Você está na página 1de 12

Tribology International 133 (2019) 55–66

Contents lists available at ScienceDirect

Tribology International
journal homepage: www.elsevier.com/locate/triboint

A new model for textured surface lubrication based on a modified Reynolds T


equation including inertia effects
Michael Rom∗, Siegfried Müller
Institut für Geometrie und Praktische Mathematik (IGPM), RWTH Aachen University, Templergraben 55, 52056, Aachen, Germany

ARTICLE INFO ABSTRACT

Keywords: For textured surface lubrication problems, a small texture aspect ratio and a large Reynolds number can lead to
Hydrodynamic lubrication strong recirculation in the textures and significant convective inertia effects. When these effects cannot be ne-
Surface texture glected, the Stokes system and the Reynolds equation are unsuitable and the Navier-Stokes equations should be
Simulation considered instead. Particularly compared with the Reynolds equation, solving the Navier-Stokes system is
Journal bearing
computationally much more demanding due to its higher spatial dimension and nonlinearity. In this work, a new
model is presented based on a modified Reynolds equation taking into account convective inertia and re-
circulation effects while so far neglecting cavitation. The model provides an accurate and considerably less
expensive alternative to Navier-Stokes, demonstrated for two- and three-dimensional numerical configurations.

1. Introduction recirculation in the textures the Reynolds equation is not applicable in


case of short and deep textures with < 10 , regardless of the Reynolds
For simulations in the field of hydrodynamic lubrication of textured number. However, for < 10 with Re < 1 up to Re < 4 (depending on
surfaces, often the Reynolds equation [1] is applied. It can be derived λ), see Fig. 5 in Ref. [2], the influence of convective inertia effects is
from the Navier-Stokes equations based on two key assumptions: (i) the small such that the Stokes system can be used, whereas for larger Re the
fluid film is thin such that the pressure does not vary in film thickness Navier-Stokes system is recommended. For increasing λ with > 10 ,
direction and (ii ) inertia can be neglected. Due to these simplifications the Reynolds number limiting the validity of the Reynolds equation also
there is a large range of applications in which the Reynolds equation is increases.
not valid as investigated in many numerical studies, e.g. Refs. [2–4], In this paper, we deal with textured surface lubrication problems
and summarized in the survey paper [5]. with texture aspect ratios smaller than ten and Reynolds numbers larger
The first problem with these studies is the term validity because than 50 for which the Stokes system and the Reynolds equation are
there is no sharp limit at which the Reynolds equation breaks down. clearly unsuitable such that the Navier-Stokes system has to be used.
Dobrica and Fillon [2] for instance investigated the relative differences Such a problem for instance occurs in high-speed machines such as
in local pressures and load-carrying capacities obtained from solving turbines or compressors in which often tilting pad bearings are em-
the Reynolds equation and the Navier-Stokes system. They declared the ployed [6]. Since solving the nonlinear Navier-Stokes system is much
Reynolds equation to be valid if both relative differences were smaller more expensive than solving the dimension-reduced linear Reynolds
than 10 % . The second problem is the large number of parameters influ- equation, we are interested in accelerating computations while ob-
encing the results, such as the texture shape and geometry (length, taining a high accuracy of the simulation results. For this purpose, we
width, height, aspect ratio, density) and the operating conditions developed an effective Navier-Stokes model [7] for computing accurate
(Reynolds number). These problems even led to contradictory results pressure distributions which contains three computational steps:
between different studies as for instance observed in Ref. [2]. To the
authors’ knowledge, an in-depth investigation of all parameters to- 1. Solve the original Reynolds equation for the investigated textured
gether in one study has not been done up to now. application,
In their two-dimensional study, Dobrica and Fillon [2] concentrated 2. solve a linearized Navier-Stokes (Oseen) cell problem for each tex-
on varying the texture aspect ratio λ (texture length divided by texture ture incorporating data of step 1 and
height) and the Reynolds number Re . They concluded that due to 3. solve the Navier-Stokes system for a smooth domain without


Corresponding author.
E-mail address: rom@igpm.rwth-aachen.de (M. Rom).

https://doi.org/10.1016/j.triboint.2018.12.030
Received 26 October 2018; Received in revised form 19 December 2018; Accepted 20 December 2018
Available online 27 December 2018
0301-679X/ © 2018 Elsevier Ltd. All rights reserved.
M. Rom, S. Müller Tribology International 133 (2019) 55–66

Nomenclature lz bearing width (only 3D)


lt , x texture length
α gradient angle due to eccentricity lt , z texture width (only 3D)
η dynamic viscosity of lubricant modRe index for modified Reynolds problem
λ texture aspect ratio n normal vector
ρ density of lubricant n degree of Newton-Cotes formula
t texture density nt total number of textures
τ shear stress nt , x number of textures in x-direction
boundary of computational domain nt , z number of textures in z-direction (only 3D)
0 side boundaries for 2D Reynolds problem p pressure
1, …, 6 domain boundary at x min , x max , ymin , ymax , zmin , zmax Q volumetric flow rate
eff effective domain boundary in effective problem Re Reynolds number
in inflow domain boundary Re index for original Reynolds problem
out outflow domain boundary ti node in Newton-Cotes formula
per periodic domain boundary u velocity vector with components u , v , w
computational domain u component of velocity in x-direction
c radial clearance u cp, i velocity solution of cell problem i
cp index for cell problem u surf surface velocity vector
d bearing diameter usurf,abs magnitude of surface velocity
dt , x texture distance in x-direction v component of velocity in y-direction
dt , z texture distance in z-direction (only 3D) w component of velocity in z-direction (only 3D)
erel relative eccentricity W load-carrying capacity
eff index for effective Navier-Stokes problem wi weight in Newton-Cotes formula
F friction force x coordinate vector with components x , y , z
h film thickness x coordinate in flow direction
hg gap height (without textures) y coordinate in film thickness direction
ht texture height ỹ transformed coord. in film thickness direction
L mesh refinement level z coordinate in width direction (only 3D)
lx bearing length

textures but with effective boundary conditions computed from data method, which is similar to a homogenization technique, see for in-
of step 2. stance Refs. [15,16], an averaged Reynolds equation with shear and
pressure flow factors is solved on the macroscale. These flow factors are
For the journal bearing configurations investigated in Ref. [7], we computed on the microscale by solving the Reynolds equation for local
obtained load-carrying capacities with our model deviating by at most cell problems. In contrast, de Kraker et al. solve the Navier-Stokes
1.03 % from the corresponding Navier-Stokes values. In addition, we system on the microscale to incorporate convective inertia effects into
observed significant speed-ups of up to the factor 8.7 compared with the modified averaged Reynolds equation on the macroscale. Hence,
solving the Navier-Stokes equations for the original textured problem. the approach is much more expensive than the original flow factor
However, a drawback of this approach is the third step which still re- method. Furthermore, to the authors’ knowledge, studies of the per-
quires solving the nonlinear Navier-Stokes equations though for a much formance of the method regarding the accuracy of macroscale simula-
simpler domain without textures. Our objective in this paper is to tion results or computational cost have never been published.
propose a simpler model by replacing the Navier-Stokes system in the Similarly to the approach by de Kraker et al., local cell problems
third step with a modified Reynolds equation which includes flow in- have to be solved in the second step of our effective Navier-Stokes
formation regarding recirculation and convective inertia effects. approach [7] and of our new model described below. However, due to
There are several approaches to incorporate convective inertia ef- solving the Reynolds equation for the textured problem in the first step,
fects into a modified Reynolds equation. Early works are those of we can incorporate information from the Reynolds solution to linearize
Constantinescu and Galetuse [8,9] who developed a popular model the Navier-Stokes system for the cell problems: the Reynolds velocity
which in addition to inertia effects can account for turbulence. Many computed from the gradient of the pressure solution enters the non-
extensions were published, e.g., to consider temporal inertia effects linear convective term in the Navier-Stokes system. In addition, we
[10,11]. Drawbacks of all these models are that they do not account for make use of the Reynolds velocity by prescribing it as boundary con-
textured surfaces and that they contain several empirical coefficients dition at the inflow boundaries of the cell problems. The cell problem
and mathematical assumptions regarding the integration of the pro- solutions provide pressure gradients which contain recirculation and
ducts of velocity components. inertia information and are used to set up a modified Reynolds equa-
Arghir et al. [12] presented a simple approach for cases with dis- tion. The latter is solved in the new third step of our approach. Since the
continuous film thickness, e.g., induced by textures, where inertia ef- modified Reynolds equation is the main ingredient, we refer to the new
fects are taken into account by computing a pressure correction from a approach as new Reynolds model in this work. The procedure contains
generalized Bernoulli equation. The validity of the Reynolds equation the following steps.
can be extended by this pressure correction, while the complexity of the
model is only slightly increased. However, as described in Ref. [2], the 1. Solve the original Reynolds equation for the investigated textured
extension is not applicable if the texture aspect ratio λ is smaller than application,
ten, whereas for = 10 it can be applied up to Re 30 and for = 100 2. solve a linearized Navier-Stokes (Oseen) cell problem for each tex-
up to Re 300 . ture incorporating data of step 1 and
The multiscale approach by de Kraker et al. [13] is based on the 3. solve a modified Reynolds equation for the textured problem in-
widely used flow factor method by Patir and Cheng [14]. In the latter corporating data of step 2.

56
M. Rom, S. Müller Tribology International 133 (2019) 55–66

The first two steps are the same as in our effective Navier-Stokes bearing surface is at the bottom ( 3, y ht , where ht is the texture
approach discussed above. height).
For the evaluation of our new Reynolds model, we investigate three The gap height hg is given by
different test cases: a two- and a three-dimensional journal bearing and
2x
a three-dimensional plane-inclined slider bearing. We qualitatively and hg (x ) = c 1 + erel cos ,
quantitatively compare Navier-Stokes, Stokes and Reynolds simulation
d (1)
results with those of our new model. Furthermore, we study the com- where c denotes the radial clearance, erel the relative eccentricity and d
putational performance of the new approach by comparing its simula- the diameter of the bearing. The film thickness h is then determined by
tion times with the Navier-Stokes simulation times. Even though in
practice usually texture aspect ratios > 10 are used, with < 10 we h g (x ) + h t (x , z ) at texture,
h (x , z ) =
can investigate whether our model is valid for extreme setups, in which h g (x ) otherwise (2)
case it will be suitable for the majority of applications.
Note that up to now we have not taken into account cavitation for any (x , y , z ) . With the gradient angle α defined by
because we have focused on the fundamental feasibility of our new h g (x )
Reynolds model. Hence, the model in its current form can only provide tan = ,
x (3)
physically valid results for lubrication problems without cavitation. An
extension would have to account for macro-cavitation in divergent gaps the rotational velocity of the shaft surface is given by
as well as micro-cavitation in the textures. As demonstrated in Ref. usurf,abs cos
[17], such a cavitation model would have to be mass-conserving be- u surf = usurf,abs sin ,
cause the textures lead to a repeated film rupture and reformation. For 0 (4)
the Reynolds equation, mass-conserving models are based on the JFO
cavitation boundary conditions published by Jakobsson and Floberg where usurf,abs is the surface velocity magnitude.
[18] and Olsson [19]. Implementations are often further developments
of the first algorithm efficiently making use of these boundary condi- 2.2. Original problem: incompressible Navier-Stokes system
tions which was presented by Elrod and Adams [20].
The structure of this paper is as follows. Section 2 introduces the For the setup of Sect. 2.1, the nonlinear stationary incompressible
setting which is used for the derivation of our new Reynolds model. Navier-Stokes equations
Section 3 gives a brief overview of our previously published effective
u = 0,
Navier-Stokes model which is the basis for the new Reynolds model. 1
The latter is derived in Sect. 4. For the quantitative evaluation of the (u )u + p Re
u=0 (5)
new approach, we are interested in characteristic values such as the
have to be solved for the velocity vector u with components u, v and w
load-carrying capacity, the friction force and the volumetric flow rate.
and the pressure p. The Reynolds number is defined by
Section 5 shows how these values can be computed. In Sect. 6, we
Re = usurf,abs c/ , where ρ and η are the density and the dynamic
describe the solvers we implemented for the different Navier-Stokes and
viscosity of the lubricant, respectively. On 1, …, 6 , we prescribe the
Reynolds problems. Section 7 evaluates the new Reynolds model re-
boundary conditions
garding simulation results and computational cost. Finally, Sect. 8
summarizes the paper and gives an outlook on future work to improve 1
( u) n pn = 0 on 1, 2, 5, 6,
the model. Re (6a)

u = 0 on 3, (6b)
2. Setting for the derivation of the new Reynolds model
u = u surf on 4. (6c)
The exemplary application we use for the derivation of our new In the natural condition (6a) on 1 (which is the plane x = x min in the
Reynolds model is introduced in Sect. 2.1. The governing equations for domain ), 2 ( x = x max ), 5 (z = zmin ) and 6 (z = zmax ), n is the par-
this problem are the Navier-Stokes equations, specified together with ticular outward-pointing normal vector. Note that 3 contains the whole
boundary conditions in Sect. 2.2. bearing surface, i.e., it includes the texture surfaces. Physically, the
In this and the following sections, all equations are given in di- natural condition (6a) means that viscous and pressure forces on the
mensionless form. boundary cancel out such that the flow can pass the boundary stress-
free and therefore unhindered.
2.1. Exemplary application: textured journal bearing
3. Overview of the effective Navier-Stokes model
As an exemplary application throughout this paper, we investigate
lubrication for textured journal bearings. However, our models are not Starting from an upscaling strategy as presented in Ref. [21], we
limited to this particular application. The three-dimensional setting is developed an effective Navier-Stokes model. The detailed derivation
schematically shown in Fig. 1. Note that the system is unfolded. The can be found in Ref. [7]. In this section, we give a brief overview of the
rotating shaft surface is at the top ( 4 ), while the stationary textured model because our new Reynolds model presented in Sect. 4 is based on
it.
Our effective model replaces solving the Navier-Stokes system (5),
(6) with a three-step process, see Fig. 2. The three steps are described in
the following.
1. Reynolds problem:The dimension-reduced linear stationary in-
compressible Reynolds equation

h3 1
Re p + u surf h = 0
Fig. 1. Unfolded domain for the investigation of lubrication for textured journal
12 Re 2 (7)
bearings. with boundary conditions

57
M. Rom, S. Müller Tribology International 133 (2019) 55–66

which is similar to the Navier-Stokes system (5), is solved for (u cp, pcp ) .
The only but important difference to (5) is the convective term which
contains the Reynolds velocity (9) and therefore is linear. We impose
the boundary conditions
u cp = uRe on in , (12a)

1
( u cp) n pcp n = 0 on out ,
Re (12b)

u cp = 0 on 3, (12c)

u cp = u surf on 4, (12d)

where in consists of the three boundaries at the planes x = x min ,


z = zmin and z = zmax of the particular cell problem domain cp for
y > ht , and 3 describes the texture, i.e., all boundaries below the dotted
surface in Fig. 2 for which y ht .
3. Effective problem:The domain eff for the effective problem
does not contain the textures, i.e., the bearing surface is smooth. To
obtain the final pressure-velocity solution (u, p) , the Navier-Stokes
system (5) with boundary conditions (6) and the additional effective
boundary condition

u (x ) = u cp, i (x ), x eff (13)

is solved. Hence, the bearing surface is divided into texture parts eff
and non-texture parts 3. The index i in (13) refers to the particular cell
problem from which the solution u cp has to be taken, depending on the
Fig. 2. Effective Navier-Stokes model: three-step solution procedure in Re , cp
coordinate x . The solutions of the cell problems are incorporated into
and eff as a replacement for solving the Navier-Stokes system in . the effective problem only via (13). Hence, only data at y = ht are
needed from the cell problems.
All aforementioned computations can be done analogously in a two-
pRe periodic on per , (8a)
dimensional setting. Then, the bearing side boundaries do not exist.
pRe = 0 on 0 (8b) Hence, 5 and 6 in the original or effective problem as well as 0 in the
Reynolds problem, which is one-dimensional, vanish. In addition, in in
is solved for the pressure pRe . Due to unfolding the journal bearing at a
the cell problems only describes the x min -boundary.
cutting plane which in 2D corresponds to the line per , we use the
periodic condition (8a). Since this leads to ambient pressure ( pRe = 0 )
on per , we could also impose pRe = 0 as done for other setups without 4. Derivation of the new Reynolds model
periodicity or for the bearing sides 0 . Similarly to the derivation in Ref.
[7], the Reynolds velocity uRe , which is needed in the second step, can With the effective Navier-Stokes model in Ref. [7], we observed
be computed from the gradient of the pressure solution pRe by significant reductions of computation time compared with solving the
Navier-Stokes equations. However, to further improve the performance,
1 p y˜
uRe = Re Re y˜ (˜y h) + usurf , our objective is the replacement of the nonlinear effective problem
2 x h (9a)
computed in the third step with a dimension-reduced linear problem.
h h y˜2 y˜ Therefore, in Sect. 4.1 we derive a modified Reynolds equation which
v Re = usurf + wsurf 2 incorporates information obtained from the cell problems. Our new
x z h2 h
three-step process is then summarized in Sect. 4.2.
1 pRe h pRe h 2 y˜
+ Re + y˜ 1 ,
2 x x z z h (9b)
4.1. Derivation of a modified Reynolds equation
1 p y˜
wRe = Re Re y˜ (y˜ h) + wsurf .
2 z h (9c) In the effective Navier-Stokes model, we only use the velocity so-
lution from the cell problems in the effective problem in step 3, see
The coordinate ỹ is defined by
(13), but no data regarding the pressure. However, the pressure gra-
y (x , z ) at texture, dients resulting from solving the cell problems and the Navier-Stokes
y˜ (x , z ) =
y ht otherwise (10) system agree well. This is demonstrated in Fig. 3 which shows a com-
parison of p / x (a) and p / y (b) on the shaft of a two-dimensional
for any (x , y , z ) . Although the Reynolds equation is dimension-re- journal bearing with 15 textures presented in Ref. [7]. Since this is a
duced and misses the film thickness direction, a complete velocity general observation, the specifications of the test case are not important
distribution can be computed since ỹ occurs as variable in (9). here.
2. Cell problems:For each texture, a cell problem has to be solved. The good agreement of the pressure gradients leads to the key idea
For this purpose, the domain cp , see Fig. 2, is chosen correspondingly for the derivation of a modified Reynolds equation, namely to in-
to the particular film thickness such that the boundary 4 is always on corporate the pressure gradient pcp of the cell problems into the new
the shaft surface. For each cell problem, the linear system equation. In the following, we concentrate on a two-dimensional setting
u cp = 0, to keep the derivation short but emphasize that the three-dimensional
1 derivation is analogous. We start by rewriting the momentum equation
(uRe ) u cp + pcp u cp = 0, (11)
Re from system (11) such that for the component pcp / x we obtain

58
M. Rom, S. Müller Tribology International 133 (2019) 55–66

We extend (16b) such that

p 1 2u pcp 1 2u
cp
= +
x Re y 2 x Re y2 (17)
The cell problem gradient pcp / x , see (14), contains extended flow
information in the term 2u / x 2
cp and inertia effects in
uRe ucp / x + v Re u cp/ y . Hence, these data are introduced into (17)
and later into the modified Reynolds equation. The subtraction of the
last term in (17) is necessary because the second derivative of u with
respect to y is already incorporated from the standard Reynolds ansatz,
see the first term on the right-hand side of (17).
For the following derivation, recall the film thickness (2) and the
coordinate (10).
Starting from (17), the derivation of the modified Reynolds equation
is analogous to the derivation of the standard Reynolds equation, which
can for instance be found in Ref. [6]. The first step is to integrate (17)
twice with respect to the film thickness direction to obtain an equation
for the velocity component u. The first integration

y˜ p y˜ 1 2u pcp 1 2u
cp
dy˜ = + dy˜
0 x 0 Re y˜2 x Re y˜2 (18)
due to (15) and (16c) yields

p 1 u 1 u pcp 1 ucp 1 ucp


y˜ = + y˜ + .
x Re y˜ Re y˜ y˜ = 0
x Re y˜ Re y˜ y˜ = 0 (19)
The second integration leads to

1 p 2 1 1 1 u 1 pcp 2 1
y˜ = u u| y˜ = 0 y˜ + y˜ ucp
2 x Re Re Re y˜ y˜ = 0
2 x Re
1 1 ucp
+ ucp |y˜ = 0 + y˜.
Re Re y˜ y˜ = 0 (20)
Due to the boundary conditions on 3, which consists of the sta-
Fig. 3. Comparison of Navier-Stokes and cell problem solutions for the pressure tionary bearing and texture surfaces, we have u|y˜ = 0 = ucp | y˜ = 0 = 0 .
gradients p / x (a) and p / y (b) on the shaft of a 2D journal bearing with 15 Hence, the second and the sixth term on the right-hand side of (20)
textures, see Ref. [7]. vanish. The constant ( u / y˜)|ỹ = 0 is determined by inserting the
boundary conditions for the rotating shaft surface 4 at y˜ = h into (20),
i.e., u|y˜ = h = ucp |y˜ = h = usurf , such that
pcp 1 2u
cp
2u
cp ucp ucp
= + uRe + v Re . u 1 p 1 pcp ucp
x Re x2 y2 x y (14) = Re h + Re h+ .
y˜ y˜ = 0
2 x 2 x y˜ y˜ = 0 (21)
Furthermore, we apply the key assumption for the derivation of the
Inserting (21) into (20) and rearranging terms to obtain an equation
original Reynolds equation, namely a vanishing pressure gradient in
for the velocity component u leads to
film thickness direction y, i.e.,
1 p 1 pcp
pcp u= Re y˜ (˜y h) Re y˜ (˜y h) + ucp .
= 0, 2 x 2 x (22)
y (15)
The crucial step in the derivation is then the integration of the
even though this assumption is not correct as visible in Fig. 3 (b). Be- continuity equation (16a) over the film thickness h, resulting in a re-
tween the textures, however, (15) is justified since pcp / y there is close duction of the spatial dimension. Applying the Leibniz integral rule
to zero and much smaller than pcp / x . Hence, pressure peaks at texture leads to
edges and corners cannot be captured entirely with our new Reynolds
model. h u h v h h
dy˜ + dy˜ = u| y˜ = h + u dy˜ + v|y˜ = h v| y˜ = 0
The starting point for the derivation of the original Reynolds 0 x 0 y˜ x x 0

equation is the Navier-Stokes system (5). With the assumption (23)


= 0.
p / y = 0 and after neglecting several further terms, this system is
simplified to Due to (3) and (4), we have u|y˜ = h h/ x = usurf h/ x = vsurf = v|y˜ = h .
Since in addition v| ỹ = 0 = 0 , we obtain
u v
+ = 0,
x y (16a) h
u dỹ = 0 .
x 0
(24)
p 1 2u
= ,
x Re y 2 (16b) Inserting (22) into (24) leads to

p h 1 p 1 pcp
=0 Re y˜ (y˜ h) Re y˜ (y˜ h) + ucp dy˜) = 0.
y (16c) x 0 2 x 2 x (25)

59
M. Rom, S. Müller Tribology International 133 (2019) 55–66

Resolving the integral in (25) results in

h3 p pcp h
Re + ucp dy˜ = 0.
x 12 x x 0
(26)

For the integral term in (26), we can follow the same argumentation
as above in (23) and (24) such that

h
ucp dỹ = 0.
x 0
(27)

Hence, our modified Reynolds equation finally reads

p pcp
h3 =0.
x x x (28)

In vector formulation, it can be written in dimension-independent


form as
(h3 ( p pcp )) = 0 . (29)

The first difference to the standard Reynolds equation (7) is the


additional pressure gradient pcp from the cell problems which leads to
a pressure correction. The second difference is the term 1/2 u surf h
which does not enter the modified Reynolds equation.
For a fixed x , the pressure gradient pcp has to be evaluated at one Fig. 4. New Reynolds model: compared with the effective Navier-Stokes model
particular height in the cell problems. Due to the assumption in Fig. 2, the cell problem domain is changed (step 2) and the effective problem
pcp / y = 0 , this position should be arbitrary. However, as mentioned is replaced by a modified Reynolds problem (step 3).
above this assumption is usually not true for a textured lubrication
problem, and the main variation of the gradient is around the texture
considering only positive pressure values [p]+ = max(0, p) , the friction
edges and corners, cf. Fig. 3 (b). In our simulations, it has proven suf-
force
ficient to evaluate pcp at the shaft surface y˜ = h , or equivalently
y = hg + ht , away from any possible pressure singularities induced by F= d
(31)
sharp texture corners. 4

with the shear stress τ and the volumetric flow rate


4.2. Modified three-step process h
Q= u dy˜ d .
Our modified three-step process incorporating the modified
1,2 0
(32)
Reynolds problem in step 3 is schematically illustrated in Fig. 4. The Here, 1,2 denotes the left ( x = x min ) or right domain boundary
three steps are detailed below by pointing out the changes compared ( x = x max ), which for an unfolded journal bearing corresponds to the
with the effective Navier-Stokes model described in Sect. 3. left or right plane of intersection, respectively. The flow rate can of
course also be computed for an arbitrary slice inside the flow domain.
1. Reynolds problem: The Reynolds problem (7), (8) is unchanged. Due to a potential squeeze effect, it does not necessarily have to be
2. Cell problems: The domain cp is extended such that the union of constant throughout the flow field.
all cell problem domains equals the original domain . This is ne- For the standard Reynolds equation, the values are determined by
cessary because data from the cell problems ( pcp ) are needed all
over the domain Re in the modified Reynolds problem in step 3. WRe = [pRe ]+ dx ,
(33)
Re
System (11) and the boundary conditions (12) are unchanged.
3. Modified Reynolds problem: The modified Reynolds equation (29) 1 uRe (9a) h pRe 1 usurf
FRe = dx = + dx
is solved for the pressure p. The computational domain Re and the Re y˜ 2 x Re h
Re
y˜ = h
Re
(34)
boundary conditions (8) on per and 0 are carried over from the
standard Reynolds problem in step 1. and
h (9a) h3 pRe 1
Analogously to the effective Navier-Stokes model, the new Reynolds QRe = uRe dy˜ d = Re + usurf h d .
0 12 x 2
model can also be used in a two-dimensional setting, for which the
per per

Reynolds problems are one-dimensional. (35)


Similarly, for our modified Reynolds equation we obtain
5. Computation of characteristic values
WmodRe = [p]+ dx ,
Re (36)
For a quantitative comparison of simulation results in the next
section, we are interested in three characteristic values and first specify 1 u
FmodRe = dx
how these are computed for a Navier-Stokes simulation. The values are Re Re y˜ y˜ = h
the load-carrying capacity
(22) h p pcp 1 u cp
= + dx
W= [p]+ d , 2 x x Re y˜
(30) (37)
Re
4 y˜ = h

60
M. Rom, S. Müller Tribology International 133 (2019) 55–66

and Table 1
Parameters for the simulations of the 2D and 3D journal bearings and the 3D
QmodRe =
h
u dy˜ d plane-inclined slider bearing.
per 0
test case 1 2 3
(22) h3 p pcp h
= Re + ucp dy˜ d . spatial dimension 2D 3D 3D
per 12 x x 0
(38) bearing type journal journal slider
bearing parameters
We compute QmodRe at the right periodic boundary per of the diameter d [cm] 15/ 0.15/ –
modified Reynolds problem, i.e., at x = x max . Therefore, the integral length lx [cm] – – 2
h width lz [cm] – 0.024
0
ucp dỹ has only to be computed at the outflow boundary out at
width to diam. ratio lz /d [−] – 0.503
x = x max of the last cell problem. This cannot be done analytically but
total number of textures nt [−] 1,000 30 200
approximately by applying a Newton-Cotes formula for numerical in- number of textures nt, x [−] 1,000 10 200
tegration, i.e., number of textures nt , z [−] – 3 1
n texture height ht,max [μm] 30 30 25
h
u cp dy˜ h wi ucp (y˜ = ti h) texture length lt, x [μm] 75 70 50
0
i=0 (39) texture width lt , z [μm] – 40 50
distance betw. text. dt , x [μm] 75 80 50
with degree n, weights wi and nodes ti . Hence, the velocity solution ucp
distance betw. text. dt, z [μm] – 40 50
from the last cell problem has to be evaluated at several heights ỹ . We texture aspect ratio λ [−] 2.5 2.333 2
use Simpson's rule with n = 2 , weights 1/6,2/3,1/6 and nodes 0,1/2,1 texture density t [−] 0.5 0.233 0.25
such that lubricant parameters
density ρ [kg/m3] 880 880 880
h 1 2 1 1 dynamic viscosity η [Pa s] 0.01 0.01 0.01
ucp dy˜ h ucp (0) + ucp h + ucp (h)
0 6 3 2 6 operating conditions
radial clearance c [μm] 20 15 –
(12c), (12d) 2 1 1
= h ucp h + usurf . max. gap height h g,max [μm] 26 22.5 20
3 2 6 (40) min. gap height h g,min [μm] 14 7.5 10
relative eccentricity erel [−] 0.3 0.5 –
surf. vel. magn. usurf,abs [m/s] 50 50 50
6. Solvers for the Navier-Stokes and Reynolds problems Reynolds number Re [−] 88 66 66

All our solvers have been implemented making use of deal.II [22], a
finite element library written in C++. For the (Navier-)Stokes and the convergence study, are given in Sect. 7.2. In Sect. 7.3, we present
cell problems, we use quadrilateral (2D) or hexahedral (3D) Q2 / Q1 qualitative and quantitative comparisons of simulation results of our
Taylor-Hood elements, i.e., biquadratic (2D) or triquadratic (3D) ve- new Reynolds model with Navier-Stokes, Stokes and Reynolds results.
locity elements and bilinear (2D) or trilinear (3D) pressure elements. Finally, in Sect. 7.4 we investigate the computational performance of
The original as well as the modified Reynolds problem can be solved our new Reynolds model by studying computation times.
with linear (1D) or bilinear (2D) Q1 pressure elements. However, we use
Q2 elements to obtain continuous pressure gradients since these gra-
7.1. Simulation setups
dients are needed for computing the Reynolds velocity (9) or the fric-
tion force (34), (37) and the volumetric flow rate (35), (38).
Table 1 summarizes the simulation parameters for three config-
The nonlinear Navier-Stokes system is linearized by applying Picard
urations. With these test cases, we can show the suitability of our ap-
linearization and solved iteratively, i.e., in the nonlinear convective
proach for two- as well as three-dimensional computations and for
term (u ) u the solution of the previous iteration is inserted such that
different applications, here journal and slider bearings. Furthermore,
for iteration k the term is given by (uk 1 ) uk . In the first iteration, the
the 3D slider bearing of test case 3 is infinitely wide which requires
linear Stokes system, which does not contain the convective term, is
periodic boundary conditions at the bearing sides in the Navier-Stokes
solved. The Picard iteration stops if the maximum absolute difference
problem and in the two Reynolds and the cell problems. Hence, we can
between two consecutive solutions falls below a prescribed threshold.
investigate whether our model is also appropriate for modified
In this work, 1,500 Pa for the pressure and 0.1 m/s for the velocity
boundary conditions.
components were sufficient threshold values.
The diameter and the width of the journal bearing of test case 2 are
The linearized Navier-Stokes systems and the cell problems of the
unrealistically small regarding a practical application, but these di-
new Reynolds model both are preconditioned with the BFBt precondi-
mensions allow to keep the number of textures moderate. As shown in
tioner, see for instance Ref. [23], and solved by the GMRES method
Sect. 7.4, solving the Navier-Stokes system for this test case with only
implemented in deal.II.
30 textures already takes more than two days. Hence, a Navier-Stokes
For solving the original and the modified Reynolds problem, both
simulation of a realistic 3D bearing of finite width, needed to obtain a
being linear, the CG solver of deal.II is used together with the SparseILU
reference solution to validate our new Reynolds model, is out of reach
preconditioner which is based on incomplete LU decomposition for
without a parallelization of the Navier-Stokes solver. However, as dis-
sparse matrices.
cussed in Sect. 7.3.2, with decreasing bearing dimensions the influence
All simulations of Sect. 7 were conducted on a computer with
of the textures on the pressure distribution increases such that the as-
256 GB RAM and 32 CPUs (Intel Xeon E5-2667 v3, 3.2 GHz). Besides
sumption of no pressure variation over the film thickness is more likely
some thread-parallel parts in the deal.II solvers, all implementations are
to be violated. Therefore, test case 2 is even more of a challenge for our
serial.
new Reynolds model.
The textures of the 2D journal bearing have an elliptical shape, for
7. Numerical setups and results
the first texture described by

The simulation setups for the investigation of our new Reynolds (x 0.5(dt , x + lt , x ))2 (y ht ,max ) 2
model are summarized in Sect. 7.1. Information on the chosen dis- + =1.
(0.5 lt , x ) 2 ht2,max (41)
cretizations for the test cases, based on a two-dimensional mesh

61
M. Rom, S. Müller Tribology International 133 (2019) 55–66

In contrast, the textures of the 3D journal and 3D slider bearing are


cuboidal. The shapes can be seen in Sect. 7.2, where the final dis-
cretizations are presented.
The linearly decreasing gap height hg of the 3D slider bearing is
independent of z and given by

hg ,min h g,max
h g (x ) = x + h g,max .
lx (42)

The texture aspect ratios 2 2.5 of our three test cases are
smaller than those considered optimal by many studies, see for instance
Refs. [5,24]. However, the optimal texturing is hard to find since it
depends on the particular application, materials, lubricant and oper-
ating conditions [24]. In addition, in literature it is discussed con-
troversially whether convective inertia has a positive or negative effect
on the hydrodynamic performance [5]. Many numerical studies con-
sider a large texture aspect ratio λ to be optimal, e.g., 20 100 in Fig. 6. Comparison of the pressure distributions for the 2D Navier-Stokes si-
Ref. [25], but they are often conducted using the Reynolds equation mulations of test case 1 with uniformly increasing mesh refinement level L.
which cannot account for inertia. We choose very small aspect ratios for
our numerical tests to investigate whether our new Reynolds model is towards the texture corners as illustrated for the first texture in Fig. 7
valid even for extreme configurations, in which case the model would (a) showing how the mesh points are distributed over the film thick-
allow for thoroughly studying the influence of inertia effects on the ness. For the cell problems, we choose the same discretization L = 2 as
hydrodynamic performance. With increasing λ, inertia is reduced but for one texture element in the Navier-Stokes mesh, i.e., putting all cell
can still be significant. Consequently, our model is useful also for large problem meshes together would result in the Navier-Stokes mesh. An
values of λ. automatic mesh transformation adapts each cell problem mesh to the
Another important parameter for an optimal texturing is the texture current position such that the correct gap height is obtained. Finally,
density t which is the ratio of textured area to total area. Our 3D test the mesh for the modified Reynolds problem has to be chosen finer than
cases with t 25 % are in the range of 20 % to 40 % regarded as optimal the mesh for the original Reynolds equation since the results in terms of
by many studies [24]. the load-carrying capacity converge more slowly with increasing mesh
refinement. Here, 40,000 finite elements are sufficient. The distribution
7.2. Two-dimensional mesh convergence study and final discretizations of the mesh points can be seen in Fig. 7 (b). For a check of the overall
mesh convergence of the complete three-step process, we compare re-
For the journal bearing of test case 1, we study mesh convergence sults obtained from the discretizations described above with results
for both the Navier-Stokes simulation and the new Reynolds model si- obtained from the following setup: 40,000 finite elements for the ori-
mulation. For the latter, three meshes are needed: one for the original ginal Reynolds problem, L = 3 for the cell problem meshes and 400,000
Reynolds problem, one for the cell problems and one for the modified finite elements for the modified Reynolds problem. The deviation in the
Reynolds problem. For the three-dimensional simulations, due to the load-carrying capacity is 0.03 % .
high computational effort an extensive mesh convergence study cannot For the three-dimensional simulations, we construct meshes with
be conducted. However, we construct the three-dimensional meshes similar average spacings between the mesh points in x- and z-direction
according to the findings of the two-dimensional study. as in the two-dimensional case in x-direction. In film thickness direction
A fragment of the final mesh for the 2D Navier-Stokes computation y, we use an even finer average spacing because we apply a stretching
can be seen in Fig. 5, showing a close-up of the first two textures. We technique to adequately resolve the corners and edges of the cuboidal
refer to the refinement level of this mesh as L = 2 . Fig. 6 compares the textures and therefore need more cells over the film thickness.
pressure distribution on the shaft in the area of the pressure maximum However, due to the higher number of textures the mesh for the 3D
resulting from using the L = 2 mesh with the distributions obtained slider bearing (test case 3 with 200 textures) is coarser than the mesh
from computations with two coarser meshes of levels L = 0 and L = 1 for the 3D journal bearing (test case 2 with 30 textures). Otherwise, the
and one finer mesh of level L = 3. All mesh refinements are done uni- Navier-Stokes simulations would be too expensive.
formly. The deviations between the L = 0 and L = 3 pressure maxima
and load-carrying capacities are only 1.9 % and 1.8 % , respectively.
However, from Fig. 6 it can be seen that the pressure distributions of the
L = 0 and L = 1 meshes are not resolved adequately such that our
choice for the final mesh is refinement level L = 2 . For the latter, the
pressure maxima and load-carrying capacities differ from the values of
the L = 3 mesh by 0.03 % and 0.04 % , respectively.
For the new Reynolds model simulation, we first set up the mesh for
solving the original Reynolds equation to contain 20,000 finite elements,
i.e., 20 per texture element, which is sufficient to obtain a mesh-in-
dependent pressure distribution. We use a stretching of the mesh points

Fig. 7. Final meshes used for the 1D original Reynolds (a) and modified
Fig. 5. Fragment of the final Navier-Stokes mesh used for the 2D test case 1. Reynolds problem (b) of test case 1, shown with film thickness information.

62
M. Rom, S. Müller Tribology International 133 (2019) 55–66

Fig. 8. Final cell problem meshes for the 3D simulations.

Table 2
Final discretizations of the three test cases (tc): numbers of finite elements (FEs)
and degrees of freedom (DOFs).
tc Nav.-St. new Reynolds model

Reynolds cell pr. mod. Re.

1 FEs 184,000 20,000 184 40,000


DOFs 1,816,023 40,001 1,839 80,001

2 FEs 215,040 18,000 7,168 54,000 Fig. 9. Pressure on the shaft of the 2D journal bearing of test case 1.
DOFs 5,795,652 72,721 198,012 217,141

3 FEs 204,800 80,000 1,024 320,000 Reynolds model shows a good agreement with a deviation between the
DOFs 6,069,540 328,041 31,084 1,296,081 pressure maxima of 1.1 %.

7.3.2. Test case 2: three-dimensional journal bearing


A contour plot of the pressure distributions for the three-dimen-
The cell problem meshes are again chosen such that putting them sional journal bearing of test case 2 is depicted in Fig. 10. The Navier-
together would result in the corresponding Navier-Stokes mesh. Stokes solution (a) is asymmetric as can be seen in the pressure legend.
Examples for both 3D test cases are visible in Fig. 8. For the two-di- This asymmetry, also observed in former publications, for instance in
mensional Reynolds meshes, we also use a stretching towards the tex- Ref. [26], cannot be reproduced by the Stokes (b) or the Reynolds so-
ture edges and corners. lution (c), whereas the new Reynolds model solution (d) is asymmetric.
The final discretizations of the three test cases are summarized in The latter shows a good agreement with the Navier-Stokes pressure in
Table 2, listing the numbers of finite elements (FEs) and degrees of the left half, i.e., for x < 0.75 mm, whereas the quality of the solution
freedom (DOFs) for each mesh. deteriorates in the area of the negative pressures in the right half. The
Note that for test case 3 the number of finite elements for the deviations between the pressure maxima in the left half and minima in
modified Reynolds equation is larger than the number for the Navier- the right half are 2.6 % and 14.6 % , respectively.
Stokes system. This is due to the rather coarse resolution of the three- The qualitative results above also apply to the pressures on the
dimensional Navier-Stokes mesh. However, due to the reduction of center line of the shaft at z = 0.12 mm which are presented in Fig. 11.
dimension in case of the modified Reynolds problem, the number of In this line plot, one can observe that the new Reynolds model solution
degrees of freedom is about five times smaller. This is, together with the is a good approximation in the positive pressure part but that also local
less complex equation which has to be solved, the crucial point for the differences occur, in particular at the texture edges at which the pres-
computational cost. sure variation in film thickness direction, which is neglected in the new
Reynolds model, is maximal.
7.3. Simulation results In the center of the textures away from the leading or trailing edge
(with respect to the flow direction), the approximation quality of the
In the following, we compare the pressure distributions and char- new Reynolds model is better than at those edges. This can be seen in
acteristic values resulting from the Navier-Stokes, Stokes, Reynolds and Fig. 12 depicting the line on the shaft through the center of the fourth
new Reynolds model solutions for the three test cases. column of textures at x = 0.525 mm. Again, the new Reynolds model
pressure shows a better accordance with the Navier-Stokes pressure
7.3.1. Test case 1: two-dimensional journal bearing than the Stokes or the Reynolds solution.
The pressures on the shaft of the two-dimensional journal bearing of Overall, this three-dimensional test case shows that the new
test case 1 are depicted in Fig. 9, on the top showing the complete Reynolds model loses approximation quality for applications for which
bearing and on the bottom the area of the pressure maximum, visua- the textures have a crucial influence on the pressure distribution. Then,
lizing the effect of the textures. While the Stokes and the Reynolds the assumption of no pressure variation over the film thickness is
solution strongly differ from the Navier-Stokes pressure, the new clearly violated. This is in contrast to test case 1 or 3 where the

63
M. Rom, S. Müller Tribology International 133 (2019) 55–66

Fig. 13. Pressure on the shaft at z = 50 µ m (center line) of the 3D slider bearing
of test case 3.
Fig. 10. Pressure on the shaft of the 3D journal bearing of test case 2.

influence of the textures can only be seen in close-ups of the pressure


distribution. However, the new Reynolds model still provides sig-
nificantly better results than the Stokes system or the Reynolds equa-
tion.

7.3.3. Test case 3: three-dimensional slider bearing


For the three-dimensional plane-inclined slider bearing with con-
verging gap, no negative pressures occur. The Navier-Stokes, Stokes,
Reynolds and new Reynolds model pressures on the center line at
z = 50 µ m are shown in Fig. 13 on the top. The new Reynolds model
solution approximates the Navier-Stokes solution well, also visible in
the close-up on the bottom of the figure. The deviation of the pressure
maxima is only 0.7 %. As observed for the other two test cases, the
Fig. 11. Pressure on the shaft at z = 0.12 mm (center line) of the 3D journal Stokes and Reynolds pressure deviate considerably from the Navier-
bearing of test case 2. Stokes pressure.

Table 3
Load-carrying capacity W (2D: [N/m], 3D: [N]), friction force F (2D: [N/m],
3D: [N]) and volumetric flow rate Q (2D: [cm2/s], 3D: [cm3/s]) for the three
test cases (tc).
tc Nav.-St. Reynolds Stokes new Re. model

1 W 2.156 106 21.9 % 12.7 % 1.5 %


F 4,103.26 4.0 % 1.8 % + 1.2 %
Q 4.63 + 3.5 % + 2.1 % 1.8 %

2 W 0.0229 29.3 % 16.4 % + 0.003 %


F 0.0139 4.7 % 1.4 % 0.15 %
Q 0.1326 + 3.8 % + 1.3 % + 1.6 %

3 W 28.09 23.5 % 6.9 % + 0.51 %


F 0.0744 4.5 % 1.2 % + 0.73 %
Fig. 12. Pressure on the shaft at x = 0.525 mm (line through the center of the
Q 0.0344 + 2.4 % + 1.3 % 1.2 %
fourth column of textures) of the 3D journal bearing of test case 2.

64
M. Rom, S. Müller Tribology International 133 (2019) 55–66

7.3.4. Quantitative evaluation of the new Reynolds model 77 % for test case 3 and 96 % for test case 2. Even though test case 2 has
After the qualitative comparison of simulation results above, we only 30 textures and, hence, cell problems, the computation time of
now investigate the characteristic values according to Sect. 5 for a 3,786 s is higher than for test case 3 with 200 cell problems. This is due
quantitative evaluation. Table 3 lists the results for all three test cases. to the finer discretization of the cell problems of test case 2 with about
For the Reynolds, Stokes and new Reynolds model simulations, the six times the number of degrees of freedom compared with test case 3,
percentage deviations from the corresponding Navier-Stokes values are see Table 2. Consequently, it is desirable to keep the effort for the cell
given. problems as small as possible. The cost can be reduced by several
With more than 20 % deviation for the load-carrying capacity W in means, e.g., by reducing the mesh resolution and accepting an in-
each test case, the Reynolds pressure solution is infeasible. The Stokes creased approximation error or by accelerating the computations by
solutions are better but still far away from being accurate, in particular applying a reduced basis method. We comment on this in the conclu-
for test case 2 with 16.4 % deviation. In contrast, the new Reynolds sion in Sect. 8.
model leads to load-carrying capacities which differ 1.5 % from the Due to large differences in the resolution of the meshes for the
Navier-Stokes values at most. Considering the visible differences be- original as well as the modified Reynolds problems, the computation
tween the Navier-Stokes and new Reynolds model pressure plots in times for these problems also differ significantly, which in particular is
Figs. 10–12, the almost perfect result of test case 2 with 0.003 % might noticeable for the two three-dimensional test cases 2 and 3 with three
be coincidental. However, this result indicates that the averaged pres- and 49 s, respectively, for the original Reynolds problem and 58 and
sure behavior is met adequately even though the pressure distribution 599 s, respectively, for the modified Reynolds equation.
cannot be exactly reproduced over each single texture. The aforementioned differences in the discretizations lead to large
Regarding the choice of the equation or model, the friction force F variations of the speed-ups obtained by the new Reynolds model be-
and the volumetric flow rate Q are less critical than the load-carrying tween 50 for test case 2 and 676 for test case 1. The smallest speed-up of
capacity W. Even with the Reynolds equation, the deviations are 50 results in a reduction of computation time from 53 h for the Navier-
smaller than 5 % and, hence, acceptable. However, an improvement of F Stokes simulation to one hour for the new Reynolds model simulation.
and Q can be achieved by using the Stokes system. With the new The speed-up of 123 for test case 3 means a decrease from 95 h to
Reynolds model, a further improvement of F is obtained, while the 47 min, while the highest speed-up of 676 for test case 1 corresponds to
values for Q are comparable with those of the Stokes simulations. a reduction from 30 h to three minutes. These values underline the
The quantitative results show that the new Reynolds model provides benefit of the new Reynolds model, especially when repeated simula-
accurate approximations of the Navier-Stokes solutions since the largest tions with slightly changed parameters are necessary, e.g., in the con-
deviation of any of the three values W, F and Q is 1.8 %. Especially for text of an optimization task regarding the geometry of the textures.
the load-carrying capacity W, the model is a great improvement com- Computing the modified Reynolds equation in the third step instead
pared with the Reynolds equation or the Stokes system. of the Navier-Stokes system for a smooth domain with effective
In general, even more accurate results for W, F and Q can be ex- boundary conditions as done in our effective Navier-Stokes model
pected from our effective Navier-Stokes model [7] because in the third presented in Ref. [7], increases the overall speed-ups enormously: the
step of that model the Navier-Stokes equations are solved instead of the highest speed-up we measured for the four test cases in Ref. [7] was 8.7.
modified Reynolds equation. For the four test cases in Ref. [7], the
effective model provided characteristic values deviating from the Na- 8. Conclusion
vier-Stokes values by 1.03 % at most. However, the better computational
performance of the new Reynolds model investigated in the following For the simulation of hydrodynamic lubrication of textured surfaces
justifies preferring the new Reynolds model. with small texture aspect ratios ( < 10 ) and high Reynolds numbers
(Re > 50 ), our new Reynolds model is proposed as an accurate and
7.4. Computational performance of the new Reynolds model computationally significantly less expensive alternative to the nonlinear
Navier-Stokes system.
In the following, we compare the Navier-Stokes and new Reynolds Our numerical test cases have shown the suitability of the new
model computation times for the three test cases and measure the model for different applications (journal or slider bearing), for two- or
speed-ups. The results are summarized in Table 4 by listing the times in three-dimensional tests, for infinitely or finitely wide bearings, for el-
seconds. The new Reynolds model times are split into the times spent liptically shaped or cuboidal textures and for textures with small or
for the three steps, i.e., for solving (i) the original Reynolds equation, large influence on the overall pressure distribution. For 2 2.5 and
(ii ) the cell problems and (iii) the modified Reynolds equation. The 66 Re 88, the deviations of characteristic values such as the load-
speed-up of the new Reynolds model in the last column is the ratio of carrying capacity from the corresponding Navier-Stokes values have
the particular Navier-Stokes computation time to the sum of the three been smaller than 2 % and speed-ups of up to 676 have been achieved.
times of the new Reynolds model. For larger texture aspect ratios and/or smaller Reynolds numbers for
Due to the higher spatial dimension and the more complex equation which convective inertia is reduced, our new Reynolds model will be
system, the cell problems cause the highest computational cost of the suitable as well. Hence, the model allows to investigate textured surface
new Reynolds model. Their share in computation time lies between lubrication in a much wider parameter range than possible with the
Reynolds equation or the Stokes system. A full study of the validity of
the model, however, is beyond the scope of this paper since it requires
Table 4 varying a large number of parameters concerning the texture shape and
Computation times for solving the Navier-Stokes system and the three steps of geometry (length, width, height, aspect ratio, density) and the oper-
the new Reynolds model (given in seconds) and total speed-up of the new ating conditions (Reynolds number).
Reynolds model. A crucial advantage of our three-step process is that all three pro-
tc bearing Nav.-St. new Reynolds model speed-up blems are linear. At first, the dimension-reduced original Reynolds
equation is solved for the original textured problem. The Reynolds
Re. cell mod. velocity, computed from the resulting Reynolds pressure gradient, is
1 2D journal 107,512 3 144 12 676 used as boundary condition and in the momentum equation of the cell
2 3D journal 190,673 3 3,786 58 50 problems, which are solved in the second step. In turn, the pressure
3 3D slider 343,387 49 2,143 599 123 gradients computed in the cell problems enter the modified and again
dimension-reduced Reynolds equation in the third step, hence

65
M. Rom, S. Müller Tribology International 133 (2019) 55–66

incorporating flow information regarding recirculation and convective Programme for SMEs (ZIM) of the German Federal Ministry for
inertia effects into that equation. Economic Affairs and Energy (BMWi), grant number: ZF4151601LP5.
The most complex part of our new Reynolds model is the second
step in which the cell problems have to be solved. In contrast to the first References
and third step with the original and modified Reynolds equation, re-
spectively, the cell problems are not dimension-reduced and a Navier- [1] Reynolds O. On the theory of lubrication and its application to Mr. Beauchamp
Stokes-like though linear system has to be solved. This makes the cell Tower's experiments, including an experimental determination of the viscosity of
olive oil. Phil Trans Roy Soc Lond 1886;177:157–234.
problems computationally costly, in particular for applications with [2] Dobrica MB, Fillon M. About the validity of Reynolds equation and inertia effects in
many textures and, consequently, cell problems. Another drawback is textured sliders of infinite width. Proc Inst Mech Eng Part J: J Eng Tribol
the necessity to repeatedly adapt the cell problem mesh to the geometry 2009;223:69–78.
[3] Cupillard S, Glavatskih S, Cervantes MJ. Inertia effects in textured hydrodynamic
at the particular position of the texture. contacts. Proc Inst Mech Eng Part J: J Eng Tribol 2010;224:751–6.
Our objective is to reduce the effort for solving the cell problems [4] Qiu M, Bailey BN, Stoll R, Raeymaekers B. The accuracy of the compressible
while preserving the accuracy of the method as well as possible. One Reynolds equation for predicting the local pressure in gas-lubricated textured par-
allel slider bearings. Tribol Int 2014;72:83–9.
option would be to only compute a small number of cell problems and
[5] Gropper D, Wang L, Harvey TJ. Hydrodynamic lubrication of textured surfaces: a
to interpolate between the resulting pressure gradient solutions which review of modeling techniques and key findings. Tribol Int 2016;94:509–29.
are needed as input in the modified Reynolds equation. Another pro- [6] Khonsari MM, Booser ER. Applied tribology: bearing design and lubrication. third
ed. John Wiley & Sons; 2017.
mising approach would be to accelerate the computation of the cell
[7] Rom M, Müller S. An effective Navier-Stokes model for the simulation of textured
problems by applying a reduced basis method. Then, only very few cell surface lubrication. Tribol Int 2018;124:247–58.
problems would have to be solved in an expensive offline stage to [8] Constantinescu VN, Galetuse S. On the possibilities of improving the accuracy of the
construct a reduced basis in which all other cell problems could be evaluation of inertia forces in laminar and turbulent films. J Lubr Technol
1974;96:69–77.
solved much more efficiently in the subsequent online stage. This would [9] Constantinescu VN, Galetuse S. Operating characteristics of journal bearings in
reduce the computational cost significantly while only introducing a turbulent inertial flow. J Lubr Technol 1982;104:173–9.
certified error. Another improvement could be obtained by parallelizing [10] Tichy J, Bou-Said B. Hydrodynamic lubrication and bearing behavior with im-
pulsive loads. Tribol Trans 1991;34:505–12.
the implementation. The cell problems could be computed in parallel [11] Dousti S, Allaire P, Dimond T, Cao J. An extended Reynolds equation applicable to
very efficiently because they do not depend on each other. In case of high reduced Reynolds number operation of journal bearings. Tribol Int
three-dimensional simulations, a further acceleration could be achieved 2016;102:182–97.
[12] Arghir M, Alsayed A, Nicolas D. The finite volume solution of the Reynolds equation
by setting up two two-dimensional cell problems instead of the three- of lubrication with film discontinuities. Int J Mech Sci 2002;44:2119–32.
dimensional one by only using two slices of the computational domain, [13] de Kraker A, van Ostayen RAJ, Rixen DJ. Development of a texture averaged
one with texture and one without, and to compute the needed pressure Reynolds equation. Tribol Int 2010;43:2100–9.
[14] Patir N, Cheng HS. An average flow model for determining effects of three-di-
gradient distribution pcp from these reduced data.
mensional roughness on partial hydrodynamic lubrication. J Lubr Technol
In addition to an acceleration of the cell problems, the computa- 1978;100:12–7.
tional cost of the two Reynolds problems can also be reduced. In Ref. [15] Almqvist A, Dasht J. The homogenization process of the Reynolds equation de-
scribing compressible liquid flow. Tribol Int 2006;39:994–1002.
[27], we have already shown how to homogenize the original Reynolds
[16] Almqvist A, Fabricius J, Spencer A, Wall P. Similarities and differences between the
problem and how to apply a reduced basis method to the cell problems flow factor method by Patir and Cheng and homogenization. J Tribol
resulting from this homogenization process. An analogous procedure 2011;133:031702.
could also be derived for the modified Reynolds equation. Similarly to [17] Ausas R, Ragot P, Leiva J, Jai M, Bayada G, Buscaglia GC. The impact of the ca-
vitation model in the analysis of microtextured lubricated journal bearings. J Tribol
such an approach, the flow factor method by Patir and Cheng [14] or a 2007;129:868–75.
modification of it could be used. [18] Jakobsson B, Floberg L. The finite journal bearing, considering vaporization. Trans
To obtain realistic simulation results in the context of lubrication Chalmers University of Technology Gothenburg; 1957. p. 190.
[19] Olsson KO. Cavitation in dynamically loaded bearings. Trans Chalmers University
problems with cavitation, an extension of our Reynolds model is re- of Technology Gothenburg; 1965. p. 308.
quired. It has to be investigated whether cavitation must be taken into [20] Elrod HG, Adams ML. A computer program for cavitation and starvation problems.
account in each or only one of the three steps of our model and, if In: Dowson D, Godet M, Taylor CM, editors. Cavitation and related phenomena in
lubrication. New York: Mechanical Engineering Publications; 1974. p. 37–41.
necessary, how the JFO boundary conditions [18,19] can be used to- [21] Deolmi G, Dahmen W, Müller S. Effective boundary conditions: a general strategy
gether with the modified Reynolds equation. In addition, the cavitation and application to compressible flows over rough boundaries. Commun Comput
model extending our new Reynolds model should be consistent with the Phys 2017;21:358–400.
[22] Bangerth W, Hartmann R, Kanschat G. Deal.II - a general purpose object-oriented
cavitation model used for the corresponding Navier-Stokes simulations
finite element library. ACM Trans Math Software 2007;33. 24/1-27.
to obtain comparable results for validation purposes. A starting point to [23] Elman HC. Preconditioning strategies for models of incompressible flow. J Sci
achieve the latter objective could be the work by Bayada and Chupin Comput 2005;25:347–66.
[24] Yu H, Huang W, Wang X. Dimple patterns design for different circumstances. Lubric
[28] in which a JFO-similar model is derived from the compressible
Sci 2013;25:67–78.
Navier-Stokes equations. [25] Etsion I, Kligerman Y, Halperin G. Analytical and experimental investigation of
laser-textured mechanical seal faces. Tribol Trans 1999;42:511–6.
Acknowledgments [26] Arghir M, Roucou N, Helene M, Frene J. Theoretical analysis of the incompressible
laminar flow in a macro-roughness cell. J Tribol 2003;125:309–18.
[27] Rom M, Müller S. A reduced basis method for the homogenized Reynolds equation
The authors would like to thank Dr. K. Backhaus and Prof. G. Knoll, applied to textured surfaces. Commun Comput Phys 2018;24:481–509.
IST mbH, for fruitful discussions of simulation parameters and results. [28] Bayada G, Chupin L. Compressible fluid model for hydrodynamic lubrication ca-
vitation. J Tribol 2013;135:041702.
This research project is funded by the Central Innovation

66

Você também pode gostar