [At 00050
0 5
/ JI 025 r~~
1
08- 0 0045 /
'
\% 06 j
04-
[ 0 3025
0 0
I
il 11 i
-04
11
1
1
if
mphcit Explicit
'Stiff' elements
1 and 2 are implicit)
\
I \
j 1
-08 - 1
\\
-1 2 --
\
-1 6 I 1 1 1
00 01 02 03 04 05
FIG. 13. Twenty-one element, bar impact problem. Comparison of damped and
undamped, implicit-explicit computations.
REFERENCES
This list of references is not intended to be complete. Rather, they have been
selected so that an individual looking to gain a definitive understanding of
transient time integration operators can confine his study to a small cross-
section of the literature. This is not to say that there are not other articles
and books which contain this information. The bulk of these articles deal
with the analysis of time integration operators rather than the results which
they produce in applications. In this area it is all to easy to make erroneous
judgements based on applications alone. A study of these references will
reveal a heavy reliance on them in these chapters, particularly the work of
Thomas J. R. Hughes and his colleagues. Our primary goal here is
instructional rather than exhaustive inclusiveness.
1. Leech. J. W.. Hsu. P. T. and Mack, E. W. ( 1965). 'Stability of a finite-difference
method for solving matrix equations'. AI AA J!. 3(11). 2172-3.
TRANSIENT RESPONSE BY TIME INTEGRATION 95
1. INTRODUCTION
{} = lB]{d} (8)
\\ = [C] [] (9)
We will here develop the discrete equations from the principle of the
virtual rate of work, which corresponds to a special case of the Galerkin
method or a weak form of the partial differential equation. The principle of
virtual work states that for any arbitrary kinematically consistent
displacements, the rate of internal work must equal the rate of external
work
Wm = WZM (10)
where 7"j is a surface traction and \b' \ isa body force. For the purpose of
applying the principle of virtual work to dynamics problems, we introduce
inertial body forces through the d'Alembert principle, so that the total body
force is given by
W\ = {b\p{\ (12)
WM = {\{} (13)
TIME INTEGRA TION OF STRUCTUREMECHA NICA L SYSTEMS 101
[B]T\a}dV (14)
= {<5>}T{F}in' (15)
where
[M] = \ [L(c)]T[m(e)][/Jc)]
] ] (19)
Here lmM] is the mass matrix of the element and [A/] the total mass matrix.
Combining eqns. (10), (15) and (18), we obtain
{D}J{Fr" = {(]\ >DJ}{F'"}) (20)
102 T. BELYTSCHKO
{/ , e ) }= f [BVlC][B]dV{cf^\ (22)
Ji-io
so
{F""} = Y[L , e , ] T {/>}
{Em} = V [Z.,C)]T[A-,C,][/.<C|]{>}
or
{Fm} = [K]{D] (23)
where [A] is the total assembled stiffness matrix. Note that the assembly of
[K] and [M] is identical.
Substituting eqn. (23) into eqn. (21), we obtain
IM]{D} + IK]{D} = {F} (24)
a system of linear, second-order ODE.
TIME INTEGRA TION OF STRUCTUREMECHA NICA L SYSTEMS 103
This difference formula is called explicit because this expression does not
involve any drivtes at the last time step.
If we write eqn. (21) at.time i and combine with eqn. (30), we obtain
D , + 1 ] = / 2 [ ] ~ ' ( { ^ " } {FinU}) + 2{D'} I D ' 1 j (31)
104 T. BELYTSCHKO
It can be seen from the above that time integration by the explicit central
difference method does not require the solution of any equations provided
that the mass matrix is diagonal. Krieg and Key 12 have shown that, in fact,
the spectral errors of central difference integration and mass lumping are
compensatory, so that mass lumping is preferable in explicit integration
both from the viewpoint of computational efficiency, in that no matrix
inversion is involved, and in accuracy.
When a linear viscosity is desired in eqn. (21), a backward difference form
must be used for the velocity term to maintain the explicitness of the
equations. The viscous force is then given by
2 (
/ < ^ 1 } (43)
which corresponds to a reduction of about 60% in the stable time step.
The highest frequency of a mesh is bounded from above by the maximum
unconstrained element frequency in a mesh (Hughes 10 ). This can easily be
shown by means of Rayleigh quotients. By unconstrained element
106 T. BELYTSCHKO
where c is the wave speed. Substituting eqn. (45) into eqn. (50) gives
/
At<(^'2+ 1 ) (47)
which limits the time step to the traversal time of the elastic wave across an
element. This corresponds to the well known CourantFriedrichsLewy
condition (CFL condition). 7 It should be noted that the inequality of eqn.
(47) holds for both uniform and nonuniform meshes; in the latter case the
time step is governed by the element for which l/c is the minimum. This, of
course, makes extreme mesh refinement in one portion of a mesh somewhat
undesirable, for the time step for the entire mesh is governed by the
minimum traversal time.
TIME INTEGRA TION OF STRUCTURE MECHA NICA L SYSTEMS 107
This yields that the maximum frequency and time step stability limit are
, 12c2
or = p (50)
/
At < = (51)
By comparing eqns. (48) and (51) it can be seen that the use of a consistent
mass reduces the stability limit on the time step below the CFL condition,
which, along with the need to invert [M], constitutes a serious drawback in
the use of a consistent mass matrix with explicit time integration.
We will now examine the characteristics of higherorder elements by
considering the threenode, quadratic displacement, linear strain element.
The mass and stiffness matrices for this element are (the middle node is
associated with the second row and column)
4 2
cons Al
lm] = 2 16 (52)
1" 2
1 0 o-
Al
[m] lump
0 4 0 (53)
~6~
0 0 1 _
7 -8
AE
[*] -8 16 (54)
1 -8
where the lumped mass here is the form obtained by scaling according to the
diagonal terms of the consistent mass.
The eigenvector for this element is given by
{rf}T={l,-il} (55)
and the eigenvalues for the lumped and consistent mass are, respectively,
, 24c2
I2 (56)
60c 2
(57)
108 T. BELYTSCHKO
210
lump Al x2
[m] (61)
42 10
2
al
12 6/ 1 2 6/ '
EI 4/2 - 6/ 2/:
[k] (62)
2 12 61
4/2.
W ! T = {,., '2/ (63)
where / is the section modulus and is a scale factor for which we usually
use a value of 175.
The highest frequency and associated mode of a beam element is then
given by consistent mass
-3; A . - A r - W l K / l (64,
/ 21 Orr, /
TIME INTEGRATION OF STRUCTUREMECHANICAL SYSTEMS 109
lumped mass
\92c2r2 /3/2 1
w2=^crA. Ai = N^LW}T = i / [ 1 6 / _ 1 6 / ] (65J
/ 12crg /
Here rg is the radius of gyration of the cross-section, which is given by
r\--A (66)
The lumped mass again yields a larger stable time step.
The important feature of the beam element is that varies with the
square of the element length. This is a property of systems that are
parabolic, such as heat conduction and the Euler-Bernoulli dynamics of
beams. In the standard engineering analysis of frames and beams in which a
linear axial displacement field is combined with a cubic transverse (bending)
displacement field, eqn. (48) governs as long as it is smaller than the bending
stability limit, eqns. (64) or (65). For example, for a lumped mass matrix, if
/ > 4^38 (67)
while
H/CX...+ 1 = WMi + i{A Z>; T ({F exu } + {Fe"'+1}) (71)
The kinetic energy. T, is given by
= \\[][\ (72)
Energy balance then requires that
Wm + - Wa < (73)
where d is a specified tolerance.
For two- and three-dimensional problems, analytic determination of the
frequencies of various elements becomes quite difficult. However, Holmes 9
has shown numerically for a variety of meshes that for two-dimensional,
constant strain elements, eqn. (54) still holds provided that /is the minimum
distance from any node to an opposite side. Considerable computational
experience also seems to indicate that the flexural criterion can be used for
plate and shell elements with cubic bending displacement fields using for /
the minimum characteristic element dimension.
5. IMPLEMENTATION
TABLE 1
FLOW CHART FOR EXPLICIT INTEGRATION PROCEDURE
TABLE II
EQUATIONS FOR COROTATIONAL FORMULATION
L
njvcX, +cX-,1
77v) =\cX,ATV^scX
+ JvVn (II-4)
e = Bm^; \) = [B][^<) (II.5)
(5) Stress-strain law:
j=J{iJ,MlJ) (116)
(6) Nodal force-stress relations:
h = aL L/'i-kl'l/'i (11-8)
112 T. BELYTSCHKO
TABLE III
EQUATIONS FOR CONTINUUM FORMULA TION
= Bt]Udu (III.3)
(3) Stressstrain law:
\Xj cx, J
(4) Nodal forcestress relation:
7M = B^TdV (III.6)
and that they are very simple in structure. Branching among elements
occurs only in the computation of the element nodal forces, and here the
program can be modularised very effectively, since the operations in various
elements are completely unrelated.
The introduction of geometric or material nonlinearities introduces no
complications, since there is no need for a predictor. Thus, the only
difference between linear or nonlinear material is that the computation of
the stresses, step (3.ii), involves a nonlinear relationship. Similarly,
TIME INTEGRATION OF STRUCTUREMECHANICAL SYSTEMS 1 13
geometric non-linearities only affect the form of steps (3.i) and (3.iv). They
add no fundamental complications to the process.
However, for the purpose of streamlining computations as much as
possible, it is desirable to choose strain and stress measures so that the
strain-displacement equations, the constitutive law and the nodal
force-stress relations involve a minimum number of computations. Tables
11 and 111 give formulations recommended for structures and continua,
TABLE IV
COMPUTATION ESTIMATES FOR A TWO-DIMENSIONAL PROBLEM
7 x 1 5 mesh
of 12 in elements (1000 Hz)
per time step 105 104 1-2 104
6
per ms HP 10 2-5 105 6 104
15 30 mesh
of 6 in elements (2000 Hz)
per time step 1-3 106 5 104 6 104
per ms 1-3 106 1-3 107 2-5 106 6 105
30 60 mesh
of 3 in elements (4000 Hz)
per time step 1-6 10' 1-8 105 2 105
per ms 1-6 107 1-6 10 8
1-7 10' 4 106
6. MESH PARTITIONS
'4 '
o o o o
o o o o
- <} <> -
, 0
+ KAa Ka + G KbG G KBb (79)
KBbG KB
To simplify the notation, we will write the explicit linear multistep formula
in the form
d" +1 =/J,d" + a 1 d"rh" 1 (d) (80)
where and a, are scalar factors which depend on the method and time
step (i.e. [J] = 2 ,] = 2 for the central difference method) and h"(d)
designates the historical values of d and its derivativesthat is, those
pertaining to time steps preceding or including /;. By use of the same
notation, an implicit linear multistep method can be written as
d" +1 =/J 0 d n + 1 +h"(d) (81)
If part of the mesh isto be integrated with a smaller time step, At/m (misan
integer), then we write the explicit difference formula as
n+ p - l ) / m , un + l p - a i / m / j .
/M n
(pm) _ - ( P - Dim
+ ,d (82)
'KA ^Al)
+ (83)
K.AB
If a consistent mass matrix is used, d'A+ ' and d'+ ', are coupled through the
118 T. BELYTSCHKO
/< (85)
<*>mai
F o r a disjointed partition the mass matrix becomes non-diagonal even
when the original mass matrix is lumped because of the presence of the
G 1 M b G term. However, because this part of the matrix is usually smaller
than the other p a r t s , it is still convenient to work with a lumped mass
matrix. Substituting eqn. (80) into eqn. (79), we obtain
where the last term represents a d'A lembert force corresponding to the
accelerations of nodes in b. Note that it must be shifted back one time step,
because in explicit methods, db is known only after db+ ' is known. Equation
(85b) is then replaced by
da+ ' = Ma '(f, + CTfb + fA B Kada K A a d a ) + h " ' (db) + , AdA
(87)
so that a complete separation of algorithms is feasible. However, the
transfer of inertial terms with a time lag which is indicated by eqn. (86) has a
destabilising effect that is demonstrated by Belytschko and Yen. 6
We next consider an explicitexplicit jointed partition where domain A is
integrated in time with a time step At I in, while domain is integrated with a
time step A t. We will immediately restrict our attention to M A B = 0. The
governing equations are then
dB+ ' = IBMBHF'B KA Bd'B KBdB) + a1Bd'B + h " '(d B ) (88)
dA+"""> = /j 1 A M A "'(F A + , " 1 , ' m K A B d A +,
"" /m K B d B +(n 1)/m )
+ aiBd B +,n """" +h'"' 1 1 ""(d B ) (89)
It can be seen that eqn. (88) can be performed independent of eqn. (89), so
that domain is easily updated first. However, for m > 1, eqn. (88) requires
the value ofdA+,"~ " "'forali nodes in A that are connected to elements in RB;
let these nodes be designated by the subscript . If dA + ' is obtained first,
then the fractional time step values can be obtained by interpolation.
Because these partitions with interface interpolation do not ensure that
all of the difference equations in time are met at all nodes, the usual stability
criteria no longer apply.
For a jointed explicitimplicit partition, (A , explicit; B, implicit), the
equations for updating the displacements are
' = /J1A IVI '( KA d"A KA Bd"B) f a 1A d A + h"" '(d A )
\"V
< ' = Ku + TT" M B 1 / mF++ 1' WKA Bd" +
J / AI + 1" + ^ MBh"(dB) (90)
V POB / V POB /
Thus, if the explicit domain is updated first, the implicit domain may be
updated without any interpolation. A s expected, the stability of this
procedure then depends only on the highest frequency of the explicit part of
the mesh.
The essence of eqns. (85) can be graphically deduced by considering the
flow of information in explicit and implicit time integrations. For this
120 T. BELYTSCHKO
A. EXPLICIT-EXPLICIT
i+1
w~ X
r-^. ft,
B. EXPLICIT-IMPLICIT
/+;
INTERFACE
FIG. 4. Flow ofinformation in explicit-explicit and explicit-implicit partitions.
REFERENCES
1. INTRODUCTION
these will form the basis for nonlinear solution procedures. The effects of
geometric nonlinearities arising from large deformations will then be
introduced into the numerical algorithm. For materially nonlinear
problems the relevant expressions from the theory of plasticity are
presented, and, finally, the concept of viscoplastic solids, which is more
appropriate to timedependent plasticity analysis, is included in the time
integration process.
element mesh band or front width. However, there are severe restrictions on
the permissible time step length. Implicit methods demand many more
operations per time step but the time step length can be an order of
magnitude larger than for explicit algorithms. Both methods have
substantial advantages for certain classes of problems, and the merits of
various options have been compared and discussed by many in-
vestigators. 2 " 7 In forming guidelines for an appropriate choice of method,
it is important to note that certain problems require small time steps
regardless of the numerical stability requirements, whereas others do not.
Generally, problems can be categorised 8 as (a) wave propagation problems
and (b) inertial problems.
Wave propagation problems are those in which the behaviour at the wave
front is of engineering importance, and in such cases it is the high-frequency
components that dominate the response. Problems in this category include
shock response from explosive or impact loading. Dynamic problems
which are not wave propagation problems can be considered inertial
problems, and here the response is governed by the low-frequency
components. This class includes seismic response and large deformations of
elasto-plastic structures under ramp or step loading. The rise time and
duration of the load relative to the time required for a wave to traverse the
structure offers one approach to categorising the problem: if the rise time
and duration exceed several traversal times, the problem is generally of the
inertial type. As a broad guideline, wave propagation problems are best
solved by explicit techniques, whereas implicit techniques are more
appropriate for inertial problems. However, the relative economy of both
approaches is also influenced by the topology of the finite element mesh and
by the architecture of the computer to be employed.
>" Na (4)
IMPLICIT FEM FOR DYNAMIC TRANSIENT ANALYSIS OF SOLIDS 127
-o-
$=-!
1-0 0-577
'I
1
)
O r^
1 T
Io
0-577
*
o o
where N, are the element shape functions for each of the m nodal points of
the element. The change in strain associated with a displacement increment
Aa is defined as
= Ba (5)
where B is the appropriate strain matrix, generally composed of derivatives
of the shape functions. For geometrically non-linear problems B is
displacement dependent and its explicit form isconsidered in Section 7. The
corresponding stress change is obtained by use of an appropriate
constitutive law as
= (6)
The explicit form of D is considered in Section 8 for plasticity problems
128 D. R. J. OWEN
ML N T pN dr (8)
C = N T CNdi' (9)
Pn = [B'fV.dr (10)
In this section we describe the use of three implicit integration methods for
linear analysis. In particular, the Houbolt. Wilson 0 and Newmark
methods are presented. Use of these methods in linear situations also forms
a basis of non-linear solution algorithm by the same techniques.
= lla 18a 9a
"+' 7^T,
/ "+' - + ,-i " - : }
+ , = ^ 2 { 2 a + 1 - 5 a + 4 a _ , - a _ : } (12)
where the subscript /; + 1 stands for time i+i + t + At for which solution
is sought. In this implicit method equilibrium conditions are considered at
time /+, in order to obtain a + ,. so that eqn. (1) becomes
M + 1 + C (I+l + Ka + , = F +l (13)
Substituting eqn. (12) into eqn. (13) and rearranging,
2 11 \ (5 3 \
_ M + C )a _ 1, ++[.;M+-C)
+ V2 (14)
27 (7 3,
Thus, the solution for a + ] requires knowledge of a, a,,., and a_ ; . The
process can be started from a knowledge of a 0 , 0 and 0 , but it is more
common and accurate to calculate the first two time steps by use of some
other algorithm such as the central difference explicit scheme with a fraction
of / as the time step.
The appearance of the stiffness matrix Kon the left-hand side of eqn. (14)
implies that a full simultaneous equation solution is required for every time
step. The H oubolt method, being an implicit method, is unconditionally
stable, and. hence, there is no critical time step limit and . can, in general.
be selected many times larger than for the central difference method.
If mass. M. and damping, C. terms are neglected, it can be seen from eqn.
( 14) that the H oubolt method reduces directly to a static analysis for time
dependent loads.
a
t.at t.eit
tAt tet
O WILSON- METHOD.
b. NEWMARK METHOD.
a
r +i a; + Q . (2, + M , ,) (15)
Or. for the particular time = OA t, eqns. (16) and (17) give
OAt
a
/ + UA/ a
, + j ( a / + ()A/ + a
;) (18)
. 2
a, + ( M , = 3 , 4 - 0 - /a, + - (a, + ., 4- 2a,) (19)
6
from which we can solve for , +(M , and l+0.A l in terms of a, + UA,:
6 6
3,+ =
" ' W (a,+"A' ~ a,) - f>/7 a ' ~~ ' <20)
a
3 0-At
/ + A, = g 7 ^ ( a , + A , - a , ) - 2a, a, (21)
a, + A/= , + N ( , + A, + a,)
3, +/ = a, + /, + 1 0 (, + d ; + 2,) (26)
where
bb = b0iO; h1=-b2/0; 8 =l-3/0; 9 = . / 2 ; 1 0 = , 2 /6
v27)
132 D. R. J. OWEN
It is noted from eqn. (26) that no special starting algorithm is required, since
a, + A l + & , + A, are all expressed in terms of the same quantities at time t
only.
0
- '
aAt2'
b
'
- S
/'
b2 =/
-;
2: 5
-e
By use of eqns. (28) and (29) all quantities at time / + A t can be finally
2/2 (32)
expressed as
, + M = bb(al + Al a,) + 7 , + b8,
I + A, = , + M . + toa< + 4,
a, + M = a , + (a, + A, a,) (33)
IMPLICIT FEM FOR DYNA MIC TRA NSIENT A NA LYSIS OF SOLIDS 133
where
bb = b0; 7 = b2; bs = b3; bg = / ( 1 ); bl0 = A t (34)
Comparing eqn. (24) with eqn. (31) and eqn. (26) with eqn. (33), it can be
seen that the numerical time integration algorithms for the Wilson and
Newmark methods are identical, provided that the appropriate values of
the constants blb10 are employed: defined by eqns. (25) and (27) for the
Wilson method and by eqns. (32) and (34) for the Newmark method.
Thus, both methods can be easily incorporated in the same computer code.
MX + CX + KX = F (36)
where
M = S T MS; C = S T CS. = S T KS, F = S T F (37)
(I
The solution to each equation in (43) can be obtained by use of the direct
integration algorithms described in Section 2. To obtain the complete
response, the solution to all equations in (43). / = 1. 2, 3 , . . . . N, must be
IMPLICIT FEM FOR DYNA MIC TRA NSIENT A NA LYSIS OF SOLIDS 135
calculated and then the finite element nodal point displacements obtained
by superposition of the response of each mode, so that
a(0 = ) ( 44 )
= 1
For many situations the applied loading only excites the lower
frequencies of the structure and therefore only a fraction of the total
number of decoupled equations (37) need be considered in order to obtain a
good approximation to the true structural response. Generally only the first
equations need be considered in eqn. (43), where <s: TV. This means that
we also need to solve only for the lowest/? eigenvalues and corresponding
eigenvectors in eqn. (40) and we only sum in eqn. (44) the response in the
first modes.
For example, in seismic loading situations in some cases only the ten
lowest modes need be considered, although the order of the equation
system may well approach 1000. On the other hand, for blast or shock
loading many more modes are generally needed and may be as large as
2/V/3. Thus, for selected situations the mode superposition technique offers
an economic solution alternative to direct integration.
The modal analysis technique can also be extended to nonlinear
dynamic problems and a summary of such applications can be found in
reference 34.
P = 'adv
K= B,[DBdi
J
in eqn. (52).
,38 D. R. J. OWEN
1
.+ 1 = ^ 7 2 i 2a fi 5a + 4a, 1 a _ , |
NO \A'\ tolerance
Return to Step (7a) + a'I factor, (59)
YES
Step (8) Return to Step (4) to process the next time step.
The solution scheme described above can be interpreted as a generalised
Newton process.
IMPLICIT FEM FOR DYNA MIC TRA NSIENT A NA LYSIS OF SOLIDS 139
A 6
h 3 1 .b
bo /2' ' oc
A t
* ~ (0,) ' ~ 0/
b2 = 2 , ; 3 = 2 2
oAt 2
A4 = 2; 5 = 0//2 4 = 1 ; 6 5 = , ( 2
6 = o / 0 ; 7 = 2/0 b6 = 0 ; 7 = 2
bs = 1 3/0; bg = //2 b8 = b3; b9 = A t(\ )
2
, = / /6 b
io = A t
K = JB T D n B n di;
3
/ + / = a, + a
NO lAa'l
<
Return to Step (8a) la, + a'|
YES
Step (9) Return to Step (5) to process the next time step.
in the global x directions and does not follow the particle; this makes it
difficult to write the appropriate constitutive equations. Furthermore, since
we are concerned with true stresses, all integrations in the finite element
process must be made over the deformed volumes.
For these reasons it is far more common to work in terms of the second
PiolaKirchhoff stresses whose reference coordinates move with the
particle and are based on the undeformed area. Thus, for small strains the
second PiolaKirchhoff stresses are the engineering stresses. The
corresponding strains are the G reenLagrange strains, defined as
cu, cu cukcuk
CX: CX: CX: CX:
J
, ' ' , > (67)
_ cu_ 1 cuk cuk
oxj 2 ex, xj
where u are the displacement components and x are the undeformed
coordinates. However, it is difficult to use this description for large strain
problems, since the Green strains are not true strains in this case and the
PiolaKirchhoff stresses lose their physical interpretation. Geometrically
nonlinear problems may be formulated in either:
(1) a total Lagrangian coordinate systemwhich is expressed in terms
of the initial position; or
(2) an updated Lagrangian coordinate system in which the initial
position changes and becomes the current equilibrium state prior to
some incremental change.
An updated Lagrangian formulation restricts displacements from a
current initial position to satisfy the criterion of small rotationsand, hence,
can be employed for the solution of large strain problems. The
straindisplacement relationship can be incrementally expressed as
de = Bda (68)
= B 0 + BL (69)
In this, B 0 is the usual operator for infinitesimal strains and B, is the non
linear contribution which is a function of the current displacement state.
Matrix B, can be explicitly expressed but it is often more convenient to
express directly. The matrix partition associated with node / can be
142 D. R. J. OWEN
ex c N cycNj czcN,
, i
~z ^
cx cx cx cx cx cx
cx c N cycN, czcNi
> N
> " \
cy cy cy cy oy cy
cx cN oy c N, c cN
_ _. ~^ ^ ^
B, c c cz cz c c (70)
cx c ,' cx c N,
cy c N, cy CN; ( ON c c/V,
^ ~^ + ^~ h ;
cz cy cy cz dz cy cy c cz cy cy c
cxcNj cx DN, cy oN c y cN c cN czcN
1 ^ + h ;
(
c cx cx c cx cx ( cx cx c
cx c N, cx cN ) c N, cy C ; f cN czcN
- ^ + -^ + ^ oy h ;
_ cy cx cx cy <~) cx cx oy cx cx c y _
The only influence that large displacements have on the numerical solution
sequence described in the previous section is in the form of the matrix B to
be employed. In the total Lagrangian approach the original coordinate
system is employed throughout the analysis and the full nonlinear strain B
defined by eqns. (69) and (70) is employed. In an updated formulation the
nodal point coordinates are recalculated (or updated) at the end of each
time step and the infinitesimal strain matrix, B 0 , is employed.
and
cF
=q (77)
( a
which define the yield surface after initial yield, eqn. (75) may be written
qTd<x-/ld;. = 0 (78)
Also, eqn. (72) can be expressed as
D-'dff + d / ^ - (79)
oa
144 D. R. J. OWEN
D
- = D -ZTdv d = Dq (82)
It can readily be shown that A = H', where H' is the local slope of the
uniaxial stress-plastic strain curve.
Elastoplastic material behaviour affects the numerical solution algor-
ithms described in Section 6 in two ways:
(1) In Step (4) of Section 6.1 (Step (5) of Section 6.2) D becomes D ep
for the evaluation of K. It is noted from eqn. (82) that D e p is stress-
level-dependent, since q = cF/ca.
(2) The stresses a at each plastic point must satisfy the yield criterion.
During application of load, either at the beginning of a time step or
during equilibrium iteration, the stresses at some points may exceed
the permitted level. These stresses must be brought back to the yield
surface by the numerical process shown schematically in Fig. 3.
Thus in Step (7) of Section 6.1 (Step (8) of Section 6.2) an additional
operation is required after operation (b):
Step (Tb)(i) (Step 8(/)) For all yielded points, reduce the stresses to
satisfy the yield surface condition.
These modified stresses are then employed in the evaluation of the
equivalent nodal forces P. + 1, in Step (7c).
A major disadvantage in using classical plasticity theory for dynamic
problems is that time rate effects in the material deformation process are
ignored. This can be partly overcome by allowing the instantaneous yield
stress of the material to be a function of the current rate of straining.
However, time-dependent plastic flow phenomena are physically better
modelled by means of viscoplastic theory, and such an approach is
described in the next section.
IMPLICIT FEM FOR DYNAMIC TRANSIENT ANALYSIS OF SOLIDS 145
bdXS.=Dde^
- Inactive if
L
vp
il
nrrrrrrm
Fie. 4. Basic viscoplastic model.
viscoplastic components. The total stress rate depends on the elastic strain
rate according to
= D c (84)
The onset of viscoplastic behaviour is governed by a scalar yield condition
of the form
F(a, ) F0 = 0 (85)
in which F0 is a reference yield value.
The rate of viscoplastic straining can be defined by the following
associated viscoplastic flow rule:
OF
= -,'<0(F)> 7<<P(F)>q (86)
( a
in which y is a fluidity parameter controlling the plastic flow rate. The term
(.) is a positive monotonie increasing function for > 0 and the notation
< > implies
<(.)> = (.) for > 0
(87)
<(.)> = 0 . < 0
IMPLICIT FEM FOR DYNAMIC TRANSIENT ANALYSIS OF SOLIDS 147
Different choices have been recommended for the function . The two most
common versions are
where
D = ( D ' + C J ' (96)
Matrix D is a symmetric matrix when the viscoplastic law is associative. In
eqn. (95) the notation B is employed to denote the possible inclusion of
large deformation effects.
During a time increment the equilibrium equations which must be
satisfied are given by the following incremental expression:
= (97)
K = BlDBdt (100)
The displacement increments a when substituted back into eqn. (95) give
the stress increments A a and thus
+ = ff, / It
+
(101)
a
M+ 1 = a + J
Use of eqn. (95) gives
. = BtJa D ' (102)
and then
.+ 1 = . + , (103)
The changes required to include viscoplastic material response in the non
linear solution algorithms described in Section 6 are given below.
(1) In Step (4), Section 6.1 (or Step (5), Section 6.2) the form of K
given by eqn. (100) must be employed.
(2) A dditional operations are required to evaluate the stresses as
follows:
IMPLICIT FEM FOR DYNA MIC TRA NSIENT A NA LYSIS OF SOLIDS 149
REFERENCES
1. INTRODUCTION
where a represents the crack length, is the applied traction, / is the time
and the loading rate. Then, for a propagating crack,
where denotes the crack velocity. If the parameters Kd and KD are known,
no further experimental fracture data are needed to obtain dynamic
fracture mechanics solutions.
An equivalent formulation of the conditions of crack propagation can be
made in terms of the strain energy release rate, G. Then the criterion for
dynamic crack propagation can be expressed as an equality between G and
its critical value, the energy dissipation rate required for crack growth.
Denoting this latter quantity by R, we have the crack propagation
condition:
G(a, , /) = R() (4)
For a propagating crack, arrest will occur at time t. when, and only when,
G < R for all / > E,. The dynamic energy release rate (or the driving force
for crack extension) can be expressed as
1
C= -^-()2 (6)
156 D. R. J. OWEN
where E and are, respectively, the elastic modulus and Poisson ratio, and
plane strain conditions have been assumed. Also, () is a geometry-
independent function of the crack speed, given by
a^2
(1 -a2/c12Y'
.
2
(7)
(1 _ v)[4(i _ a2/c2)12(\ - a2lc\y<2 - (2 - 2/c)2]
where c, and c2 are, respectively, the dilational and shear wave speeds in the
material.
As a consequence of eqn. (6) we have, for linear elastic conditions,
f E R V'2
(8)
M^W
Thus both Kt and R express the material's resistance to cracking and are
taken to be material properties which are independent of geometry and
applied loads. Equation (4) is attractive, since its basis lies in a fundamental
energy balance and can be readily extended to nonlinear situations. An
alternative criterion which has been employed for dynamic crack
propagation, in view of its simplicity, is one based on a limiting value of the
normal tensile stress at the crack tip. 2 Crack initiation is assumed to occur
at a stress level , and continuing propagation takes place if the stress
component normal to the cracking plane at the crack tip attains a value
uc = <7 The factor 5 can be interpreted as a measure of the initial
bluntness of the crack tip, and a c quantifies the material toughness.
opening profiles of a pipe having a stationary crack tip using a thin shell,
finite element, large elastoplastic deformations computer program.
and its advantages are counterbalanced by the large number of time steps
required. Further, in the usual finite element approximation the matrices M
and C do not appear generally in a diagonal form and a further
approximation known as 'lumping' must be introduced to make the scheme
viable.
For many problems involving short transients (i.e. very rapid
loadingdeformation sequence) extremely efficient solutions using such a
scheme can be obtained, and fast fracture problems generally fall into this
category. The internal forces for use in eqn. (11) are most conveniently
determined from
in which <r are the current stresses which satisfy all nonlinear conditions in
the structure. For linear elastic situations obviously
P = Ka (13)
obvious that it is not necessary to assemble the full vector of forces before
the calculation beginsand that the computation can be carried out'point by
point' to determine a"+1.
At any time stage the sequence is, therefore:
(1) Calculation of and assembly of forces P" and F" for a node ; = 1.
(2) Computation of a" +1 from known values a" and a"'1.
(3) Repetition of the process for i = 2, etc.
(4) Computation of P"+ ' and F"+ ', etc.. for the next time step.
The storage requirements are thus directly proportional to the number of
unknowns. /and not as when an inversion is necessary to r o r / 3 . Further,
the computational time is also directly proportional to ;. Herein lies the
great computational simplicity of the process.
It is noted that to start algorithm (14) the values of a" and a"~ ' are
necessary. Often at the start of a dynamic problem the values of af and J'
(the velocity) are prescribed and apparently a starting difficulty exists.
However, writing
a = (a -~1)/(2) (16)
allows a special algorithm to be written for the first time step by eliminating
1 from eqn. (14).
In practice we seldom find it necessary to use this starting algorithm if.
say before the commencement of the transient, the structure was at rest and
all the displacements prescribed.
a = "',., = (17)
= 3 (18)
,^,,^+,, (19)
where
= (20)
TRANSIENT DYNAMIC NUMERICAL METHODS 161
On use of the principle of virtual work, the appropriate terms of eqn. (9) are
found to be
M N T pNdu (21)
N T CNdi' (22)
Aa" (25)
in which is the material density, C stands for velocity proportional
damping forces, b are the body forces per unit volume and d are the
boundary tractions. All integrations are performed numerically with Gauss
quadrature formulae.
shell formulation the yield function is based on the stress resultants, and
particular forms are described in Section 7.2.
The process of computation proceeds, therefore, as follows:
(1) From steps (l)(3) of the previous sequence, a is found by use of
the elastic matrix, D.
(2) A check is made to determine whether a is above the yield limit. If it
is not, computation proceeds to step (4) of the previous sequence; if
yes:
(3) The elasticplastic matrix, D e p , is computed by use of _,.
(4) Steps (l)(4)ofthe previous sequence are carried out with D = D t p .
A slight elaboration is necessary in principle if _ , is below the plastic
limit. The matrix DL.p is then computed with a part of the increment allowed
to be made in elastic mode and part in elastoplastic mode. Details of such an
elaboration are given in reference 24, but it is found that in using small time
steps this procedure is not essential.
Little has to be said concerning the computation of F. If the forces band
d are conservative, then their magnitudes are simply calculated by eqn. (24).
the original element coordinates being used. However, if the forces are not
conservative (i.e. if, for example, a pressure loading is applied to grossly
deforming boundaries), then the computation given by eqn. (24) must be
performed on the deformed geometry of the element. This modifies not only
the regions of integration, but also the shape functions themselves, which
will depend now on updated element coordinates.
li = 1 (29)
This is equivalent to requiring the total mass of the element to be preserved
and within this constraint permits many possible systems of lumping.
TRANSIENT DYNA MIC NUMERICA L METHODS 163
in which
When all element diagonal terms have been determined, we find the total
diagonal term by summationthat is,
c2= h (34)
164 D. R. J. OWEN
for dilatation waves and shear waves, respectively, and since always the first
of these is the higher one, the time step is given by
p(\ + v)(l - 2 v )
, = y'L"lcx = )'" (35)
(1 - v) "
where y' is some number less than unity and L" is the smallest distance
between adjacent nodes of any parabolic element used. Rock 27 finds that
y' = 0-45 is a safe value to use, and the present studies corroborate the
point. Indeed, for many problems y' values in excess of this limit provided
stable solutions, the maximum observed being 0-5.
If linear elements are used, then for stable solutions a value of 0-91 for
y' was found suitable.
These limitations can often be imposed with some severity even if one
small size element is included in the analysis. The choice of mesh
subdivision is thus critical.
Figure 1 shows the DCB specimen to be considered where one of the finite
element meshes employed is illustrated. Owing to symmetry only one half
of the solid need be analysed. The loading-pin holes are ignored in solution,
and instead selected nodal points are assumed to be displaced a prescribed
L = 360 72 m m
@> m
r
23 1mm
12 CD
22 36 m m each
. : 92 <mm = 268 32 mm
Mesh "B"
distance apart. As the loading rate is slow, static conditions can be assumed
before crack propagation commences. Numerically the steady state
deformed configuration is obtained by performing a transient dynamic
analysis and employing a suitable mass-proportional damping term as
described in Section 4.4. Node 100 at position (H/2,L) is subjected to a
prescribed lateral displacement of H. This lateral steady state displacement
is consistent with the values assumed in references 2 and 4, and is employed
here for comparison purposes. In order to avoid numerical instability
associated with the large mesh distortion which takes place, this
displacement is assumed to be achieved in a selected number of time steps
(100 for the problems presented) and to be applied at a constant rate. Steady
state conditions are deemed to have been achieved when the change in
displacement norm for successive time steps of all nodal points is
sufficiently small.
(Wo c =2 S) "
A
a
<S I(""Voc= 17 )
jS
< ^ ^ l 0 / 0 < : = 1 Ol
Time ( M- s)
FIG. 2. Computed crack growthtime plots using stress criterion for different
aqaQ values (DCB. mesh B).
energy absorbed at the crack tip as the crack propagates. It is noted that the
external work done is zero in this case, since the loading takes the form of a
constant prescribed displacement.
The strain energy, u, at any instant can be calculated from
,.., = + (P+, +)(3) (36)
The kinetic energy contribution, K, at any instant can be directly calculated
from
* + i = ( + 1 ) T M n + 1 (37)
TABLE I
SUMMARY OF RESULTS FOR DCB PROBLEM
Experimental [2]
Kanninen - energy criterion [21
Sn
h im
i iu
uee li yy &
& Peretz
r e r e t z -- ss tt rr e
e ss ss c rnitteerr i o n
e
{ c r i t e r iion
O - Energy (Mesh '
Present formulation i o -- S
o S tt rr e
e ss ss ce rni tt ee rr ii co n (Mesh 'A')
0 15
0 20 3 0
a
Relative arrest length /a.
FIG. 3. DCB problem. Plot of crack velocity against relative arrest length.
u 0 , and kinetic energy, K0, are calculated. (In fact, K{l = 0.) The damping
term is then eliminated, the restraint at the crack tip removed and u and K
determined for each time step. The quantity ((w0 + K()) (u + K))/b da is
calculated and propagation of the crack to the next node allowed when this
term attains the specified value of R. Similar to <xc in the stress criterion, is
a measure of the specimen toughness under dynamic conditions.
250 -i
(Arrest )
E 150
D (Arrest)
g" 100 o
(R= 11K3 k g / m m
(Arrest)
(R = 1 6 7 U kg/mm)
FIG. 4. Computed crack growthtime plots using energy balance criterion for
different toughness (R) values (DCB mesh B).
The energy change (du + dK) required for crack advance by the energy
criterion can be produced in two ways:
(1) If the material behaviour is assumed to be elastic, an energy sink
can be provided by gradually releasing the crack tip node as soon as
a critical energy level is reached. In this way work is done (and
energy dissipated) at the crack tip.
(2) If elastoplastic material behaviour is assumed, plastic deformation
in the crack tip region automatically provides the necessary energy
sink.
In Fig. 4 the relationship between crack growth and time is presented for
three toughness values, and assuming elastic material behaviour. A gain
a/a0 and v/c0 values are recorded in Table I and incorporated in Fig. 3.
Good agreement between the eight points A from the present
TRANSIENT DYNAMIC NUMERICAL METHODS 169
10 V
p,f'" X- 0 5 ^ ^ V s _
P.e-'
0 I I I I
I1 1 2 3
t(m.s)
FIG. 5. Opening profile of a pressurised pipe with an initial crack and the pressure
distribution in the pipe.
-J
C
3
-
^-
C
( 20 noded I elements
er of nodes 254
rule 2 H 2 2 gauss
FIG. 6. Mesh used for dynamic large-deformation clastic-plastic three-dimensional analysis of a pipe (with a stationary or
propagating axial crack) under pressure loading.
TRANSIENT DYNA MIC NUMERICA L METHODS 171
this work the problem is uncoupled and the steady state gas pressure
distribution is assumed to exist at all times, which clearly introduces an
element of approximation, particularly at commencement of crack
propagation. Furthermore, the pressure distribution is assumed to be
axisymmetric with respect to the pipe axis. The initial pressure, P is
assumed to drop linearly to a value P0 in a specified time and to remain
constant at this value thereafter. This is achieved by use of an appropriate
function of time J(t) (see Fig. 5) in the relation
P0 = Pc JO) (40)
18,20
The variation in pressure in the cracked zone is taken to be
= p o e '^.,) (4 j)
32.000 I b / i n *
(crack tip reached end ot pipe section)
aq = 70.000 lb/in 2
q/oc =15
(Arrest)
p.= 2250 l b / i n 2 . p i t ) = p ( i . e . t l
E = 27 5 IO 6 lb/in 2 . V--03
= 0 734x10" 3 lb s 2 / m ' , e. =-/^7,
2
R = 15' ; t = 2 5" , L = 7 5 " : a.= 30"
= 2 5ps
o .y = stress at crack (propagation) initiation
Oc = critical stress for subsequent propagation
It) = half angle of crack opening
WO 500 600
crack growth Initiation
criterion being used, and the results are shown in Fig. 8. The calculated
crack velocities in this case are in agreement with experimentally measured
speeds for pipelines. On the other hand, the values based on a limiting stress
propagation criterion are excessive and can only be attributed to the values
of aq and assumed.
Crack propagation in a pressurised pipeline is a problem in which
5
E = 27 S I O 6 I b / i n 2
to v = 03
= 0 734 1 0 " 3 l b . s 2 / i n 4
35 At= 2 5 s
Rj = toughness for initiation of crack propagation
30 R d y n = toughness for subsequent propagation
(Arrest )
0-057
AZ
central node are combined to create a deflection normal to the element. The
rotations and normal displacement at the centre, as well as the 8 rotations
0Jy: at the loof nodes are then eliminated by imposing 11 constraints on the
shear behaviour. The quantities employed in the definition of the various
matrices involved can then be expressed in terms of the degrees of freedom
retained, a", as
G = Fa'' (42)
in which
CU CU CV CV cu cv cu C2u cv
-.2
(43)
c X cY c c YCZ CZ C XC c YCZ c XC C YCZ
The terms CujcX, etc., are the displacement derivatives with respect to the
local axes. Matrix F is a 13 32 array and contains the final shape
functions for use in the evaluation of the strain and mass matrices.
Subject to the conditions of small rotations and finite membrane strains.
the strain-displacement relationship for large-displacement problems can
be written as
( ;/ < u
CX 'Z
Cv
CY
Cu cv u cv
+
CY CX Tzc~z
(B 0 + BL)a'' = Ba (44)
=
2
+
c U
o
CXCZ
-< 7
cv o
cYcZ
cu
-, >
CYCZ +- o
xcz
Matrix B 0 can be formed from rows 4-7 and 10 3 of F, and B L , which
contains the non-linear terms, is formed from rows 8 and 9. For
geometrically non-linear behaviour the strain matrix, B. has to be
determined at every time step according to eqn. (44) if a Lagrangian
formulation is adopted. For the updated Lagrangian procedure new values
TRANSIENT DYNA MIC NUMERICA L METHODS 175
= N. + N u (46)
= )'<(>(^ (48)
<a
where a is the stress vector composed of the membrane stresses and bending
moments and F is a scalar yield function. Also,
F=Q,+% + L I , J 1 (50)
in which
1
/O _ t M \l .. M _ // /
0
(51)
2
where // = is the uniaxial yield force, M0 = 0 /4 is the uniaxial
yield moment and 0 is the uniaxial yield stress of the material. The
parameter is defined in terms of the viscoplastic curvature y^ as
a=l_ie^; /7 = f ^ (52)
30
where fi is a constant. E is the elastic modulus and
^ - f - ^ - F . (58,
F l t l = 0 005 U.DL.
i i i i i u i i i i i i m i i i i i i n u m u i i n uiiiiiii
Material properties
E =105
v= 0 3
/ h=05
; Mo= 0
a 21 =32 0
/ M =0 035
/' o Y' = 0 0001
y = O00C05
_ _ . D _ . '=
/
/
where the results obtained are compared with those of reference 30. Large-
deformation effects are included and solutions for both a total Lagrangian
and an updated Lagrangian formulation are presented. The considerable
difference between the present results and those of reference 30 is
attributable to the boundary conditions assumed. In reference 30 a 10-node
isoparametric element was employed with two out of three nodal points
being allowed to slide along the inclined support, whereas a fully fixed
condition was assumed in the present analysis. Further analyses employing
different boundary conditions confirmed this observation, but a direct
comparison was not possible, since the nodal distribution along the
boundary differed for the two studies.
For engineering problems the limiting time step length for the
viscoplastic algorithm, eqn. (58), is generally far greater than either of the
limits imposed by eqns. (56) and (57). For thin shells the time step length
restriction imposed by eqn. (56) is not severe, owing to the presence of// in
the denominator, and the critical time step is provided by the membrane
action restriction, eqn. (57), for most practical problems. It should be noted
that for pure fiexural action the time step limit imposed by eqn. (57) must
still be observed even though the membrane stresses will be sensibly zero.
For this reason it is advantageous to develop a plate bending element for
TRANSIENT DYNAMIC NUMERICAL METHODS 179
m
o ,-. o \
l * / ?\
.1 :
"\\ !
w
* I
\^% \ I !
9 \ i' /
\\
\ \
/ / /
rP I
dynamic analysis, since in this case eqn. (56) provides a much larger limiting
time step for plates which are appreciably thin. In fact, the solutions
illustrated in Fig. 10 were obtained by use of such a plate element and the
permissible time step length was some seven times greater than that allowed
by a full shell formulation.
f '
9.20
s / /
2000
9-60
ZONES OF VISCOPLASTIC
FLOW
-
t 0 - 4 I N.
303 PS
I.
c
rr.
7.
8-120
-180 .J . I I I I
Z.O 5 10 15 JO 25 30 35 40 45 50 I N.
FIG. 12. Finite element mesh and development of viscoplastic zones for pipe fracture modelCase I
TRANSIENT DYNA MIC NUMERICA L METHODS 181
TABLE II
Case I Case I
the author. Details of the geometry and material properties are included in
Table II. In each case both elastoplastic material behaviour and large
deformation effects are considered. The drop in crack tip pressure is based
on the plan area of the crack configuration. Equation (40) is replaced by the
following semiempirical relationship:
A
^ = 1 _ 07 (59)
AV
in which A is the crack opening area and A p is the crosssectional area of the
undeformed pipe.
The axial variation in pressure in the cracked zone is again assumed to be
given by eqn. (41). In each case the symmetric quadrant of the model was
idealised by a total of 44 semiloof shell elements, and the developed mesh is
shown in Fig. 12.
The static solution was first obtained by carrying out a damped transient
dynamic analysis, and Fig. 13 shows that steady state conditions are rapidly
achieved. The starter crack of prescribed length was then introduced by the
release of selected nodal points, and at this stage the damping constant was
set to zero. The energy criterion of crack propagation was employed, and
the energy dissipated in the crack tip zone was directly calculated as
a1,pdvdt (60)
in which a are the total stresses, vp the viscoplastic strain rate and the
volume integration taken over the material in the crack tip region
contributing to the cracking process. For neartip plasticity problems this
sc
t J
t = 10 SE
C S
D A M P I N G C ONSTANT - - 1 2 0 0 0 S E C '
400 500
TIME STEPS
FIG. 13. Static solution of uncracked pressurised pipe by means of a damped transient dynamic solution.
CRACK PROPAGATION
564 - ENERGY LEVEL FOR CRACK
PROPAGATION CASE II )
-
>
/.
-
<
>
s
c
'.
m
5
>
r
i:
m
C
D
:
3000
N OF TIME STEPS
Fia. 14. Development of energy available for crack propagation with time.
GC
SC
-U
CASE 1 t - 0-4 I N
- 303 PSI .
I
t-
p
Q AVERAGE CRACK VELOCI TY - 470 FT/SEC.
C
(3 - 0-02
<4
-6 6 O
52 L
2300 2400 2500 2600 2700 2800 2900 3000 3100 3200 3300 3400
N OF TI M E STEPS
Fio. 15. Computed crack growth-time plot for Case 1 pipe study.
TRANSIENT DYNAMIC NUMERICAL METHODS 185
t = 0 - 3 8 9 IN.
P = 1 7 5 PSl.
t = 10 SEC
ALL DISPLACEMENTS M A G N I F I E D BY 5
R (61)
b da
in which b is the pipe wall thickness, da the internodal distance and R the
fracture toughness of the material. The variation in fracture energy with
time is shown in Fig. 14 for both cases. For Case 1 the energy builds up until
the critical level is attained and crack propagation occurs. However, for
Case II the fracture energy reaches an upper limit well below the critical
value and the induced starter crack does not propagate at all. These results
are in agreement with experimental observations, where Case I showed
considerable crack propagation and Case II none whatsoever. Figure 15
SC
TANGENTIAL
FORCE/UNIT LENGTH
12000
6000
THEORETICAL
4000 INITIAL
C
FIG. 17. Tangential stress distribution along crack axis for Case I.
TRANSIENT DYNA MIC NUMERICA L METHODS 187
shows the extent of crack propagation with time for Case I, and solution
was terminated at 3000 time steps for reasons of economy. The average
crack velocity is calculated to be r/c 0 = 0 028.
The development of zones of plastic flow with time is included in Fig. 10
for Case I. and it is seen that the initially localised plastic flow becomes
w idespread as the flanks of the starter crack deform appreciably. The extent
of deformation in the cracked zone is illustrated in Fig. 16, where the
deformed shape of the central section (Z = 0) is shown for various times.
The greater deformation of Case I is immediately apparent.
Finally, the tangential (hoop) stress along the crack axis is shown plotted
in Fig. 17 for Case I for two times prior to crack extension. The relatively
small departure from zero in the starter crack region can be attributed to the
fact that the stresses plotted are the Gauss point values nearest to the crack
axis and not the values on the axis itself.
9. DISCUSSION
REFERENCES
1. INTRODUCTION
- S T O R A G E RACK
MECHANISM
SUBASSEMBLY GRIPPER MECHANISM
SHIELD COOL
ING DUCT
P R I M A R Y TANK
COVER
SUBASSEMBLY
S T O R A G E RACK
MECHANICAL PUMP.
HEAT E X C H A N G E R -
NUCLEAR INSTRU
M E N T THIMBLE
PRIMARY TANK-
REACTOR VESSEL-
-BLAST SHIELD
-AIR BAFFLE
loads are sufficient to cause plastic yielding (outer tank strains ~ 5 %). For
the core barrel the loading is quite different, since this structure is first
pressurised internally, which causes plastic deformation, and then
externally, which possibly gives rise to buckling.
For most reactor designs and for moderate HCDAs there then follows an
interlude when the core bubble pressure has dropped considerably,
pressure waves have been reduced and the general level of fluid pressures is
insufficient to cause further plastic deformation of the vessel. The core
expansion work has been partitioned roughly equally into structural
deformation and kinetic energy of the fluid, and the coolant is moving
vertically both inside and outside the core barrel.
After the coolant free surface has risen a distance equal to approximately
15% of the total height of the reactor vessel (by which time the core bubble
has expanded to touch the core barreland almost fills the core barrel itself).
it impacts the vessel roof, thrusting the cover gas into a narrow annulus
where roof and vessel meet, and subjecting both to a loading with a short
rise time. The vessel deforms radially, mostly at its upper end, and pressure
waves propagate both downwards into the coolant from the roof region and
upwards from the bottom of the tank as the upper portion of the tank
contracts longitudinally during its radial expansion.
When these secondary pressure waves have become so reduced as to
cause no further deformation, the containment loading phase of the
accident is over and the deformed reactor, which is still pressurised by the
hot core bubble and the containment capability of which has possibly been
compromised, is handed on to the next phases of analysis, that of the
assessment of heat removal capabilities and of radiological consequences.
across the core bubble and uniform pressure within it. It is also customary
to assume a uniform pressure within the cover gas space (significant
pressures exist there only when the cover gas has been compressed into a
small annular volume), and under these assumptions the fluid flow is purely
single-phase with time-varying boundaries and time-dependent pressure
boundary conditions.
Because of the large scale of the system, surface tension and viscosity
effects are negligible, and significant turbulence is confined to regions small
enough for its effects to be modelled adequately by friction coefficientsfor
example, for flow through perforated structures. It appears from recent
work 2 that at least for dip plates pressure drop formulae suitable for steady
flow can also be applied to transient situations with the time scales existing
in containment loading problems.
For a significant proportion of the evolution of the accident the
compressibility of the fluid is important, as witness the account of the
loading process in the preceding section in terms of wave propagation. The
fluid also undergoes displacements of the same order as the largest length
scale in the system, culminating in roof impact.
Structures
The structural elements of the system arealmost all of the same material,
austenitic stainless steel. There is weld material at the seams of the outer
vessel and elsewhere, and irradiation-embrittled steel in the neighbourhood
of the core.
Deformations are, in general, elastic-plastic with strain-hardening and
significant strain rate effects. The principal structures are roughly classified
below:
(1) Axisymmetric thin shells. These include the outer tank and core
barrel. Junctions between thin shells are found where the core
support skirt joins the outer tank in some designs.
(2) Strong structures through which no flow takes place. This class has
as chief' representative the roof and its hold-down system.
(3) Strong structures through which there is flow. This class includes
the diagrid and the dip plate, and the heat exchangers and pumps in
pool designs.
(4) Weak structures through which no flow takes place. Examples are
heat shields and insulation.
(5) Weak structures through which there is flow. This includes most of
the remaining structures, such as breeder zones, shield rod arrays
and the above-core structures.
198 P. FASOLI-STELLA AND A. V. JONES
AXIS OF ROTATIONAL
SYMMETRY
HOLD DOWN SYSTEM
COOLANT
ABOVE-CORE
STRUCTURE
(flow resistance) I
.OUTER VESSEL (shell)
J
CORE BARREL (shell j
than one or two heroic calculations. There are good reasons to simplify still
further by ignoring the inertial constraints near the core, treating the
diagrid as rigid, and omitting or drastically simplifying the flow resistances
above the core. Such simplifications not only aid understanding, but also
allow the validation of calculational techniques and of the codes in which
they are embedded. Since the predictions of containment codes play an
important role in the licensing procedure for commercial LMFBRs. such
validation is taken seriously by the reactor safety community. In one such
validation programme, COVA , 7 60 model experiments are being
performed, beginning with extremely simplified geometries and pro
gressively adding more and more features of real reactors, each case being
calculated by the codes to be validated. In this way the suitability and range
of validity of individual approximations, either physical or numerical, can
be assessed.
T R A N S I E N T F L U I D S T R U C T U R L PROBLEMS IN RF.A CTOR SA FETY 201
This section presents those aspects of fluid dynamics most relevant to the
containment problem. It begins by discussing the hydrodynamic equations,
their Lagrangian and Eulerian formulations and the effect of compres
sibility on their mathematical character; carries on to survey the numerical
methods available for their solution ; and concludes by reviewing the special
problems posed by physical discontinuities in fluid flow.
II
p\e + + OX; pu, e +
where is the specific mass, u, is the components of'the velocity vector 0,p is
the fluid pressure, e is the internal specific energy, Fis the temperature,
and are the viscosity coefficients, and represents the coefficient of heat
conduction. The tensor notation on the i,J,k indices has been used to
indicate the summation.
In the applications which we are considering, the viscous terms and heat
conduction may be regarded as negligible compared with the other terms in
eqns. (1) and (3), which leads to the simplified Euler equations:
^ 4 ^ = 0 (5)
ct cx,
CpU: C Cp
+ (pu,Uj)= ~ (6)
</
c / u \ c u
P\e + T) + Pl' I e + ~ipu) (7)
C'.Y,
et \ 2 cx,
P=f(e,p) (8)
The NavierStokes and Euler equations are nonlinear partial
differential systems. The nonlinearity is due to the advection terms
'CpU, cpu,uk
etc
ex, CX:
where c is the sound speed and c2 = Cp/Cp. p,. px, u,. ux indicate the local
derivatives of and u.
In order to recast this system in the form of eqn. (10), we have to define a
matrix A (u, p):
It f>
A = (12)
c2/p It
(u )/2 + p/J = 0
c22/
/?.322 +, ,..
(u .)/.
, , .1 _ r
=0 (13)
= (u + c )
, <14>
= ( c)'/j
By differentiating u and p along these directions the equations in the so
called characteristic form are found:
U. + C/P Px = 0)
, c/p pp = 0J
204 P. FASOLI-STELLA AND A V. JONES
These equations express the fact that the characteristic curves in the .v, /
plane represent motions of possible disturbances whose velocities are
dx
= u+c
1 >
The characteristic curves play a fundamental role in defining the
properties of one-dimensional flow, such as uniqueness of the solution,
propagation of discontinuities and separation between regions of different
type. They can also be used directly to calculate the propagation of
disturbances in one-dimensional compressible systems. 10
In two-dimensional problems the characteristics are surfaces, and the
theory becomes very complicated, although it is possible to define the bi-
characteristic curves.
The time-dependent Euler equations for compressible fluids are then of
hyperbolic type, whereas it has to be noted that the Euler equations for
incompressible cases exhibit a hybrid elliptic-parabolic character, which
gives rise to quite different numerical problems.
In general, because of the non-linearity of F, non-continuous solutions of
system (9) will exist for all time, even for perfectly reasonable initial and
boundary conditions. Experience indicates the occurrence of shock waves
in such cases and the corresponding solutions of the Navier-Stokes
equations for vanishing viscosity and heat conduction exhibit regions of
very high gradients, tending to local discontinuity. Mathematically
satisfactory descriptions of such discontinuous flows are obtained by means
of the concept of generalised solutions introduced by L a x . 1 1 1 2
The vector function w is called a generalised solution of eqn. (9) if it
satisfies the integral form of this set of conservation laws:
d
u, d V + F,n,dS = 0 (17)
di
Ro
R D
l_ i. _L.
. "2
R
Vi
1 \
\
L_
t= 0 t.T
cl
+ -V (18)
d7 cl
206 P. FASOLI-STELLA AND A V. JONES
p-=-V, (20)
de
p-r= -/>V-0 (21)
dt
To solve the problem numerically, the region R (in our case the plane /, z) is
discretised into a rectangular mesh of cells with sides parallel to the
coordinate axes or into a set of arbitrarily shaped elements. All variables of
the fluid are calculated for subsequent times at a discrete set of mesh points.
In the Lagrangian form the mesh points follow the particle paths of the
fluid elements. The independent space variables refer to a coordinate
system fixed in the fluid (e.g. the masses or initial positions of the fluid
elements) and the mesh points (or nodes) move in a way determined by the
approximate solution of hydrodynamic equations. The Lagrangian point
of view has frequently been preferred, especially in one-dimensional
problems, because: (1) this form of equation is simpler; (2) the positions of
boundaries and of material interfaces are automatically determined in the
calculation, since we follow each bit of fluid: (3) during the calculation the
mesh following the fluid flow becomes finer in the region of greater density.
Difficulties with this approach arise in multi-dimensional problems
when, because of sliding or shear of different parts of fluid, the mesh
becomes highly distorted (Fig. 5). In this case the calculation becomes very
inaccurate and often must be terminated prematurely.
In the Eulerian formulation the independent space variables refer to a
coordinate system fixed in space, so that the fluid streams through the cells.
3 6 9 12
2 5 8 11
1 t. 7 10
2.3. Compressibility
The hydrodynamic equations are applicable to a very broad range of
fluids. In the safety calculations which we are considering the coolant must
be treated as compressible during the wave propagation phase, but it can be
considered as quasi-incompressible in the second stage of the flow before
roof impact.
In the compressible case the situation is described by the Euler equations
and the specific energy of the fluid consists of two components:
E = e + \u2 (22)
14
namely the specific internal energy and the specific kinetic energy. In
compressible fluids an interchange can occur between the internal energy
and the kinetic energy, so that in some region of the fluid most of the local
energy can be internal energy, whereas elsewhere it is the energy of flow
which is dominant. On the other hand, in many cases of interest the energy
of flow is small: the flow can then not compress the fluid significantly, no
appreciable interchange of energy takes place and we say that the fluid
behaves incompressibly.
We can consider the comparison between compressible and incom-
pressible fluids from a different point of view. In the absence of viscosity and
heat conduction the hydrodynamic equations are hyperbolic and contain
two characteristic speeds, the particle's velocity, 0", and the sound speed, c.
We can distinguish between classes of problems, depending on the ratio (the
Mach number) of the flow speed to the sound speed
M = M/c (23)
208 P. FASOLI-STELLA AND A V. JONES
dw
= Lw (26)
d/
where w is the column vector w = (p, pu pe) and L is a spatial differential
operator. FD methods replace the spatial operator L by some difference
operator L* operating on w, where w is defined only on a discrete mesh.
Hence, the righthand side of (eqn. 26) is replaced by some time average
"+'=<;" (28)
|c"+1l<|c"l (29)
\gQ"\ < Ic"l => (30)
|g|<I (31)
For any difference scheme condition (31 ) must be satisfied if the scheme is to
be stable.
Unfortunately, the investigation of such properties of FD schemes is
possible only when the differential equations are linear except for some
special cases. Possible tools of analysis are linearisation and heuristic
procedures. 16
If in eqn. (27) is zero, the method is explicit, since iv" +1 is defined
explicitly by w". In this case the computation of w"+i is very simple for
compressible fluids, but the magnitude of the time step / is limited by the
Courant-Lewy-Friedrichs 1 7 stability criterion. If c is non-zero, the method
is implicit and system (27) leads to
5+iu+i = -p u (32)
Lagrangian Methods
In Lagrangian coordinates the compressible hydrodynamic equations
assume a particularly simple form, which allows one to overcome the
difficult problem of the treatment of advective terms (19)(21). For this
reason, at least in one-dimensional problems, the Lagrangian formulation
may be regarded as the basic approach. In multi-dimensional problems,
however, shear flow in addition to compressible motion is to be described : a
Lagrangian mesh can be difficult to define and can quickly become
unacceptably distorted. Nevertheless, several quite successful attempts
have been made to extend the application of the Lagrangian formulation to
two-dimensional cases also.
In FD two-dimensional axisymmetric codes using quadrilateral meshes
(Fig. 6) the only essential difference between the various schemes lies in the
TRANSIENT FLUIDSTRUCTURE PROBLEMS IN REA CTOR SA FETY 21 1
3
FIG. 6. Lagrangian difference mesh.
P_
-=[tl Pi)(=2 4) (Pi P 4 X 1 J ]
Cr
2, (33)
Cp_
C
-^-[(Pi P3)(r2 ''4) (Pi ' - U)]
2,
or as a function of the fluid variables at points 5, 6, 7 and 8:
1
-2-2[^ -*- --)- - ( ~)(5 - -")]
Cr
(34)
Cp _ (
C - ' -Pi)Oc- ' 8 ) - -(Pe- -Pa)(r5 - ''s)]
The new velocities give the nodal positions, which, in turn, give the new
volumes of the cells and their densities. At this stage it is possible to solve the
internal energy equation and to update the pressures by means of the
equation of state. The time loop is then complete and may be repeated foi
further time steps. The Courant condition imposed on the time steps tc
ensure stability of the Lagrangian mesh is:
/ < min (AlI c) (36)
where A l is the smallest dimension of a cell, c is the sound speed in that eel'
and the minimum is taken over all cells.
As mentioned previously, a severe limitation of Lagrangian schemes is
mesh distortion, which often produces growing errors in the results after
certain time. Many attempts have been made to apply remapping 01
rezoning procedures with the aim of reestablishing a regular mesh during
the calculation. 19 We shall return to this topic later.
Eulerian Methods
The most general FD approaches for the solution of multidimensiona
problems in compressible fluid dynamics use Eulerian meshes. Owing to thc
difficulty in treating advective terms appearing in the Eulerian formulatior
(l)(3) and to the great variety of features displayed by fluid dynamic:
problems, many different algorithms have been proposed and applied. Th<
conservative form (eqn. 9), is generally preferred because the conservatior
laws of mass, momentum and energy are then automatically satisfied
Under particular conditions such as exist in high Mach number flow, the
use of a nonconservative equation for the internal energy can be
advantageous in limiting inaccuracies associated with large kinetic
energies. In several codes both formulations for the energy equation art
provided.
In Eulerian schemes pressures, densities and energies are always definec
at the centre of the cells, but the velocity components may be defined at the
cell corners, in the middle of the cell sides or at the centre of the cells.
The most significant feature of any Eulerian scheme is the approximatior
adopted for the advective terms. If the conservative vector form (9) is
adopted, the divergence operator on the righthand side is replaced by the
sum of fluxes through each cell surface: 14
2
+l
w" = \w[ > F^A t/A V (38)
j=l i=l
13
Expanding in Taylor series, as in Hirfs stability analysis. Lax" s scheme
is seen to introduce an effective diffusion coefficient which can give rise to
inaccuracies in the results. In many cases second or higherorder accuracy
schemes must be used for the solution. Necessarily, in any secondorder
accurate method two storage locations are required for each space point on
the mesh. These may be employed to store two adjacent time levels in the
calculation (twostep procedures) and the time derivative may be time
centred. In the 'leapfrog" scheme (Fig. 7)
1
+ 1
step (1) iv" ' = " \ F"Aa t/A V
r=J
IN
> (39)
step (2) 2
u " ++ 2 = ,,_y^ +Fl,A
= uu " ) +[
i/
A V
1=1
FN
FW w C E FE
S
,
FS
1> ii
1
!
EVEN TIME (main
mesh)
X ODD TIME (auxiliary mesh)
P
^~TLl =
"" ;>/,,> 2, (PO? + + ( l .,2,/ i + ( .,2,]
+ T:[(/'":)"H,2|W1', 1 2,]
Ar
V')^ 1
')MI> 2> (F:)" J + ,2,1 (41)
(pur)i:l\2).j(p^i"+o2,J_ ..
At A r(P'J Pi+iJ
'
. V)
,. (P'.j - P'+ l.j) + '+(1 DJ
(P.-)"jf'|l 2, - (.-)",-+11,2 _ ,- _ - .
Al ~ Az{P'-> PiJ+)
where R"+l 2)J and 5" + ( , 2) contain nonadvanced time factors; the
weighting functions and 0 with magnitudes between 0 and 1 denote the
relative levels of timecentering of corresponding terms; and is a hybrid
function, approximating p"+ ' and defined by
j + 1/2
i+1/2
j -1/2
FIG. 8
where c" = (Cp/Cp)'' is closely related to the square of the speed of sound.
Eliminating by algebraic substitution (pMr)"+(', 2 ) J , (pwz)"J+,i 2) in equation
(41) and substituting p"* '' in eqn. (41), by means of eqn. (42), an implicit
equation in pu is obtained:
Pi.] = A
iPil.j + A 2pi.j 1 + iP+ \.j + A A P.J* 1 + , (44)
1 ,( ! 1
BLJ = + 202 [ + -
c,j \Ar- _-
Equation (44) becomes the usual approximation to the elliptic Poisson
equation c,; -> co. In some circumstances a direct solution method can be
used; for others it may be necessary to employ a relaxation technique. The
determination of for every cell allows the calculation of p"+ [ by means of
eqn. (42) and the calculation of new values of veloci ties from eqn. (41). The
energy equation can then be solved explicitly.
In the ICE method there are, of course, restrictions upon the time step
(although they are less onerous than the Courant condition), the chief of
which is the so-called Euler condition At < min(A.v/|w|).
The reactor containment codes SEURBNUK-2 2 8 and ICECO 2 9 employ
this technique.
the characteristic form and the discretisation is made along these lines (Fig.
9). This method becomes very cumbersome in multidimensional cases and
applications to twodimensional situations are few. 10
Semianalytical techniques are used in particular when properties of the
flow allow simplification of the original system of partial derivative
equations (Fourier analysis, integral equations, methods of lines, etc.).
' (45)
where i/'0' is the approximation of the true solution, the 'shape functions'
N{kci are linear or bilinear functions ofthe spatial coordinates in the element
(e) and u.c) are the values of u at the element nodes.
Under this hypothesis the pressure, density and internal energy are
218 P. FASOLI-STELLA AND A. V. JONES
|
U
2
procedures: for example, the FD code YA QUP ,:> employs a nodal force
technique as in FE, whereas some FE programs have incorporated /,/
systems to number fluid elements.
In conclusion, FE fluid dynamics compressible codes give results, having,
in general, the same degree of accuracy as FD codes. Major programming
effort is required but iscompensated by advantages in the treatment of fluid
structure interactions. We shall come back to this point later.
without degrading the results over the remainder ofthe flow field. One of
the most common forms of the artificial viscosity is
(rtQpA(VQ)2 ifV-fl<0
= (46)
^ J0 ifV-B>0
This term is to be added to the hydrostatic pressure ofthe cell; A is the cell
area; aQ is a constant whose value controls the spread ofthe shock and is
determined empirically (a standard value is 2-56).
Frequently, and particularly in the treatment of solids and liquids, a so-
called linear pseudoviscosity is required: 33
3.1. Structures
In view of the remarks concerning the customary simplifications of the
fast reactor containment problem in the first section of this chapter, we shall
specialise immediately to structures which are figures of revolution about
the vertical axis of symmetry of the reactor. For thick structures such as the
diagrid, one may make use of the fact that their displacements are always
small, and calculate them by Lagrangian techniques, either finite
difference33 or finite element. 36 The equations to be solved are similar to
eqn. (33) but with pressure replaced by the full stress tensor and more
complex equations of state including elastic-plastic behaviour. 37
Whichever type of modelling is employed, since motion parallel to, rather
than through, the structure is small, the use of slide-line techniques is not
necessary.
222 P. FA SOLISTELLA A ND . V. JONES
The other important structures which are figures of revolution are the
inner and outer vessels or, in different terminology, the core barrel and the
primary envelope. Both these structures may be considered as thin shells for
analytical purposes, since in all commercial reactor designs and in most
model experiments the ratio of thickness to radius curvature is
approximately 1:50. We may to the approximation employed here consider
them to be symmetrically loaded, and the chief theoretical complications
arise from the fact that they are not, in general, of the same thickness
throughout, that they may suffer quite large elasticplastic deformation
under popular loading assumptions, with strains up to 10%, and that
considerable fluid flow tangential to the shells takes place, necessitating the
numerical treatment of sliding (for which see later in this section, and also
Section 4) in Lagrangian coordinates and of irregular cells with moving
walls in Eulerian coordinates.
In the early stages of development of containment codes it was customary
to model the reactor vessels as continua, subdividing the vessel thickness
into about five Lagrangian cells. This computational device was the
simplest possible, allowed considerable freedom in modelling the
elasticplastic behaviour ofthe vessel material, and could easily be inserted
into the Lagrangian containment codes containing slideline logic, which
then represented the state of the art. Its sole disadvantage was the
consequent very low value of the stability time step for these explicit codes,
set by the Courant condition in the shell material:
At < Ah (48)
presented in this section will be brief, and will serve mostly to define the
notation for the account of fluid-structure coupling which follows.
Taking the usual path of development, 3 K J 9 we assume that the thickness
ofthe shell is small compared with its principal radii of curvature, that shell
deformations are small (again compared with these radii), that stresses
normal to the thin dimension are negligible and that material lines originally
normal to the shell reference surface (usually the shell mid-surface) remain
normal to it and unstrained. This last assumption is sometimes called the
hairbrush hypothesis. Finally, in accordance with our assumptions about
axisymmetric loading, we assume that the shell is originally a figure of
revolution and retains axial symmetry during deformation. A useful
reference to the somewhat simplified theory which obtains in this case is
reference 40.
Under these assumptions we may write the equations for the dynamic
equilibrium of an infinitesimal element ofthe shell. Figure 11 shows such an
element, and the various forces and couples acting upon it. The shell
224 P. FASOLI-STELLA AND A. V. JONES
(rN^de^dsde^rNf,)
--daf, (, - + ^
rM+de + d s d 8 ^ ( r M ^ )
Pj;rd9ds
(rQ)d8+ (rQ)dsde
ds
Ng d 8 d s
P 0 rdSds
Mgdds cos
we may derive expressions for the resultants ofthe forces on the element per
unit facial area, P and P,\
Pn = N^+NuXl)--(rQ) (51)
v
r d.v
1d cos d)
, = -^(')-0~^- + (52)
where Pn is the stress in the direction and P, that in the s direction (normal
and tangential stresses).
If P0 and P, are the fluid pressures outside and inside the shell,
respectively, and we assume that both act normal to the shell (i.e. that
viscous drag on the shell is negligible), we may write down the equations of
motion of the shell element:
hPsan = P0 - , + P n ; A p A = P, (53)
where p5 is the local density ofthe shell material and an, a, are the normal
and tangential accelerations. In the global r-z coordinate system we then
have
d2r d2z
-J-T = -a n sin<p + a, cos ; - = an cos a. sin (54)
d/ - dt
for the radial and axial accelerations of the element.
To close the system of shell equations it is necessary to express
N0, , Mu, in terms ofthe shell stressesand to relate the shell stressesto
the deformations by some constitutive law. If a coordinate is introduced
which is zero in the shell midplane and which increases in the direction of ,
and if the meridional and hoop stresses at any point in the shell (s, ) are
(5, ) and a0(s, ), respectively, then the forces and couples on an element
are
Ch 2
KW = aa(s, )
h, 2 = 0,} (55)
M.(s) = (., ) d i
- ft/2
The strains are assumed to vary linearly through the thickness of the
226 P. FASOLI-STELLA AND . V. JONES
=^- = , (59)
The most popular yield function in containment codes is that ofthe von
Mises criterion:
= 2 + 2 - 0 - 2 < 0 (60)
in which is the yield stress, which is combined with the Reuss flow rule to
obtain the (elastic)' stresses 0, at any time.
For plastic-hardening materials (and the stainless steel of containment,
vessels is plastic-hardening) several different models have been proposed,
including isotropic hardening, kinematic hardening4-"' and the mechanical
sublayer model. 41 This last model, 42 which is now employed in several FD
and FE structural codes. 4 3 , 4 4 has generally been preferred for containment
calculations because of its relative success in simulating material behaviour
in load reversal situations.
TRANSIENT FLUIDSTRUCTURE PROBLEMS IN REA CTOR SA FETY 227
/fluid 1
shell / fluid 2
j+i -.
P
j -___^ /
/ / /
j-i J M /W"~~~
/
i-1
+1
fc.
FIG. 13. Discretisation of shell in REXCO-H.
shells are 'almost vertical' or 'almost horizontal', and breaks down for
curved vessels such as the outer tank in Fig. 3. It is, however, adequate for
the analysis ofthe simpler containment experiments and is worth describing
both on this account and to give the flavour of more complex treatments.
Figure 13 shows the discretisation and physical variable allocation for a
portion of a shell in contact with fluid on both sides. Note that shell nodes
and fluid nodes are aligned, and that the shell is 'almost vertical'. The
equations of motion for the shell nodes (ij) are somewhat analogous to eqn.
(34). The radial acceleration is given by
M
i/ij= -(Pij-,-- +Pi)(-lJ+l ii 1 + Zf-1- '+ \j)
*-p*-i /X/^M--^
(N^^fN^-i
i
- (rosing.,, 2>][rk(sk+ sk_l)/2]~ (ha0/r\
(67)
(P:AzrA0)k = (/A ^sin) k + ( , / 2 ) )_( 2)- [0'cos(/;) t + ( , 2|
1
-(rcos</.) t _ ( W 2 ) ][r k (5 k + 1 -sk^)/2]- (68)
230 P. FASOLI-STELLA AND . V. JONES
(where sk is the total length ofall the segments from some reference circle to
node k) which are related to the normal and tangential components of eqns.
(51) and (52) by
Pn = P. cos Pr sin
(69)
, = P. sin + F r c o s
For the calculation of the forces , Nu, Q at the half-points one may
calculate stresses at all the points, both crosses and dots. This ensures that
() + (, , takes account of changes in the length ofthe segment, but entails
the calculation of stresses at points where they are not really required in
order to ensure spatial centring of the difference equations. Warham 5 0 has
obtained good results with a method in which the + (1 2) required by
Witmer's method are replaced by ak values. The loss of centring is
compensated by a halving of the effort expended in calculating stresses, and
numerical results differ little from those obtained with the fully centred
scheme. With this modification the difference equations become
l
- (cos - )k + ( , 2, [(hM0)k + (hMu)k + , ]/: (74)
The primes indicate that '. Q' are non-centred quantities.
This completes a rather detailed description of a common FD
formulation of the thin shell equation as used in such containment codes as
REXCO-HEP, 4 7 ASTARTE 5 1 and SEURBNUK. 2 8 In discussing FE
formulations we shall be brief, since they will be more familiar to those
working in the field of structural mechanics.
Finite element discretisations of shell structures are usually based on
TRANSIENT FLUIDSTRUCTURE PROBLEMS IN REA CTOR SA FETY 231
15). In the simplest treatment the total fluid forces acting on a shell segment
AB are determined by forminga vector sum ofthe products of pressure and
directed area for the intersections of AB with fluid cells, in this case with
cells XY W and YZV. The shell internal forces are calculated from the shell
stresses and the combined forces used to advance the shell in space (eqns.
54) by one complete fluid time step, substepping the shell motion if this is
considered desirable. Ina purely explicit fluid calculation, the shell internal
SLAVE MASTER
^y s l i d e l i n e
. s h e l l nodes
internal f l u i d
nodes
Interface f l u i d nodes
FIG. 15. Slide line treatment of fluid shell coupling in Lagrangian codes.
pressures together with the pressures in surrounding cells are used to move
the fluid nodes by one complete time step, a vector sum again being used to
obtain the total force on a cell face. In the case of cell Y VZ this sum includes
contributions from shell segments AB and BC. At the end ofthe fluid time
step nodes which were originally on the shell surface will no longer be so in
general and they are moved so as to bring them into contact with it again in
a direction normal to the shell surface. The fluid pressures may now be
calculated from the new cell volumes and the whole process repeated.
A coupling mechanism such as that described above is conceptually
simple, allows fully Lagrangian treatment of the shell and ensures
continuity of normal velocities in the long term, since nodes originally in
contact with the shell remain in contact. It is, however, rather complex from
the point of view of computer logic, and the arbitrary displacement of fluid
nodes to lie on the slide line at the interface produces perturbations which
may show up as pressure oscillations at the shell (see example of Section 4).
Some more gradual matching of shell and fluid normal velocities is clearly
desirable, but it has proved difficult to achieve a satisfactory scheme from
this point of view without an explosion in logic requirements.
TRANSIENT FLUID-STRUCTURE PROBLEMS IN REACTOR SAFETY 235
SHELL
taken to be the timeadvanced values. The mean normal velocity ofthe shell
over the time step includes contributions from these pressures as well as
contributions from the shell internal forces. The two contributions are
separated, and matching of normal velocities at the shell as in the previous
paragraph again leads to a Poisson equation but with the A coefficients
taking account of the acceleration imparted by the fluid to the shell and with
the shell internal forces included in the source term. Substepping is possible
with this technique, the A in this case being again modified to take into
account the change in shell configuration over the fluid time step. 2 8
The logical problems with this technique arise from the fact that in some
cells not all of the required information is available. In some cells with a
Lagrangian boundary only one or two ofthe velocity field components will
be defined, the remaining midside points ofthe Eulerian cell being outside
the fluid. It is therefore necessary to extrapolate from surrounding cells,
taking care in the logic that these do in fact have their full complement of
defined midside velocitiesthat is, that they are not themselves boundary
cells. The logic for such extrapolation procedures can become exceedingly
involved when all possible configurations of shell and fluid are considered,
and this logical complexity forms the basic drawback to Eulerian codes as
applied to fluidstructure problems.
These descriptions of fluidstructure coupling techniques conclude this
section.
Z
H
r
r
TABLE I
7
y.
H
Code Fluid Flow type Discretisation Slide line Rezoning Structure Thin shell 7:
mesh type continuum treatment C
equations H
r
REXCOEUR Lagrangian compress. II) limited no no yes 0
m
ASTARTE Lagrangian compress. ID yes no yes yes -i
C
ARES Lagrangian compress. FD yes yes yes yes B)
K)
U)
238 P. FASOLI-STELLA AND A. V. JONES
1 ) Method of e q u a l areas
x 0 is moved to x 0 0 in o r d e r to make t h e
areas of A as equal as possible
2 ) C e n t r o i d of c e n t r o i d s method
Rezoning based on v e l o c i t i e s
important point is that the parameters ofthe slave material are associated
only with the slave material and that material providing the fixed boundary
(master) is treated as if the boundary were a moving surface on which
pressure forces act.
Slide lines can be used in the fluid when large transverse motion along / or
j lines is expected at predetermined positions. In such cases it is also
possible, in principle, to use a different mesh size on the two sides.
1 SLave ine
Master line
1 o Master points
1 Stave points
1
1
I
1
I
I
1
FIG. 19. Slide line.
Unfortunately, experience shows that any mesh distortion tending to
'wrinkle'a slide line blocks all tangential motion at the line and so defeats its
object.
In the Lagrangian codes which we are considering, the shock
discontinuities are treated by means of a combination of quadratic and
linear pseudoviscosities (eqns. 45 and 46). At least in the first phase of our
containment studies, these mechanisms are found to give a good
representation of the wave propagation.
Figures 20 and 21 present specimen results of a calculation of COVA
experiment IT9 made with the code ASTARTE. Figure 20 shows the
Lagrangian mesh just after roof impact. The vessel deformation is clearly
distinguishable, and incipient gross mesh distortions may be seen in the
vicinity of the charge. Figure 21 displays experimental and calculated
pressure and impulses for an Eulerian wall station at the point marked A in
Figure 20. It is clear that, while the perturbing effects ofthe fluid-structure
coupling mechanism mentioned in Section 3 have produced very spiky
pressure predictions which are difficult to compare with experiment, the
predicted impulse shows quite good agreement, especially at later times.
Figures 22 and 23 show the mesh configuration of time = 5 ms without
rezoning and with rezoning (centroid of centroids method) in the COVA
TRANSIENT FLUID-STRUCTURE PROBLEMS IN REACTOR SAFETY 241
IT5 calculations (rigid vessel, LDE charge). This result has been obtained
with the code ARES. As it is possible to see, the application of rezoning has
had a limited effect on the mesh distortions. In the rezoned mesh the
tangling of mesh lines has disappeared but at the same time a certain non-
orthogonality has spread over most of the mesh.
The limitation of Lagrangian codes in the description ofthe second phase
of the containment problems (inertial phase and roof impact), especially
when internal structures are present, has necessitated the development of
more powerful and sophisticated codes, such as pure Eulerian or mixed FD
or FE codes.
At present SEURBNUK-2, developed jointly by UKAEA and JRC, is
one of the most advanced codes in the field of reactor containment.
SEURBNUK-2 is an Eulerian two-dimensional axisymmetric FD code,
based on the ICE technique for the calculation of the (compressible)
coolant motion. Some configurations which the code can tackle are
IJ
4-
I J
2.40r 1.20
Pressure Impulse
(W7 Pai (lO^Pa.s)
2.00 100
1.60 _ EXPERIMENT_ /
0.60
/ ^ JUSTARTE r'
*~
1.20 0.60
ASTANTE >
>
V.
0 60 -
0.20 c
SEXP
\
0.00
2 4 6
L Ll-l-.-
9 10
0.00
12 '
time (ms)
Fie;. 21. ASTARTE predictions and experimental results for pressure and impulse
TRANSIENT FLUID -STRUCTURE PROBLEMS IN REACTOR SAFETY 243
=jrffi
^ ^ - - ^
,
>\^-^
||j|
jky
-~~~
-
j
...
PIG. 22. ARES mesh configuration FIG. 23. ARES mesh configuration
without rezoning. with rezoning.
presented in Fig. 24. The internal and external structures are treated in
Lagrangian coordinates by FD and approximated by thin shell theory (see
Section 3) Perforated plates are treated as porous rigid structures.
SEURBNUK-2 is a development of the incompressible Eulerian code
SURBOUM, which was developed jointly by Belgonucleaire and
UKAEA 5 9 and is based on the MAC 2 7 technique.
244 P. FASOLI-STELLA AND . V. JONES
i gid diagrid
Cover gas
Thin tank-.
\ \
) ) \
A\V \\
)
c) di e) f)
) :>
A ^
h) i)
J ji
isl- S3 3 DE3
5
3 KV
J \
s
)
3
)
)
k) l) mi
FIG. 24. SEURBNUK-2 configurations.
TRANSIENT FLUID-STRUCTURE PROBLEMS IN REACTOR SAFETY 245
I''jH
I I ' I I I | l I Internal c e l l s
B
"j^xW | I lx I Boundary cells
"Ht"
I i | iBj
^__1J_L_LJ
FIG. 25. Cell labelling in Eulerian codes.
the coolant region (Fig. 25) and the bubble-coolant interface represents an
external boundary on which the gas bubble exerts a pressure force. The
pressure values in the boundary cells are obtained by means of
interpolations, and the uniform pressure of the gas bubble is a function
(through the equation of state) of the bubble volume.
The ICE scheme introduces an implicit diffusive mechanism: in the
containment problems of the COVA series we have not needed to introduce
an explicit quadratic or linear pseudoviscosity.
In Figures 26 and 27 are presented some results of an IT9 COVA
calculation with SEURBNUK for comparison. Figure 26 shows the
pressure and velocity distributions calculated for experiment IT9 well after
roof impact (7 ms). The outward motion of the deformed vessel and the
pressurised regions close to the roof and close to the curved bottom of the
vessel are clearly indicated. Figure 27 presents calculated and experimental
pressures and impulses for the same gauge station as for Fig. 21 (with which
this figure may be compared). The damping and smoothing effect of the
numerical diffusion present in the SEURBNUK code may be seen in the
pressure records, while good agreement is observed for the overall impulse.
246 P. FASOL I-STEL L A AND . V. JONES
"J
..::::::: _,. i -
I"t ; : : : : ; : : :
:" " : _--,
___.---.
* -H
^^-=^
;::::::: 2::^::::
::
: :: \\
\ I
:::::::-i
: 11 inti
1
\
:::::::;:
--- - -
" -
r tt
- r I
i -
S :r I
::
.Ij . . . . . . --
5
1I
^::
:_._: /
I
-
S _
/
. : ; : : ; ; : ; :
~< I
I
..c:-: _ :
"s::;;::
s : " ;
; / J
f /
"^1 /
S;:::::::J-
ff -ax/s
FIG. 26. SEURBNUK: pressure and velocity distributions.
EURDYN-1M 2 6 is an FE mixed coordinates fluid-structural code under
development at JRC for the treatment of both containment and
subassembly dynamics problems. It holds great promise of being able to
overcome many difficulties of Lagrangian and Eulerian schemes, to treat
complicated structures and to simplify the fluid-structure coupling
problem. The numerical schemes employed in this code and the first
significant tests form the subject of the chapter by Dr Doea.
16 - 16
r.
.
7>
12
-
70
C
ra
r
-
S
7=
4 >
H
C
c/:
>
16 g
time (ms)
H
<
FIG. 27. SEURBNUK: pressure and impulse predictions and experimental results.
IJ
t-
248 P. FASOLI-STELLA AND . V. JONES
seek to explain the directions taken by recent research work inspired by the
limitations ofthe various techniques currently available.
As mentioned previously, the codes most frequently used in containment
problems are either axisymmetric Lagrangian or axisymmetric Eulerian.
and each type of code has its own built-in limitations. Lagrangian codes
have proved very suitable for the calculation of the early stages of the core
bubble expansion, before much fluid motion takes place, since the
calculation ofthe propagation and reflection of'shock waves is their forte.
They are also well adapted to the detailed analysis of wave propagation
effects in subsections of the whole containment geometry, such as the
volume enclosed by the diagrid, the underside of the core support skirt and
the curved bottom ofthe outer vessel. On the other hand, when relative
motion is large, their accuracy declines and the calculation time required
increases rapidly because of mesh distortions limiting the time step.
Effective termination of the period of validity of Lagrange codes can
occur distressingly early when the core bubble is quite close to the
perforated diagrid, to give but one example. It is also difficult to insert the
resistance effect of nearly fixed perforated structures into Lagrangian
codes, so that both the direct modelling of perforated structures and their
substitution by distributed resistances come up against code limitations at
an early stage.
Eulerian codes are ideally suited to the calculation of the overall
progression of the fluid flow from initial core expansion to roof impact and
after. Their limitations appear most clearly when flow through complicated
geometries or deformable structures must be calculated. The insertion of
distributed resistances in Eulerian codes is quite straightforward, but
separate experiments may be necessary in order to determine what type of
resistance law should be used. In addition, when coupling with complicated
structures must be modelled, the computer logic required can become
extremely heavy.
In accordance with the practical requirements of cheap computing for
long real times of logically simpler but accurate fluid-structure coupling,
and of modelling of large fluid displacements at all flow speeds from
supersonic to far subsonic (almost incompressible), current efforts are
being devoted to developing more general techniques with both Lagrangian
and Eulerian features. This class of technique has two main representatives
in the shape of hybrid coordinates and mixed coordinates. In hybrid
coordinates the integration field is discretised by a fixed Eulerian mesh, but
selected portions ofthe field may be handled by means of a Lagrangian
mesh which is superimposed on the Eulerian mesh and fully coupled to it.
TRANSIENT FLUID-STRUCTURE PROBLEMS IN REACTOR SAFETY 249
REFERENCES
1. INTRODUCTION
DISPLACEMENT
EXPLICIT FORCES AND STRESS
FIELDS
STRUCTURAL RESPONSE
V
MOTION RESPONSE
INTERFACE
AT
PRESSURES
INTERFACE
I \
FLUID RESPONSE
PRESSURE AND
VELOCITY FIELDS
The above basic choices will be discussed in detail in the next paragraphs.
n
/ . \
/Rit) / P \ N
** 1
/
U
* ^
^,^ /
~ ~ ~ * * ^ * ^ s / ^ - * "
, > < * ) <. C (t) /
J><^. j/^
^* 5
. / **
j< **
j^**" J
s ,^**" ^ /
C(0) = R(0)
C = C ontinuum
1
R = Reference frame
FIG. 2. Kinematics in the ALE description.
(.,,.,, .v3). The necessary and sufficient condition for the inverse functions
(5) to exist is that the Jacobian determinant
J = \cx.\da\ (6)
should not vanish.
Since eqns. (4) and (5) are the inverses of" one another, any physical
property of the continuum that is expressed with respect to a specific
particle (Lagrangian description) may also be expressed with respect to the
particular location in space occupied by the particle (Eulerian description).
For example, if the Lagrangian description of the density is given by
= P(a r) (7)
the Eulerian description is obtained by replacing a, in this equation by the
function given in eqn. (5). Thus, the Eulerian description of the density is
p = p[F,(xt),t] = p*(xt) (8)
where the symbol p* is used to emphasise that the functional form of the
Eulerian description is not necessarily the same as the Lagrangian form.
Both the Lagrangian and the Eulerian approaches present advantages
and drawbacks. A clear delineation of fluidfluid and fluidsolid interfaces
and wellresolved details of the flow are afforded by the Lagrangian
approach, but it is limited by its inability to cope easily with strong
distortions, which often characterise flows of interest. By contrast, in the
Eulerian formulation strong distortions can be handled with relative ease,
but generally at the expense of precise interface definition and resolution of
detail.
Because ofthe shortcomings of purely Lagrangian and purely Eulerian
approaches, techniques have been developed that succeed to a certain
extent in combining the best features of both the Lagrangian and Eulerian
approaches. One such technique is the A rbitrary Lagrangian Eulerian
(ALE) method. A LE treatments have been developed by Noh, 1 Trulio. 2
and Hirt, A msden and Cook 3 in finite difference formats. Dona, Fasoli
Stella and Giuliani 4 and Belytschko and Kennedy 5 have reported finite
element ALE methods. The ALE formulation treats the mesh as a reference
frame which may be moving with an arbitrary velocity w in the laboratory
system. Depending on the value of the velocity w, three basic viewpoints
may be listed:
(1 ) w = : The reference frame is fixed in space and this corresponds to
the Eulerian viewpoint in which the motion is described in terms of
260 J. DOEA
where | indicates that the material coordinates are held constant in taking
the derivative.
If g is given in Eulerian form, its material derivative contains two terms,
since the instantaneous position vector ofthe particle varies with time:
Dg_cg*(x,Q cg*(, t)
+ 'V (12)
Dt t t CX:
where , are the components ofthe particle velocity. The first term in eqn.
(12) is the spatial derivativethat is, with the point fixed in position. The
second term represents the contribution of movement and is called the
convective term.
Finally, if g is given in mixed representation, one may write, according to
eqn. (9),
g**(X. t) = g**(J*(H, t), t) = g(, ) (13)
By analogy with eqn. (12), the material derivative again contains two
terms and is given by
Dg = cg**(X,t) Bg**(C,t) V:
(14)
D/ cl <W
Thus, noting that cCJct = ii,, we may rewrite the material derivative (14) in
the form
D g ^ cg(,/)| g**(X,t)
+ O, 'F) (16)
Dr ~ et ,
262 J. DONA
Expression (16) contains the partial derivatives of g** with respect to the
mixed coordinates ,; these can be related via eqn. (13) to the partial
derivatives of g with respect to the material coordinates a,. To this aim use
must be made of the mixed Jacobian (10) to write
(17)
Substitution of eqns. (17) into eqn. (16) enables one to express the material
derivative in mixed representation in terms of material variables only.
The Jacobian J provides a mathematical link between the current volume
element d V (function of the mixed variables) in the reference frame and the
volume element d F 0 (function of the material coordinates) in the initial
configuration:
dV=JdVQ (18)
From relation ( 17) and making use of the derivation rule of a determinant,
one may easily show that the following identity holds:
~ = Jdivw (19)
ct
By use of this identity, eqn. (15) can be rewritten as
eg eg** , ,. , ., , 1 c J
= ( div (?**w) g**
ct cl J cl
or equivalently as
^ +^ =0 (23)
Ct CXj
8
^ + ^(pv,vj) = pg,-^- ( , = 1,2,3) (24)
c t cj CXj
epe c c
- - + (pvj-e) = pvjgj - (pvj) (25)
Ct CXj CXJ
In these expressions is the density,/) the pressure, the fluid velocity and g
the gravity. The total specific energy, e, is defined by
e = li- 2 + (26)
where /' is the specific internal energy. For compressible fluids, the pressure
is defined by an equation of state of the form = p(p, i).
In an incompressible fluid the pressure is an intrinsic and independent
variable of the motion and is not related to any thermodynamic equation of
state; it is an implicit variable which instantaneously'adjusts itself' in such a
way that the incompressibility constraint, div v = 0, remains satisfied.
Introducing the Eulerian forms (23)-(25) of' the conservation laws into
the right-hand side of eqn. (20), we obtain the local form of these laws ina
moving reference frame:
C C
- (pv~J) + J ~ (pv,(vj - wj)) = J\pg, - J-\ (28)
ct oxj V ex, J
the same symbols are used to represent both the Lagrangian and the mixed
functional forms of the physical quantities. Moreover, the mixed position
vector is written instead of X It should thus be reminded that the symbol
/t indicates a time derivative with the material coordinates being held
fixed. It is easily verified that the particular cases of purely Lagrangian and
purely Eulerian descriptions are contained in the above equations. Taking,
for example, the continuity equation (27), we note that
(1) If w = v, it reduces to
D
TtM = m(pJ) = 0 (30)
where n denotes the outward normal to the surface 5. The integral form of
the conservation laws for mass, momentum and energy are readily obtained
by replacing g** by p, pv and pe, respectively, in eqn. (34). By use of eqns.
(23)-(25) and applying the divergence theorem, the following conservation
statements are obtained:
dM _ d
pdV = C) p(w - v) n d S (mass) (35)
~dT~d7 I IM Mii
dQ d^
p v d F = () p\(\v V)-dS gradpdF pgdF
d7~d7 V(l) Mo F(i) Jl(l)
(momentum) (36)
d_ d
pedV = O pe(vi-\)-dS - O p\dS+\ pv-gdF
d7"d7 no Mn su) Jfin
(total energy) (37)
One may also write an integral expression governing the time rate of change
of the internal energy contained in the volume V(t). Such an expression is
obtained in the form
d/ _ d_
pidV = p/'(w - ) d S pdivvdF
d7~d7 V(i) JS(r) no
(internal energy) (38)
When w = v, eqns. (35)(38) are Lagrangian and the time rate of change,
d/dt, coincides with the material derivative, D/Di. In fact, the material
derivative of C7(/) in eqn. (31) is given by eqn. (34), provided that the surface
S(t) displaces with the particle velocity V. Thus,
DG(t) eg" dV + O g**v dS (39)
Dt } v u ) ct MO
When w = 0, eqns. (35)-(38) are Eulerian and the time rate of" change,
d/dt, now represents the spatial derivative as defined in eqn. (12). In fact,
when w = 0, eqn. (34) reduces to
dGQ) eg**
dt -4 dF (41)
ct
To achieve the discretisation of two-dimensional fluid domains, we shall
employ 4-noded quadrilateral elements with rectilinear sides and bilinear
local velocity field, as well as 3-noded triangular elements with linear
velocity field. The density and the specific internal energy (and consequently
the pressure) will be assumed uniform over each of these elements. It follow s
that direct use can be made ofthe integral forms (35) and (38) of mass and
internal energy conservation to update the element density and specific
internal energy, provided that V(i) represents the current volume of the
element. The situation is quite different as far as conservation of
momentum is concerned.
Since the velocity field in each element depends on six or eight parameters
(the nodal values of the velocity), the integral statement (36) is clearly
unable to furnish enough equations to update the element momentum. This
situation is overcome by formulating a variational statement associated
with the differential form (28) of the momentum equation.
and make use ofthe continuity requirement, eqn. (27). and ofthe property
div (B) = a div B 4- B grad a (44)
TRANSIENT DYNA MIC F L U I D S T R U C T U R E INTERA CTION 267
ep
v,^JdV + O v,(T,S,)dS (46)
VO) CX:
We now integrate by parts the pressure term in eqn. (46) and apply the
divergence theorem; this yields the principle of virtual power:
cv,
, dV ,( ) grad , d F 4 viPg,dV
et VU)
( ;) d V + , , dS (47)
m ( SU)
-
C' .
= ( ) grad dV \\ (52)
l IO et IO VU)
Equation (2.52) is the weak variational form ofthe internal energy equation
which replaces the integral statement in eqn. (38) when higher-order
elements are employed. In practical applications and /will be defined over
each element by
p = 1"JPJ]
(53)
where NjS are appropriate shape functions and the weighting function
would be = )V.
Computational Schemes
As shown in the previous sections, the following conservation laws form
the basis of numerical solutions based on constant pressure elements:
dM d r r
= p d F = O p(w v ) d S (mass) (54)
dt dt , s
d/_ d
pidV = p/(w v) n d S pdivvdF
d7"d7
(internal energy) (55)
d_ d
pedV= O pe(w v j d S OpvdSf pgvdF
dT ~ d7
(total energy) (56)
+ (6,) + O ovTaS
Jr ex, j s
This yields the following matrix system which approximates the momentum
equation:
cv
[M]<^}={F (60)
l\ + \Fb] + >F\ + [F\
( I
Mj = N,pNjdV (61)
Fl N,pgxdV
(63)
n.i = N,pgvdV
Ff= NTdS
(65)
' = o NTdS
Here, again, suitable extrapolations are used for the evaluation of the
transport terms.
An interesting computational scheme based on explicit time integration
and valid for both compressible and incompressible flows can be obtained
272 J. DONA
following the ideas developed by Hirt et al.3 and Stein et al.9 in a finite
difference context. In deriving such a scheme, one notes that the convective
terms in eqns. (54)(57) vanish in the Lagrangian formulationthat is,
when w = v. It is this reduction that leads one to first solve the Lagrangian
integral equations in the first step of the calculational cycle, phase I, and
then to add the convective contributions as a separate step in the last phase,
phase III. The middle step, making up phase II, adds an implicit pressure
calculation that permits solutions to be obtained at all flow speeds.
In describing the above three phases of the calculational cycle, we shall
closely follow Stein et a/.9 and translate into finite element terms their finite
difference formulation.
The tilde superscript indicates an intermediate time for this phase of cycle
+ \, and the superscript /; indicates previous cycle values.
The specific internal energy, i. is advanced to an intermediate time from
the integral equation (55) with w = v:
Ai f" +
p"div dV (73)
M".
In this equation. A /L. is the element mass and a meanvelocity has been used,
as suggested by Pracht,' in order to conserve total energy exactly in the
finite element formulation.
Let the superscript L denote advanced Lagrangian time values, and let
the superscript n denote values at the beginning of a cycle.
The relation for density is the Lagrangian mass conservation relation:
J(PLJ) (75)
and velocity is advanced by
At
\F(p'-)\ (76)
M'
where pL = pL p".
Equations (74)-(76) are solved iteratively by use of a Newton-Raphson-
like scheme.
Following the convergence of the iteration, a final adjustment of the
'Lagrangian phase' internal energy is made by
At
'' + '
dV (77)
M"
zkOjYdV (78)
At O (w -fL)dS (79)
274 J. DONA
lar 1 ) 2 d F (84)
Nprdrdz
M,s=a.,M"; a = (90)
rdrdz
The nodal loads (63) induced by body acceleration, g, are evaluated at the
element level as
(92)
i:Hfc}
where M, is the nodal mass defined by eqn. (90).
Applying the Green-Gauss theorem and neglecting the virtual power due
to the discontinuities in pressure between elements, the nodal loads (64)
induced by the element pressure, p, are found to be, at node /',
FIG. 4. Hourglass type of mesh instability.
modes. A finite element form of' Maenchen and Sack ' 6 rotational antihour
glass viscosity has been proposed by Belytschko and Kennedy. 5 Their work
should be consulted for the details of the formulation.
At the element level the internal nodal forces are given in terms of the
stresses by
Here [B] is the strain-nodal displacement matrix, {} the stress matrix and
F e the volume of the element.
In modelling the structure we use either continuum (triangular) elements
or structural (rectilinear beam, conical shell) elements based on the
Kirchhoff hypothesis. Non-linearities due to large linear and angular
displacements are treated by a corotational technique developed by
Belytschko and Hsieh. 1 ' This technique consists in formulating the
equations of motion in terms ofconvected coordinates that rotate but do
not deform with the elements. Such a procedure linearises the strain-
displacement relationships (provided that the strains can be assumed small)
and simplifies the nodal force-stress relations (97) within the elements'
convected coordinates.
Elastic-plastic relations of the incremental type are used for material
description in connection with the von Mises criterion of yielding. 18
Isotopie and kinematic hardening models are easily implemented. Several
structural materials show a significant dependence ofthe stress-strain curve
on the strain rate. Halleux 19 recently developed a pointwise mathematical
model for treating this effect.
-ti- 3 "~^
^- s C
2
Ir Structural nodes
0 Fluid nodes
S = Structural element
F= Fluid element
w, = v, and v. (98)
where w indicates the grid velocity ofthe fluid and the velocity ofthe solid.
These conditions are imposed directly at each time step and serve, together
TRANSIENT DYNA MIC FLUIDSTRUCTURE INTERA CTION 281
where the tangential loads are obtained by transformation, into the (t,n)
system of the assembled internal nodal forces (F*, P):
F = F"* sin 4 Focosa
F'n = Ff sina + F], cosa (105)
F2 = F,2 sin a + P{2 cos a
The only case where the A LE finite element formulation does not allow
the simple treatment of sliding just described isin the vicinity of a fluid free
282 J. DOEA
4. APPLICATIONS
FIG. 7(b). Hexagonal tube, water pressure in the element in contact with the tube
on horizontal line of symmetry.
FIG. 7(C). Hexagonal tube, longitudinal strain at external surface ofthe tube on
horizontal line of symmetry.
TRANSIENT DYNAMIC FLUID-STRUCTURE INTERACTION 285
- 1
^
1
s
u-iJ-_LL
>m~uL
<CKVM-
>sxSt lJ
FIG. 9. Primary containment model, (a) Deformed mesh at / = 0-5 ms; (b) deformed mesh at/ = 10 ms;(c) deformed mesh at t
= l-5ms; (d) deformed mesh at = 2-0ms; (e) deformed mesh at = 4-0ms.
H
>
S
y
H
O
<
>
2
5d
-
H
C
je
rr
Z
H
rr.
7:
>
H
C
2
FIG. 9contd.
rO
oc
288 J. DOEA
0 00
0 250.50 0751001251 50 1 75 2 00 2 252 502753.003 253 503.754 0 0
3
Time ( s x l O )
FIG. 10(a). Containment model, water pressure in the element in contact with the
shell on axis of symmetry.
FIG. 10b. Containment model, meridional strain at external surface ofthe shell on
axis of symmetry.
ACKNOWLEDGEMENTS
REFERENCES
4. Dona, J.. Fasoli-Stella, P. and G iuliani. S. (1977). Paper Bl/2, 4th SMIRT
Conference, San Francisco. California.
5. Belytschko, T. and Kennedy. J. M. (1978). Nucl. Eng. Design, 49, 17-38.
6. Truesdell, C. and Toupin. R. (1960). 'The classical field theories', in
Encyclopedia of Physics, Vol. III/l, [Springer- Verlag, Berlin.
7. van G oethem, G .. Doctoral Thesis. University of Louvain-la-Neuve. Belgium.
to be submitted.
8. Connor, J. J. and Brebbia. C. A. (1977). Finite Element Techniques For Fluid
Flow, Newnes-Butterworths, London.
9. Stein. L. R., G entry, R. A. and Hirt. C. W. (1977). Comp. Meth. Appi. Mech.
Eng.. I l , 57-74.
10. Pracht. W. E. (1975). J. Comp. Phys.. 17, 132-59.
11. Dona, J.. Fasoli-Stella. P.. G iuliani. S., Halleux. J. P. and Jones. A. V. 'An
Arbitrary Lagrangian Eulerian finite element method for transient dynamic
fluid structure interaction problems', paper submitted for presentation during
SM1RT-5 Conference. Berlin. August 13-17. 1979.
12. Zienkiewicz, O. C. (1971). The Finite Element Method in Engineering Science,
McGraw-Hill. London.
13. Amsdcn. A. A. and Hirt. C. W. (1973). Los Alamos Scientific Laboratory
Report LA-5100.
14. Wilkins, M. L.. Blum. R. E., Cronshagen. E. and Grantham. P. ( 1975). UCRL-
51574 Rev. 1. Lawrence Livermore Laboratory, May 30.
15. von Neumann. J. and Richtmyer. R. D. (1950). J. Appi. Phys.. 21, 232.
16. Maenchen.G .and Sack. S. (1964). In Methods in Computational Physics. Vol. 3
(Ed. B. Alder er al.), p. 181. Academic Press, New York.
17. Belytschko. T. and Hsieh, B. J. (1973). //;/. / . Num. Meth. Eng.. 1, 255-71.
18. Dona. J.. Giuliani. S. and Halleux. J. P. (1976). Nucl. Eng. Design,37,95-\ 14.
19. Albertini, C. and Halleux. J. P. (1978). 'Material behaviour and modelling in
transient dynamic situations". lspra Course on Advanced Structural Dynamics.
October 9-13.
20. Wilkins. M. ( 1964). In Methods in Computational Physics. Vol. 3 (Ed. B. Alder
ci al.), pp.211-63. Academic Press. New York.
21. Cameron. 1. G.. Hoskin, N. E. and Lancefield. M. J. (1977). Paper 2.. 4th
SMIRT Conference, San Francisco. California.
Coupled Fluid-Structural Dynamics in Blowdown
Suppression Systems: Numerical Schemes and
Applications
R. KRIEG
Kernforschungszentrum Karlsruhe GmbH, Institut fr
Reaktorentwicklimg, Karlsruhe. West Germany
1. THE PROBLEM
27 m
13 m
7.5 m
which is similar to that used in this chapter. Basic differences concern the
types of singularities (sources and dipoles, respectively) and their
arrangement (panels). A similar method of solution for fluid-structural
coupling in a blowdown suppression system was proposed by Schweiger
and Mayr. 5 To describe the inertial effect of the fluid, they also use a
singularity method, but modelling ofthe geometry is not carried through in
detail. A large number of contributions to the dynamics of pressure
suppression systems were presented at Sessions B5 and J6 of the 4th
International Conference on Structural Mechanics in Reactor Technology
held in San Francisco in August 1977. There was one paper where another
singularity method had been used. 6 More recently, too, Di Maggio. Bleich
and McCormick have dealt with the fluid-structure coupling aspects of
pressure suppression systems, but under axisymmetric conditions. 7 Class
tackles the same problem. 8 Finally, some preliminary results have been
published by the author. 9 , 1 0 A more detailed paper on the treatment of
coupled problems in fluid-structural dynamics, which also includes
additional references, is under preparation. 11
loads cannot be given explicitly, but are results of fluid dynamics and
depend on the unknown structural displacements, which, in the wetted
region, are identical with boundary displacement of the fluid. In other
words, fluid dynamics provides a feedback of the unknown structural
displacements, s, on the loads, q.
Therefore, in a second step we shall distinguish between those structural
displacements and loads which at the same time describe the fluid dynamic
behaviour in parts ofthe fluid boundary and those structural displacements
and loads which are not directly related to the fluid. This will be done as
follows.
The degrees of freedom normal to the fluid structure interface and
describing those structural parts which are wetted by the fluid may be
numbered separately n = 1,2, Nl. The corresponding displacements
may be the components of the vector s 1 . The corresponding loads per unit
area of the wetted structural parts may be the components of the vector q i .
This is designated structural region 1 (the coupling region).
All other degrees of freedom may be numbered n = 1 , 2 , . . . , '2. The
corresponding displacements may be the components of the vector s 2 . The
corresponding loads may be the components of the vector q 2 . This is
designated structural region 2.
Figure 2 characterises a coupled problem in fluidstructural dynamics,
where the structure consists of a thin shell. In this case vectors s1 and q1
describe the normal displacements and loads of the wetted shell region (Fig.
3a). The vectors s 2 and q 2 describe the tangential displacements and loads in
the wetted region as well as all displacements and loads in the unwetted
region (Fig. 3b).
Withs 1 a n d s 2 as the second time derivatives of s 1 a n d s 2 , the equations of
motion read:
A n s ' + A 12s2 + B n s ' + B 1 2 s 2 = q ' i
A 2 V + A 22s2 + B 2 l s ' + B 2 2 s 2 = q 2 j
Now. A 1 1 , . . . , B 2 2 are submatrices ofthe stiffness matrix. A, and the mass
matrix. B, of the whole structure. The integers Nl and A 72 denote the matrix
dimensions.
1
\Ail\Ai2}\ N]
A= 21 r " 2 2" 2;
A ! A !J/V
I
I L
296
In eqns. (2) the loads q1 are due to the fluid dynamics and depend on the
structural displacements s 1 , which, at the same time, are displacements of
the fluid boundary. The loads q 2 will be assumed to be given.
>
r
<
>
:
r
C
q C
c/i
C
--3
-rr
:/:
v:
C
FIG. 3. (a) Structural degrees of freedom coupled with fluid dynamics, (b) Structural degrees of freedom independent of -
fluid dynamics.
J
298 R. KRIEG
CX = a +
(3)
DX = +
t A dipole can be generated by two sources having the same intensity but opposite
signs, with a distance approaching zero. The direction ofthe distance becomes the
direction of the dipole.
The application of dipole panels provides some advantages in comparison with
source panels widely used in aerodynamics:
(1) A cross dipole panels a pressure step occurs which allows for simulation of
pressure differences across walls.
(2) A t the common edge of two dipole panels with different orientations no
leakage peak occurs.
(3) The coefficient matrices describing the superposition of the flow fields due to
the particular dipole panels have dominant diagonals, which facilitates the
numerical treatment.
As a disadvantage ofthe dipole panel against the source panel, the solution for the
flow field is somewhat more complex.
% Owing to the highly transient flow field with relatively small velocities, the only
essential effect of neglecting the fluid viscosity is vanishing fluid damping of the
system. A ccording to GUer, 15 neglecting the fluid compressibility is tolerable.
FLUID-STRUCTURAL DYNAMICS IN BLOWDOWN SUPPRESSION SYSTEMS 299
requires that the locations of each field point and the given direction in this
point must be described in local coordinate systems attached to each ofthe
panels. Usually these calculations require considerable effort. In more
complex cases a single component of X may describe the intensities of
several dipole panels, thus reducing the number of unknown quantities. On
the other hand, the dipole intensity of one panel may depend on several
components of X, thus reducing pressure discontinuity at the panel edges.
The weights of these components for a particular panel must begiven. The
same weights are necessary to determine the matrices C and D. The
components of the vectors and describe the influence of given fluid
sources, if present.
Now the reduction ofthe three-dimensional flow problem to a boundary
problem is easy. We specify appropriate boundary points as given field
points. The required reference directions in the field points are normal
vectors ofthe boundary. If the number of boundary points is the same as the
number of components of the vector X (in simple cases, the same as the
number of panels), the unknown vector X can be eliminated in eqn. (3),
leaving a linear relationship between the normal accelerations and the
pressures in the boundary points. There are no other unknown quantities of
the fluid field.
In the following, the above procedure will be implemented in detail.
However, analogous to structural dynamics, we will distinguish between
those regions of the fluid boundary which are formed by the flexible
structure and regions with prescribed normal accelerations and pressures,
respectively.
Points representing boundary regions wetting the flexible structure may
be numbered consecutively m = 1, 2 , . . . , M1. The corresponding normal
accelerations may be the components ofthe vector a 1 . The corresponding
pressures may be the components of the vector p 1 . This is designated fluid
boundary region 1 (the coupling region).
Points representing boundary regions with prescribed normal accele
rations may be numbered consecutively m = 1,2,..., M2. The given
normal accelerations may be the components of the vector a 2 . The
corresponding pressures may be the components of the vector p 2 . This is
designated fluid boundary region 2.
Finally, points representing boundary regions with prescribed pressures
may be numbered consecutively m = 1,2, ...,M3. The corresponding
normal accelerations may be the components of the vector a 3 . The given
pressures may be the components of the vector p 3 . This is designated fluid
boundary region 3.
300 R. KRIEG
Boundary Region 3
a3
Boundary Region 2
2
a (given! Boundary Region I
2 1
P a
FIG. 4. Boundary regions ofthe fluid: coupled with shell (1). given acceleration
(2), free fluid surface (3).
A special case of fluid boundary region 2 is rigid walls, where the normal
accelerations a 2 vanish. A special case of fluid boundary region 3 is free fluid
surfaces, where the static pressures are constant. With usually low fluid
velocities in this boundary region, the dynamic pressure can be neglected
and, instead ofthe static pressure, the total pressure p 3 can be assumed to be
constant.
Again, the coupled problem of Fig. 2 is used as an example. The three
different regions ofthe fluid boundary are shown in Fig. 4. Region 1 wets the
flexible shell, region 2 is rigid or the normal acceleration is prescribed, and
region 3 is a free fluid surface.
With the accelerations and pressures introduced above, eqns. (3) split up
into
C'X = a' + y '
D ] X = p ! +'
> (4)
C 2 X = a 2 4-y 2
D3X = p 3 + 3
As in eqn. (3), matrices C 1 , D 1 , C 2 , D 3 describe the influence of the dipole
panels (with intensities represented by X) on the accelerations and pressures
FLUIDSTRUCTURAL DYNA MICS IN BLOWDOWN SUPPRESSION SYSTEMS 301
"
D1 [M 1
F = C2 M2
D3 }/ 3
= F*
M1 = A'1
Provided that the positive normal accelerations of the fluid boundary are
directed inside the fluid region, the equilibrium and compatibility
conditions at the fluid-structural interface read:
p1 = - q 1 and a1 = s 1 (7)
Here, analogous to free fluid surfaces, it has been assumed that the dynamic
pressure at the fluid-structure interface (coupling region) can be neglected
in comparison with characteristic pressures. Consequently, the static
pressure acting on the flexible structure could be replaced by the total
pressure p 1 .
Now in fluid dynamics equation (6) vectors a 1 and p 1 can be replaced by
the structural accelerations s and the loads q'. Then, with the resulting
relation, the unknown loads q1 in the structural dynamics equations (2) can
be eliminated.
FLUIDSTRUCTURAL DYNA MICS IN BLOWDOWN SUPPRESSION SYSTEMS 303
Finally, the following equations of motion for the coupled problem are
obtained:
(C 1 D* 1 A 1 1 )s 1 r(C 1 D* 1 A 1 2 )s 2 4(C 1 D* 1 B u rE)s 1 +(C 1 D*'B 1 2 )s 2 "]
= y 1 4 C ^ D * 1 ^ 1 4 C* 2 (a 2 4 y2) + D* 3 (p 3 + 3)] > (8)
A 2 1 s' 4A 2 2 s 2 + B 2 1 s 1 + B 2 2 s 2 = q 2 J
E is the identity matrix.
Equations (8) are of the same type as the structural dynamics equations
and the matrices governing the coupled problem have the same dimensions
as the stiffness and mass matrices for the structural dynamics. Only the
symmetry of the matrices disappears for coupled problems. Together with
appropriate initial conditions eqns. (8) yield unique solutions for the
structural displacements s1 and s 2 . Now the other unknown quantities
describing the coupled problem may easily be obtained. By use of solutions
s1 and s 2 , the first of eqns. (2) yields the load q1 of the wetted structural
parts. Then, with eqns. (7) and (5). the vector X can be calculated, which
allows for the determination of pressure and accelerations at any fluid point
by means of eqns. (3).
For further treatment of the problem, eqns. (8) may be rewritten as
Gs + Hs = r (9)
with
s", :Y'
N2
.V' Y2
^D^A11 ^D^A12;}^1
G =
21
"A" ""
"""22"""jjyv2
N'
2
I ciD*iBn + E ; c'D^B'^jA"
2 2
A n s l + A 1 2 s 2 + [B 11 -f ( C ' D * 1 ) - 1 ^ 1 + B I 2 s 2 =
Comparison with the first equation in (2) shows that coupling with fluid
dynamics can be described by introduction of the added mass matrix
(^D*1)"1
Added masses are defined only for those degrees of freedom which describe
structural parts in contact with the surrounding fluid. The transfer of
prescribed forces or accelerations from the fluid to the structure as
described by the right-hand sides of eqns. (8) is not included in the added
mass concept.
H ~ ' inverse of H
r = H1G
p = H 'r
r s - F S = p' (10)
FLUIDSTRUCTURAL DYNA MICS IN BLOWDOWN SUPPRESSION SYSTEMS 305
If, on the righthand side, cos ; is replaced by sin it, the solution of eqn.
(13) reads
1
/ ~ .
sin / sin for
(15)
sinjt /cos,,/ for =
From eqns. (14) and (15) it follows that, in general, the solution is aperiodic.
Exceptions are cases where the /,, ratio assumes a rational number. For
excitations with a frequency , which agrees with the eigenfrequency OJ,
there is resonance, i.e. the amplitudes increase monotonically (at least for
long times).
(2) Systems conditions are known for the time T, i.e. a(F)and rjn(7")are
known.
The righthand side p disappears. Then the solution for / > F of eqn.
( 13) is
= A cos 1 + B sin ,,/ ( 16)
with
B
= ( ) sin F + () cosa;,,
C
= 1
() cos OJ ,,() sinojF
With these basic solutions the response of the system due to excitations
occurring only within one time interval can be investigated. Let this time
interval be 0 < / < F a n d represent the righthand side pn(t) by a Fourier
expansion with the frequencies = 2/ , 2(2/ ). 3(2/7),... ; the system
response may then be obtained as follows: During the time interval
0 < / < Fthe solutions (14) and (15) for the different frequencies must be
multiplied by the corresponding Fourier coefficients and then super
imposed. For time / > T. formulae (16) apply with the values () and
(T) due to the response obtained at the end ofthe preceding time interval.
Once the components a(t) are known, the vectors s of the structural
displacements can be calculated by eqn. (11). In general, the structural
displacements will be an aperiodic function of time.
water pool of the pressure suppression system was assumed and SING-S
was used to compute the dynamic response ofthe spherical containment
shell, the coupling with the fluid dynamics of the adjacent water pool being
taken into account. Major input data are the stiffness and mass matrices of
the flexible structure.
In earlier work Gller investigated the structural dynamics of the
pressure suppression system of the Brunsbttel reactor by finite elements. 13
Coupling with the fluid dynamics of the water pool was not included.
Instead pressure distributions measured during steam relief tests 14 were
given. Comparing the calculated displacements with corresponding values
measured during the same tests. Gller found that structural modeling by
standard finite elements (STRUDL/DYNAL) was inadequate. Therefore
the spherical shell has been investigated by a semi-analytical method taking
into account about 2000 mode shapes, which is far beyond the computation
capacities of standard finite element programs. 9 , 1 5 Now the computations
agreed very well with the experimental data. Furthermore. Gller was able
to show that major deformations occur only in the loaded regions.
Consequently, for application ofthe SING-S program to the spherical
containment, the stiffness matrix was determined by the above semi-
analytical method. However, only that part ofthe spherical containment
was taken into account which at the same time constitutes the outer wall of
the condensation chamber. (At the upper and lower edges of that part the
shell was assumed to be simply supported.) The other parts of the spherical
containment could be neglected, since no dynamic loads act there and
according to Gllers' findingno major deformations are to be expected.
Also, the flexibility ofthe other walls ofthe condensation chamber had been
neglected, since investigation of those elements was not the primary goal of
the work. Furthermore, these flexibilities are relatively slight; hence, their
influence on the dynamics of the spherical containment was estimated to be
rather small.
For excitation of the system three simultaneously collapsing steam
bubbles were assumed to be located at the ends of condensation tubes at the
circumferential angles of 0, 120 and 240. Consequently, the in-
vestigations could be restricted to a 60 section of the containment shown in
Fig. 1. The initial bubble radius was assumed to be about 0-5 m. The bubble
collapse (i.e. the decrease of the bubble radius versus time) was modelled
according to Class 1 and is shown in Fig. 5. The collapse time is 0-1 s. The
times for bubble growing are much longer by comparison. Therefore,
dynamic effects due to this period are negligible.
In the SING-S program the decreasing bubble was simulated by a
308 R. KRIEG
5 02
0 04 0 06 0.10
Time (sec)
FIG. 5. Steam bubble radius versus time during bubble collapse.
transient point source (sink). The dipole panels necessary to describe the
fluid dynamics of the water pool constituted the pool boundary as shown in
Fig. 1. The number of dipole panels was 696, while the number of the
components of X was 247.
The eigenfrequencies calculated with SING-S are
9-4, 13-0, 17-5. 17-8, 19-3, 21-4, 2F9, 24-4, 24-8, 25-0,... Hz
The mode shapes ofthe shell vibrations (coupled with the fluid dynamics of
the adjacent water pool) which belong to the four lowest eigenfrequencies
are shown in Fig. 6. For simplification only the wetted shell region is
plotted. The radial shell displacements versus time at six different points are
shown in Fig. 7 (location of these points. Fig. 1). It can be seen that after
bubble collapse (which takes 0 1 s ) the amplitudes of the radial
displacements are roughly constant. Decreases due to damping (material
damping in the shell, viscosity in the fluid) are not included. Comparing the
results for different points shows that with increasing distance from the
prescribed source (simulating the bubble collapse) a short time delay can be
observed. This is an indication that local disturbances propagate at a finite
velocity in coupled systems, although fluid compressibility is neglected, i.e.
the velocity of sound is finite.
The lowest eigenfrequency measured during the Brunsbiittel tests was
about 11-5 Hz, 1 4 which is some 20% above the calculated value of 9-4 Hz.
The reason for this discrepancy is supposed to be additional stiffening ofthe
lower containment section ofthe Brunsbiittel reactor. This stiffening, which
spreads over the three bottom rows of panels, was not taken into account
FLUID-STRUCTURAL DYNAMICS IN BLOWDOWN SUPPRESSION SYSTEMS 309
17-5 Hz
TWTTTT
I f - 1 "4
it I t
T_:_T#t
itti
\\\ V HTT
mi- xijffi
1 3 0 Hz 17-8 Hz
FIG. 6. Vibration modes of the spherical shell coupled with the adjacent water
pool of the pressure suppression system (components of X: 247).
310 R. KRIEG
-3
02 04 06 08 10 0 02 04 06 io
_. 3
E Boundary Point 4 Boundary Point 3
E
^ V ;
-=^ ^ T
02 04 0.6 1.0 02 04 06 08 10
^v/u
3
Boundary oint 2 Boundary Pomi 1
V
VV
-3
02 04 06 02 04 06 08 io
Timels ) Time ( s )
FIG. 7. Radial shell displacements versus time for points 1-6 shown in Fig. 1
(components of X: 247).
12-5HZ 17-9 Hz
14-9 Hz 2 0 4 Hz
FIG. 8. Vibration modes of the spherical shell coupled with the adjacent water
pool of the pressure suppression system (components of X: 394).
pool on the structural dynamics, the above spherical shell was investigated
without this coupling. The corresponding eigenfrequencies are
520. 56-0. 58-8, 59-1, 59-5, 6 0 0 . 60-2. 60-9, 6 1 0 . 61-1. Hz
The mode shapes of the four lowest eigenfrequencies are shown in Fig. 9.
Comparing the above results shows that coupling with the fluid dynamics of
312 R. KRIEG
mmE *
" 11-1
mti T T T TWU
!
.
ff ';' T
T T
,|T T TT
T T T
.
TTT TTT * *
^ -r T T IT "I*
T.
T1 T T * T T T T TT - * TT TT
" : , T TTT * i
T T
' T TTTTT *
' t T )Tlfi
4 * TT T T
T
r * ! * ; t IT IT jrj
m im T
! *T T t T
WTULUUI'I * * T * * ;3$0^
H $ t t t fr fri
VMXV
UUIH
u\UUU WtL
+ttttw~.
TuutL
mWWW mmfw
520 Hz 588Hz
56 0 Hz 591 Hz
FIG. 9. Vibration modes of the spherical shell without coupling.
the adjacent water pool greatly reduces the eigenfrequencies ofthe spherical
shell. Furthermore, coupling completely changes the mode shapes.
Displacement peaks occur close to the pool bottom, where the influence of
inertia due to the fluid masses is stronger than close to the free fluid surface.
In the unwetted region the displacements disappear almost completely. (In
the figures, displacements lower than about 10 % ofthe maximum value are
FLUID-STRUCTURAL DYNAMICS IN BLOWDOWN SUPPRESSION SYSTEMS 313
not plotted.) Moreover, it can be noticed that in both cases, without and
with fluid-structure coupling, the eigenfrequencies are very close to each
other (while the corresponding vibration modes, in general, are entirely
different). According to Gller. 15 this is due to the local flexibility
behaviour ofthe system, where the displacements at a shell point essentially
depend only on the load at the same point and immediately around it.
REFERENCES
1. INTRODUCTION
GATE VALVE
JUNCTI ONS
boundary conditions (pumps, blowers, etc.) and those with fixed boundary
conditions (fixed obstacles in flow, geometrical variations of pipes).
Many studies have been devoted to obstacles in flow as heat-exchanger
bundles and to pumps or blowers.
Pipe singularities as sudden enlargements, diffusers, valves. T junctions,
etc.. have been less studied (see refs. 6-9). However, their contribution to
the fluctuating pressure level in piping systems is important. Moreover,
their characteristics are perhaps simpler. A methodical study of these
singularities has been performed by the C.E.A., D epartement des Etudes
Mcaniques et Thermiques.
Experiments have been worked out on atmospheric air rigs and water
loops. Theoretical considerations have also been developed. The results
obtained have been collected in a formulary (see refs. 1, 2 and 5 for more
details of methods used). Some examples are presented in this paper.
&(s) =
1 + (.v/.vc)2
PERMANENT FLOW PARAMETERS
GEOMETRICAL PARAMETERS
Enlargement ratio: D/d Reference velocity: V
d D V - mean value of the flow velocity in the
Reference length: D-d
Upstream cross-section: S, upstream tube
Downstream cross-section : S2 Reference pressure: AP
AP = permanent pressure jump across the
enlargement
SPECTRAL QUANTITIES
y ( v) F 1 F
PSD:.^(.v) = PSD for pressure discontinuity: &&F(s) = Sr&P( v) 2
() 2 D-d AP D-d
Phase, [ ,
Al 2 Al
^ ( - 0 = ^?()
cv
I' (F7 v
y S, ; AP2 D-d
ZONE I
fis)
/(%)
1
10-'-=.
20
d/D = 0 5 ^ /^^**^
/V d/Q=0-625
/ //' d/D=0 71
i -
2 - _v(D-d)
L/(D-d)
1 2 3 6 7 9 10 t)-2 2 U 6 8 )-' 2 4 6 8 1
FIG. 3. Local pressure fluctuations.
where sc is the 'cutofT Strouhal number: ,vc ~ 0205 for the zone
of maximum fluctuations (see Fig. 3).
(c) CrossPSDs are analysed in the form of coherence functions (ratio
between PSD modulus and the square root of the product of the
PSD of both correlated signals) and phase functions. The
coherence functions of pressure fluctuations present a plateau, the
level of which varies exponentially with the distance / between
both correlated sensors (c = ~ //; ).The correlation length,/has the
same order of magnitude as the characteristic length . A longitudinal
and transversal correlation length can be obtained according to
whether the correlated sensors are located parallel or transverse
to the mean flow. For example, the longitudinal correlation length
ofthe fluctuation downstream from a sudden enlargement is equal
to the diameter of the upstream tube. The 'longitudinal' phase
curves are straight lines. This shape is typical of a fluctuating
'transport' phenomenon.
The slope ofthe phase lines allows us to obtain the characteristic
velocity, v, of this phenomenon. The dimensionless value of t' is
equal to about v/V = 07.
Transverse fluctuations are in phase.
scM
where M = mean flow Mach number.
As s < 05 and M ; 1 for the most current industrial circuits, then
/.JA : 1 : and the tube diameters of the circuit having the same order of
magnitude. vc is much lower than the acoustical cutoff frequency of tubes.
Consequently, only plane waves are transmitted through the circuit and
each singularity can be characterised by two spatial discontinuities: a
discontinuity of the acoustical pressure (averaged in a tube crosssection).
Ap(/)t and a discontinuity of the acoustical mass flow rate, A q(t).
Integral expressions of these functions have been written in ref. 2, using
the local fluctuations of pressure and velocity.
Ap(t) and A q(t) are random functions of time which are characterised by
their PSDs, r^(v) and y^(v).
'log
Vs)
2-W5>
log s
Strouhal number range, the acoustical source can be represented by a A q
function.
All these theoretical results have been experimentally verified. A method
has been developed in order to deduce the A p and A q functions from the
measurements of acoustical pressure at different locations in the circuit. In
particular, the acoustical transfer functions ofthe test circuit are given by
the computer code VIBRAPHONE. 4 Good isolation from other acoustical
sources (especially pump or blower) and welldefined boundary conditions
are needed. Figure 4 illustrates the results obtained in the case of a sudden
enlargement.
IO" 3
8
6 <~d/D=0 5
^ " ^ d/D = 062S
"N_ d/D = 0 71
d/D=0 75
"'
8
6
2 . v(D-d) .v(Dd)
"' _L -L -io-' X
4 6 810-2 2 6 8TJ-' IO" 2 2 4 6 8 10-1 2 4 6 8 1
3.1. Introduction
3.1.1. Statement of the Problem
The third section of this chapter is devoted to the 'response problem' of a
piping system to the pressure fluctuations which have been analysed above.
Both local and acoustical fluctuations lead to fluctuating forces which act
on the walls ofthe circuit. The resulting vibrations may be evaluated for any
given piping system, by computing its mechanicalacoustical eigenmodes
and by using results of ref. 1. We present below a theoretical formulation
for the computation of the modal characteristics and of the transfer
functions of piping systems conveying fluid. Finally, an experimental test of
the whole method is given.
Im f)
increasing values
of velocity V
R e (Q)
| v=o
data). Furthermore, in the case of piping systems the mechanical-acoustical
calculation is simplified by using the fact that generally the acoustical
wavelengths of vibrations are very large with respect to the cross-
dimensions of the tubes.
t That is not rigorously verified near the extremities ofthe tube. However, necessary
corrections remain small, provided that LjD 1 (L = length of the tube;
D = diameter of the cross-section).
324 R. J. GIBERT
In the Appendix an example is given in the simple case of a rigid bent tube
with open extremities and containing a quasi incompressible fluid. This
example shows the relative influence of the longitudinal and transversal
effects in fluid on the translation modal masses of the system.
Let R(s) be the local radius of curvature, dS/ds (s) the local variation rate
of the cross-section, and I and t the local unit vector system associated with
the osculatory plane ofthe neutral fibre. The local coupled equations can be
written as
dS
KJ + MJ + P o So, - -St + p \= 0
R ds
(1)
\ Sc2p at dS-
[S^f- --^^TT- PcjS + Po - = 0
s\ es J c2 dt2
(Re)
ds
A"mand M,,, represent the stiffness and mass operators ofthe tube in absence
of fluid; is the acoustical pressure averaged over the crosssection S(s);
and c is the velocity of sound.
From these equations simplified expressions may be easily derived for
several particular cases, such as a junction between tubes, a local variation
in the curvature or the crosssection of the tube.
3.2.2. Description of the Computer Code TED EL
Short description ofthe mechanical version of TED EL. TEDEL is designed
to compute threedimensional piping systems. The structure is modelled by
connecting straight beams or more complex elements (e.g. bends. T
junctions, etc.). Computations of several types may be performedin
particular, computation of the resonance frequencies and mode shapes.
and computation of the time response to any excitation.
The code uses the classical finite element method. The elements have two
nodes and there are six variables per node (three displacements and three
rotations). The form function which is used is a cubic polynom.
Elementary stiffness matrix. Fluid does not provide any stiffness; then
the only additional terms are given by the discretisation of the equation
2 = 0
Elementary mass matrix. Three kinds of elements are distinguished:
t The transversal inertial coupling between coaxial tubes separated byfluidcan also
be taken into account. The corresponding terms ofthe mass matrix are obtained by
an analytical bidimensional calculation. Note that this coupling effect is very
important in the case of a narrow gap between tubes.
326 R. J. GIBERT
Local and acoustical effects of such singularities have been outlined above.
The local pressure fluctuations generate forces on the walls, (), of the
locally perturbed volume, ( V), which takes place next downstream from the
singularity. For the first resonance modes of the system the strains over ()
are small enough to be neglected. In addition, acoustical wavelengths are
also large in respect ofthe dimensions ofthe perturbed volume. Therefore,
by integrating the equation ofthe fluid motion, it can be shown that the
force field acting on () is equivalent to the force field acting on the cross
sections, (S), which separate ( V) from the rest ofthe circuit (see the diagram
below).
As the force field acting on (5,) depends only upon the acoustical pressure
averaged over (S,), the local excitation problem can be reduced to an
acoustical plane wave problem. Therefore, excitation effects on the whole
circuit, caused by a singularity, can be derived from the characteristics of
the acoustical source which is associated with the singularity. In the case of
low Mach number flows, such as liquid flows encountered in nuclear reactor
circuits, their acoustical sources may be characterised by a discontinuity of
the acoustical pressure (see Section 2).
The TEDEL program allows the calculation ofthe transfer functions
H(v,s) associated with a unit pressure discontinuity, located anywhere in
the circuit. Using such functions, it is also possible to compute the power
spectral density ,9"R of the response (displacement, stress, acceleration,
pressure) at any point of the piping system, to several decorrelated
singularities, according to the relation
y ff,(v,i)|Vjv) (3)
3025 > t_
rf~\ gate valve
extensible springs (steel mass of the whole is M s = 250 kg; fluid mass is
= 300 kg). This pipe is inserted in the 'Gascogne' water loop. Connections
are made by using rubber bellows which isolate mechanically the rig from
the rest of the circuit. In the same way two cavities with a compressed air
level are located upstream and downstream from the rig, providing an
acoustical isolation of it (see Fig. 5).
Vibrations are induced by a half-closed gate valve, inserted between the
downstream bellows and the downstream cavity. The level ofthe acoustical
source associated with the valve is much higher than that from any other
source in the whole circuit, at least in the low-frequency range, which is of
interest here (0-10 Hz). Pipe motions are measured by accelerometers, and
pressure fluctuations by wall pressure transducers (their location is given in
Fig. 5). Pressure measurements are used, in particular, for evaluating the
spectral characteristics of the acoustical source.
MODE 1 =382
m =112 Kg
battows
/ C > v . ^
max
^\,
, /c, acoustical pressure field
MODE 2 v=5B8Hz
mr460Kg
the flanges. The latter introduce a reduction ofthe flexibility ofthe bends
which it is worth emphasising. This effect has been calculated separately, by
using the threedimensional shell computer code TRICO.
The fluid boundary conditions are assumed to be fluctuating pressure
nodes, located at each cavity. TEDEL provides the resonance frequencies
and the mode shapes (displacement and pressure) of the system, in the
010 Hz frequency range. Mode shapes fall into two categories:
Normal to the pipe plan. It is expected that such modes are not excited by
the source. Indeed, the level of such resonances is always very low in the
experimental data.
and is only a function of the valve gate level. Consequently, p(s) proves to
be a specific quantity which characterises the singularity. The shape of
tFAp(s) is given in the diagram below.
10"
<v> h/D=0.2
h/D = 0.
10 " 3
h/D = 0.4
10 - 4
10-3 10-2 10-1
PSD(gVHz)
4 5 6 7 8 9 10 5 6 7 8 9 10 4 5 6 7 8 9 '
Frequency Hz 3 Frequency Hz 3 Frequency Hz
, PSD(g 2 /Hz)
6 7 8 9 6 7 8 9 5 6 7 ~8
Frequency Hz Frequency Hz Frequency Hz
CONCLUSION
REFERENCES
F2 = - 2 *
Fn - 5
COMPUTATION OF FLOWINDUCED VIBRA TIONS IN PIPING SYSTEMS 333
(c) The inertial force due to the fluid can be separated into two
parts:
(i) The force acting on the straight parts (due to the transverse
motions =( N /2/2) x ) projected on OX is:
^2 J
-$ p0Sl^ dx 3L_ = pSlx
pido + cp\ = 0
III Hi ct
The second term of thisequation is a volume term and can be neglected. The
surface, (), ofthe volume (V) includes the wall, (), of the tube and the
crosssections, (S) and (S2):
pnda + pn2 da pn do 0
(,) i.v :) i.S'|l
The first term is the force F, 3 ; the two other terms only depend on the
"plane wave' pressure p(l) in the circuit at the level of the bend:
F,3= p(l)S(n2n,) (A.l)
Symmetrically the motion of the bend generated a mass flow rate
discontinuity, A q, which can be deduced from the mass conservation
equation integrated in (V). The term due to the compressibility being
neglected, we obtain
Aq + ,,^ = 0
(/)
Using
0^ = 0
we have
</ = p0S<5(n2 - , ) (A.2)
This discontinuity, Aq. generates plane waves in the circuit. An acoustical
monodimensional calculation, taking into account the boundary con
ditions = 0 at each extremity ofthe tube, gives the 'plane wave' pressure
p(s) at any point of the tube. In particular.
11 , -
P(l) = 2^ A 4 = :bcA(rl, - n , )
assuming that the wavelengths of fluctuation are large in comparison with /
(quasi-incompressible fluid hypothesis).
Using this expression of p(l) in eqn. (A.l). the projection of F l 3 on O.v is
given by
F3 = ~Pi)Sl'x
COMPUTATION OF FLOWINDUCED VIBRA TIONS IN PIPING SYSTEMS 335
P(S)
P(D
I 21
Fn + Fu + F,.
Mox = = Ms + 2pSl
2K 2K
Mn Ms + 2pSl
(2) OF translation (y):
(a) The stiffness force acting on the tube is F r = 2Kdr
(b) The inertial force due to the tube itself is Fn = MsSy
(c) The inertial force due to the fluid is
(i) force acting on the straight parts: F,2 = pSl.
(ii) force acting on the bent part: F i 3 = 0 (since (n2 n,)
= 0).
The modal mass ofthe 0y translation mode is:
M o v = Ms + pSI
The associated natural frequency is
2K
Ms + pSI
CONCLUSION
This example shows the mechanism ofthe coupling between the motions of
a circuit and the contained fluid. It particularly shows the importance of
336 R. J. GIBERT
1. INTRODUCTION
2. LOAD FUNCTIONS
2.1. Survey
The current specifications relating to the load case of an aircraft crash are
characterised by the fact that they are not defined by the authorities
ANALYSIS OF AIRCRAFT IMPACT PROBLEMS 339
150r-
indirectly in terms ofthe type of aircraft, speed in flight, etc., but are laid
down directly in the form of a load-time diagram including a binding
specification of the impact area and of the angle of impact (Fig. 1). Such
data do. indeed, provide unambiguous load definitions, but these
definitions are only valid if one assumes that the concrete structure at the
point of impact is infinitely rigid. In view of the conservative nature of these
load data, and the consequences which this has not only for the
dimensioning of the load-carrying building structures, but also for the
design and the dimensioning of the mechanical equipment components
installed in the building, these load data require revision.
In order to define at least a provisional basis for the consideration of this
load case, a load function has also been defined in the Federal Republic of
340 N. J. KRUTZIK
Germany, and has in the meantime been specified as binding for locations
in Germany (Fig. IB).
This load function postulates the impact of an aircraft ofthe Phantom
RF-4 E type with a full-load mass of 200000 N and an impact velocity of
215m/s. The further stipulations are that the structure against which the
aircraft crashes is infinitely rigid, that the impact is normal to the surface
and that the load is applied to a circular region with an area of 7 m 2 . No
consideration is given to the building region, components, stiffness ofthe
region, or point of impact.
parameters. The load reduction rate is always equal to the load increase rate
(slope) in the elastic range.
The equations of motion describing the impact are solved by the finite
difference method (FDM).
The total force, F, acting on the building for the load case with
detachment of fragments is obtained by superposing the forces exerted by
the body ofthe aircraft (braking force) and those exerted by the impacting
engines (fragment impact loads).
D D
FLAT 1
i PLATE I
This may result ina reduction in the amplitude ofthe load-time function by
about 10%, and in a reduction of the global thrust acting on the
investigated reactor building by about 50 %.
boundary of the local region considered was also taken into account. In
accordance with the resultsofa separate static calculation, the constants at
this boundary were so chosen that the deformations induced by a load of
llOOOMPa amounted to about 13mm. It was found that taking into
account the deformability of the global structure beyond this demarcation
boundary has almost no effect on the results of the calculations. This
deformability may therefore be neglected for practical purposes.
;so
100-
GRID PLOT
50
AIRCRAFT
00
50 00 50 10 0 15 0
FIG. 3. Modelling of the aircraft and the crash zone of the shell.
2.3.2. Results
The events and the forces within the plastified or partially destroyed
region in the vicinity ofthe impact area are of no interest in investigations ofthe
dynamic response ofthe global system, provided we can define the forces
and moments acting on the global structure along the boundary of an
appropriately defined (e.g. circular) zone enclosing this plastified and partly
destroyed region. The governing load for the global system is the thrust
actually acting along this boundary, including the temporal characteristics
of this thrust. This boundary can easily be defined by following the
344 N. J. KRUTZIK
deformations and the strain conditions within the plastified and partly
destroyed region.
GRID PLOT
40-
20r 1
/ 77 / 7 7 1
/ I ! '\ l|l|i|i|i|
- NIH"
LrtTtS !
''
- M
!! l
, 'ili
1 1 1 1
0 0 ' 0-0 50
The deformations of the zonal cells of the concrete and of the body of the
aircraft for the case of a plane slab are shown in Fig. 4.
1 5" + 4
5 10"+4
5 0^+3
S 00
CO 0 02 0 04 0 06
/ (s)
I 5if + 4 -
10" + 4
, 50F+3
00
0 0 002 0 04 0 06
/ (s)
FIG. 6. Time history of shear force at 7-3 m, spherical shell of 1-8 m thickness.
346 N. J. KRUTZIK
2 O
5
't 'h-
\ o.. ^ -
-M V.
05
10 20 30 40 50 60 70 80 90 100
MHZ]
0 50r
040
030-
020-
\
o io-
\
Duration of load function Cd = 0 0 7 s
009:
0 08- /
0 07:
0 06:
0 05-
0 04-
0 03--
,7"
1 l i l i l
001
10 20 30 40 50 60 70
FREQUENCY (Hz)
0 40
*_Q 4 = 2 (Hz)
7"s = 0 50 ( s )
-0 4 0 L
0 00 0 05 0 10 0 15 0 20 0 2 5 0 30 0 35 0 40 0 45 0 50
TIME (s)
040
T-f~j /s = 5 0 ( H z )
030 71 - 0 0 2 ( s )
0 20
O 10
-0 00 JL. l l . . L - . . , l -1 J. l 1 I
- O 10
-O 20
- 0 30
-0 4 0 L
000 005 010 015 020 0 25 0 30 0 35 0 40 0 45 0 50
Time ( s )
22
DLF
20
i=L
1
/ M- AIRCRAFT C RASH
V \\
/ / > \
12
/ // vi
i o
08
- 1 '/ / ^
06
04
0 2
00 I I I I I I I I 1
1 2 3 5 7 10 20 30 5 0 7 0 100
/[Hz]
FIG. 11. Dynamic load factor for different types of load functions.
ANALYSIS OF AIRCRAFT IMPACT PROBLEMS 349
Figure 11 shows the dynamic load factor (DLF) curve for various load
functions. It can be seen that the load functions based on an elastic material
behaviour result in smaller DLF values.
For nuclear power plants located in regions with low earthquake intensities,
the design of load-carrying structures as well as of mechanical equipment
components may be governed by the load case of an aircraft crash. The
impact may take place at any point ofthe building. Starting from the point
of impact, there takes place a wavelike propagation of vibrations. In
monolithic buildings these vibrations are transmitted to the coupled
structures directly, whereas with separate structures they excite the internal
structures indirectly via the common foundation slab.
The response ofthe building structures to the impact load induced by an
aircraft crash always definitely depends on the point of impact and on the
assumed plane of action ofthe load. The first step in the investigation ofthe
consequences of this load case is therefore to find the governing points of
impact and to determine their topological arrangement in the building
structure as well as the direction of action ofthe impact load. This step may
be very laborious and time-consuming, and in some cases this problem
cannot be solved at all. If the coupling points of the most important
mechanical equipment components are not known in advance and/or if the
investigations of the effects of loads applied at several assumed points of
impact cannot be limited to only a few points ofthe structure, it becomes
necessary to carry out repeated analyses of the dynamic response of the
global system for various points of impact. This leads to extensive and costly
studies.
The level of stress of the system outside the area of impact is relatively
low. so that, as a rule, the global structure should occur outside the near
vicinity of the impact area; this could have a significant effect on the
response of the structure. In such a case it would be necessary to carry out a
non-linear investigation ofthe entire system. However, the state of stress
can bedefined and checked at any time by following the time histories ofthe
deformations. On the basis of state of strain of the elements in the local
region, it can also be definitely established whether partial or total
destruction is to be expected in the impact area.
350 N. J. KRUTZIK
Node
1 Shell elements
1981 Mass elements
114 Springs
0 O Harmonic
1 1 Harmonic
V Vertical
Torsion
Horizontal
60
50
INNER CONCRETE PRESSURE VESSEL
Biological
199 399
40 ^shield
198
398
<>197
196 T 295
30 tl95
194 397
r.
?193 294 -
20 192
{>99 t 2 9 3
98 396
10 9 7 129
96
95
94
T 93
92
*039
-10
REACTOR BUILDING STEAM GENERATOR
FIG. 13. Idealised structure of a rectangular reactor building.
ANALYSIS OF AIRCRAFT IMPACT PROBLEMS 353
D,ELn
D m
F,.
^MTJ*^4-" il
FIG. 14. Global and local mode shapes of a PWR building (standard).
^iteliii^ 1 ' 1
. : * c\, ' "1 1
Tr-^ r - f u i r - i ^q ^
.iv T" r-'F-
.-rtrr-4-
- - - =U'^1^-X--i^.pv./ .'
l^-tf'-fcrffT '1 "
certain regions and directions of impact which cover all other regions and
directions within a practically acceptable range of scatter. On the other
hand, it is only in rare cases that it is possible, by means of parametric
studies, to define one single region of impact resulting in approximately
maximum responses at all the relevant locations.
In the studies of the box-shaped building it was proved possible, on the
basis of eleven case investigations, to find one impact region which was
finally used as the basis for'complete calculations and for obtaining relevant
data for detailed investigations of the coupled substructures. This impact
region corresponds to a corner in the top part ofthe building (Fig. 13). For
the axisymmetric building, however, it is necessary to investigate four to six
points of impact (Fig. 12).
ref. 4 and on the function determined taking into account the non-linear
behaviour (Fig. 5). The reductions in the dynamic response ofthe global
structure by virtue of taking local plastic effects for characteristic regions
are compared and discussed. The computations were carried out using the
programs MESY and STARDYNE.
20
0 o
2 0 : \
2 0
0 0
-20
20-
J f
i *./W'
O)
00
t/>
E -20-
tr 20
o 00 ^
o
< -20
20
0 0 >Ued\j\pQ&*\f*ev~^*
-20
20
0 0 ^&/^Ae^
-2-0
RADIAL
VERTICAL
0-08 024 0-40 056 072 0 88 104
00 016 032 048 0 64 0 80 096
TIME(s)
FIG. 16. Time history of acceleration at the level ofthe foundation plate.
ANALYSIS OF AIRCRAFT IMPACT PROBLEMS 359
FIG. 17. Time history of acceleration at the level ofthe inner cylinder, upper edge.
360 N. J. KRUTZIK
002
000
-002
004
002
000
-0 02
0-02
?
F- 000
2
S -002
UI
o
< -004
_l
o.
o 002
0-00
-002
-004
002
000
-0-02
-004
0-02
000
-002
-004
HORIZONTAL
VERTICAL
I I I
0 0 0 0 0 8 016 024 0 32 0 40 0 48 0 56 064 0 72 0 8 0 088 096 104
TIME! s )
J L J_
50 000 100 000 50 000 10O0O0
m /s2
FIG. 21 Maximum value acceleration.
ANALYSIS OF A IRCRA FT IMPA CT PROBLEMS 363
J L _L J_J L J _
50000 100000 5 0 0 0 0 100 000
T tt7~t
il 1 x" '
:
JL,h^n
,
'yf- "
/ 1 I
50 0 0 0 100000 50000100000
O 50 100 tm/s2)
LOCATIONS
0o ,15- ,45 ,90u 180
of eleven harmonics were taken into account. The time interval was 1 s and
the time step was 2-5 ms.
The maximum and minimum values ofthe accelerations, obtained from
the time histories for each impact area considered, are presented as
horizontal and vertical vectors for the nodal point concerned (in the
azimuthal plane 0; Figs. 21 to 23). In Fig. 24 relating to an impact at point
5 taken as an example, the distribution ofthe accelerations in the peripheral
direction (0. 15, 45, 90 and 180) is presented. It can be seen that
ANALYSIS OF A I R C R A F T I M P A C T PROBLEMS 365
the accelerations decay rapidly beyond the immediate vicinity ofthe point
of impact.
F
o
\ t
24
22 idi
20
18
STIFF PLASTIC
14
12
10
8
> \
6
-1
p>^
4
'V /"*^
2
cSSS ---" .-_-: - t - ' "-"-v -
3 4 5 6 7 8 9 1 0 20 30 40 5060 8 0 f (Hz)
FIG. 25. Dynamic reaction at the level foundation plate (A) (maximum radial and
vertical).
Figures 25-30 show the response spectra for characteristic regions ofthe
building structure and of the steam supply system coupled to it, for both
load functions. It may be stated without limitation that when the local
elasto-plastic effects are taken into account, the maximum values of the
acceleration spectra are reduced by about 30-50 %, and it can also be seen
that the frequency regions ofthe maximum acceleration peaks are slightly
shifted. The acceleration peaks are generally lower. Owing to continued
cracking, which induces brief shocks in the structure, acceleration peaks
also occur at higher frequencies. Considered overall, however, the response
spectra are smoother.
366 N. J. KRUTZIK
26
24
I
22
Jdi
fijt .
STIFF PLASTIC
z-\
/f\\l/"N
// \ Li?
,^ N
FIG. 26. Dynamic reaction at the level of the connection point of the RPV (B)
(maximum radial and vertical).
E
26
24
|
20 \ Idi
DI
18
STIFF --PLAS TIC
16
14 j
2X
12
IO
I A
L ^
/ / /"\ '*
Itz ^ *
8 \
v
'
y '//
t\
/ ^-v
^ _
2
"^ 5= -f - /
2 3 4 5 6 78910 20 30 40 5060 80 MHz)
FIG. 27. Dynamic reaction at the level of the inner cylinder (upper edge) (E).
ANALYSIS OF A IRCRA FT IMPA CT PROBLEMS 367
\J \ o
1
12 -Wl
V, f
10
-STIFF PLASTIC
4 __1. _
'"VL
/ \(2>
2
-^v <. ^zzr-zzTzSF-j*:
~~L~ ^ i \Z =
I 2 3 4 5 6 78910 20 30 40 5060 80 )
FIG. 28. Dynamic reaction at the level of the foundation plate (A).
- J
1
'W
\ ~ ^-.
24 j\ I
\
22
J '
- S T I F F - - P L AS TIC
18
\
\
\
/
. V\\ \
/ /
10
8
/
II
\>
\
*\ \\ \
J>
6
-\ ?
^ a
2 '/
.'
1 1
1 2 3 4 5 6 7 8 910 20 3040506080 A (Hz)
FIG. 29. Dynamic reaction at the level of the concrete ring (B).
368 N. J. KRUTZIK
(
~~| .-
^
12
.
10
STIFF - -PLASTIC
/
/
/
/ 2 z_g
;>> ~ ^
y
= = a.
1 2 3 4 5 6 78910 20 30 40 50 60 80 /(Hz)
FIG. 30. Dynamic reaction at the level of the RPV support skin (D).
"0 8 16 24 32 40 48
Damping = 0 0 2 C FREQUEN Y (Hz)
Shell model Beam model
40
f
32
i
f /
24
1 I
j j \
1 \ f " \
1 \ JI
u 16 *i
/ * \\
/' / / " \
f /
08
i f
0 8 16 24 32 40 48 56 64 72 80
Dompinq=0 0 2 C FREQUEN Y ( Hz )
Shell model Beam model
8 00 16 00 24 00 32 00 40 00 48 00 56 00 64 00 72 00 80 00
FREQUENCY (Hz)
Damping =0 02
100 16 00 24 00 32 00 40 00 48 00 56 00 64 00 72 00 80 00
FREQUENCYlHz)
Damping 0 02
100 16 0 0 24 0 0 32 0 0 40 00 48 0 0 56 0 0 64 0 0 72 0 0 80 0 0
FREQUENCY (Hz)
Damping =0 02
32 40 48
FREQUENCY (Hz)
are often quite complex. Minor deviations from the exact axisymmetric
shape, due to gaps, openings, branchings or annexes, can, however, also be
represented axisymmetrically provided only the shear areas, moments of
inertia, and moment distributions of the actual and the idealised structures
result in the same inherent behaviour.
10
A
1
_ 8
O
6
<
cr
L
_l
LU .
4
<
y
2
/
/ y
.s
0 8 16 24 32 40 48 56 64 72 80
Damping =0 0 2 C FREQUEN Y ( Hz)
11 Harmonics , 2 Harmonics ,
FIG. 38. Response spectra for load case aircraft crash (PWR, standard).
20
_^ 1 6
h*
rr
UJ
_j
UJ s
808
s
< s
"s
^v s
04 / ^
// \y
0 8 16 24 32 40 48 56 64 72 80
Dompmg = 0 0 2 % C
FREQUEN Y [Hz]
I l Harmonies , 2 Harmonics-
FIG. 39. Response spectra for load case aircraft crash (PWR, standard).
374 N. J. KRUTZIK
" V
Si 12
d
A \
'
J XV
04 //
^=r ,
/ r
0 8 16 4 32 40 48 -S 64 72
FREQUENCY(Hz)
s:
Damping = 0 0 2 %
FIG. 40. Response spectra for load case aircraft crash (PWR. standard)
u
o
1,I.I /
/\
1
1 A \
\ \
/ 1 v
" ~--
\ /- *-^. ^
16 24 32 40 48 56 64 72 80
FREQUENCY(Hz)
FIG. 41. Response spectra at the level ofthe foundation plate ( 1 % damping), X\.
u*
28
24
o
\ , /
E 20
'i
I i
o I 1
= 16
< /
A
^ N A
12
< 1
8
i?"C.
4
0
16 24 32 40 48 56 64 72 80
FREQUENCY (Hz)
FIG. 42. Response spectra at the level ofthe top ofthe building (1 %damping). Xi.
376 N. J. KRUTZIK
6
*
5
f
\
\ /
A k I
i 'i ^
, -' \ *'r"*
h'
7 \
,~\^ \
0
8 16 24 32 40 48 56 64 72 80
FREQUENCY(Hz)
FIG. 43. Response spectra at the level restraint location RPV (I % damping) X\.
06
>l
J05
O
1c
Lu
I
LU
03
02
Ol
1 I I L
02 04 060810 20 4 0 6080100 20 0 040060
FREQUENCY(Hz)
results were obtained in investigations based on the beam model for the
box-shaped building.
il
-ri
SA
MH MH
ws
DG
25 00
2000
5 15 00
en
1000
Aclive
5 00
Possivi
000
...
K>0 100 0
FREQUENCY(Hz)
.SSE
2
\s
\. AC C
^ s \
<*
r
>*^3
L *"
01 10 2 4 6 8 10 20 50 100
FREQUENCY (Hz)
FIG. 47. A ccelerations at the level ofthe foundation plate due to earthquake (SSE)
and aircraft crash (A CC) loadings.
25 00
M 22 50
2 0 00
O SINGL
17 50
y 15 0 0 l\
ui
12 0 0 I ^-f
T D--0
g loco
F-
7 50
. ^ ^
on
5 00 ^s,
2 50
0 00
!0 '0 0 100 0
FREQUENCY(Hz)
20
I 1 C_(,
.
Atrcrrjf 1 Crosh
Explosion
f\
\\
(K- 8 00
^h- -
0 00 16 00 24 00 32 00 40 00 48 00 56 00 64 00 72 00 80 00
FREQUENCY (Hz)
Damping, -O 0 2
FIG. 49. Comparison of response spectra due to external dynamic loads. PWR
reactor building/foundation plate, radial.
FIG. 50. Comparison of response spectra due to external dynamic loads. PWR
reactor building/foundation plate, vertical.
ANALYSIS OF AIRCRAFT IMPACT PROBLEMS 381
The excitation functions of all these load cases are of a transient and
stochastic nature. Their effects, however, are quite different, because of
different durations and different frequency contents.
None of these load cases can be considered on its own as representative
for the design of all the structures and equipment components. Figures
0 00 8 00 16 00 24 00 32 00 40 0 0 48 00 56 00 64 0 0 72 CO 8000
FREQUENCY(Hz)
Damping =0 02
FIG. 51. Comparison of response spectra due to external dynamic loads. PWR
reactor building/upper interior cylinder, radial.
20
\\
\
\
// \\
// / \\
12
/ k* !/ \
/ \
I
I \
I
1
\\
< Ol
t
1 \
.
/
.//*;
v
8 16 24 32 4C 48 56 64 72 80
Damping = 0 2 Safety earthquake - FREQUENCY (Hz)
Aircraft crash Explosion
FIG. 52. Comparison of response spectra due to external dynamic loads. PWR
reactor building/upper interior cylinder, vertical.
the load case of an earthquake is governing for the design, whereas in the
high-frequency range (above 10 Hz) the load case of an aircraft crash
definitely governs.
plastic deformations, are much lower than those calculated assuming linear
elastic behaviour and infinite rigidity at the point of impact.
The findings of the investigations carried out so far may be summarised
as follows:
(1 ) The accelerations of building structures which may be expected in
the event of an aircraft crash reach in some structure regions very
high values compared with those resulting from other load cases of
external origin.
(2) The maximum values of the accelerations lie in a relatively high
frequency range (usually above 20-30 Hz) in which the excitation of
most reactor components and systems is only marginal.
(3) The high acceleration peaks occur only in the immediate vicinity of
the point of impact, and decay very rapidly. With redundant
duplication and physical separation of the components installed in
the building and relevant from the safety point of view, these high
peaks are only of hypothetical interest and are not governing for the
design.
(4) The plastification of the area of impact results in considerable
dissipation of energy and in reduction in the dynamic response of
the structure regions.
(5) The global thrust at the transition boundary from the plastified to
the elastic zone is reduced by about 50%, referred to a wall.
(6) Accordingly, the acceleration peaks in characteristic regions ofthe
structures, referred to the frequency range ofthe highest resonance
peak, are lowered by 30-60%, and in the rigid body domain by
about 50 %.
(7) The frequency range of the highest resonance peak is shifted by
about 10% towards higher frequencies.
(8) The section forces and moments in the governing locations of the
building structure are reduced by about 50%.
In view of these findings, of the several conservative assumptions
incorporated in the load function specified at present, and ofthe specificity
and low probability of occurrence of the load case of an aircraft crash,
calculations taking into account local plastification phenomena constitute
a realistic and economically necessary basis of design.
Until the deformation behaviour of reinforced concrete is better known
on the basis of large-scale experiments, this procedure is the only
technically sound alternative making it possible rationally to solve the
problems in reactor design posed by the load case of an aircraft crash.
384 N. J. KRUTZIK
8. PROSPECTS
BIBLIOGRAPHY
22. Kaiser, ., Krutzik, . and Schrader, .-. 'Local and global response of
reactor buildings at the load case aircraft impact', 4th S MIRT Conference, San
Francisco, 1977.
23. Schrader, K.-H., Kaiser, A. and Krutzik, N. 'Comparison of the dynamic
response ofthe structures of a typical DWR-reactor building to the load cases
earthquake, aircraft and explosion', Conf. on Structural Analysis Design and
Construction in Nuclear Power Plants, Porto Alegre, Brazil. April, 1978.
12
Pipe Whip Analysis
L. LAZZERI
AMN, Genova, Italy
1. INTRODUCTION
Safety requirements force the designer to consider the most severe and
extreme loading conditions for pipes, components and systems, the failure
of which could jeopardise the safety ofthe plant. Leaving to the specialised
literature the description and discussion of the safety problems and the
implications in terms of fracture mechanics, ' " 5 - ' here only the calculation
aspects of the pipe whip problem will be discussed. Then we shall assume
that a break takes place in a given location in a pipe and that it will be
accelerated by the emerging fluid; the calculation ofthe pipe movements
and the problem of sizing the restraints will be discussed.
Particular importance will be given to the different ways of solving the
problems, from the simplified methods for an initial solution to the most
sophisticated.
2. FORCES C O M P U T A T I O N
C
pA (la)
ofthe fluid, etc. It should be mentioned that the hypotheses made in order
to compute the jet forces according to eqn. (la) are quite pessimistic and
extreme. Various phenomena such as crack propagation and final size, and
fluid flow change of direction around the crack, are physical factors which
could be claimed to mitigate the effects of the forces as computed by eqn.
(la).
In any event, some experimental research appears to be necessary to
make as quantitative as possible such qualitative considerations.
Sketches of possible longitudinal and circumferential cracks are shown in
Fig. 1. The force computation by the Moody method has been automated
and a computer code named G E T T O is available; it computes the forces
due to a circumferential and a longitudinal break for a given pipe line.
Effects due to flow area restriction, change of directions, friction, etc., are
included.
PIPE WHIP ANALYSIS 389
3. SOLUTION ANALYSIS
In ref. 4 some criteria are given to decide whether restraining ofthe broken
pipe is necessary. No protection is required if any of the following
conditions is fulfilled:
(1) The broken pipe is physically separated from any other component
whose failure could impair the safety of the plant.
(2) No damage can result to any other related component, whichever is
the movement ofthe damaged pipe postulating a hinge at each end.
(3) The kinetic energy of the pipe (due to its size and/or pressure) is not
enough to cause damage to structures or other components
important for the safety of the plant.
If these criteria are not fulfilled the uncontrolled movement of the pipe must
be prevented and restraints (or similar other devices) are necessary.
Solid stainless steel bars working in the plastic range, dampers, friction
elements and other devices have been proposed as restraining units.
It should be further mentioned that the above-listed requirements might
not match at all. The problem of the study of the influence of various
parameters and of dynamic optimisation was discussed by Dini and
390 L. LAZZERI
force
displacement
11
1 1
10
10
WV-| Ri
9
?
R8
8
8
WH
7
7 R6
6 VW-f
6
5 >
5
4
4
-WV-i R4
3 "
3
2 " V W - | R2
2
1 I'
5 10 15
initial clearance (cm)
FIG. 4. Influence of clearance.
392 L. LAZZERI
J 10
8.
o- 8
04 12 20 28
overhang length ( m )
the ends ofthe pipe, particularly for the case of short runs, where generally
the restraint cannot work and all the force is directly absorbed by the pipe
end. When the computation ofthe forces on the restraint and ofthe strain
of the pipe has been performed, one should compare the actual with the
allowable ones. The allowable values for the restraining unit should be based
on their maximum load capability. For dampers in the form of hydraulic
snubbers load capabilities as large as 1 IO6 lb have been reported by
/ = \ E,
i=
where IF is the external work, E the maximum energy per restraint unit, N
the number of units per restraint, i the number of restraints and av the
correction coefficient (percentage of total energy in the restraints).
394 L. LAZZERI
Once a preliminary solution has been obtained, the final appraisal can be
performed by means of a more refined finite elements code. In these
paragraphs some indication will be given of some of the finite elements
available, with particular reference to the codes developed by the author
and his colleagues.
decided to make a full dynamic simulation only of the pipe spans adjacent
to the break.
With reference to the piping sketch (Fig. 7), if a longitudinal break occurs
at location 1, then only the two spans 01 and 12 will be fully simulated
(statically and dynamically), while the remaining part of the pipe will be
simulated only statically by means of suitable dummy elements. The
author 7 has already shown that in many cases even the static simulation of
(a
(b)
the pipe part far from the break is not important in the dynamic response of
the structures. The simulation ofthe pipe is performed by the use of four
basic elements: (1) a pipe element with complete elasto-plastic behaviour;
(2) a dummy element with indefinite elastic response ; (3) a restraint element
with bilinear stiffness characteristics; and (4) a damper element whose
restraint force is proportional to the velocity ofthe damper location node.
Two different geometries can be studied: (1) the cantilever geometry of
Fig. 8(a) (for circumferential breaks); and (2) the double fixed geometry of
Fig. 8(b) (for longitudinal breaks).
The general behaviour of these elements and the integration schemes and
techniques will be discussed below.
(4) In the plastic domain : = ' (where A is the plastic constant and
m the plastic exponent).
(5) Ovalisation occurs if the maximum strain in the section is larger
than a predetermined amount.
(6) Instability occurs when the specific energy overcomes a pre
determined value.
(7) Crushing of the pipe is possible when it impacts a hard flat surface.
moment,.
elastic unloading
-instability
ovalisation
E3
se<l elastic
, < s c <2 plastic
> 0 2<Sse< ovalisation
&3 < *., instability and flattened behaviour
r5se < 0 elastic, if E < E3, i.e, elastic unloading is followed. Furthermore,
once unloading has occurred, elastic behaviour is assumed until the energy
for which unloading first happened is overcome. In Fig. 9 the behaviour just
described is schematically drawn, The values of E,E2 and 3 are
calculated as follows:
Ex is the energy value for which the elastic and plastic moments
computed according to the plastic law are the same.
PIPE WHIP A NA LYSIS 397
E2 is the energy value which corresponds to the value of the moment for
which the maximum strain is equal to , i.e. the value of the strain for which
ovalisation occurs.
3 is assumed to be some multiple ofthe E2 value; in the present version it
is assumed that
E3 = nE2 =3 (2)
Any different value for could easily be assumed to match the result of
experimental data; however, this problem does not appear to be decisive, as
will be discussed later. The value of is computed by means of the
equation 9 , 1 4
= 06yt/R (3)
where t is the thickness, R is the radius and is a factor which typically
ranges from 1 to 2.
For each subelement a constant value ofthe bending moment is assumed,
so that if M a n d Fare the moment and shear at one end ofthe subelement
and / is its length, the average moment which is considered is given by
3 = |05/ (4)
An incremental approach is followed and increments of displacement W
and rotation at each step at the subelement level are given by
cW cW
AW = ^A T + ^ A M
T CM
(5)
n
0 = + A M
M
where the coefficients of the tangent flexibility matrix can be computed by
the following formulae directly in terms ofthe strain energy. In the plastic
range it is
2 / l \ 2 n i ; l +m / \1 m'l+m
cW 1 1
+1
~df mEIm
1 m/1 + m
dW 1 1 ml 1 + m
+1
M 2 \ mEIm m
> (6)
2m/ 1 + m / 1 l mi 1 + m
1 1 m 1 + m
1 4
M ,mElm m
Elm = A R2t
R\2 sin 'OdO
398 L. LA ZZERl
moment
not crushing
contact ;> In
rotation
ace (10 m / s 2 )
displ(10"' m )
mom (10 kg m )
IT .fe
at instability
1 2 3
displacement ( 1 0 " ' m )
force
elastic unloading
displacement
FIG. 12. Restraint properties.
PIPE WHIP ANALYSIS 401
added directly to the pipe element at each step for the computation ofthe
overall stiffness matrix:
K, = 0, if X < XQ
K = FK0, if XG < X < XE
K=FK if X > AV (8)
where XG is the initial gap, XE the elastic limit and X the displacement ofthe
node where restraint is applied.
Unloading is supposed to be elastic, so (if impact has occurred) one has
X < 0, KT = FK0
Once unloading has occurred, the stiffness is given by eqn. (1 l),even if ->0,
until overcoming of the value of displacement for which unloading first
takes place. Obviously no traction by the restraint is possible, so that the
stiffness is equal to zero if contact is lost between pipe and restraint, the Xc
value is consequently changed to 2 (i.e. the value of displacement for
which contact is lost).
This model of restraint is a relatively simple and crude one; however, the
author believes that it can be satisfactorily used for practical situations.
Furthermore, a maximum force for the restraint can be specified. Should
the actual force overcome it, the restraint is assumed to have been broken
and no force can be given by it.
In practice this is done by assuming a flat characteristic for the stiffness
and by applying a continuous force opposite to maximum.
Pipe crushing is simulated by means of a proper restraint (see subsection
2.2), which simulates the pipe distortion (see Fig. 13) during the impact; a
value of Ehm must be specified at the restraint level. It could be discussed
force
pipecrushing displacement
5.2.6. Masses
The mass of the system can be given as both a distributed and a
concentrated quantity directly lumped on each node. Even if the mass is
distributed on each element, it will be directly lumped on the nodes by the
program. The masses are consequently assembled as diagonal matrix.
5.2.7. Forces
The jet force can be applied at each node, but no more than one node can
be loaded (this is due to the hypothesis made to simulate the action of the jet
leaving the broken pipe, while no difficulty obviously exists at the program
level should a different simulation become necessary). The jet force is
assumed to change with time in a stepwise manner, 5 i.e.
F=Fy t<t,
F= F. /,,_, < / < ! = , (10)
PIPE WHIP ANALYSIS 403
where [K] is the total matrix, [K] = [Kp] + [Kr]; [Kv] is the pipe stiffness
matrix; [KT] is the restraint stiffness matrix; [C] is the damping diagonal
matrix; [M] is the mass diagonal matrix; and |F(/)| is the applied force
vector (in this version only one term is different from zero). An incremental
approach is followed and the 'constant average acceleration' scheme is
used:
Here the subscripts 0 and 1 denote the value ofthe selected quantities at the
beginning and end of each time step. I f eqn. ( 12) is substituted into eqn. (11),
an incremental integration scheme results.
The stiffness matrix (and possibly the applied force vector) are re-
evaluated at each step, because the pipe and restraint stiffnesses change
during the integration phase. However, the effect of change of geometry is
not included.
A check on the actual moment, as compared with the theoretical
moment, is performed and then the maxima (positive and negative) ofthe
parameter are given by the program. A check is also performed on the
actual energy accumulated by the restraint as compared with the theoretical
one. Provided the time step is small enough, the discrepancies between
'calculated' and 'true' values are quite limited and negligible. Finally, a
check is performed on the overall energy (some discrepancies could arise as
a consequence of the abrupt change of properties during the integration);
again optimum convergence was generally achieved.
With this model (named 'mixed model' by analogy with the stress
and strain field hypotheses) one can immediately compute ,, 2 , 3
as functions of F2, F3, F 4 (or their increments).
(3) The relationships between stress and strain tensor components are
formulated in terms of PrandtlReuss equations; 17 then for the
present case,
a,,, z being the components of the stress tensor
being the components of the strain tensor
y,j being the engineering shear strain
E, G being elastic constants
being equivalent stress and strains (plastic components):
di:: = (a. 0-50) + \E(da: ^0)""
0 = (0- 0-5.)/ + \/(0 - da.) ^ (15)
dy-o = 3. 0 / + dT_0/G
dy rr = 3. / + dx J G
(4) von Mises equations are used to compute the equivalent stress. This
equation is formulated as:
2 = a2 + a2 - 0: + 3(xl, + 20) (16)
Together with the above assumptions about the strain-hardening rule,
these hypotheses do completely define the pipe section model. However, as
far as the material strain-hardening model is concerned, one may note that
the general formula ( 13) can be further processed to account for the effect of
elastic unloading, by specifying that:
= if < 0 or a < a
where the dot indicates differentiation with respect to time and ays is the
406 L. LA ZZERI
TABLE I
H. = cajcap
H:r = cijcap
H
0; = czJdav
g(0) = l/apdf/ap
l/(0) = (a. 0-5aO)g(0)IH.(0) + \E
l/%(0) = (<r, 05a.)g(0)IH~.'B) v/E
1/G=O(0) = 3g(6)T:0/H.0(9) + 1/G
1/G.r(0) = 3*(0)r_ r /#,(0) + 1/G
(17)
dy s 2 = rdy.0cos(id0 rdO >
,
= [b,k]-l\AF,\ (18)
}',
\AF\ being the increment of generalised load in each section. T h e meaning
o f t h e b;k p a r a m e t e r s is shown in Tables I and II. In the case o f t h e mixed
model, while the u p p e r portion of the equation still holds, the A y,
parameters are related to the increments of loads by means of
riv.
rt cos :.() 12 = smrt E.JO) dll >13 = riE::()dO
b)) =
b2 = sin cos 0(G:r(U) + G ; (0))d />,_, = ,/(G. r (0)cos 2 + G .(0) sin2 0) d b2 = r(G.(0) sin d -
-c
-0 m
2" r-2n
34 = rT[C7rr()sin2 + G\ () ()cos 2 ]d F, 5 = r/ sino cos 0(G:r(0) + G:l(0))dU , = rtGJO)cosOdU r
Jo 0
>
2
"
2 >
/,2,sinG..tf()d riG.()d
r2tcosOG:U(0)dO /) 45 = />4 = <
0 s.
b5l = ricos2 E ..(0) dll bi2= ri E:JU)cosU sin UdU > s ., = rtE,JO)cosUdU
Jo "0
fin ^2
-
408 L. LA ZZERl
TABLE III
where the a,k terms are shown in Table III. In any case the section
p a r a m e t e r s , Ay, will be related to the increments of loads in any section
by m e a n s of an expression of the type
,
= [,*]'lA F*, (20)
7;
Then if the local loads \Fk\ are related to the loads applied at the end o f t h e
element \Nk\, a n d if the correlation between local strains and element
displacements |vt'| is expressed by means of the o p e r a t o r Lik, one o b t a i n s
\AWk\=L,k([;ki][n,k])\ANk\
where the matrix [rk] and the o p e r a t o r L are defined in Tables IV and V;
then the local stiffness matrix is defined by
TABLE IV
TABLE V
(5) An energy approach has been used, rather than formulating the
equilibrium equations.
(6) The influence of large displacements on geometry and orientation
matrix is considered.
6. APPLICATIONS
8 5
TABLE VI
JET FORCES
Break code Fi ^3
C) (ms) U) (ms) (0
CV 148 0 103 0 186
1 CT 148 16 103 104 115
L 266 1 186 104 271
CV 148 16 103 32 144
2 CT 148 26 103 72 115
L 266 1 186 72 233
CV 148 26 103 84 120
3 CT 148 10 103 20 115
L 266 1 186 84 212
CV 148 10 103 104 114
4 CT 148 0 103 0 115
L 266 1 186 104 205
CV 148 0 103 0 115
7 CT 148 6 103 24 154
L 266 I 186 24 242
8 CV 148 6 103 12 115
CT 148 6 103 12 170
L 266 1 186 12 256
TABLE VII
RESTRAINT PROPERTIES
TABLE VIII
RESTRAINT CONDITIONS
I 2F
2 41- IL 2L
3 IF 8F 3L 4L
4 3F 8F 5L 6L
7 3F 81- 5L 6L
8 3F 7L 8L
TABLE IX
TENTATIVE DIMENSIONING OF RESTRAINT
factor' (i.e. the ratio ofthe actual to the maximum force for the total
number of moduli) is given. However, it should be mentioned that a
fully elastic response corresponds to a usage factor of roughly 0-63.
The allowable values are determined under the assumption of a
maximum strain of 25 %.
It is fairly obvious that in terms of energy the margins are somewhat
higher than those which can be derived from the consideration ofthe 'force
TABLE
MAIN RESULTS OF DETAILED F R U S T A A NA LYSIS
" This value was used because such a number was required by break 2C.
U)
414
35 1 05 15 05 0 5
-o-
8
H f S
(d)
FIG. 15. (a) Geometry of case 2L13. (b) Geometry of case 2C12. (c) Geometry of
case 3L28. (d) Geometry of case 5L38. (e) Geometry of case 5C89. (f) Geometry of
case 6L38. (g) Geometry of case 8L39.
5 3 5 1 0 5 15 0 5 0 5
5
ZT
S 1
3^VVVf^ fr 3 4 8 ^ 9
- O
2 co
o
(f)
1
(e)
08 11 11
(g)
FIG. 15conici.
It should also be mentioned that in all these cases the maximum strains
are localised near the ends, so that a decrease in the number could cause an
increment both in the pipe strain and in the restraint strain. This is shown by
cases 5L38 and 6D38, where the run was repeated assuming five moduli for
both restraints; an increase in the pipe strain resulted in these cases.
In both cases a reduction ofthe restraints moduli is clearly possible (see
Tables IX and X); so from the tentative solution a reevaluation is made on
the basis of results of the FRUSTA B3 code, and the final solution is
obtained with the maximum number of restraints which are compatible
416 L. LAZZERI
TABLE XI
GENERAL CHARACTERISTICS FOR CASE A
with the applied loads and the allowable values of restraint force and strain
on the pipe.
As far as the residual kinetic energy is concerned, it is generally quite low,
apart from the 3L28 case, where the residual kinetic energy is relatively high
and further consideration of the phenomenon is necessary. This was true
also for case 8L39, where the phenomenon was followed until practically no
kinetic energy was present in the system. However, the usual criterion
previously described is generally satisfactory and rebound phenomena are
not usually very important; the somewhat irregular behaviour of these two
cases is due to the large inertia of a valve which retains large kinetic energy
even at low velocities.
The models used in the FRUSTA code analysis are shown in Figs.
15(a)-15(g), while the detailed results for each model are shown in Tables
VIIIX. Figures 16(a)-16(c) show the main features of case 8L39; the
energy percentage, node displacements and restraint force are given as a
function of time.
7 node displacement
(10"1m)
(b)
5.5 7.5
time ( 1 0 - 2 sec )
FIG. 16. (a) Energy for case 8L39. (b) Node displacements for case 8L39.
418 L. LAZZERl
force
(10 5 kg)
le)
J_ -L.
1 2 3 4 5 6 7
time U0~ 3 sec)
FIG. 16. (c) Restraint forces for case 8L39.
XI ; it should be mentioned that the restraint located near the break has been
lumpedon two successive nodes nearby. This was done to take into account
the relatively large axial size of each restraint unit.
The time sequence of events is as follows:
(1) After 16-3ms contact is reached on the restraint located at node 4
and a further contact is reached after 18-2ms on node 5. Contact
takes place at slightly different times on the different units of the
restraint; it should be noted that large yielding is present on
elements 1 and 2. The displacements after 20ms (i.e. just after
contact) are shown in Fig. 18(a).
4 5 io
10
3 4 5 6 7 8 10
(a)
125 mm
(2) After 32-8 ms ovalisation begins at one end (node 0) and propagates
for a few subelements until a plastic hinge forms on the same node
after 47-2 ms. Unloading rapidly occurs and the moments on 1 and
2 rapidly decrease an order of magnitude; the displacement
pattern is shown in Fig. 18(b) after 50ms, i.e. a few ms after the
hinge formation time.
(3) The transient ends after 85 ms, when inversion of the velocity is
obtained on the node where the force is applied and on the nodes of
the restraint. The geometrical situation is depicted in Fig. 18(c).
Finally, one should observe that:
(a) The total energy given to the system is distributed as follows : 47-5 %
on the restraint; 51-5% on the pipe; and 1 % residual kinetic energy.
Furthermore, the first element where a hinge is formed absorbs
roughly 30 % of the total strain energy, while the remaining part is
absorbed in the restraint zone. It is then obvious that the correct
simulation of energy during and after the development ofthe hinge
is a decisive factor in assessing the total behaviour and that
simplified hypotheses (a hinge from the beginning) would yield
incorrect results.
420 L. LAZZERI
(b) The behaviour ofthe system is such that the absorption of energy is
different on the two restraints located on nodes 4 and 5; in fact,
more than 70 % of the total restraint energy is given to the restraint
closer to the break location. It reaches a maximum force of 83 % of
the allowable, while the other one is roughly 62 %. This makes clear
that a very refined model may be necessary in the predimensioning
phase.
(c) The difference in the displacement pattern between Figs. 18(a) and
18(c) in the zone around mode 0 (the tangent changes inclination
from horizontal) clearly shows the hinge formation mechanisms.
6.2.2. Example
The case depicted in Fig. 19 is examined. A flat surface is supposed to be
present at node 2 and a further restraint is applied split between nodes 4 and
3
I4 5 6 ^JL^.
5 (a total of six modular units was supposed; three were lumped at each
end); furthermore, a valve is present and is simulated by element 7.
The pipe is 609-6mm in diameter and 25 mm thick, and a critical
crushing force of 34-4 ton was assumed (this value is quite irrelevant and its
only use is to show the performance of the program). Typical data are given
in Table XII.
TABLE XII
GENERAL CHARACTERISTICS FOR CASE
I 20 m sees
I 4 0 msecs
54msecs
1 m1
JlOOr
FIG. 20. Displacement of the pipe at various times for case B.
(a) The total energy given to the system is 59-5 % on the restraint, 37 %
on the pipe and 5 % residual kinetic energy. Most of the energy is
tJ
io
RUNS Type
Load P ercentage O simili /led
no pressure
K3/emi
\ 150 ru//cmi
o .
o . . "
eo -
Cp R- OOO m
t m
r
>
v N
Ky N
rr
Sx f<2 7:
f > o.rtx*e>
m ,
-1
so
30
Displacement (cm)
In all the cases the geometry which has been considered is a cantilever beam,
with various end loads. Obviously the PAPS T2G code can handle much
more complicated situations; however, a comparatively simple one has
been selected in order to enable comparison with simpler models and for an
understanding of the phenomenon.
Some runs have been depicted in Figs. 21 and 22. The general results can
be summarised as follows:
RUNS TV/SC
Case P rete Shear Torque Bend
-I tD O.IS O O
load Percent \ 2 tO G /3 O
V 3 tO O.IB OSO Oto
o simplified (pure bending)
m
I*
v* i
? - O. das m
r
iC3}t ** t O.O m >
N
N
fi- 0.43*10*fy/m'
axial m o.ie>
/
.--.
-~ mixed model
Di*picement (cm)
7. CONCLUSIONS
A general discussion of the pipe whip problem has been given. Basically
three types of codes are available:
REFERENCES
1. Irwin, G. ( 1957). 'Analysis of stress and strain near the end of a crack traversing
a plate", J. Appi. Mech., 24(3).
2. ASME 3, Nuclear Components, 1976.
2. Wessel, Clark and Pryle. 'Fracture mechanics technology to heavy section steel
structures', 2nd Conf. Fracture Mechanics, Brighton, Paper 72/VI.
4. AEC Regulatory Guide 1.46, US Regulatory Commission.
5. Moody, L. F. 'Prediction of blowdown thrusts and jet forces', ASME, paper
69-HT-31, 1969.
6. Esswein, G. A. et al. 'Systems criteria for protection against pipe break for the
mark III containment", N E D O 10990, GEC, September, 1973.
7. Dini, D. and Lazzeri, L.'An analysis ofthe dynamic elasto plastic behaviour of
a pipe during the pipe whip accident". 2nd SMIRT Conf., Berlin 1973, paper
F5/4.
8. Palusamy. S., Patrick, W. L. and Cloud, R. L. (1974). 'Dynamic analysis of
non-linear pipe whip restraints', Nuil. Eng., 31, 106.
9. Gerard, G. Handbook of Stability, NACA-TN-3783.
10. Bisconti, N.. Lazzeri. L. and Strona, P. (1976). 'Pipe whip analysis for nuclear
reactor applications', Nucl. Eng. Des., 37, 347.
11. Bathe. J. K. and Adira, M. 'A finite element program for automatic dynamic
incremental non-linear analysis', Rep. 884499-1, MIT.
12. Sanders, L. 'An improved first-approximation theory for thin shells', NASA TR
R-24.
13. Dini. D. and Lazzeri, L. (1973). 'Descrizione del programma di calcolo
FRUSTA', CNEN Rep. RT/PROT, 15.
14. Flgge, W. (1932). 'Die Stabilitt der Kreiszylindershaie", Ing. Arch., 3.
15. Hodge, P. G. (1959). Plastic Analysis of Structures, McGraw-Hill, New York.
426 L. LAZZERI
1. INTRODUCTION
It has long been known that the mechanical properties of metals and
composite materials, such as plain, reinforced and prestressed concrete,
depend on the strain rate. Therefore, the computer codes for structures
subjected to dynamic loading require the knowledge of material
constitutive relationships which correlate the stress increments with the
strain and strain rate increments.
In order to give the experimental basis to the formulation of constitutive
relationships for metals, we have developed some techniques based on the
principle ofthe split Hopkinson-Davies bar 1 which permit the deformation
at high strain rate of a short specimen of the material.
427
428 C. ALBERTINI, J. P. HALLEUX AND M. MONTAGNANI
CALIBRATED
ORIFICE VALVE 1
MEMBRANE \ ^<<%^
HIGH
PRESSURE
GAS LINE
DISPLACEMENT
TRANSDUCER
STRAIN GAUGES
i:''"'///-.
^EAR_TH
FIG. 1. Schematic of hydraulic-pneumatic machine for dynamic tensile tests at
constant strain rate.
Membrane
Valve i.
^-j< ^ufjftW Valve 1
BS r*
:..,..,..
AL 1mm/sq.
M&SHHK
*- 5 m s/sq.
116 MN/m2
420 - . (0 6- -12 1
r^
IRO 5- - i n
b ^E(t)
?Rn LU
- s- u- -OR
z S
tr
/ 210 <i* -0.6
LO CE
LU
IAO -td J2- -0.4 <
cc J(t) CE
LO 70 1--0.2 LO
AO 80 120 160
TIME ( m sec)
The recording of a testat a constant strain rate is shown in Fig. 3; one can
observe that during plastic deformation the slope ofthe displacement-time
curve is nearly constant; therefore the strain rate is also constant.
Figure 4 shows the recording of a test at constant stress. In this case, when
the applied stress is lower than the ultimate tensile strength ofthe specimen,
it is possible to observe also the effects of interrupted loading.
Incremental loading tests can also be carried out, connecting the elastic
stress bar of the hydropneumatic machine to the moving cross-head of a
slow tensometer. In this case the specimen is deformed initially at a low
strain rate (10 _ 4 -10~ 2 /s), which is suddenly increased to a value of the
order of 102/s by superimposing the action ofthe gas piston on the action of
the slow cross-head.
BAR ELEMENT
>
r
OS
II HALF-BAR TRANSMITTED IT
PULSE E T
-
X
>
r
r
rr
C
X
>
/.
S
STRIKER BAR O
>
O
>
STRAIN GAUGES
C 2C
4 = -~-(]- f:R - ) ~ 9-,, (2)
'o 'o
, A A
,= {E (, + 4- ) = E (3)
t + ER - &r (4)
Earth
Loading mechanism by
hydraulic piston
Gripping jaws
Earth
Bar I
Strain gauge I
Strain gauge 2
Bar 2
screwed to the two split bars, causing deformation and being partly
reflected and partly transmitted from there, as shown in Fig. 7. Strain gauge
1 records the incident wave , and reflected wave , whereas strain gauge 2
will record the transmitted wave . The device generates impulses having a
2-5 10~ 3 s duration; it is consequently suitable for strain rates of
100-1000/s. The two propagating split bars have the same length as the
prestressed tract, thus affording a wave train in the specimen for 2-5
10" 3 s without any superimpositions.
In order to vary the strain rate it is sufficient to vary the incident pulse
amplitude by changing the stress in the tension bar and, consequently, the
resistance of the brittle notched bridge used as a lock.
Each of the two strain stations is made of two strain gauges connected in
series and mounted at 180 from each other on the bar in order to cancel all
signals following bending ofthe bar, which cannot be avoided completely.
Both strain gauge stations are statically calibrated by the following
method: the two split bars are connected by a strong specimen; the
complete device is then brought into tension, which is measured by a
calibrated dynamometer.
61 -
) ,_.
Ak 100-10-6/sq.
200-10 _6 s/sq.
200 MN/m2
b)
200 KT 6 s/sq.
0-173/sq.
FIG. 7. Modified Hopkinson bar with prestressed barloading device, (a) Record
of test on specimen UQH 2 DIN 4306 steel, = 320s" 1 ; (b) integration of ,
and , cR.
MATERIAL BEHAVIOUR AND MODELLING 437
SPECIMEN Scale 4M
FIG. 8.
take into account the contribution ofthe two connections to the threaded
extremities of the specimen (Fig. 8). This correction is carried out by
checking the deformation of the specimen with a lattice printed on the
specimen and measured by microscope before and after testing. For the
correction the following formula was used, based on the conservation of the
volume ofthe deformed part ofthe specimen:
A! - /
ih = i0 + (5)
/
FIG. 9. Modified Hopkinson bar with prestressed bar-loading device and with
pulse shaper for very short rise time.
(Fig. 9), which transmits only tensional waves, avoids reflected compressive
waves, successive to the incident tensile pulse, acting on the specimen.
Incremental loading tests are also possible connecting the second half-
bar to the moving cross-head of a slow tensometer; in this case the specimen
is deformed initially at a low strain rate (10~ 4 -10~ 2 /s) which is suddenly
increased t o a value of the order of 103/s by superimposing the action ofthe
prestressed bar on the action of the slow cross-head.
2.3. Results
Tests at constant strain rate were performed on various austenitic
stainless steels (AISI 316, AISI 304, AISI 321, AISI 347) in virgin, welded
MATERIAL BEHA VIOUR A ND MODELLING 439
STRAIN
G = E A L = E' AU
L L
CORRECTION FOR STRAIN VALUES k'
L
AL
L " m)
FIG. 10. Correction of stressstrain curves to the slope of static elastic modulus.
Hopkinson bar) were then corrected to the static elastic modulus. This
means a correction of the values of strain, performed as shown in Fig. 10, in
function of stress. This correction takes into account the contribution to
deformation measurements from the connection pieces between specimen
and test machine. A s mentioned in subsection 2.2.1, the experimental
elastic modulus determined with the Hopkinson bar is not exact, because of
the finite time the specimen takes to arrive at stress equilibrium along its
length, through several wave reflections.
The results obtained on virgin AISI 316L tested at ambient temperature,
are reported in Fig. 11, where one observes that with increasing strain rate
the flow stress at a given strain increases and fracture strain decreases.
Some tests at constant stress and rapidly changing strain rate were
performed at ambient temperature on the same steel, in order to study the
influence of continuous change of strain rate during one loading cycle on
the stressstrainstrainrate relationships. Taking instantaneous values of
440 C. ALBERTINI, J. P. HALLEUX AND M. MONTAGNANI
,
eoo
/ "',-- \ %
/
600
f
f/.-
'
, \ \ \
400
TEST TEMPERATURE 20 C
CURVE STRAIN RATE
sec-1
n . m-2
200 -
Ut.
420
n I I 1 1 1 1 .
30 45 60 75 90
ENGINEER STRAIN'/.
FIG. 11. Stress-strain curves for AISI 316 L stainless steel, virgin collaboration
Euratom-C.N.E.N.
20 30 40 50 60 70
ENGINEER. STRAIN 7.
FIG. 12. Stress-strain curves for AISI 316 L stainless steel, virein.
MATERIAL BEHAVIOUR A N D M O D E L L I N G 441
stress, strain and strain rate from the recordings of these tests (Fig. 4), we
have constructed part of the flow curves of AI SI 316L for three strain rates.
From Fig. 12 it is possible to compare the short flow curves obtained with
tests at constant stress and the flow curves obtained with tests at constant
strain rate.
Making the comparison at strain rates of the same magnitude, one
generally observes a good agreement. This means that imposing two
different dynamic deformation histories on the virgin material yields the
same response.
SG2 1 1 ^ 5 0 - 1 0 ~ 6 A
SGi.
500 yusec/sq
-SG6
-^SG7
FIG. 14. Test results on long specimens of AISI 304 steel.
The end ofthe test strip is fixed to a metal block which reflects the incident
wave with the same sign.
Some results are presented in Fig. 14, where strain gauge 2 gives the
incident pulse on the impacted end and strain gauges 6 and 7 show the
doubling of the stress due to the reflection at the fixed end of the strip.
Hence, the experiment is suited to measuring the deformation wave
propagating along the specimen.
The propagation of the deformation wave can also be calculated by
solving the set of equations:
cax dv , .
= (equation of motion)
ox at
dv ce
(equation of continuity)
cx et
/(,) (constitutive equation)
where .vis the Lagrangian coordinate measured on the unstressed specimen.
The boundary conditions are measured experimentally at the impacted end
of the strip; the conditions for a one-dimensional wave propagation are
realised.
Comparison ofthe calculated deformation wave and the measured one
gives a measure of the validity of the proposed material constitutive
equation.
The measurement ofthe first elastic-plastic wavefront as it travels along a
long specimen enables us also to better determine the first part of the
dynamic stress-strain curve corresponding to the transition from elastic to
plastic flow than with the Hopkinson bar. In fact, in the Hopkinson bar the
process of wave reflections, which produces a uniform stress distribution in
MATERIAL BEHA VIOUR A ND MODELLING 443
Section A - B
52x52
46
40
16.7
0.7 0.7 10
^
7Z&Z
22 Wa - CM r-
r=3,
1A-rSlllIL 2 B ^
the short specimen, takes a short finite time which affects by a degree of
uncertainty the first part of the flow curves.
The stress, strain and strain-rate diagram under dynamic biaxial loading is
obtained by a machine which is capable of loading the four arms of a
DISPLACEMENT TRANSDUCER
WATER
PISTON
DETONATOR
MEMBRANE
ORIFICE
LOADING DEVICE
PRESTRESSED CABLE
SHOCK ABSORBER
FIG. 17. High load dynamic biaxial testing machine. Load = 3MN; pulse
duration = 0-04 s; rise time = -200 10"6 s.
446 C. ALBERTINI, J. P. HALLEUX AND M. MONTAGNANI
4. INTRODUCTION
5. FIELD EQUATIONS
5.1. Kinematics
When a continuum undergoes deformation, the particles (material
points) of the continuum move along various paths in space. This motion
may be expressed by equations of the form
x = x(X,f) (6)
5.2. Kinetics
The establishment in the Lagrangian description of the equilibrium
equations is rather awkward. This is due to the 'Eulerian' character ofthe
stresses.
If xis the spatial position where the stress is experienced and we choose
as independent variable (spatial or Eulerian description), one easily shows
that the equilibrium equations are:
cj.j+f=0 and cu = cM (9)
where c,} are the true stresses (Cauchy stress tensor);/,' are the internal
volumetric forces (including intertiai forces); and _} designates the derivative
with respect to x. Now, transcribing these equations to a Lagrangian
description involves the Jacobian matrix ofthe transformation between
and X and requires a choice for the stress tensor. Depending on the stress
definition, various forms of equilibrium equations are obtained.
Unfortunately, none of them is simple. The most convenient for our
purpose seems to be
(o,j + akjuLk) j + F, ~ 0 and ,} = (10)
where , is the second PiolaKirchhoff stress tensor, obtained through
ij
i,j.n nin
6. PROBLEM DEFINITION
7. CONSTITUTIVE RELATIONS
lead to the field equations : they characterise the structure, not the material.
As a matter of fact, it is clear that for these relations also a certain
constitutive choice has been made concerning the material. For instance, it
was mentioned that the material must be continuously divisible without
losing its defining properties. However, one prefers to look at the field
equations as the fundamental mathematically derived equations of a
theoretical discipline (structural mechanics) dealing with the description of
problems involving various subclasses of'structural" materials, each of them
being defined by its constitutive equation. This explains that constitutive
relations have to fulfil various conditions in order to be consistent with the
general theory they fit.
In order to ensure reasonable solutions to correctly posed problems,
some mathematical principles have been studied for a long time. ' 2 1 3 They
can be used to help in formulating proper constitutive equations, but until
now no exhaustive reliable procedure for 'constititutive relations
construction' exists. It is, of course, not our purpose here to enter into the
axiomatic foundations of constitutive equations.
A constitutive equation defines an ideal material and therefore is a
mathematical model for a particular class of materials encountered in
nature rather than an exact representation for a specific unique material.
Thus, in practice, to simulate a real material the analyst first chooses an
analytical model suited to some general features (those he is interested in) of
the material and then tunes the parameters ofthe model in order to fit the
real material as accurately as possible.
We only considered purely mechanical aspects until now. It is most
probable that a separation between mechanical and thermal aspects is
unrealistic in the description of material behaviour in structural problems.
Other physical phenomena, such as electromagnetics, for instance, are
also correlated to mechanics, but are negligible in most cases. A better
structural model can be obtained by introducing thermal variables through
energy conservation, which involves thermomechanics.
Clearly, a constitutive relation should not depend on the actual structure
one studies: it is, by definition, common to all structures made out ofthe
same material. This suggests designing test structures for which it is easily
possible, by means of forces, torques, pressure, and so on, to produce some
desired stress patterns; adequate strain measures (by strain gauges) permit
then to find stress-strain relations that can be used to build up a constitutive
equation. For example, a common macroscopic experimental con-
figuration used in static plasticity is the combined traction, torsion and
internal pressure tubular test specimen with which it is possible to realise a
454 C. ALBERTINI, J. P. HALLEUX AND M. MONTAGNANI
large range of stress states, with the further advantage that the stress state is
about homogeneous in the specimen. 18
In transient dynamic behaviour large strain rates can be generated. If
many metallic materials exhibit only limited rate dependence when low
strain rates occur, this is not the case when strain rates go up to, say, 103/s.
This is particularly true for austenitic steels used in nuclear reactors.
Therefore, the constitutive relations obtained in static experiments have to
be adapted in order to simulate also the rate dependence. If the study'of
non-linear material behaviour in quasi-static situations is obviously a
complicated subject, what about dynamic situations?
The experimenter testing to determine the mechanical properties of'
materials under dynamic loading conditions encounters immediately at
least two difficulties : first, the obligation to measure in real time phenomena
of very short duration; secondly, the fact that disturbances due to wave
propagation effects will appear. The first will render the experiment rather
cumbersome to perform; the second will complicate the design of the
experiment.
Some practical difficulties are as follows: friction effects in the joints
present in the experiment ; unwanted reflection of waves on boundaries ; and
inertial effects due to non-continuous geometry (mainly concentrated
masses) which produce oscillations in the stress patterns.
It is therefore evident that the combined traction torsion pressure test
tube cannot be easily used in dynamics. Researchers have therefore tried to
design and study very simple configurations capable of producing one-
dimensional dynamic stress-strain curves (simple traction or compression).
Apart from the precursor work of Hopkinson (father 8 and son 9 ), extensive
work on dynamic properties of materials has only been done during the last
30 years. A survey of the work performed in Europe in the first part of this
period (until the 1960s) can be found in a paper by Cristescu; 1 ' it is also in
this period that Kolsky's famous book on stress waves in solids appeared. 1 0
Although it does not seem that much progress has been made in the
second period, 17 at least some more reliable reproducible results now exist.
Also, very helpful numerical experiments 23 have been performed with the
tools offered by the developing computing business. At present bi-
dimensional tests are also designed, but their use is at an early stage. More
about experimental aspects can be found in the first part of the chapter.
will thus have two distinct forms, depending on what regime it applies in:
(1) Linear elastic form Pijkl = Dijkl, the classical elastic matrix
depending only on material constants: E the elastic modulus,
Poisson's ratio. This form applies to OA or BC, for instance;
(2) An elastic-plastic form where F depends on material constants but
also on the current stress state. This type of behaviour is called
456 C. ALBERTINI, J. P. HALLEUX AND M. MONTAGNANI
Compared with other criteria (for instance. Tresca 18 ), the von Mises'
MATERIAL BEHA VIOUR A ND MODELLING 457
2 1
= X/M 2) + (2 3 ) 2 + 3 J (21)
>/2(1 + )
where is Poisson's ratio.
There are some difficulties in handling eqn. (21) in the elasticplastic
range; fortunately, the constitutive relation that we shall establish makes no
use of it. However, we shall have to introduce it for the strainrate
dependence.
Put in this form the problem consists in the determination of df.pm. This is
obtained through the law of normality:
=-^- (24)
this is unsatisfactory and one can only hope that the extrapolation has a
sense in real life.
Two-dimensional experiments have not delivered sufficient information
yet to be used in deriving a constitutive relation, but can be used in the
framework of numerical experiments performed with extrapolated
monodimensional constitutive relations so as to confirm the extrapolation :
this is current research.
Thus, in our view, at this stage it is best to try to keep out as much as
possible of the characteristics of static elastic-plasticity, trying to adapt
only some quantitative parameters: a strain-rate-dependent elastic-plastic
material appears, then, as belonging to the subclass of elastic-plastic
materials. From a computational point of view it would certainly be nice to
use relation (25) where only K depends on the strain rate.
The test features mentioned above suggest the consideration of the
strain-rate effect as a simple additional hardening; let us write a formal
constitutive relation, capable of describing the monodimensional tests
during plastic loading:
da = + ' (27)
where da is the stress increment; d is the strain increment; dt; is the strain
rate increment; and Ep and E' are, respectively, the hardening at constant
strain rate and the hardening at constant strain, both depending on the
actual state of the material.
In order to simulate an increment of stress by using an adapted 'static'
relation, one needs the form
da = Epdt: (28)
where E'p becomes an equivalent plastic modulus. Now,
d
E'p = Ep + E' (29)
dt:
there is no evidence that strain rate should follow the same rule. A more
attractive alternative missing, we arbitrarily choose to represent strain rate
as we represent strain (eqn. 21). Owing to the different Poisson's ratios in
elastic (0-3 for steel) and plastic (0-5) deformations, it is more convenient to
write:
3 = a v /(d8 1 - 2) + (2 - 3)2 + (3 - derf (30)
During elastic-plastic deformations the elastic part (eqn. 1 22) is often
very small compared with plastic deformation (plastic modulus <sc elastic
modulus), and one can use:
3- V(E> - )2 + (2 - 3 ) 2 + (3 - .) 2 (31)
Strain-rate comparison values are obtained by simply replacing e by ' in
eqns. (30) and (31).
Of course, more extended expressions for dynamic constitutive relations
where the strain rate dependence appears explicitly in all components can
be derived; in particular, this is easily done for Prager's kinematic
hardening model by simply adding strain-rate dependence to Prager's
hardening rule. The fundamental difficulty encountered above remains: it is
not known how strain-rate effects in various directions combine, and
therefore the specific coefficients of the strain rate components cannot be
correctly derived.
8. C O M P U T A T I O N A L ASPECTS
1
Calculate the comparison stress a"
-YES-
NO
YES NO
d(T
ij=Dijkl^kl
Calculate equivalent
plastic modulus E'p
_n _n 1
+ da,
462 C. ALBERTINI. J. P. HALLEUX AND M. MONTAGNANI
REFERENCES
1. Davies. R. M. (1948). 'A critical study ofthe Hopkinson pressure bar". Phil.
Trans. Roy. Soc, London. Ser. A. 240, 375.
2. Albertini. C. and Montagnani, M. (1976). 'Wave propagation effects in
dynamic loading'. Nue!. Eng. Des.. 37, 115-24.
3. Lindholm. Ulric S. (1971). "High strain rates tests', in Techniques of Metals
Research. Vol. 5. Part 1. Wiley. New York.
4. Albertini. C. and Montagnani, M. 'Testing techniques based on the split
Hopkinson bar". Inst, of Phys.. Conf. Ser. No. 21. London, 1974; Oxford. 2-4
April. 1974.
5. Kolsky. H. (1953). Stress Waves in Solids. Clarendon Press. Oxford.
6. Campbell. J. D. (1973). 'Dynamic plasticity: macroscopic and microscopic
aspects'. Mat. Sei. Eng.. 12, 3-21.
7. Lindholm. U. S. and Yeakley. L. M. (1968). 'High strain rate testing: tension
and compression', J. Exp. Mech.. January.
8. Hopkinson. J. (1872). Collected Science Papers. Vol. 2. p. 316.
9. Hopkinson, B. (1905). 'The effects of momentary stress in metals'. Proc. Roy.
Soc. ., 74, 488-506.
10. Kolsky. H. (1953). Stress Waves in Solids, Clarendon Press, Oxford.
MATERIAL BEHAVIOUR AND MOD ELLING 463
Earthquake
d'Alembert excitation, 59-65
force, 119 loads, 381
principle, 100 resistance. 349
INDEX 467
Variational equation, 25
Variational form of continuity, 237 Wall pressure fluctuations, 316-17
Variational methods, 1-41 Wave
numerical applications. 31-40 equations, 72, 80
Variational principle, 1 propagation problems, 126
based on complementary energy, 19 Wilson 0 method, 129-32, 139-40,
for small and large displacements in 149, 150
elastodynamics. 10-18
of elastodynamics, finite element
implementation of, 18-31 Yield criterion, 175
CONTENTS