Você está na página 1de 24

Experimental Identification of the Nonlinear

Parameters of an Industrial Translational Guide


for Machine Performance Evaluation1
JASPREET S. DHUPIA
A. GALIP ULSOY
REUVEN KATZ
NSF Engineering Research Center for Reconf igurable Manufacturing. System University
of Michigan, Ann Arbor, 2250 GGBL, 2350 Hayward Street, Ann Arbor, MI 48109, U.S.A.
(jdhupia@umich.edu)

BARTOSZ POWALKA
Technical University of Szczecin, Piastow 19, 70-310 Szczecin, Poland
(Received 8 August 20061 accepted 30 March 2007)

Abstract: Prediction of machine dynamics at the design stage is a challenge due to lack of adequate methods
for identifying and handling the nonlinearities in the machine joints, which appear as the nonlinear restoring
force function of relative displacement and relative velocity across the joint. This paper discusses identification of such a nonlinear restoring force function for an industrial translational guide for use with the
Nonlinear Receptance Coupling Approach (NLRCA) to evaluate machine dynamic characteristics. Translational guides are among the most commonly used joints in machine tools. Both parametric and nonparametric
techniques have been employed to identify the nonlinearities. A novel parametric model based on Hertzian
contact mechanics has been derived for the translational guide. A nonparametric method based on twodimensional Chebyshev polynomials is also used. The models derived from the two techniques, i.e., parametric and nonparametric, are fitted to the experimental data derived from static and dynamic tests to get the
restoring force as a function of relative displacement and relative velocity across the joint. The nonlinear representation obtained from both techniques is later converted into the describing function representation which
is needed for evaluation of machine dynamic characteristics using the NLRCA. The describing function representations obtained from the two approaches are compared. The design of experiments for evaluating the
nonlinearities in such industrial machine tool joints is a challenge, requiring careful alignment and calibration, because they are typically very stiff. This constrains the dynamic experiments to be carried out at high
frequencies (e.g. 20007000 Hz) where the experimental readings are very sensitive to errors in geometry
and calibration.

Keywords: machine dynamics, nonlinear joints, receptance coupling, frequency response functions.
1

This paper was contributed by Professor N. G. Chalhoub

Journal of Vibration and Control, 14(5): 645668, 2008


1 2008 SAGE Publications Los Angeles, London, New Delhi, Singapore
2

Figures 312, 14 appears in color online: http://jvc.sagepub.com

DOI: 10.1177/1077546307081325

646 J. S. DHUPIA ET AL.

1. INTRODUCTION
Accurate prediction of machine dynamics at the design stage, before the machine is constructed, is an important goal, which has become even more important with introduction
of high-speed and high-performance machines. Research has been done on automatically
generating the possible machine configurations for a given kinematic task (Moon and Kota,
2002).However, the related work (Moon, 2000) on finding the dynamic performance of each
machine configuration is preliminary and is based on the linear receptance coupling theory
(Bishop and Johnson, 1960). The nonlinearities present in the machine structure make it
difficult to predict machine dynamics. Most current techniques are limited by linear model
assumptions, and are not suitable for handling nonlinearities such as those arising in joints.
Research has been done to formulate the effects of nonlinearities on machine dynamics in
the frequency domain (Ferreira and Ewins, 19961 Yigit and Ulsoy, 2002). However, identification of the nonlinear parameters in machine tool joints in a manner suitable for use
with frequency domain techniques can be viewed as a major barrier that prevents successful
industrial application of such techniques.
The load experienced at the machine tool joints appears as the restoring force developed
due to joint deflections. The nonlinearity in the joint manifests itself as a nonlinear restoring
force function of relative displacement and velocity at the joint. Much work has been done to
model the stiffness and damping components of this restoring force. The stiffness has been
modeled based on material properties along with Hertzian mechanics (Johnson, 1982) or the
asperity model (Greenwood and Williamson, 1966). The energy dissipation, or damping,
can be modeled in many ways. While sliding motion in a translational slide may need to
be modeled through the static or dynamic friction models (Dahl (Dahl, 1976), LuGre (de
Wit et al., 1995) and Lueven (Swevers et al., 2000)), our work is focused on the dynamic
properties in the normal direction. The damping is quite small and a viscous damping model
was experimentally found to be sufficient, as will be described later.
System identification techniques can be broadly classified into parametric or nonparametric methods. While parametric methods seek to determine the values of the parameter in
an assumed model structure for the system to be identified, nonparametric methods seek to
determine the functional representation as well as parameters of the system to be identified.
Most of the parametric and nonparametric identification methods employ the least squares
approach, in which the square of the errors between the measured response and that of the
identified model is minimized, to provide the best estimate. The most commonly used nonparametric methods employ the Volterra series (Lee, 1997) and two-dimensional Chebyshev
polynomials to represent the restoring force surface (Masri et al., 1982a,b1 Worden and Tomlinson, 2001). Most of the parametric identification methods are time based (Kapania and
Park, 19961 Mohammad et al., 19921 Yasuda and Kamiya, 1999). Time-domain techniques
have the advantage of requiring less time and effort for data acquisition than the sine-dwell
techniques used for frequency domain identification, and can be used for the identification
of strongly nonlinear systems. Potential drawbacks of such time-domain approaches include
problems of differentiating noisy signals and inability to accurately estimate the coefficient
of terms which are small (Malatkar and Nayfeh, 2003).Frequency domain techniques include
approaches based on the backbone (or skeleton) curve and limit envelope (Benhafsi et al.,
19951 Fahey and Nayfeh, 1998) and harmonic balance method (Yasuda et al., 1997).

MACHINE PERFORMANCE EVALUATION 647

This paper considers identification of the nonlinear restoring force function parameters
for an industrial translational guide and its subsequent conversion to describing function
representation (e.g., Atherton, 1975) for use with Nonlinear Receptance Coupling Approach
(NLRCA) (Ferreira and Ewins, 19961 Yigit and Ulsoy, 2002) to evaluate the dynamic characteristics of a machine structure. The paper considers both a parametric and a nonparametric
approach to identify the joint parameters. In the parametric approach, a nonlinear stiffness
relationship has been derived for translational guides with preloaded ball bearings based on
Hertzian mechanics. The parameters for the model are experimentally obtained from static
and dynamic tests on the translational guide. The results from this approach are compared
with a generalized nonparametric approach using a two-dimensional Chebyshev polynomial
developed by Masri, Sasri and Caughey (Masri et al., 1982a,b). The paper emphasizes the
design of experiments for estimating the nonlinearities in the translational guide. This is a
challenge because the machine tool joints are typically very stiff. Most literature on estimation of nonlinearities in the joints considers systems with low natural frequency (less than
50 Hz), while the isolated machine tool joints (e.g. translation guide) have high natural frequencies (e.g., 2000-7000 Hz). Thus, the dynamic experiments are constrained to be carried
out at high frequencies, where only small displacements may be obtained and therefore the
system is very sensitive to alignment.
Subsequently, the paper discusses conversion of the experimentally obtained nonlinear
restoring force function in the time domain to a describing function representation in the
frequency domain. The parametric approach for nonlinear joint identification will typically
yield a system of equations describing the relationship between the restoring force and relative displacement and relative velocity across a joint. As is the case here, it may not always
be possible to analytically convert such a representation into a describing function. Thus,
one may compute the describing function numerically for use with the NLRCA for evaluating system dynamics. However, a nonparametric approach, employing the two-dimensional
Chebyshev polynomials for nonlinear joint identification, yields a polynomial representation
for the nonlinear restoring force function. This paper presents a derivation for obtaining the
closed form describing function for such a nonlinear restoring force function. Finally, the
evaluated describing functions are compared in the frequency domain within the frequency
range of interest. The effect of these nonlinearities in the joints on overall machine performance is beyond the scope of this paper, but is described separately in Dhupia et al. (2007).

2. TRANSLATIONAL GUIDE MODELING FOR PARAMETRIC


ANALYSIS
2.1. Geometrical Description of Translational Guide

Translational guides were chosen for experimental evaluation as they are among the most
common joints in machine tools. The translational guide chosen for the experiment was a
Bosch-Rexroth linear guide system: R1621, size 30. The cross-section of the translational
guide used for the experiment is shown in Figure 1. The joint consists of two components,
the rail and the runner block and the contact is made through the preloaded ball bearings. A
model for joint stiffness under normal load P is developed using Hertzian mechanics in this

648 J. S. DHUPIA ET AL.

Figure 1. (a) Cross-section of the translational guide, (b) Detailed view of ball bearing contact.

section. The model accounts for joint stiffness as a property based on material properties,
geometry and preload, but can be directly estimated from experiments.
2.2. Hertzian Contact Model for Translational Guide

The model is derived based on the assumption that the runner block and the rail are rigid
and all compliance in the system may be attributed to the ball bearings. The ball bearings
are located in four tracks as shown in Figure 1. The lower track bearings are denoted with
subscript L and the upper track by subscript U. The objective of the model is to describe the
relationship between external load P and relative displacement 1z, taking into account the
stiffness relationship at the ball bearings:
P 3 f 21z3
or ,
1z 3 f 41 2P3
where, 1z 3 z 1 4 z 2 4

(1)

MACHINE PERFORMANCE EVALUATION 649

The contact deformation of the single ball within two grooves can be found by Hertzian
analysis as dependent on the normal load Pball as (Rivin, 1999).
1
dball 3 2 5 1441n 1

2
Pball

2R1 4 R2
R1 R2

1 4 5 21 1 4 5 22
6
E1
E2

32
(2)

where,
dball 3 Deformation of a single ball
Pball 3 Force acting on a single ball
R1 3 Radius of the ball
R2 3 Radius of the groove
E 1 6 5 1 3 Youngs modulus, Poisson ratio of ball material
E 2 6 5 2 3 Youngs modulus, Poisson ratio of groove material
n 1 3 Parameter that depends on the ratio R1 and R2 . If R2 7 8 (i.e. ball in contact
with flat surface), then n 1 7 1
Or, the deformation-load relationship in equation (2) may be defined simply by the model
283
dball 3 7 Pball

(3)

where, 7 depends upon material and geometrical properties of the joint.


Since the bearings are preloaded, a nominal deformation d0 is present even when no
external load P is applied. When load P is applied, the lower set of bearings experience
deformation d L and the above set of bearings experience deformation dU . When, the balls in
all tracks are in contact with the grooves, the amount of compression the balls in the lower
tracks experience must be equal to the amount of decompression the balls in upper tracks
experience (see Figure 1) and they can be related to the relative displacement 1z through the
following equation:
2d L 4 d0 3 sin 9 3 2d0 4 dU 3 sin 9 3 1z4

(4)

Let PL be the normal load experienced by the balls in the lower track and PU be the
load experienced by the balls in the upper track. Then, the relationship between the joint
deflection and the normal loads developed at the ball bearing may be obtained by using
equation (3) and equation (4):
4
5
4
5
283
283
283
283
1z 3 7 PL 4 7 P0
sin 9 3 7 P0 4 7 PU sin 94

(5)

Finally, to relate normal loads at the ball bearing (PL and PU ) to external load P, we
consider the free body diagram of the runner block and equate the forces in the vertical
direction to get the relationship:
P 6 2PU sin 9 3 2PL sin 94

(6)

650 J. S. DHUPIA ET AL.

Using equation (6) to substitute for PU in equation (5), we obtain the relationship between load at the lower bearings, PL and the external load, P. This equation may be used to
evaluate PL
2

2PL sin 9 4 P
2 sin 9

3283
283

3 2P0

283

4 PL 4

(7)

Equation (7) is solved numerically for PL as a function of the external load P and preload
P0 . Putting the resulting PL into equation (5), the relationship between external load P and
joint deflection 1z is obtained which is valid when both upper and lower tracks balls are in
contact.
Similarly if PL is substituted in equation (5), using equation (6)
2
PU283

2P0283

2PU sin 9 6 P
2 sin 9

3283
4

(8)

Equating, PU 3 0, we get
9
P 3 4 2P0 sin 94

(9)

If P exceeds this value, the balls in the upper track will cease to be in compression
and the joint stiffness is provided by the balls in the lower track only, i.e., the relationship
between P and PL becomes P 3 2PL sin 9. Putting this result into equation (5) gives the
relationship between external load and joint deflection which is valid when only balls in the
lower track are in contact.
Our parametric analysis assumes the model structure described by equations (5)(9).
The difference between the experimental and analytical results is attributed to any errors
associated with geometrical and material properties as well as the change in preload from
the factory settings, which is expected from regular wear and tear of the slide. Therefore, the
values of 7 and P0 are estimated from experimental data for 1z vs. P.

3. NONPARAMETRIC ANALYSIS USING TWO-DIMENSIONAL


CHEBYSHEV POLYNOMIALS
A procedure described by Masri et al. (1982a,b) has been used for estimating the governing
nonlinear joint relationship when the joint dynamics can be represented by a single degreeof-freedom system. It was assumed that only the translational motion is present between
the slide and the runner block of the translational guide. Thus, there is no relative rotation
between the slide and the runner block and the joint can be modeled as a single degreeof-freedom system as shown in Figure 2(a). The restoring force which is a function of
relative displacement and velocity may be computed using equation (10), where z
2 is directly
measured by accelerometers mounted on the rail.
f 21z6 1 z 3 3 4m 2 z
2 4

(10)

MACHINE PERFORMANCE EVALUATION 651

Figure 2. (a) Mass-spring-damper representation for the relative motion of the translational guide, (b)
Mass-spring-damper representation of translational guide with steel block.

The translational guide in itself has a very high natural frequency for normal translational
mode (e.g. 60007000Hz). However, as the dynamic experiments are carried out at higher
frequencies, the displacements observed reduce significantly. This is because the energy
required for same displacement is proportional to
2 where the dynamic experiments are
carried at the input excitation frequency
. However, high displacement range is desired
for nonlinear parameter estimation to capture the nonlinear trend effectively. Therefore, the
natural frequency of the system is lowered by attaching a steel block (see Figure 2(b) and
Figure 8) of mass m3 to the rail. Due to the high frequency nature of the experiment, the rail
and the steel block cannot be assumed to act as a single rigid body. Thus, the experimental
evaluation of the restoring force function is modified as:
f 21z6 1 z 3 3 4m 2 z
2 4 m 3 z
3

(11)

where, z
3 is the measured acceleration of the steel block (Figure 2(b)). This equation may be
used to calculate the restoring force value at every relative displacement and velocity value
found through experiment. Once the surface is obtained, the restoring force may be expressed
as a nonlinear function expanded in the form of two-dimensional Chebyshev polynomials:
f 21z6 1 z 3 3

Ny
Nx 6
6

Ci j Ti 21z3 T j 21 z 3

(12)

i30 j30

Ti and T j being the ith and jth order Chebyshev polynomials, the Ci j being the associated
coefficient in Chebyshev expansion of the function.

652 J. S. DHUPIA ET AL.

The property of orthogonality may be used to calculate the Ci j coefficients, which give
results similar to the minimax polynomial fit, where the largest deviation in error is made
smallest (Worden and Tomlinson, 2001). However in this research we use least square estimates using the Chebyshev polynomial basis. The advantage of the Chebyshev basis function is that we can avoid the ill-conditioned matrices that are obtained when fitting regular
polynomial basis with higher order.

4. EXPERIMENTAL PROCEDURE AND RESULTS


4.1. Hertzian Model Fit to Static Experiments

While static experiments were done to determine the parametric model, dynamic experiments
are needed to determine damping properties and for the nonparametric analysis. It was found
experimentally that the net energy loss due to damping per cycle was less than 2% of the
maximum potential energy, and thus damping had an insignificant role in the overall system
dynamics. This low damping is because it is of mostly material nature rather than frictional.
Frictional damping is commonly observed in bolted or riveted structures and is much larger.
Also, for these experiments the translational guides did not have any lubricant, which may
increase damping. Therefore, the viscous damping model was assumed sufficient to model
the joint damping when using Hertzian analysis. Static experiments allow large input forces
to be transmitted to the joint, and allow an understanding of the overall nonlinear relationship
between the restoring force and displacement. The restoring force function, f 21z6 1 z 3, can
be related to the force exerted during the static experiments by:
f 21z6 1 z 3 3 P 21z3 6 c1 z

(13)

where, P is the restoring force developed due to static deflection or due to stiffness alone and
c1 z represents the contribution due to the viscous damping.
The translational guide was clamped inside a two-piece vise (Figure 3). A Brinell calibrator was used to measure the amount of force transmitted to the joint. Two displacement
sensors were aligned symmetrically to measure displacement on either side of the runner
block to ensure that only normal translation was observed when external load is applied.
Care was taken to minimize any rotation effects in the joint and to ensure a uniform distribution of force on the joint surface.
The experimental results and the estimated nonlinear stiffness curve are shown in Figure 4. These results were obtained by sweeping the preload values and finding 7 using a
least square estimate. The Root Mean Square (RMS) error for each fit is calculated and the
fit giving the minimum RMS error is chosen (Figure 5). The parameters found for the given
results were P0 3 630 N, 7 3 448403 5 1047 m/N283 with a RMS value of 40.1 N.
4.2. Modal Simulation and Experiments

The formulation for both parametric and nonparametric methods assumes that the rail-runner
block system responds only in the translational direction when an external load P is applied

MACHINE PERFORMANCE EVALUATION 653

Figure 3. Top view of the experimental setup for determining static stiffness.

Figure 4. Comparison of analytically derived Hertzian model for joint with the static experimental data.

654 J. S. DHUPIA ET AL.

Figure 5. Variation in the RMS error as preload is varied for the analytically derived Hertzian.

Figure 6. (a) Three degrees-of-freedom model for translational guide, (b) Translational mode along
Z -axis at 4800 Hz, (c) Rotational mode around Y -axis at 2882 Hz, and (d) Rotational model along
X -axis at 2656 Hz.

to it. This is correct, when the system is symmetrical and free from any imperfections and
the load P is applied through the center of mass of the rail-runner block system. However,
in reality the load P can cause rotational motion around the X and Y-axes apart from the
translational motion along the Z-axis (Figure 6). Thus, modeling these three degrees-offreedom, i.e., relative translation about Z-axis and relative rotation around X and Y-axes
between the rail and the runner block is needed. Therefore, a corresponding six degrees-offreedom model, which included the translation displacement and the two rotational motions

MACHINE PERFORMANCE EVALUATION 655

for both the rail and the runner block was developed for simulation (for the purpose of this
simulation, the steel block was assumed to be rigidly connected to the rail) and later verified
experimentally to determine the natural mode corresponding to the translational response of
the system. The combined system will have three rigid body modes and three relative motion
modes. One of the relative motion modes corresponds to the relative translation between the
rail and the runner block. The dynamic experiments for estimating the nonlinearities of the
joint may be done around this natural frequency as the single degree-of-freedom assumption
for parametric and nonparametric model can be well justified near this natural frequency.
The various motions that need to be modeled are the translational displacements z 1 and
z 2 of the runner block and rail respectively and the angular displacements x 1 and x 2 around
the X-axis and y 1 and y 2 around the Y-axis respectively. The mass of the runner block, the
rail and the steel block are denoted by m 1 , m 2 and m 3 . These were measured and found to
be m 1 3 04942 kg and m 2 6 m 3 3 242 kg. The corresponding moment of inertia around
the X-axis was evaluated as Ix1 3 549927 5 1044 kg-m2 and Ix2 6 Ix3 3 746785 5 1044 kgm2 . Similarly, around the Y-axis they are evaluated as I y1 3 941761 5 1044 kg-m2 and
I y2 6 I y3 3 664867 5 1044 kg-m2 . The effective stiffness due to the ball bearings is k, and is
assumed to be distributed symmetrically at the four points as shown in Figure 6. The length
between equivalent springs along the X-axis is l x 3 1245 mm and along the Y-axis is l y 3 21
mm. After preliminary investigation of the joint, k was chosen to be 600 N/ m. The system
dynamics can be represented by the mass-spring system equation as
M
x 6 Kx 3 f
where,
x3

z 1 x1 y1

(14)

z 2 x2 y2

8T

is the state vector,


M 3 diag

m1

Ix1

I y1 m 2 6 m 3

Ix2 6 Ix3

I y2 6 I y3

is the mass matrix, and,


9

0
0
l x2 k

0
0
l y2 k

K3

0
0

4k84

0
0
4l x2 k

0
0
4l y2 k

4k

4l x2 k

l x2 k

is the stiffness matrix. The input force vector for this system is
f3

0 0 0

0 0

8T





2
4l y k



0


0

l y2 k
0

656 J. S. DHUPIA ET AL.

Figure 7. The frequency response magnitude of the translational guide from sweep sine tests.

It may be noted that the ideal system is decoupled in translation and for each rotational
direction. However, as mentioned earlier, the imperfections in the joint and excitation of
the system causes response in the rotational directions also. Modeling them is necessary to
verify that these do not lie close to the natural frequency corresponding to the translational
motion, in which case the single degree-of-freedom approximation will no longer be valid
even near the natural frequency corresponding to the translational mode.
The system has six eigenvalues, three of which correspond to the rigid body motions and
are zero. The remaining eigenvalues correspond to rotation around the X-axis at 2656 Hz,
rotation around the Y-axis at 2882 Hz, and translational motion along the Z-axis at 4800 Hz.
The natural frequency for translation was experimentally verified by doing a sine sweep
test and the frequency response magnitude is shown in Figure 7. The natural frequency
corresponding to the translational mode along Z-axis is found to be 4819 Hz from the sine
sweep test.
4.3. Hertzian Model Fit to Dynamic Experiments

The Hertzian model derived from the static test, along with the viscous damping coefficient
obtained from the data is used to obtain the restoring force function. Since the energy loss per
cycle is less than 2% of the maximum potential energy, damping has little role in the overall
dynamics of the joint and can be modeled adequately using a viscous damping assumption.
The setup for the dynamic test is shown in Figure 8. An electromagnetic shaker is used to
excite the translational guide at 4800 Hz which is close to the natural frequency of normal

MACHINE PERFORMANCE EVALUATION 657

Figure 8. Experimental setup for dynamic tests.

translational mode. Since displacement and velocity are orthogonal, the viscous damping c
may be calculated as:

c3

f 21z6 1 z 3 41 z
 2
3 503N 4s8m
1 z

(15)

where n is the number of data points. Figure 9 shows the evaluated restoring force surface
from the Hertzian model vs. the experimental data from the dynamic test used to derive the
viscous parameter for parametric analysis.
4.4. Polynomial Fit to Dynamic Experiments using Chebyshev Polynomials

The experimental setup for nonparametric analysis of the translational joint was same as
that for determining the viscous damping coefficient in parametric analysis (see Figure 8).
The experimental results described here were carried out at 4800Hz, the natural frequency
attributed to the normal translational motion. The amplitude of the input excitation was
manually changed continuously during the experiment. This is done to distribute the data
points over the entire velocity and displacement range and hence improve the quality of
estimation of nonlinear restoring force function. A third order polynomial fit in displacement
and first order polynomial fit in velocity was chosen as it achieved most of the improvement
in the Root Mean Square (RMS) of residues. The evaluated nonlinear restoring force function
is as follows:

658 J. S. DHUPIA ET AL.

Figure 9. Experimental data and evaluated nonlinear restoring force surface from parametric analysis.

f 21z6 1 z 3 3 47046689 6 65441640 1z 4 28458201z 2 4 1470831z 3


6 463406201 z 4 22437441z1 z 4 19484151z 2 1 z 6 44430151z 3 1 z (16)
where 1z is in m and 1 z is in mm/sec. The RMS error was evaluated by calculating the
square root of the mean of the sum of squares of the residues at each data point. The RMS
error of the polynomial fit is 166N, which is less than 3% of the maximum of restoring force
measured. The experimental data used for determining the restoring force function and the
evaluated restoring force function are shown in Figure 10.

5. NONLINEAR RECEPTANCE COUPLING APPROACH USING


DESCRIBING FUNCTIONS
After determining the nonlinearities in the translational guide these characteristics are then
used to evaluate the dynamic characteristics of a machine structure. The relevant dynamic
characteristics of a machine structure can be obtained using the Frequency Response Functions (FRFs) and the Stability Lobe diagrams (SLDs) of the machine structure. However,
the SLD itself is found analytically by using the FRFs along with the information about the
cutting process. Thus, the evaluation of the FRF for the machine structure, including the
nonlinear joint, is discussed here. The displacement-force FRF is also referred to as receptance. The receptance for the machine structure can be found by the Nonlinear Receptance
Coupling Approach (NLRCA) described by Ferreira and Ewins (1996).

MACHINE PERFORMANCE EVALUATION 659

Figure 10. Experimental data for polynomial fitted nonlinear restoring force surface and evaluated
nonlinear restoring force surface.

The NLRCA represents the response of a post-coupled system to various input excitations based on relations determining response of the pre-coupled system to various input
excitations. The post-coupled system contains the pre-coupled system along with all the
connections. The local nonlinearities that may be contained within the connections are
approximated by describing functions to represent them in the frequency domain.
Two coordinate sets are defined for this approach: an internal coordinate set and a connection coordinate set. While those coordinates related to connections are in the connection coordinate set (referred by subscript n or N), the coordinates not related to connections
are in the internal coordinate set. The connection coordinates are always a union of a pair
of coordinates ( p j 6 q j ), which represent the locations across a connection. Index i refers to
the ith connection among a total of M. In the following equations lowercase indices are used
to represent the pre-coupled configuration, while uppercase indices are used to represent the
post-coupled configuration. The pre-coupled response of the system is defined through the
receptance matrices of the system and given by:

 9
xn 
Hnn






xp
3

H pn






xq
Hqn

Hnp

Hnq

H pp

H pq

Hqp

Hqq



fn 






f
 p 




fq

(17)

where Hmn is the receptance matrix between the coordinate sets m and n. In the post-coupled
system, the response can be written as

660 J. S. DHUPIA ET AL.



xN



xP



xQ









HN N

HP N

HQ N

HN P

HN Q

HP P

HP Q

HQ P

HQ Q


f

 N


  fP


fQ









(18)

Ferreira derived the relationship for the post-coupled relationship (equation 18) in terms
of the pre-coupled receptance matrices as

xN



xP



xQ






9

Hnn

Hnp

3 
 H pn



Hqn

2Hnp 4 Hnq 3

H pp

2H pp 4 H pq 3
H pq
4



Hqp

Hqq

2Hqp 4 Hqq 3

T  
 fN





2H
4
H
3
[B]41


pp
pq

   fP


fQ
2Hqp 4 Hqq 3
9

Hnq

2Hnp 4 Hnq 3









(19)

where
B 3 H pp 6 Hqq 4 H pq 4 Hqp 6 1 with 1 3 diag218G i 3

(20)

G i is the describing function (e.g., Atherton, 1975) representation for the local nonlinearities
in the connections. Under the assumption that the output of the nonlinearity to a harmonic
input is primarily dominated by the first harmonic, describing functions are a good approximation for the nonlinearity in the frequency domain. The describing function gain is the
fundamental component of the Fourier series representation of the periodic output obtained
from the nonlinear function to a sinusoidal input of frequency
. Let f i 21zi 6 1 z i 3 represents
the ith nonlinear restoring force function for a joint in the machine structure, where 1z i is
the relative displacement for the joint along the direction of the restoring force. Thus,
gi 2t3 3 f i 2A sin
t6 A
cos
t3

(21)

is the output to the sinusoidal input at the joint and can be expanded in a Fourier series. The
describing function is then evaluated as:
G i 2
6 Ai 3 3

Ai

gi 2t3 2sin
t 6 j cos
t3 dt4

(22)

In the subsequent sections, the describing function evaluation from the nonlinear parameters
obtained from the two different approaches, i.e., parametric and nonparametric approach is
considered. It will be shown, that the even though the describing functions obtained from the
two approaches are different because the translational guide is modeled differently, however,
they yield similar results when evaluating a machines response. The detailed procedure

MACHINE PERFORMANCE EVALUATION 661

to evaluate a machines dynamic response in terms of Frequency Response Functions and


Stability Lobe Diagrams is described in Dhupia et al. (2006, 2007).

6. DESCRIBING FUNCTION FROM PARAMETRIC ANALYSIS


RESULTS
The NLRCA approach for determining machine dynamic characteristics involve representing all modules by their pre-coupled receptance matrices H and have the governing nonlinear
restoring force relationships, f i 21zi 6 1 z i 3, of the interface between them. Equation 19 converts the nonlinear differential equations into a nonlinear algebraic approximation using the
describing function representation of nonlinearities at the joint. This nonlinear algebraic
system of equations needs to be solved using some numerical technique, e.g. the NewtonRaphson method. Use of the NLRCA to determine the receptance matrix of the structure
for different nonlinearities is described in (Ferreira and Ewins, 19961 Yigit and Ulsoy, 20021
Dhupia et al., 2006). All these have an assumed and simple analytical expression for nonlinear restoring force expression. However, as in this case, it may not be straightforward to
derive a simple analytical form from the evaluated nonlinearities for the joint, especially in
the parametric approach. This may happen because the parametric description of the nonlinearity cannot always be reduced to a simple nonlinear restoring force function description
described by one explicit equation. Thus, at every Newton-Raphson iteration step, the describing function has to be numerically evaluated for the chosen amplitude, which may be
obtained from the response variables 2xn 6 x p 6 xq 3 at that iteration and the chosen frequency
for the iteration. Figure 11, shows the describing function plot as the amplitude and frequencies are changed. At different amplitude levels of relative vibration amplitude, A, the
nonlinear stiffness of the joint leads to different DC gains. The roll-off frequency is determined by the stiffness-damping ratio. Since the damping is assumed to be viscous and the
change in stiffness is very small, the roll-off frequency,
c , remains almost a constant at
about 1.2 MHz.
Besides the numerical computation of the describing functions, the Newton-Raphson
technique to find a solution of nonlinear equations requires the computation of the Jacobian
of the system of equations describing the system (equation (19)). Because of the lack of
closed form representation of describing function this Jacobian has to be numerically computed by perturbing every variable in the connection coordinate set around the nominal value
of that iteration. The solution to this nonlinear equation is prone to the ill conditioning of
the Jacobian matrix and may require tuning of several parameters before the desired solution may be obtained. In the next section, the NLRCA approach using the nonparametric
analysis is discussed. Because of the representation of the nonlinearity in the form of a two
dimensional polynomial function, a closed form describing function can be obtained, which
avoids several of the above mentioned difficulties in applying the results obtained to the
NLRCA.

662 J. S. DHUPIA ET AL.

Figure 11. Describing function response for parametric analysis,


c 3 142 MHz.

7. DESCRIBING FUNCTION FOR NONPARAMETRIC


ANALYSIS RESULTS
The two-dimensional Chebyshev polynomial obtained from the nonparametric analysis can
be expanded as a two dimensional polynomial representation, as in equation (16), of the
nonlinear restoring force function as:
f 21z6 1 z 3 3

Ny
Nx 6
6

Cixjy 1z i 1 z j

(23)

i30 j30

where Cixjy represents the coefficient of the term containing ith power of relative displacement
and jth power of relative velocity across the joint. Substituting 1z 3 A sin
t and 1 z 3

A cos
t
g2A sin
t6 A
cos
t3 3

Ny
Nx 6
6

Cixjy
j Ai6 j sini 2
t3 cos j 2
t3 4

(24)

i30 j30

Thus, the describing function representation is given by:


G 2
6 A3 3

Ai


0

2sin
t 6 j cos
t3

Ny
Nx 6
6
i30 j30

Cixjy
j Ai6 j sini 2
t3 cos j 2
t3 dt4 (25)

MACHINE PERFORMANCE EVALUATION 663


Substituting 3
t
Ny
Nx 6
6
1 x y j i6 j41
G 2
6 A3 3
C
A
ij
i30 j30

2

i61

sin

cos d 6

sin cos
i

j61

d 4

(26)

To evaluate the above describing function consider the integral:




82

I 3

sinm cosn d 4

(27)

Substituting,  3 sin2 ,

I 3

m41
2

21 4  3

n41
2

d 4

(28)

The beta function (Kreyszig, 1999), B2x6 y3, is defined as:




 x41 21 4  3 y41 d

B 2x6 y3 3

and satisfies the property:

B 2x6 y3 3

 2x3  2y3
3 B 2y6 x3
 2x 6 y3

where, the gamma function (Kreyszig, 1999),  2x3 3


erties:

(29)
8
0

x41 e4 d and satisfies the prop-

 2x 6 13 3 x 2x3
 213 3 1, and
 21823 3

9
4

(30)

 82
Thus, the integral I 3 0 sinm cosn d , can be evaluated using gamma function properties for integers m and n, and may be written as:

I 3
0

82

2
sin cos d 3 B
m

m61 n61
6
2
2

3
3

 m61

 n61
2
2
4
24 m6n62
2

(31)

The describing function integral [equation (26)] can thus be evaluated using the result obtained in equation (31) as:

664 J. S. DHUPIA ET AL.

Figure 12. Describing function response from nonparametric analysis,


c 3 142 MHz.

G 2
6 A3 3

! 044N x 044N y
6 6

2
Cixjy
j Ai6 j41 B

i3odd j3een
044N
6x 044N
6y
i3een j3odd

2
Cixjy
j

Ai6 j41 B

i
j 61
6 16
2
2

3"
i 61 j
6 61
4
2
2

(32)

Thus, the describing function representation for the evaluated nonlinear restoring functions
using nonparametric analysis (equation (16)) is:
2
3
A2
3A2
46
46340620 4 1948415
6 j4
410
G 2
6 A3 3 65441640 4 147083 5
4
4

(33)

The describing function response from the nonparametric approach is shown in Figure 12.
In the nonparametric approach using Chebyshev functions, both the stiffness as well as
damping is modeled. While damping is quite small and leads to energy dissipation of less
than 2% of the maximum potential energy in the translational guide per cycle, the nonlinear
representation indicates that the damping function is itself quite nonlinear. This results in
varying roll-off frequencies for each of the describing function plots for different vibration
amplitudes. However, the small nonlinearity in stiffness, which affects the machine dynam-

MACHINE PERFORMANCE EVALUATION 665

Figure 13. Single degree-of-freedom system used for comparison of evaluated joint description.

ics significantly, affects the DC gain of the describing function response. The describing
function response plot from both approaches show similar trend till the roll-off frequency.

8. DISCUSSION
The nonlinear restoring force function has been evaluated using two different approaches,
i.e., parametric and nonparametric approach. Any joint may be evaluated by either of these
two techniques. The parametric approach requires considerable modeling effort and it may
be difficult to derive a physically relevant model for each joint. However, it is easier to
identify experimental errors from the parametric approach. Modeling of the joint also gives
considerable insight for the design of experiments. The nonparametric approach does not
require such extensive modeling and can be extended to different joints. But problems with
experiments may go unnoticed and yield unexpected or incorrect results later on. The modeling of the translational guide during the parametric approach is beneficial for the design
of experiment. However, with a good design of the experiment, the nonparametric approach
also gives good results. The objective of the nonlinear estimation of machine joints is that
the evaluated nonlinear parameters should be used to evaluate the machine dynamics at the
design stage. We are currently using a nonlinear receptance coupling approach to evaluate
the machine performance at the design stage. This approach uses the describing functions
to represent the joint nonlinearities. It is possible to obtain good closed-form describing
function representation for the polynomial fitted nonparametric approach presented in this
paper. This is an advantage because this approach may be used to solve for the dynamic
performance of machine tools efficiently and several problems regarding ill-conditioning of
Jacobian matrices and evaluation of each iteration step for the nonlinear algebraic equation
solver can be avoided.
The describing function evaluated for the translational guide using the two approaches
has a slightly different representation because of the different approaches that have been employed. In the Hertzian, or the parametric approach, the nonlinearity has been assumed only
in stiffness and damping has been assumed to be viscous. In the Chebyshev approach, or
the nonparametric approach, a two-dimensional polynomial function models nonlinearity in

666 J. S. DHUPIA ET AL.

Figure 14. (a) SDOF response with parametric describing function, (b) SDOF response with nonparametric describing function.

MACHINE PERFORMANCE EVALUATION 667

both stiffness and damping. Thus, while the describing function response at lower frequencies is quite similar, the higher frequencies response, i.e. after the roll-off frequency is quite
different. However this roll-off frequency is very high (
c 3 142 MHz) since the joint is
very stiff and the damping is quite small. To compare the effect of this describing function
on the machine structure a single degree of freedom system with the evaluated parameters
of the translational guide and a mass of 60 kg as shown in Figure 13 is considered. The
frequency response function for the SDOF system for parametric vs nonparateric approach
with constant displacement amplitudes vibration is shown in Figure 14(a) and (b). Despite
differences in the describing function response plot the FRFs of the SDOF system are very
similar. Even though the natural frequencies shift significantly for a small amplitude change
from 1 m to 5 m due to the nonlinearity, the difference between the describing functions
at high frequencies does not affect the SDOF system whose frequency range of interest is
around its natural frequency and has the range up to kHz.
It can be observed from the static experiment results (Figure 4) as well as subsequent
discussion of results in the frequency domain, that the relative displacement vs. normal load
follows almost a linear trend. The dominating linear terms can also be seen in the restoring
force function representation in equation 16. Thus the joint is only weakly nonlinear. However, it should be noted that even such small nonlinear terms as observed in this joint can
cause significant variations in the machine dynamic performance. This important point is
discussed in detail in (Dhupia et al., 2006) and is also evident from Figure 14.

9. SUMMARY AND CONCLUSIONS


The nonlinear parameters for restoring force function of an industrial translational guide
have been evaluated and an approach to use these parameters for measuring the machine
dynamic performance in the design stage has been described. A translational guide has been
selected because it is among the most common joints found in machine structures. The
joint was found to have a weak nonlinear stiffness term. Two approaches: a parametric
approach based on Hertzian mechanics and a nonparametric approach based on Chebyshev
polynomials has been used. The parameters are later converted into their describing function
representation for use with the Nonlinear Receptance Coupling Approach to determine the
machine dynamic characteristics in the design stage. An algorithm to find the closed-form
describing function from the nonparametric approach using polynomial representation is also
described.
While the modeling of the translational guide via the parametric approach is essential for
proper design of experiments, the generalized nonparametric approach is desirable because
of the possibility to extend it to different joints and the nonlinear parameter results from the
polynomial fitted restoring force function yield closed-form describing function which may
be used to efficiently evaluate the dynamic performance of machine tools at design stage.
Acknowledgments. The authors are pleased to acknowledge the f inancial support of the NSF Engineering Research
Center for Reconf igurable Manufacturing Systems (NSF grant # eec-9529125) and the Foundation for Polish Science.

668 J. S. DHUPIA ET AL.

REFERENCES
Atherton, D. P., 1975, Nonlinear Control Engineering, Van Nostrand, New York.
Benhafsi, Y., Penny, J. E. T., and Friswell, M. I., 1995, Identification of Damping Parameters of Vibrating Systems
with Cubic Stiffness Nonlinearity, in Proceedings of the 13th International Modal Analysis Conference,
Nashville, TN, pp.623 629.
Bishop, R. E. D. and Johnson, D. C., 1960, The Mechanics of Vibrations, Cambridge University Press, Cambridge.
Dahl, P. R., 1976, Solid friction damping of mechanical vibrations, AIAA Journal 14(12), 1675.
de Wit, C. C., Olsson, H., Astrom, K. J., and Lischinsky, P., 1995, New model for control of systems with friction,
IEEE Transactions on Automatic Control 40(3), 419.
Dhupia, J., Powalka, B., Ulsoy, A. G., and Katz, R., 2006, Effect of a nonlinear joint on the dynamic performance
of a machine tool, in Proceedings of CIRP Second International Conference on High Performance Cutting,
Vancouver, BC, Canada.
Dhupia, J., Powalka, B., Ulsoy, A. G., and Katz, R., 2007, Effect of a nonlinear joint on the dynamic performance
of a machine tool, ASME Journal of Manufacturing Science and Engineering 129(5), 943950.
Fahey, S. OF. and Nayfeh, A. H., 1998, Experimental nonlinear identification of a single structural mode, in
Proceedings of the 16th International Modal Analysis Conference, Santa Barbara, CA, February 25, pp.737.
Ferreira, J. V. and Ewins, D. J., 1996, Nonlinear receptance coupling approach based on describing functions, in
Proceedings of 14th International Modal Analysis Conference, Dearborn, MI, pp.10341040.
Greenwood, J. A., and Williamson, J. B. P., 1966, Contact of nominally flat surfaces, in Proceedings of the Royal
Society of London. Series A, Mathematical and Physical Sciences 295(1442), pp.1442.
Johnson, K. L., 1982, 100 years of Hertz contact, in Proceedings of the Institution of Mechanical Engineers, vol.
196, pp.363378.
Kapania, R. K., and Park, S., 1996, Parametric identification of nonlinear structural dynamic systems using finite
elements in time, in Proceedings of the 37th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics
and Materials Conference and Exhibit, Salt Lake City, UT, April 1517, pp.674.
Kreyszig, E., 1999, Advanced Engineering Mathematics, Wiley, New York.
Lee, G. M., 1997, Estimation of non-linear system parameters using higher-order frequency response functions,
Mechanical Systems and Signal Processing 11(2), 219.
Malatkar, P., and Nayfeh, A. H., 2003, A parametric identification technique for single- degree-of-freedom weakly
nonlinear systems with cubic nonlinearities, Journal of Vibration and Control 9(34),317.
Masri, S. F., Sassi, H., and Caughey, T. K., 1982a, Nonparametric identification of nonlinear systems, in Proceedings of the US National Congress of Applied Mechanics, Ithaca, NY, June 2125, pp.476.
Masri, S. F., Sassi, H., and Caughey, T. K., 1982b, Nonparametric identification of nearly arbitrary nonlinear
systems, ASME Journal of Applied Mechanics 49(3),619.
Mohammad, K. S., Worden, K., and Tomlinson, G. R., 1992, Direct parameter estimation for linear and non-linear
structures, Journal of Sound and Vibration 152(3),471.
Moon, Y.-M., 2000, Reconf igurable Machine Tool Design: Theory and Application, University of Michigan, Ann
Arbor, USA.
Moon, Y.-M. and Kota, S., 2002, Design of reconfigurable machine tools, ASME Journal of Manufacturing Science and Engineering 124(2), 480483.
Rivin, E. I., 1999, Stiffness and Damping in Mechanical Design, Marcel Dekker, New York.
Swevers, J., Al Bender, F., Ganseman, C. G., and Prajogo, T., 2000, Integrated friction model structure with
improved presliding behavior for accurate friction compensation, IEEE Transactions on Automatic Control
45(4), 675.
Worden, K. and Tomlinson, G. R., 2001, Nonlinearity in Structural Dynamics: Detection, Identif ication, and Modelling, Institute of Physics Publishing, Bristol, U.K.
Yasuda, K., Kamiya, K., and Komakine, M., 1997, Experimental identification technique of vibrating structures
with geometrical nonlinearity, ASME Journal of Applied Mechanics 64(2), 275.
Yasuda, K. and Kamiya, K., 1999, Experimental identification technique of nonlinear beams in time domain,
Nonlinear Dynamics 18(2), 185.
Yigit, A. S. and Ulsoy, A. G., 2002, Dynamic stiffness evaluation for reconfigurable machine tools including
weakly non-linear joint characteristics, in Proceedings of the Institution of Mechanical Engineers, Part B:
Journal of Engineering Manufacture, 216(1), 87101.

Você também pode gostar