Applied Mathematical Modelling: Viorel Badescu

Applied Mathematical Modelling 81 (2020) 887–909
Contents lists available at ScienceDirect
Applied Mathematical Modelling

journal homepage: www.elsevier.com/locate/apm
Two classes of sub-optimal shapes for one dimensional slider

bearings with couple stress lubricants
Viorel Badescu a,b
a
Candida Oancea Institute, Polytechnic University of Bucharest, Spl. Independentei 313, Bucharest 060042, Romania
b
Romanian Academy, Calea Victoriei 125, Bucharest 010017, Romania
a r t i c l e i n f o a b s t r a c t
Article history: Shape optimization of slider bearings operating with couple stress lubricants is performed
Received 11 September 2019 here for the first time by using a novel direct optimal control approach, which defines
Revised 9 January 2020
the gradient of the film height as a control. The bearing load is maximized. One dimen-
Accepted 14 January 2020
sional Reynolds and energy equations are used. Several constraints are taken into consid-
Available online 21 January 2020
eration. They avoid the occurrence of cavitation and ensure the validity of the Reynolds
Keywords: equation. The model is validated against a known analytical solution (the Rayleigh step
Couple stress lubricants bearing). Two simple design rules are inferred, which yield two different classes of sub-
Optimum bearing shape optimal shapes: the multi-stepped bearings and the multi-sloped bearings, respectively.
Multi-stepped bearings Multi-stepped bearings consist of several steps and the couple stress parameter may affect
Multi-sloped bearings the constant value of the film height between steps. Multi-sloped bearings consist of sev-
eral inclined regions and the couple stress fluid parameter may affect the constant value of
the film height between regions. The slider bearings operation under variable load is sta-
ble. A sensitivity analysis identified the design parameters which have the highest impact
on bearing performance. The optimal slider bearing shapes obtained for Newtonian lubri-
cants do not change when most common couple stress fluids are used. Isothermal models
may be used successfully at lower values of the couple stress parameter.
© 2020 Elsevier Inc. All rights reserved.
1. Introduction
Slider bearings are widely used in the transmission systems of many engineering applications for their load-carrying
capacity, excellent stability, and durability. Applications include piston rings, machine tool ways, mechanical seals, computer
hard disks and plain collar thrust bearings [1].
In the past, optimization of slider bearings consisted either in (i) finding the optimum design parameters for predefined
bearing shapes or in (ii) finding the optimal bearing shape. Among the predefined shapes used in practice, the inclined
bearing has been investigated in a large number of studies (for instance, see [2]). Other bearing shapes have been studied as
well (see, e.g. [3]). A steady power-law curved film slider bearing with Rabinowitsch fluids has been studied in [4] where the
bearing performances are discussed in terms of various parameters including the inlet-outlet height ratio. A Rabinowitsch
fluid model has been studied in an externally pressured hydrostatic thrust bearing in [5]. Power-law film-shape bearings as
well as inclined-plane shapes and parabolic-film profiles have been treated in [6] in case of an electrically conducting fluid.
It has been shown that the parabolic-film bearing provides significantly higher steady performance than the inclined-plane
E-mail address: badescu@theta.termo.pub.ro
https://doi.org/10.1016/j.apm.2020.01.044
0307-904X/© 2020 Elsevier Inc. All rights reserved.
888 V. Badescu / Applied Mathematical Modelling 81 (2020) 887–909
bearing. Plane inclined slider bearings with micropolar fluids have been studied in [7] where the maximum steady-load-
carrying capacity has been investigated. The second research direction started with the work of Rayleigh [8] who tried to
find the bearing shape which supports the greatest weight. The geometry which is now known as the Rayleigh step bearing
(consisting of two parallel surfaces — one having a rectangular cross-sectional dam) causes the first variation of the weight
functional to vanish.
Long-chain organic additives are used in modern technology to change the physical and chemical properties of oils, in-
cluding viscosity-temperature behavior, pour point, oxidation and corrosion and to improve the extreme pressure resistance
and wear. The classical continuum theory is not suitable for describing the rheological behavior of these non-Newtonian
fluids since it neglects the particle size. The intrinsic motion of fluid constituents is described in a more appropriate way by
micro-continuum theories, which take into account polar effects associated with Cauchy stresses and couple stresses result-
ing from the spin of fluid constituents [9]. Couple stresses effects may be significant in lubrication applications involving oils
with additives and in some cases the results obtained for classical Newtonian fluids may be in need of correction. Therefore,
it is of interest to see how the optimal shapes of slider bearings change when classical Newtonian lubricants are replaced
with couple stress fluids. This is the main scope of this paper. A popular micro-continuum model taking into consideration
couple stresses and body couples has been proposed by Stokes [10] and is used here.
Starting with the seminal work by Rayleigh [8] many researchers optimized 1D slider bearings by using the calculus of
variations with the objective to maximize the carrying capacity. Rhode used the same method to minimize the coefficient
of friction [11]. A method based on Lagrange multipliers has been used in [12] in case the lubricant viscosity depends on
temperature. Also, the calculus of variations has been used in case the viscosity depends on pressure [13]. The classical
Rayleigh step has been found to be the optimum shape for all these cases. More information about previous studies on
slider bearing optimization may be found in Section S1 of the Electronic Supplementary Material (ESM).
The variational optimization techniques are not able to handle differential equations as constraints. The more powerful
methods of optimal control can do this. Maday used indirect optimal control methods based on the Pontryagin Maximum
Principle and the optimal shape he obtained was a step journal bearing [14]. A combination of optimal control methods and
parameter selection has been used in [15]. The optimal solution is again a Rayleigh step slider bearing.
Different performance criteria may be considered. The objective here is to maximize the load of slider bearings operating
with couple stress fluids. Realistic optimization of slider bearing shape should take into account restrictions occurring in
practice. Mathematically, this means constrained optimization. The most important constraints come from basic principles
and consist of lubricant mass conservation, momentum conservation and energy conservation. In the present case these
constraints are governed by one-dimensional differential equations.
Couple stress fluids in step-slider bearing has been previously considered (see [16]) but shape optimization is considered
for the first time here, to the best of our knowledge. Powerful direct optimal control methods are used. The main weakness
of the previous optimal control approaches is that they define the film height as a control. This choice is useful in case
of simple Reynolds equations. However, this approach cannot be used in case of more complex Reynolds equations or in
case of non-Newtonian fluids, where the space derivative of the film height is involved. A new optimal control approach is
proposed in this paper. The main novelty is that the control is not the film height but the gradient of the film height. This
new choice allows a convenient treatment for any kind of Reynolds equations, including that associated with a couple stress
fluid, which is used here.
A model is proposed for the variation of the lubricant velocity along the bearing and for the shear strain rate in the
lubricant. Appropriate relationships to be used for couple stress fluids are derived. Previous optimal control treatments are
performed in the isothermal case [14] or assume a priori temperature distributions [15]. The model developed here is based
on the energy equation and takes into account thermal effects. Further constraints come from manufacturing restrictions
and these are considered as well here. Simple design rules are proposed to handle these constraints.
2. Model
Optimal bearing shapes are derived here by using the optimal control theory. A mature theory is still missing for optimal
control with distributed parameters, despite few studies have been published [17]. Instead, a solid background exists for
one-dimensional optimal control, where indirect methods (such as Bellman-Jacoby dynamic programming or Pontryagin
Maximum Principle) or direct methods are available. Therefore, one dimensional slider bearings are treated here. Infinity
wide bearings in steady state operation are assumed.
The plane slider bearing shape, which defines the height variation throughout the flow strip, is generally described by
many or even an infinity of parameters. In case of simple predefined geometries (such as the inclined bearings, the Rayleigh-
step bearings, the grooved parallel slider bearings [18] and the porous slider bearings [19]) only few parameters are needed.
The geometry of the plane slider bearing considered here is generically described by three known parameters (the bearing
length (L) and the bearing inlet and exit heights (hin ,hout ), see Fig. 1). The bearing shape, i.e. the dependence of bearing
height h on x, is obtained by optimization.
The Reynolds equation, which neglects inertial forces but concentrates on pressure forces and viscous shear, provides a
convenient approach for steady-state flows of lubricants. The isothermal Reynolds equation comes from the assumption of
a constant mass flow rate along the sliding direction (x-axis in Fig. 1) and does not need the assumption h L [3]. Further
comments on different forms of Reynolds equations may be found in Section S2 of the ESM.
V. Badescu / Applied Mathematical Modelling 81 (2020) 887–909 889
Fig. 1. The geometrical configuration considered here.
Table 1
Notation. Eqs. (1.1)–(1.5) are obtained by using symbolic computation [25,29].
Equation Number Expression

h
1.1 f (h, l ) ≡ h3 − 12l 2 h + 24l 3 tanh 2l
h
A(h, l ) ≡ l 2 − 2 lh tanh
3 2
h
1.2 2l
− 12
⎡ h h h h h 2
h ⎤
3
⎢3 h l e
l + 24hl 3 e 2l cosh 2l
− 12h2 l 2 e 2l cosh 2l
− 6h2 l 2 + h4 cosh 2l ⎥
⎣ h h h h h
⎦
−24hl 3 e− 2l cosh − 12h2 l 2 e− 2l cosh − 3hl 3 e− l
1.3 B(h, l ) ≡ 2l 2l
12h2 cosh
2
( 2hl )
h
C (h, l ) ≡ 3h − 12l + 12l sech
2 2 2 2
1.4 2l
h h
D(h, l ) ≡ 24hl − 72l 2 tanh
2
1.5 2l
+ 12hl sech 2l
1.6 ∂ l ∂μ + ∂ l ∂η
VT (l, μ, η, T ) ≡ ∂μ ∂T ∂η ∂ T
1.7 ∂ l ∂μ + ∂ l ∂η
Vp (l, μ, η, p) ≡ ∂μ ∂p ∂η ∂ p
Reynolds equations apply in full-film regions. However, cavitation may occur when the lubricant pressure drops below
its vapor pressure. In this case, pressure remains constant in the cavitated region. Therefore, the momentum equations
take different forms in the full-film region and cavitated region, respectively. Different internal boundary conditions at the
interface between the two regions have been considered such as Sommerfeld-type conditions, Swift-Stiber conditions and
other more advanced conditions (for a good review see [20]). Cavitation can occur in converging-diverging shape bearing
surfaces [21] or in case the external pressure at the inlet and outlet of the bearing has different values. Cavitation effects
should be taken into account in case of textured surfaces [22] or in case of hydrostatic bearings [23]. The constraints used in
the optimization procedure used here exclude obtaining converging-diverging optimal shapes (see Section 4.1). Also, there
is no pressure difference between the bearing inlet and outlet. Therefore, cavitation effects are not likely to appear under
these circumstances.
In case of couple stress fluids, the steady-state Reynolds equation without squeeze film effects takes the form (see for
instance [24]):

d f dp dh
= 6U (1)
dx μ dx dx
where p and μ are lubricant pressure and dynamic viscosity, respectively, while U is bearing sliding velocity. The function f
in Eq. (1) is described in Eq. (1.1) of Table 1. The quantity l entering function f is defined as follows [24]:
η 1 / 2
l≡ (2)
μ
where η is the couple stress parameter. For short, l is called here “couple stress length” since it has dimension of length.
A previous optimization of a 1D slider bearing with non-Newtonian (Rabinowicz) fluid is treated in [15]. A perturbational
method has been used for the state variable, which is pressure. The pressure has been described as a linear superposition
of two state functions and the Reynolds equation has been transformed into a system of two ordinary differential equations
for these two state functions.
Here we do not use perturbational methods. In the general case μ and η depend on the fluid pressure p and temperature
T, which in turn depend on x. Therefore the couple stress length l defined by Eq. (2) depends on x. The space derivative of
the function f in the Reynolds Eq. (1) is:
df dh dT dp
=C − DVT − DVp (3)
dx dx dx dx
where the functions C, D, VT and Vp are defined in Eqs. (1.4), (1.5), (1.6) and (1.7), respectively, in Table 1. C, D have been
obtained by using symbolic computation [25]. From Eqs. (1) and (3) one finds after simple algebra:

2
d2 p 1 ∂μ DVT dT dp 1 ∂μ DVp dp 6μU − C dp
dx dh
= + + + + (4)
d x2 μ ∂T f dx dx μ ∂p f dx f dx
Slider bearing optimization is often performed under the isothermal assumption. In some cases a linear distribution of
temperature is a priori chosen [15]. The Rayleigh step bearing has been studied in [26] by using a thermo-hydrodynamic
lubrication method. Thermal effects may be described by using the following energy equation [27]:

d2 T ∂T dT
k + − ρ cu(x, z ) + μr˙ 2 (x, z ) = 0 (5)
d x2 ∂ z2 dx
where z is the coordinate along bearing height (see Fig. 1), k, ρ and c are the thermal conductivity, density and specific heat
of the fluid, respectively, u is the x-direction fluid velocity and r˙ is the shear strain rate in the fluid.
For bearings with reasonable sliding velocity, conduction is secondary to convection heat transfer. Therefore, the first
term in the l.h.s. member of Eq. (5) is usually neglected. The dependence of u and r˙ on z is given by Eqs. (A1) and (A2) of
[27]:
z
1 dp
u(x, z ) = 1 − U+ ( z − h )z (6)
h 2μ dx
du U 1 dp
r˙ (x, z ) = =− + ( 2z − h ) (7)
dz h 2μ dx
where the following boundary conditions are used:
u=U for z=0
(8a-b)
u=0 for z=h
By using these assumptions and averaging Eq. (5) on z direction one finds:
dT
ρ cū(x ) − μr˙ 2 (x ) = 0 (9)
dx
where the following averaged quantities are defined:
h
1
ū(x ) ≡ u(x, z )dz (10)
h 0
h

2
1 du
r˙ 2 (x ) ≡ dz (11)
h 0 dz
Eq. (11) is similar with that used in [28, p.45]. By using Eqs. (6), (10), (11) and Eq. (1.1) in Table 1 one finds:
U A dp
ū(x ) = + (12)
2 μ dx

2
U2 B dp
r˙ 2 (x ) = + 2 (13)
h2 μ dx
where the functions A and B are defined in Eqs. (1.2) and (1.3) in Table 1, respectively. These functions have been obtained
by using symbolic computation [29].
Eq. (12) provides a better approximation than the approach in [27] where the assumption was that the convection
heat transfer is carried at the cross-film average velocity U/2 of the fluid. A simplified form of Eq. (13) has been used
in [27] where a constant value r˙ 2 = 4U 2 /(hin + hout )2 along the bearing length has been assumed. This assumption comes
from neglecting the second term in the r.h.s. of Eq. (13) and assuming an average film height.
Usage of Eqs. (9), (12) and (13) yields:

2
U A dp dT U2 B dp
ρc + −μ + 2 =0 (14)
2 μ dx dx h2 μ dx
Eq. (14) is non-linear in the pressure gradient in good concordance with other studies (see, e.g. Eq. (2a) of [30]).
The differential Eqs. (4) and (14) act as constraints here. There are two dependent variables (p and T) since in general the
dynamic viscosity μ, the couple stress parameter η, the density ρ and the specific heat c of the fluid depend on pressure and
temperature. The following assumptions are adopted. First, there is no heat transfer at fluid-solid interface. This adiabaticity
hypothesis is often adopted [1].
The boundary conditions are:
p(x = 0 ) = pin (15)
p(x = L ) = pout (16)
T (x = 0 ) = Tin (17)
where pin , pout , Tin are known quantities. Eq. (17) is needed to solve Eq. (14). Eqs. (15) and (16) cover the general case of
hybrid sliding bearings and are used when Eq. (4) is solved. The particular case of pure sliding treated in previous studies
[31] is obtained when pin = pout .
3. Optimization of slider bearing shape
3.1. Optimization methods
The methods of bearing shape optimization are usually classified as stochastic methods and methods based on calculus.
Examples of stochastic methods are genetic algorithms, simulated annealing and artificial life algorithms. These methods al-
low global searching and near-global-solutions are usually found. Another advantage is that the solution is not dependent on
the initial configuration. The main disadvantage is that they are very slow in going towards the global solutions [15]. Also,
in some cases constraints on parameter values cannot be introduced and unrealistic solutions may be found. Genetic algo-
rithms, for instance, cannot guarantee to find an optimal or even near sub-optimal solutions [32]. Therefore, the stochastic
methods are not of very much help when many configurations should be analyzed and compared or when parametric stud-
ies should be performed. The methods based on calculus are using a diversity of mathematical techniques from the fields of
calculus of variations and optimal control, such as dynamic programming, gradient-based algorithms and algorithms based
on the Maximum Principle of Pontryagin. These methods are effective in finding local solutions and they are generally faster
and more accurate than the stochastic methods. However, their convergence strongly depends on the starting point [20].
Hybrid methods combining stochastic algorithms and gradient-based algorithms have been also proposed [33]. A combina-
tion of the harmony search algorithm with the sequential quadratic programming methods has been used in [20]. The aim
of the hybrid methods is to increase the chance of finding a global optimum and improve the convergence speed.
The optimal control techniques used in this paper are faster and more accurate than the stochastic methods and are
appropriate to perform parametric studies. The methods can be easily used in case of 1D slider bearings since a good guess
for the initial starting point is not difficult to find. This choice allows the convenient analysis and comparison of many
configurations.
3.2. Optimal control
The optimization of the bearing shape requires defining the objective function and the control(s). The objective is to
maximize the bearing load per unit bearing width, W, defined as:
L
W ≡ ( p − pout )dx (18)
0
The objective function Eq. (18) and the constraints Eqs. (4) and (14) constitute a Bolza optimal control problem, which is
transformed into a Mayer problem in two steps. First, a new dependent variable g is defined through the equation:
dg
= p − pout (19)
dx
with the boundary condition:
g( x = 0 ) = 0 (20)
Second, the objective function of the Mayer problem is defined:
g(x = L ) → max (21)

Previous optimizations of 1D slider bearings based on optimal control adopted the film height as a control (see [14]).
The difficulty comes from the fact that in the general case the Reynolds equation contains the spatial derivative of the film
height. But the optimal control theory does not operate with derivatives of the control(s). Therefore, the dynamic equations
should be put under a form which does not involve derivatives of the film height. This can be easily done in case of simple
Reynolds equations (for instance, Newtonian fluids in the isoviscous case, neglecting thermal effects (see [15])), but raises
difficulties in case of more complex Reynolds equations, containing dh/dx (such as Eq. (4) for instance).
Table 2
Dimensionless notation.
Notation Meaning of dimensionless quantity
Independent variable
xˆ ≡ xL Space
State variables
hˆ ≡ h
hre f
Lubricant film thickness
p
pˆ ≡ pre f
Lubricant pressure
Tˆ ≡ T
Tre f
Lubricant temperature
πˆ ≡ d pˆ
d xˆ
Gradient of lubricant pressure
Control
d hˆ
υˆ ≡ d xˆ
Gradient of lubricant film thickness
Objective function
g
gˆ ≡ g0
Load per unit length
Other functions
lˆ ≡ l
hre f
Couple stress length
η
ηˆ ≡ ηre f Couple stress parameter
μˆ ≡ μμre f Dynamic viscosity
ρˆ ≡ ρρre f Mass density

c
cˆ ≡ cre f
Specific heat
Here the control is denoted υ (Greek u) to follow the traditional notation. The control is taken to be the gradient of the
film thickness h, i.e.
dh
υ≡ (≤ 0 ) (22)
dx
The control υ is non-positive since the film thickness is allowed to decrease or to be constant locally but is not allowed to
increase locally. This excludes waviness or textured shapes, which are used in practice in the recent years and are associated
with positive values of υ . Also, this excludes converging-diverging configurations prone to cavitation effects.
Table 2 shows dimensionless notation. There, g0 is an arbitrary positive constant quantity while ref denote appropriate
reference quantities. Usage of Eqs. (4), (14), (19) and the notation of Table 2 yields:
d pˆ
= πˆ
dxˆ

22 Bˆπˆ 2
6μ
dπˆ 1 ∂μ ˆ Dˆ VˆT
1
1 μˆ hˆ 2 + μˆ 2 1 ∂μ ˆ Dˆ Vˆ p
ˆ
2 − Cˆπˆ
= + · · πˆ + + πˆ 2 + υˆ
dxˆ μˆ ∂ Tˆ fˆ ρˆ cˆ 1 + 22 Aˆπˆ μˆ ∂ pˆ fˆ fˆ
μˆ
1 2 Bˆπˆ 2
dTˆ 1 μˆ hˆ 2 + μˆ 2
2
= ·
dxˆ ρˆ cˆ 1 + 22 Aˆπˆ
μˆ
dhˆ
= υˆ
dxˆ
dgˆ
= 3 pˆ − pˆ out (23a–e)
dxˆ
where the dimensionless constants 1 to 3 and dimensionless functions fˆ, Aˆ , Bˆ, Cˆ, Dˆ , VˆT and Vˆ p are defined in Table 3.
Eq. (23a–e) are solved by using the following boundary conditions:

pin
pˆ xˆ = 0 ≡ pˆ in =
pre f

pout
p(xˆ = 1 ) = pˆ out =
pre f

Tin
Tˆ (xˆ = 0 ) = Tîn =
Tre f
Table 3
Dimensionless quantities.
Number Constant parameters

2ULμre f
3.1 1 ≡ h2re f ρre f cre f Tre f
h2re f pre f
3.2 2 ≡ ULμre f
L pre f
3.3 3 ≡ g0
Functions
3.4 fˆ(hˆ , lˆ) ≡ f h(h,l
3
)
re f
3.5 Aˆ (hˆ , lˆ) ≡ Ah(2h,l )

re f
3.6 Bˆ(hˆ , lˆ) ≡ Bh(h,l

2
)
re f
3.7 Cˆ(hˆ , lˆ) ≡ Ch(h,l

2
)
re f
3.8 Dˆ (hˆ , lˆ) ≡ Dh(3h,l )

re f
Tre f
3.9 VˆT (lˆ, μ
ˆ , ηˆ , Tˆ ) ≡ hre f
VT (l, μ, η, T )
Vˆ p (lˆ, μ
ˆ , ηˆ , pˆ ) ≡
pre f
3.10 hre f
Vp (l, μ, η, p)
gˆ(xˆ = 0 ) = 0
hin
h(xˆ = 0 ) = (24a-e)
hre f
where the notation of Table 2 has been used. Eq. (24a-c) have been obtained from Eqs. (15)–(17) while Eqs. (24d) and
(24e) come from Eq. (20) and notation of Table 2, respectively.
In summary, the Mayer optimal control problem is defined as follows:
• independent variable: dimensionless space xˆ;
• state variables: dimensionless film thickness hˆ , pressure and temperature pˆ and Tˆ , respectively, and pressure gradient πˆ ;
• control: the dimensionless gradient of film thickness, υˆ , and
• objective function: gˆ.
The objective function gˆ (coming from Eq. (21) and notation in Table 2) is maximized under the constraints of the ordi-
nary differential Eqs. (23a–e), which are solved by using the boundary conditions Eqs. (24a–e).
3.3. Numerical procedures and model implementation
The numerical method is shortly described now. The optimal control problem consists of the constrained extremization
of the objective function. The constraints are ordinary differential equations for the state variables and control(s). Opti-
mal control problems may be solved by using direct or indirect methods. A direct optimal control method is used here.
It transforms the optimal control problem into a non-linear programming problem. This is performed by a discretization
in the space of the independent variable, applied to the state variables and control(s), as well as the constraints differen-
tial equations. More details can be found in [34]. Direct methods are usually less accurate than indirect methods but they
are widely used in engineering applications since they are more straightforward to apply and more robust with respect to
initialization. The BOCOP programming package [35] is used here. It transforms the optimal control problem described by
the user through several C++ functions into a non-linear programming problem. BOCOP is designed for objective function
minimization. However, Eq. (21) asks for bearing load maximization. Implementation in BOCOP requires defining a new ob-
jective, i.e. − g(x = L) → min . BOCOP has several discretization methods. In this work we used two methods: Euler (explicit,
1- stage, order 1) and Midpoint (implicit, 1- stage, order 1). The first method is faster. The number of discretization steps
for the independent variable is 10 0 0. This corresponds to a space step of 0.001. The maximum allowed number of iteration
is 10,0 0 0 while the tolerance is 10−14 .
3.4. Model validation
3.4.1. Validity of Reynolds equation

Several studies analyzed the circumstances when the Reynolds equations are good approximations of the Navier-Stokes
equations. A condition is the thin film assumption, which means that the scale along the bearing is at least three orders of
magnitude larger than the scale across the fluid film [36]. A systematic study has been performed in [37] for 1D Rayleigh
step bearings. It has been found that for runner velocity of 30 m/s the difference between the carrying load predicted by
using Navier–Stokes and Reynolds equations is less that 5% for film thickness around 0.1 mm and 20% for film thickness of
Fig. 2. Rayleigh step bearing. (a) Bearing shape; (b) Distribution of fluid pressure.
0.5 mm. These errors come from the inertia effects due to the pressure drop after the discontinuity line. For lower velocity
values (e.g. 10 m/s), the difference increases by increasing the film thickness. The authors concluded that Reynolds equations
may be used for film thickness smaller than 0.2 mm, when the errors are less than 5%. In [38] the influence of the modified
Reynolds number on the fluid inertia has been experimentally and theoretically analyzed in an infinite slider bearing. The
authors concluded that for Re < 1 the Reynolds equation may be used since the fluid inertia is negligible. Fluid inertia
gradually increases and the usage of Reynolds equation is not recommended for Re larger than 10. Temperature effects are
smaller than inertia effects [32]. Also, 1D Reynolds equations may be used in case of large enough bearing width [21].
Several constraints have been used here for the geometrical configuration and the operation regime, ensuring the validity
of the Reynolds equation. First, the thin film assumption has been adopted, i.e. hin ≤ L/10 0 0. Second, films which are thinner
than hin = 0.1 mm have been considered. Third, the Reynolds number has been estimated as follows. The upper value for the
equivalent diameter of a cross- section through the fluid film is Deq = 2hin B/(hin + B), where B is the bearing width. For
infinitely large bearings B → ∞ and Deq → 2hin . The average fluid velocity is given by Eq. (12) where the pressure gradient
is neglected, i.e. ū = U/2. The Reynolds number is defined as Re = Deq ū/μ = hinU/μ. The third constraint adopted here is
Re ≤ 1. These constraints have been used during the optimization procedure in order to avoid potential solutions which are
not compatible with valid Reynolds equations.
3.4.2. Validation of the optimal control procedure

A case with known solution is used to validate the proposed optimal control procedure. Fig. 2a shows a 1D Rayleigh step
bearing.
Two flow regions of constant film thickness are identified [3]: the ridge (or step) of length Lstep and the film land of
length L − Lstep . The optimization problem asks to finding the step length and the inlet (or outlet) bearing height making the
bearing load a maximum. Next, the known analytical solution and the numerical optimal control solution are compared in
the simplest case: Newtonian isoviscous fluid and isothermal flow.
Analytical solution: For a Newtonian fluid (η = 0Ns) the couple stress length l defined by Eq. (2) vanishes. Therefore, the
function f defined by Eq. (1.1) of Table 1 reduces to f = h3 . Under the additional assumptions that the flow is isothermal and
isoviscous, the steady-state Reynolds Eq. (1) takes the simple form:

d dp
h3 − 6μUh =0 (25)
dx dx
A first integration of Eq. (25) gives:

dp
h3 − 6μUh = −CQ (26)
dx
where CQ is an integration constant (CQ = 12μQ, where Q is the volumetric flow rate per unit bearing width [3]). Integration
of Eq. (26) is performed separately for the step and land regions. It is shortly described in Table 4. In step #1 the pressure
variation is obtained in terms of the integration constants CQ , Cin and Cout . In step #2 the boundary conditions are specified.
Following tradition a null pressure is adopted at bearing inlet and outlet (pin = pout = 0 Pa). In step #3 the constants of
integration CQ , Cin and Cout are obtained.
The following notation is adopted:
Lstep
α≡ (< 1 )
L
hout
γ ≡ (< 1 ) (27a,b)
hin
Table 4
Steps used for the integration of Eq. (26).
Step Step region x ∈ [0, Lstep ] Land region x ∈ (Lstep ,L]

−CQ + 6μU hin −CQ + 6μU hout
1 Pressure variation p= x + Cin p= x + Cout
h3in h3out

0 pin = 0 Lstep p = pstep
2 Boundary conditions x= x=
Lstep p = pstep L pin = 0
−Q + 6μU hout
3 Integration constants Cin = 0 Cout = − L
h3out

h3in pstep h3out 6μU hout
−CQ = − 6μU hin −CQ = pstep − (Lstep − L )
Lstep Lstep − L h3out
Table 5
Input quantities used during validation procedure.
Parameter Value
Slider velocity, U (m/s) 2

Bearing length, L(m) 0.03
Bearing inlet height, hin (m) 0.0001
Inlet pressure, pin (Pa) 0
Outlet pressure, pout (Pa) 0
Fluid dynamic viscosity, μ(Ns/m2 ) 0.0782
Reference pressure, pref (Pa) 105
Constant g0 (N/m) 1000
The maximum pressure value pstep is found by equating the expressions of − CQ in the two columns of the last row in
Table 4 and using Eqs. (27a,b):
6μUL (1 − α )(1 − γ )γ
pstep = (28)
h2in 1 − γ + α3γ
Eq. (28) and the equations in the two columns of last row in Table 4 allows finding the integration constants CQ and Cout
but this is not of interest here.
The bearing load is pstep LB/2 (see Fig. 2b). The bearing load per unit bearing width (B = 1 m), W, is obtained by using
Eq. (28):
1 3μU L2 (1 − α )(1 − γ )γ
W = pstep L = (29)
2 h2in 1 − γ + α3γ
For given value of the ratio α = hout /hin , the optimization of 1D Rayleigh step bearings asks to find the optimum value of
γ (i.e. the optimum step position) which makes the bearing load W a maximum.
Notice that the common approach of 1D Rayleigh step optimization shows results in terms of given hout rather than
given hin , as we did here (see e.g. [3]). In that case an optimum value of α exists and the specific optimum values which
determine a maximum maximorum bearing load are α opt = 0.5357 and γ opt = 0.7182. This is already known (see e.g. [39]).
Optimal control solution: The optimal control model Eqs. (23a–e) is now particularized to the case treated analytically. A
Newtonian fluid (η = 0Ns) under steady-state isothermal and isoviscous flow is assumed. Also, pout = 0 Pa. The result is:
d pˆ
= πˆ
dxˆ
dπˆ 6 υˆ υˆ πˆ
= −3
dxˆ 2 hˆ 3 hˆ
dhˆ
= υˆ
dxˆ
dgˆ
= 3 pˆ (30a-d)
dxˆ
The optimal control procedure has been run for given value of hin and different values of α (or, in other words, different
values of hout ). The input values of Table 5 have been used.
A two step procedure has been used. First, no constraints have been used for the state variable πˆ . This allows obtaining
a first estimation for the value of the optimum ratio (Lstep /L)opt , say γõpt . Second, the value γõpt is used to obtain more
accurate constraints for the state variable πˆ , as described next. The volumetric flow rate is conserved in the step and land
regions of the bearing. Eq. (26) gives:
h3re f pre f h3re f pre f
hˆ 3in πˆ in − 6μU hre f hˆ in = − hˆ 3out |πˆ out | − 6μU hre f hˆ out (31)
L L
Table 6
Constraints used during model validation.
Quantity Lower bound Upper bound
pˆ 1 none
hˆ α 1
gˆ 0 none
πˆ 1st stage: none 1st stage: none
2nd stage: πˆ min 2nd stage: πˆ max
υˆ none 0
Here notation of Table 2 is used and we took into consideration that πˆ out is a negative quantity. πˆ in and πˆ out are the
slopes of the pressure space variation in the step and land regions, respectively. The geometry of Fig. 2b allows writing:
γ˜ πˆ in = (1 − γ˜ )|πˆ out | (32)
Usage of Eqs. (31) and (32) yields:
6μUL (1 − α )(1 − γ˜ )
πˆ in =
h2re f pre f 1 − γ˜ + γ˜ α 3
6μUL γ˜ (1 − α )
|πˆ out | = (33a-b)
h2re f pre f 1 − γ˜ + γ˜ α 3
The values πˆ out (< 0 ) and πˆ in are used as lower and upper bounds for πˆ , respectively:
πˆ min = πˆ out
πˆ max = πˆ in (34a-b)
In the present approach, href = hin . Table 6 shows the constraints used during validation.
Figures S1 and S2 in Section S3 of the ESM show results obtained by using the numerical optimal control procedure
for the particular case α = hout /hin = 0.55. The Euler and Midpoint discretization methods have been used. Both methods
estimate similar values for the ratio γ = Lstep /L (i.e. the step position) (see Fig. S2c in the ESM). The Midpoint discretization
gives at step position a more abrupt space variation (impulse-type) of the control υˆ than the Euler discretization (compare
Figs. S2b and Fig. S2a, respectively, in the ESM). More abrupt variations of υˆ yield better description for the sudden slope
change of the pressure variation at the step (see Fig. 2b).
Optimal control vs analytical approach: A comparison between analytical results (obtained by using relationships shown
in Section Analytical solution) and numerical optimal control results (obtained by using the procedure presented in Section
Optimal control solution) is shown in Fig. 3 for different values of α = hout /hin . Generally, there is good concordance between
the analytical and numerical results. The optimum step position (i.e. γ opt = (Lstep /L)opt ) is slightly underestimated by the nu-
merical procedure at smaller values of α and slightly overestimates at larger α s (Fig. 3a). Midpoint discretization is slightly
more accurate than Euler discretization in estimating γ opt . The numerical procedures are more accurate in estimating the
maximum bearing load per unit width Wmax at larger values of α (Fig. 3b). Euler discretization is slightly more accurate
than Midpoint discretization. Overall, Midpoint discretization provides a better description of the optimal bearing profile
than Euler discretization.
4. Sub-optimal classes of slider bearings
Unconstrained optimization searches over the given set of design and operation configurations and finds an optimal
solution. Constrained optimization covers a sub-set of configurations and therefore finds a sub-optimal solution.
4.1. Constraints consistent with input quantities
Eq. (23b) makes the Hamiltonian of the optimal control problem linear in the control υˆ . The control is therefore non-
regular and includes the impulse solution as a particular case. Realistic problems need constrained optimization and the
sub-optimal shapes in this case are usually more complex than the simple Rayleigh step bearings.
The solution and the convergence of the optimal control methods are dependent on the lower and upper bounds adopted
for state variables and controls. Generally, by increasing the range of variation for state variables and controls is associ-
ated (in case of maximization problems) with larger values of the objective function. However, using larger upper or lower
bounds is not always the best solution since the local sub-optimal solutions may be lost in case too large upper/lower
bounds are adopted, due to the finite sub-space of approximation.
Finding appropriate bounds is a matter of experience and trial. Few comments follow. Higher values of the objective
functions are associated with more abrupt variation of bearing height in the inlet region. However, a too abrupt variation
of the bearing shape in the inlet region puts some question marks. Indeed, assume the film thickness changes from hin
Fig. 3. Comparison between analytical results and numerical optimal control results. (a) The optimum ratio γ opt = (Lstep /L)opt ; (b) the maximum bearing
load per unit width Wmax . Several values of the ratio α = hout /hin have been considered.
to h∗in (<< hin ) in a short inlet region of length

xin ( L). Then, the mean fluid flow direction in that short inlet region is
oriented in z direction rather than in x direction (see Fig. 1). The mean fluid flow turns into the x direction after the short
inlet length
xin and this makes h∗in the effective bearing inlet height. Therefore, the inlet shape should be less abrupt,
making the mean flow to be oriented in x direction from the very bearing inlet at x = 0. Two sorts of constraints may be
used. First, not too large positive values should be imposed for the pressure derivative dp/dx. Second, not too large negative
values should be imposed for dh/dx (which is the control υ , see Eq. (22)). Combination of these constraints yield different
sub-optimal bearing shapes, as shortly described in Section 4.2.
The constraint dh/dx ≤ 0 (i.e. negative or null values of the control υ ) corresponds to a marginally monotonous decrease
of the bearing height. This constraint is important from two points of view. First, it allows proper convergence of the optimal
control method (which is effective in finding local optima) towards a global optimum. Allowing positive and negative values
of υ would correspond to textures surfaces. In this case, which is to be avoided, the optimal control solver may be trapped
in a local optimum, depending on the starting point. Second, this constraint makes impossible the occurrence of diverging
regions and avoids apparition of cavitation.
4.2. Rules for constrained optimization of slider bearing shapes
The differential Eq. (23b) is linear in the control υˆ and the optimal solution involves impulse-like jumps of υˆ between
its minimum value υˆ min (< 0 ) and its maximum value υˆ max (= 0 ). These jumps have different effects on the bearing shape,
depending on the bounds imposed to πˆ , i.e. πˆ ∈ [πˆ min (< 0 ), πˆ max (> 0 )]. Two different classes of sub-optimal bearing shapes
may be obtained, as described by the next two design rules.
Rule #1. Sub-optimal multi-stepped bearing shapes come from
• not too large negative values for πˆ min ;

• low positive values for πˆ max ;
• no bound on the negative value υˆ min .
Table 7
Input quantities and their values and boundary condi-
tions.
Parameter
Slider velocity, U (m/s) 10
Bearing length, L(m) 0.01
Inlet pressure, pin (Pa) 105
Outlet pressure, pout (Pa) 105
Inlet temperature, Tin (K) 293
Bearing inlet height, hin (m) 0.0001
Constant g0 (N/m) 1000
Lubricant
Lubricant type HM32
Boundary conditions
Inlet pressure pˆ (xˆ = 0 ) = 1
Outlet pressure pˆ (xˆ = 1 ) = 1
Inlet bearing height hˆ (xˆ = 1 ) = 1
Objective function gˆ(xˆ = 0 ) = 0
Depending on the values of πˆ min and πˆ max , different sub-optimal solutions are obtained, which may be connected with
the sub-optimization problems in the space of multi-stepped profiles considered in the earlier work [40].
Rule #2. Sub-optimal multi-sloped bearing shapes come from
• no bound for πˆ min and πˆ max ;

• low negative values for υˆ min .
These rules are used in Sections 5.3 and 5.4.
5. Results
As a first approximation we assume that the dynamic viscosity μ and the couple stress parameter η do not depend
on temperature and pressure. Therefore, an isothermal analysis is performed now. This assumption has been used in many
slider bearing optimization studies [3] and avoids known difficulties related with the solution of the energy equation at steps
[41]. It has been shown that the isothermal assumption works well for several applications like piston rings and cylinder
liner conjunctions [42] and big-end bearings [43]. The assumption of negligible thermal gradients has been used in [21].
Eqs. (23a–e) are run under these hypotheses. This means that Eq. (23c) is not used and the r.h.s. member of Eq. (23b) re-
duces to the third term. Further assumptions adopted here are presented in Section 5.1. Section 5.2 presents the constraints
including the lower and upper bounds used to implement Rules #1 and #2 of Section 4.2. Representative results are shown
in Sections 5.3 and 5.4.
Stability issues and a sensitivity analysis are presented in Section 6. In Section 7 a brief comparison is made between the
isothermal and temperature dependent approaches.
5.1. Assumptions and input quantities
The reference values of temperature and pressure correspond to bearing inlet, i.e. Tref = Tin and pref = pin . Also, ηref ,μref ,ρ ref
and cref denote appropriate quantities evaluated at bearing inlet temperature Tin and pressure pin .
Notice that for very small gap clearances, any abrupt changes in the thickness of a film may break the validity of the
Reynolds equation, which is based on lubrication theory. Cases where the classical engineering approach is justified are
studied in [44]. The input quantities and their values are listed in Table 7. The common outlet bearing height ranges between
30 and 50 μm [2]. Taking into account the value hin = 100μm adopted in Table 7, a reasonable range of variation of the ratio
hout /hin is 0.25 to 0.6. Few comments follow about the value g0 = 10 0 0 N/m in Table 7. Solving optimal control problems
may face convergence difficulties in case the state variables have different magnitude orders. These difficulties are usually
avoided by properly scaling of the state variables. Using dimensionless models is a convenient procedure. In some cases the
reference values come in a natural way from the physics of the problem (see Table 2). In other cases, such as the definition
of the dimensionless quantity gˆ, the reference value (i.e. the value of g0 ) is to be stated after several trials. The reference
value is to some degree arbitrary but the specific value g0 = 10 0 0 N/m ensures solution convergence for all cases treated in
this paper.
Appendix A shows the physical properties of the lubricant HM32 considered in this paper. Different values of the couple
stress length l (see Eq. (2)) are found in literature and they may be used to obtain values of the couple stress parameter
η, which are of the order (in Ns): 10−20 [24], 10−10 [9], 10−10 –10−9 [45], 10−11 –10−8 [46]. The range of variation of η is
extended in this paper, to give perspective to the results.
Table 8
Lower and upper bounds for dependent variables, objective function and control.
Quantity Lower bound Upper bound
All cases
pˆ 1 pˆ max (to be specified in text)

Tˆ 1 none
hˆ 0.01 1
gˆ 0 none
Multi-stepped bearings (Rule #1)

πˆ −103 10
υˆ none 0
Multi-sloped bearings (Rule #2)

πˆ none none
υˆ −10 0
5.2. Constraints
Sub-optimal bearing shapes depend on the constraints imposed to the dependent variables and control. The constraints
generally depend on case. In practice, the meaning of “low” and “not too large” in the Rules #1 and #2 of Section 4.3 is
found by trial procedures. Table 8 shows the constraints used here.
5.3. Multi-stepped slider bearings
Rule #1 of Section 4.3 has been implemented by using the input values listed in Table 8 for the lower and upper bounds
of the state variable πˆ and control υˆ . Optimization of slider bearing shapes has been made for several values of the maxi-
mum pressure pmax . Results are shown in Fig. 4 for pmax = 2 bar and 10 bar.
The sub-optimal bearing shape consists of more than one step, the last step being placed near the bearing exit (Fig. 4a
and b). Increasing the maximum pressure pmax increases the length of the inlet constant height interval (compare Fig. 4a
and b). The shape depends significantly on η only at higher values of the couple stress parameter. Increasing η increases the
height of the middle constant height interval. Most values of η in the literature range between 10−20 and10−8 Ns. In these
cases the sub-optimal multi-stepped bearing shape is similar with that obtained for a Newtonian fluid (η = 0 Ns).
Results not presented here show that the maximum objective function gˆ does not depend significantly on the value of
the couple stress parameter η .
Rayleigh journal bearings with more than one step have been studied in [47]. Series of Rayleigh steps in journal bearings
have been proposed in [48]. However, those studies assume predefined bearing shapes. The bearing profiles in Fig. 4 are
optimal since they are obtained by maximization of the load-carrying capacity.
Manufacturing Rayleigh step bearings with differences between the inlet and outlet heights smaller than 25 μm is dif-
ficult. Also, such bearings are prone to improper operation since a small amount of wear may wipe away the step [2]. The
sub-optimal stepped bearings obtained in this paper (see Fig. 4) have several height variations, some of them of the order
of 10 μm. Manufacturing and operating such bearings require special care and seem to be better suited for gas lubrication.
5.4. Multi-sloped slider bearings
Rule #2 of Section 4.3 has been implemented by using the input values listed in Table 8 for the lower and upper bounds
of the state variable πˆ and control υˆ . Results are shown in Fig. 5 for two values of the maximum pressure pmax . Notice that
the lower bound υˆ min = −10 corresponds to an angle of about 5° between the bearing shape at inlet and the axis z in Fig. 1.
The representation in Fig. 5 deforms this angle since the horizontal and vertical axes have different scales. Lower negative
values for υˆ min would increase the angle and would make the bearing shape less abrupt at inlet. This is associated with
lower values of the objective function, of course.
The sub-optimal shapes of multi-sloped bearings consist of an inclined inlet followed by several constant-height and
sloped intervals, respectively (Fig. 5a,b). This is in good agreement with previous results [49] where it has been found that
the slider bearing flow problem consists of three regions: the inlet flow, the bearing flow, and the outlet flow. By increasing
the maximum pressure pmax one obtains shorter inlet regions and larger heights of the constant-height intervals (compare
Fig. 5a and b). The sub-optimal shape may change when the value of the coupled stress parameter changes, but only for
higher values of η. The sub-optimal shape in the inlet region does not depend on η. The sub-optimal shape obtained for a
Newtonian fluid (η = 0 Ns) may be used in case of most couple stress fluids encountered in literature.
Results not presented here show that the value of the maximum objective function gˆ slightly increases by increasing
the value of the couple stress parameter η. This is in good agreement with previous results obtained on poro-elastic slider
bearings [50].
Fig. 4. Dependence of the dimensionless height hˆ on the dimensionless space xˆ for sub-optimal multi-stepped slider bearing shapes obtained by using
Rule #1 and input values of Table 8. (a) pmax = 2 bar and (b) pmax = 10 bar.
Notice that optimal 1D single-sloped slider bearings have been obtained in [51] by using the calculus of variations in case
the objective was to minimize the friction coefficient. The objective here is to maximize the bearing load and multi-sloped
optimal bearings are obtained.
Sub-optimal multi-sloped bearings are easier to manufacture than stepped bearings. Also, these bearings have smoother
operation than the stepped bearings since wear products are not accumulated near the height jumps and they may be easier
removed.
6. Stability and sensitivity analysis
6.1. Stability
Several researchers focused on bearings stability by using theoretical models and experiments [52]. This is a major con-
cern in applications of gas bearings demanding accurate positioning control (such as the flying head of hard disk drives).
Stability issues of oil-lubricated bearings are less stringent due to the higher stiffness provided by liquid lubricants compared
to gases [20]. Bearing instability is a problem peculiar to thrust bearings with fixed geometry [2].
Bearings stability under variable loads is an important issue of practical relevance. A few comments follow about the
stability of the optimal 1D slider bearings. The simplest case of isothermal and isoviscous fluid flow is considered. The
approach is as follows. First, a simple analytical model is developed in Section 6.1.1 to estimate the carrying-load capacity
of bearings with know profile. Second, the analytical model is tested in Section 6.1.2 against a numerical optimal control
model. Third, the analytical model is used in Section 6.1.3 to perform the stability analysis of a deformed slider bearing.
6.1.1. Analytical model of bearings with known profile

A simple analytical model is proposed to estimate the carrying load of a 1D bearing with known profile. The bearing
length L is divided into mp − 1 slices of equal length. The bearing inlet and outlet have the abscissa x1 = 0 and xm p = L,
respectively. The abscissa of slice j range between xj and xj + 1 . Fig. 6 shows the slice numbered j in the 1D bearing.
Fig. 5. Dependence of the dimensionless height hˆ on the dimensionless space xˆ for sub-optimal multi-sloped slider bearing shapes obtained by using Rule
#2 and input values of Table 8. (a) pmax = 2 bar; (b) pmax = 10 bar.
Fig. 6. Slice numbered j in the 1D slider bearing. x, h and p denote abscissa, film height and fluid pressure, respectively.
A linear film height variation is assumed inside the slice j. Therefore:

h = m jx + h j − m jx j (35)
where:
h j+1 − h j
mj ≡ (36)
x j+1 − x j
Usage of Eqs. (26) and (36) gives the pressure gradient inside the slice j:

dp −CQ + 6μU m j x + h j − m j x j
= 3 (37)
dx m jx + h j − m jx j
Table 9
Notation to be used in Eqs. (38) and (38).
Number Notation
9.1 aj ≡ 6μUmj , bj ≡ mj , cj ≡ hj − mj xj , dj ≡ 6μU(hj − mj xj )

a j ( 2b j x + c j ) dj
9.2 R j (x ) = − −
2b2j (b j x + c j ) 2b j ( b j x + c j )
2 2
1
9.3 H j (x ) = −
2b j ( b j x + c j )
2
9.4 Tj ≡ Rj (xj + 1 ) − Rj (xj )

9.5 Zj ≡ Hj (xj + 1 ) − Hj (xj )
Integration of Eq. (37) gives the pressure variation inside the slice j:

p(x ) = R j (x ) − H j (x )CQ + C j x ∈ x j , x j+1 (38)
where the notation of Table 9 is used and Cj is an integration constant.
The following boundary conditions are used:
p = pj x = xj
(39a-b)
p = p j+1 x = x j+1
Usage of Eqs. (38) and (39a) allows obtaining the integration constant Cj :
C j = p j − R j (x j ) + H j (x j )CQ (40)
Usage of Eqs. (38) and (39a,b) yields the following relationship between pj + 1 and pj :
p j+1 = p j + T j − Z jCQ (41)
where the notation of Table 9 is used.
The inlet and outlet pressure values are known:
p1 = pin (42a)
pm p = pout (42b)
Summing up Eq. (41) from j = 1 to j = mp − 1 and using Eqs. (42a,b) allows finding the integration constant CQ :
m p −1
pin − pout + j=1 Tj
CQ = m p −1 (43)
j=1
Zj
The volumetric flow rate per unit bearing width, Q, may be obtained from Q = CQ /(12μ). The integration constants Cj (j = 1,
mp − 1) are obtained by using Eqs. (43) and (40). The fluid pressure distribution along the bearing length is obtained by
using Eq. (38) and the known values of CQ and Cj (j = 1, mp − 1).
The carrying load per unit bearing width W is obtained by summing up the contributions of all mp − 1 slices:
1
m p −1

W = x j+1 − x j p j − pout + p j+1 − pout (44)
2
j=1
This simply evaluates the area under the curve p(x) − pout as Eq. (18) does.
The values mp = 1001 is used in next applications.
6.1.2. Testing the analytical model

The analytical model has been tested as follows. First, optimal profiles have been obtained for a multi-stepped and a
multi-sloped slider bearing, respectively, under the constraint that the maximum dimensionless pressure equals pˆ max = 2.33.
The boundary conditions and the constraints of Table 10 have been used.
Second, the analytical model developed in Section 6.1.1 has been used to estimate the pressure distribution for the (given)
optimal profiles shown in Fig. 7a and b, respectively, under the same maximum pressure constraint. In case of multi-stepped
profile, the simple linear analytical model has lower accuracy at the abscissa where the steps are located. The maximum
pressure at first step (see Fig. 7a) is lower by 3% than pˆ max . A model calibration is performed in this case by appropriately
increasing the pressure. The analytical model is performing very well in case of multi-sloped profiles. No calibration proce-
dure is needed. Pressure distributions obtained by using the optimal control procedure and the analytical model proposed
in Section 6.1.1 are shown in Fig. 7c and d, respectively. The pressure distributions estimated by the analytical model are
Table 10
Boundary conditions and constraints for optimal multi-stepped and multi-sloped slider bearings.
Quantity Boundary conditions Lower limit Multi-stepped Multi-sloped
Upper limit Lower limit Upper limit
pˆ (0 ) = 1
pˆ 1 2.33 1 2.33
pˆ (1 ) = 1
πˆ none −1000 10 none none
hˆ hˆ (0 ) = 1 0.5 1 0.5 1
gˆ gˆ(0 ) = 0 0 none 0 none
υˆ none none 0 −10 0
The optimal profiles are shown in Fig. 7a and 7b, respectively.
Fig. 7. Optimal profiles of (a) multi-stepped slider bearing and (b) multi-sloped slider bearing. Dependence of dimensionless pressure pˆ on dimensionless
space xˆ obtained by using the optimal control procedure (BOCOP) and the analytical model proposed in Section 6.1.1 for (c) multi-stepped bearing and (d)
multi-sloped slider bearing. U = 2m/s, L = 0.03m, μ = 0.07288Ns/m2 ,hin = 0.0 0 01m.
in good agreement with those provided by the optimal control procedure. In case of the multi-stepped bearing, the values
of the carrying load per unit width, W, estimated by the optimal control method and the analytical model, are 3642 N/m
and 3634 N/m, respectively. When the multi-sloped bearing is considered, the two values are 3621 N/m and 3616 N/m, re-
spectively. Again, there is good concordance between results provided by the optimal control procedure and the analytical
model.
6.1.3. Stability of optimal profiles

A bearing of optimal profile provides the maximum carrying load. In case the external load exceeds the maximum car-
rying load, the film thickness decreases and, therefore, the bearing profile is deformed. The carrying load of the deformed
bearing is estimated next by using the analytical model proposed in Section 6.1.1.
Fig. 8. (a) Carrying load per unit bearing width, W and (b) Volumetric mass flow rate per unit bearing width, Q. The dependence on the inlet height h∗in of
the deformed bearing is considered for both multi-stepped and multi-sloped bearings. (For interpretation of the references to colour in this figure legend,
the reader is referred to the web version of this article.)
Several hypothetical deformed profiles are considered. These deformed profiles are simply obtained by down parallel
translation of the top surface of the optimal profile and keeping fixed the bottom bearing surface. Therefore, the inlet height
(h∗in , say) of these deformed profiles is smaller than the inlet height of the optimal profile (hin = 100 μm) and the deformed
bearings are narrower than the optimal bearings. The deformed profiles are used as input for the analytical model proposed
in Section 6.1.1 and the pressure distributions, the volumetric mass flow rate per unit width, Q, and the carrying load per
unit width,W, are estimated. Fig. 8 shows results.
Consider a multi-stepped slider bearing. Assume that the external load equals the maximum carrying load obtained by
the optimal control procedure. Then, the inlet bearing height is hin = 100 μm and the bearing profile is optimal, as shown in
Fig. 7. In case the external load becomes larger than the maximum carrying load, the bearing is deformed. The top bearing
surface moves down towards the bottom surface and the deformed inlet bearing height h∗in becomes smaller than the initial
value hin . Fig. 8a shows that in this case the carrying load increases. As a consequence, h∗in has a tendency to get back to
the initial value hin . The higher the external load is, the smaller the deformed bearing inlet height h∗in is and the larger the
carrying load becomes. Therefore, the stronger is the tendency of the deformed bearing height h∗in to return towards the
value hin of the initially undeformed bearing. Consequently, the operation of multi-stepped slider bearings is stable. Similar
arguments allow to conclude that the operation of multi-sloped slider bearings is stable.
Higher external loads yield smaller values h∗in of the inlet height. As a consequence, the volumetric flow rate decreases
(see Fig. 8b), as expected.
6.2. Sensitivity analysis
Manufacturing slider bearings with optimal profiles may be associated with technological difficulties. Have the values of
all design parameters similar effects on the bearing performance? A short sensitivity analysis might be useful in identifying
the main design parameters. The optimal profiles generated in Section 6.1.2 (see Fig. 7) are used as examples. They are
shown with more details in Fig. 9.
Fig. 9. Optimal slider bearing profiles. (a) multi-stepped bearing; (b) multi-sloped bearing. For details see Section 6.1.2.
Table 11
Parameters selected for sensitivity analysis (see Fig. 9). Optimal values are shown in dimensionless and dimensional form. The range of
variation of these parameters is also shown in dimensionless and dimensional form and as fraction of the optimal values.
Parameter Range of variation
Optimal value Dimensionless Dimensional (mm) Fraction of optimal value (%)
Dimensionless Dimensional (mm) minimum maximum minimum maximum minimum maximum
Multi-stepped slider bearing

xˆ2 0.134 4.02 0.107 0.161 3.21 4.83 80 120
xˆ4 0.958 28.74 0.766 0.958 22.98 28.74 80 100
hˆ 4 0.653 0.0653 0.522 0.784 0.0522 0.0784 80 120
Multi-sloped slider bearing
xˆ2 0.07 2.1 0.056 0.084 1.68 2.52 80 120

xˆ4 0.141 4.23 0.113 0.169 3.39 5.07 80 120
hˆ 2 0.933 0.0933 0.746 0.98 0.0746 0.0980 80 105
hˆ 4 0.622 0.0622 0.529 0.746 0.0529 0.0746 85 120
Several parameters have been selected for analysis. For the multi-stepped bearing (Fig. 9a) we selected the abscissa of
the first two steps, x2 and x4 , and the height h3 of the constant height segment 34. The abscissa x2 and x4 of the end
points of segments 12 and 34, respectively, have been selected for the multi-sloped bearing (Fig. 9b), together with the
heights h2 and h4 of the constant height segments 23 and 45, respectively. Notice that the dimensionless representation in
Fig. 9 uses different scales on the two axes (the scale ratio between ordinate and abscissa is 1:300). Therefore, the actual
representation of the sloped segments 12, 34 and 56 in Fig. 9b have a lower inclination. Table 11 shows the optimal values
of these parameters, in dimensionless and dimensional form.
The manufacturing process is prone to technological errors. Therefore, in practice the values of the selected param-
eters have a variation range. We assume that this variation may range between −20% and +20% of the optimal values.
Table 11 shows the range of variation of the selected parameters. In some cases this variation range is smaller, in order
Fig. 10. The dependence of the carrying load of a multi-stepped bearing (expressed as percentage of the maximum carrying load W of the optimal profile)
on the values of several parameters (expressed as percentage of the optimal values). (a) height h3 ; (b) abscissa x2 ; (c) abscissa x4 (See Fig. 9).
to keep unchanged the topology of the bearing (see, for instance, the parameter xˆ4 for the multi-stepped bearing and the
parameters hˆ 2 and hˆ 4 for the multi-sloped bearing).
The simple linear analytical model developed in Section 6.1.1 has been used to estimate the bearing carrying load, for dif-
ferent values of the parameters in their range of variation, under the constraint pˆ max = 2.33. The carrying load is expressed
as fraction of the maximum carrying load, which characterize the bearings with optimal profile. The maximum carrying load
is 3634 N/m and 3613 N/m for the optimal multi-stepped and multi-sloped bearing, respectively (see Section 6.1.2).
The multi-stepped bearing is considered now. The carrying load is sensitive to the value of the height h3 , especially when
it is lower than the optimal value (Fig. 10a). Values of the abscissa x2 larger than the optimal value have neglecting effect on
the carrying load but when x2 is lower than the optimal value the carrying load may decrease significantly (Fig. 10b). The
dependence of the carrying load on the abscissa x4 is negligible (Fig. 10c). Therefore, the technology should mainly focus on
the accurate manufacturing of the optimal position and height of the first bearing step.
Fig. 11 shows the dependence of the carrying load of the multi-sloped bearing on the selected parameters. The carrying
load significantly decreases for values of the heights h2 and h4 which are lower than the optimal values (Figs. 11a,b). The
abscissa x2 has little effect on the carrying load (Fig. 11c). However, notice that the optimal value of x2 is small (see Table 11)
and, as a consequence, the variation range ± 20% constitutes a narrow interval around the optimal value. The carrying load
decreases rather strongly for values of the abscissa x4 which are either smaller or larger than the optimal value (Fig. 11d).
Therefore, the technology should mainly focus on the accurate manufacturing of the optimal height of the first horizontal
segment and the optimal position and height of the second horizontal segment.
7. Comments on the limitations of the isothermal approach
The results of Section 5 are obtained under the isothermal assumption. A simple ad hoc model, which takes into account
the dependence of the fluid properties on temperature and pressure is built and used in Section S4 of the ESM to obtain
some rough information about the limitations of the isothermal model. For small values of η (i.e. less than 10−6 Ns), the
isothermal and the temperature dependent approaches yield similar sub-optimal bearing shapes, for both multi-stepped
and multi-sloped bearings (Fig. S3a,c in the ESM). Most couple stress fluids used in practice are characterized by values of
η lower than 10−8 Ns.
Fig. 11. The dependence of the carrying load of a multi-sloped bearing (expressed as percentage of the maximum carrying load W of the optimal profile)
on the values of several parameters (expressed as percentage of the optimal values). (a) height h2 ; (b) height h4 ; (c) abscissa x2 ; (c) abscissa x4 . (See Fig. 9).
8. Conclusions
Optimization of slider bearings has been usually performed for Newtonian lubricants. Couple stress fluids are often used
in present-day technology and it is of interest to know if the optimal slider bearing shape is affected by the fluid nature.
A first answer is given here. A model consisting of one dimensional Reynolds and energy equations is developed. Direct
optimal control methods are used. In previous approaches the control is the film height. This restricts the usage of the
optimal control methods only to simple Reynolds equations. A new optimal control approach using the gradient of the film
height as a control is proposed here. It allows a convenient treatment of any form of Reynolds equation. Several constraints
for the state variables and control are taken into consideration. These constraints avoid occurrence of cavitation and ensures
the validity of the Reynolds equation. The model has been validated by comparison with a known analytical solution (the
Rayleigh step bearing).
Two simple design rules are inferred in Section 4.3. Usage of these rules allows obtaining two classes of sub-optimal
shapes: the multi-stepped bearings and the multi-sloped bearings, respectively. Sub-optimal multi-stepped bearings consist
of several steps and the couple stress parameter may affect the constant value of the film height between these steps.
Sub-optimal multi-sloped bearings consist of several inclined regions and the couple stress fluid parameter may affect the
constant value of the film height between these regions. In general, the sub-optimal slider bearing shape obtained under
the assumption of a Newtonian lubricant does not change when most common couple stress fluids are used.
The operation under variable load of the multi-stepped and/or multi-sloped slider bearings is stable. A simple sensitiv-
ity analysis identified the design parameters with the highest impact on bearing load performance (see Section 6.2). The
technology should focus on the accurate manufacturing of these parameters.
The isothermal model may be used successfully for both multi-stepped and multi-sloped bearings at lower values of the
couple stress parameter.
The four main conclusions that can be insightful for a practical engineer are as follows:
• Sub-optimal slider bearing shapes obtained for Newtonian lubricants does not change when common couple stress fluids
are used;
• Multi-stepped and multi-sloped slider bearings have stable operation;
• For the multi-stepped bearings obtained in this paper, accurate manufacturing is needed for the optimal position and
height of the first bearing step;
• For the multi-sloped bearings obtained in this paper, there is a need for accurate manufacturing of the optimal height of
the first horizontal segment and the optimal position and height of the second horizontal segment.
Declaration of Competing Interest
None.
Acknowledgments
The author thanks the reviewers for useful comments and suggestions.
Supplementary materials
Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.apm.2020.01.044.
APPENDIX A
Viscosity is the most important physical property of lubricants. In this paper the dependence of the lubricant dynamic
viscosity μ (units: Ns/m2 ) on temperature T (units: K) is approximated by the Vogel equation [53]:

b
μ(T ) = a exp (A1)
T −c
where a, b, c are constants dependent on lubricant. Table A1 shows the values of these constants for the lubricant HM32
considered in this paper.
Table A1
Constants to be used in Eqs. (A1) and (A2) for lubricant HM32 [53].
Lubricant type HM32

a (Ns/m2 ) 73.63 · 10−6
b (degree centigrade) 797.7
c (degree centigrade) 177.3
ρ 15 (g/cm3 ) 0.879
The dependence of the lubricant density ρ (units: g/cm3 ) on temperature (units: °C) is given by Knežević and Savić [53]:
ρ (T ) = ρ15 [1 − 0.0 0 07(T − 15 )] (A2)

where the value of the density at 15 °C, ρ 15 , is shown in Table A1. This density is comparable with the constant density
value 864 kg/m3 used in [27]. Measurements for the specific heat of the lubricants in Table A1 were not available so a
constant value has been used (c = 1670 J/(kgK)). It is lower than the value 1987 J/(kgK) adopted in [27].
References
[1] S. Cupillard, Thermohydrodynamics of Sliding Contacts with Textured Surfaces Doctoral Thesis, Lulea University of Technology, Sweden, 2009.
[2] D.E. Brewe, in: Slider Bearings. Chap. 27 in Modern Tribology Handbook, CRC Press LLC, 2001, p. 35.
[3] L. San Andrés. Modern Lubrication Theory. Notes 2. Classical Lubrication Theory, Appendix. One dimensional slider bearing, Rayleigh (step) bearing
and circular plate squeeze film damper, Texas A&M University, College Station TX, US, 2014; http://rotorlab.tamu.edu/me626/DEFAULT.HTM, Accessed
10 December 2014.
[4] J.-R. Lin, L.-M. Chu, T.-C. Hung, P.-Y. Wang, Derivation of two-dimensional non-Newtonian Reynolds equation and application to power-law film slider
bearings: Rabinowitsch fluid model, Appl. Math. Model. 40 (2016) 8832–8841.
[5] H. Yu, T. Zhuxin, A new derivation to study the steady performance of hydrostatic thrust bearing: Rabinowitch fluid model, J. Nonnewton. Fluid Mech.
246 (2017) 31–35.
[6] J.-R. Lin, Dynamic characteristics for wide magneto-hydrodynamic slider bearings with a power-law film profile, Appl. Math. Model. 36 (2012)
4521–4528.
[7] N.B. Naduvinamani, G.B. Marali, Dynamic reynolds equation for micropolar fluids and the analysis of plane inclined slider bearings with squeezing
effect, Proc. Inst. Mech. Eng. Part J J. Eng. Tribol. 221 (2007) 823–829.
[8] L. Rayleigh, Notes on the theory of lubrication, Phylos. Mag. 35 (1918) 1–12.
[9] A. Ruggiero, A. Senatore, S. Ciortan, Partial journal bearings with couple stress fluids: an approximate closed-form solution, Mech. Test. Diagn. II 2
(2012) 21–26.
[10] V.K. Stokes, Couple stresses in fluids, Phys. Fluids. 9 (1966) 1709–1715.
[11] S.M. Rohde, A demonstrably optimum one dimensional journal bearing, ASME J. Lubr. Technol. 94 (1972) 188–192 ASME Journal of Lubrication Tech-
nology, 92 (1970) 482-489.
[12] C.J. Maday, A bounded variable approach to the optimum slider bearing, ASME J. Lubr. Technol. 90 (1968) 240–242.
[13] A. Charnes, F. Osterle, E. Saibel, On the solution of reynolds’ equation for slider-bearing lubrication VIII. The optimum slider profile for viscosity, a
function of the pressure, Trans. ASME 77 (1955) 33–35.
[14] C.J. Maday, The maximum principle approach to the optimum one dimensional journal bearing.
[15] A. Guzek, P. Podsiadlo, G.W. Stachowiak, A unified computational approach to the optimization of surface textures: one dimensional hydrodynamic
bearings, Tribol. Online 5 (2010) 150–160.
[16] N.B. Naduvinamani, A. Siddangouda, Effect of surface roughness on the hydrodynamic lubrication of porous step-slider bearings with couple stress
fluids, Tribol. Int 40 (2007) 780–793.
[17] R.A.J. van Ostayen, Film height optimization of dynamically loaded hydrodynamic slider bearings, Tribol. Int. 43 (2010) 1786–1793.
[18] S.-H. Shyu, W.-C. Hsu, A numerical study on the negligibility of cross-film pressure variation in infinitely wide plane slider bearing, Rayleigh step
bearing and micro-grooved parallel slider bearing, Int. J. Mech. Sci. 137 (2018) 315–323.
[19] R.C. Shah, M.V. Bhat, Ferrofluid lubrication in porous inclined slider bearing with velocity slip, Int. J. Mech. Sci. 44 (2002) 2495–2502.
[20] M. Fesanghary, Topology and shape optimization of hydrodynamically–lubricated bearings for enhanced load-carrying capacity, Topology and shape
optimization of hydrodynamically–lubricated bearings for enhanced load-carrying capacity, 1644, 2013.
[21] R. Rahmani, H. Rahnejat, Enhanced performance of optimised partially textured load bearing surfaces, Tribol. Int. 117 (2018) 272–282.
[22] A. Guzek, P. Podsiadlo, G.W. Stachowiak, Optimization of textured surface in 2D parallel bearings governed by the reynolds equation including cavita-
tion and temperature, Tribol. Online 8 (2013) 7–21.
[23] R. Rahmani, A. Shirvani, H. Shirvani, Analytical analysis and optimisation of the Rayleigh step slider bearing, Tribol. Int. 42 (2009) 666–674.
[24] C. Zhang, S. Wen, J. Luo, Characteristics of lubrication at nanoscale in two-phase fluid system, Sci. China (Series B) 45 (2002) 166–172.
[25] Derivative Calculator; http://www.derivative-calculator.net; retrieved: April 23, 2019.
[26] O. Hideki, Thermohydrodynamic lubrication analysis method of step bearings, IHI Eng. Rev. 38 (2005) 6–10.
[27] L. Chang, A baseline theory for the design of oil-lubricated centrally pivoted plane-pad thrust bearings, J. Tribol. 132 (2010) 041703 1-041703.6.
[28] R.J. Bruckner, Simulation and Modeling of the Hydrodynamic, Thermal, and Structural Behavior of Foil Thrust Bearings PhD Thesis, Department of
Mechanical and Aerospace Engineering, Case Western Reserve University, Cleveland, Ohio, US, August 2004.
[29] Integral Calculator; http:www.integral-calculator.com; retrieved: April 23, 2019.
[30] A.A. Ozalp, S.A. Ozel, An interactive software package for the investigation of hydrodynamic-slider bearing-lubrication, Comput. Appl. Eng. Educ. 11
(2003) 103–115.
[31] M.H. Oladeinde, J.A. Akpobi, A comparative study of load capacity and pressure distribution of infinitely wide parabolic and inclined slider bearings,
in: Proceedings of the World Congress on Engineering 2010 Vol II WCE 2010, London, U.K., 2010, pp. 1370–1377. June 30 - July 2, 2010.
[32] A. Guzek, Optimization of Surface Texture Shapes in Hydrodynamic Contacts Phd Thesis, University of Western Australia, 2012.
[33] J. Zhang, F.E. Talke, Optimization of slider air bearing contours using the combined genetic algorithm-subregion approach, Tribol. Int. 38 (2005)
566–573.
[34] J.T. Betts, Practical Methods for Optimal Control Using Nonlinear Programming, Society for Industrial and Applied Mathematics (SIAM), Philadelphia,
2001.
[35] F. Bonnans, D. Giorgi, V. Grelard, S. Maindrault, P. Martinon. BOCOP – The Optimal control solver, user guide, April 8, 2014; http://bocop.org; Retrieved
10 December 2014.
[36] T. Almqvist, R. Larsson, Some remarks on the validity of Reynolds equation in the modeling of lubricant film flows on the surface roughness scale,
ASME J. Tribol. 126 (2004) 703–710.
[37] M. Dobrica, M. Fillon, Reynolds’ model suitability in simulating Rayleigh step bearing thermohydrodynamic problems, Tribol. Trans. 48 (2005) 522–530.
[38] J.A. Tichy, S.-H. Chen, Plane slider bearing load due to fluid inertia – experiment and theory, ASME J. Tribol. 107 (1985) 32–38.
[39] M.N. Mcallister, S.M. Rohde, G.T. Mcallister, Constructive solution of the 1918 problem of lord Rayleigh, Proc. Am. Math. Soc. 76 (1979) 60–66.
[40] S.M. Rhode, Finite element optimization of finite stepped slider bearing profiles, ASLE Trans. 17 (1974) 105–110.
[41] H. Ogata, J. Sugimura, Equivalent clearance model for solving thermohydrodynamic lubrication of slider bearings with steps, J. Tribol. 139 (2016)
034503.
[42] N. Morris, R. Rahmani, H. Rahnejat, P.D. King, B. Fitzsimons, Tribology of piston compression ring conjunction under transient thermal mixed regime
of lubrication, Tribol. Int. 69 (2013) 248–258.
[43] H. Shahmohamadi, R. Rahmani, H. Rahnejat, C.P. Garner, D. Dowson, Big end bearing losses with thermal cavitation flow under cylinder deactivation,
Tribol. Lett. 57 (2015) 1–17.
[44] H. Vosper, K.A. Cliffe, S. Hibberd, On thin film flow in hydrodynamic bearings with a radial step at finite Reynolds number, J. Eng. Math. 83 (2013)
37–55.
[45] A.E. Yousif, A.A. Al-allaq, The hydrodynamic squeeze film lubrication of the ankle joint, Int. J. Mech. Eng. Appl. 1 (2013) 34–42.
[46] J.-R. Lin, Y.-M. Lu, Steady-State performance of wide parabolic-shaped slider bearings with a couple stress fluid, J. Marine Sci. Technol. 12 (2004)
239–246.
[47] N. Morris, M. Leighton, M. De la Cruz, R. Rahmani, H. Rahnejat, S. Howell-Smith, Combined numerical and experimental investigation of the micro-hy-
drodynamics of chevron-based textured patterns influencing conjunctional friction of sliding contacts, Proc. IMechE Part J J. Eng. Tribol. 229 (2015)
316–335.
[48] G.W. Stachowiak, A.W. Batchelor, Engineering Tribology, 2nd ed., Butterworth Heinemann, US, 2001.
[49] N. Phan-Thien, R. Zheng, On the slider-bearing flow of a viscoelastic fluid, J. Nonnewton. Fluid Mech. 33 (1989) 197–218.
[50] N.M. Bujurke, N.B. Naduvinamani, G. Jayaraman, Theoretical modelling of poro-elastic slider bearings libricated by couple stress fluid with special
reference to synovial joints, Appl. Math. Model. 15 (1991) 319–324.
[51] S.M. Rohde, The optimum slider bearing in terms of friction, ASME J. Tribol. 94 (1972) 275–279.
[52] F.T. Schuller. Experiments on the stability of water-lubricated Rayleigh step hydrodynamic journal bearings at zero load. NASA TN D-6514, 1971
[53] D. Knežević, V. Savić, Mathematical modeling of changing of dynamic viscosity, as a function of temperature and pressure, of mineral oils for hydraulic
systems, Facta Univ. Ser. Mech. Eng. 4 (2006) 27–34.

Applied Mathematical Modelling: Viorel Badescu

Uploaded by

Copyright:

Available Formats

Applied Mathematical Modelling: Viorel Badescu

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Applied Mathematical Modelling: Viorel Badescu

Uploaded by

Copyright:

Available Formats

Applied Mathematical Modelling 81 (2020) 887–909

Contents lists available at ScienceDirect

Applied Mathematical Modelling

Two classes of sub-optimal shapes for one dimensional slider

E-mail address: badescu@theta.termo.pub.ro

Fig. 1. The geometrical conﬁguration considered here.

Equation Number Expression

The boundary conditions are:

p(x = 0 ) = pin (15)

p(x = L ) = pout (16)

3. Optimization of slider bearing shape

3.1. Optimization methods

3.2. Optimal control

g(x = L ) → max (21)

Notation Meaning of dimensionless quantity

μˆ ≡ μμre f Dynamic viscosity

ρˆ ≡ ρρre f Mass density

Number Constant parameters

3.5 Aˆ (hˆ , lˆ) ≡ Ah(2h,l )

3.6 Bˆ(hˆ , lˆ) ≡ Bh(h,l

3.7 Cˆ(hˆ , lˆ) ≡ Ch(h,l

3.8 Dˆ (hˆ , lˆ) ≡ Dh(3h,l )

3.3. Numerical procedures and model implementation

3.4. Model validation

3.4.1. Validity of Reynolds equation

3.4.2. Validation of the optimal control procedure

A ﬁrst integration of Eq. (25) gives:

Step Step region x ∈ [0, Lstep ] Land region x ∈ (Lstep ,L]

Slider velocity, U (m/s) 2

Quantity Lower bound Upper bound

4. Sub-optimal classes of slider bearings

4.1. Constraints consistent with input quantities

to h∗in (<< hin ) in a short inlet region of length

4.2. Rules for constrained optimization of slider bearing shapes

• not too large negative values for πˆ min ;

• no bound for πˆ min and πˆ max ;

These rules are used in Sections 5.3 and 5.4.

5.1. Assumptions and input quantities

Quantity Lower bound Upper bound

pˆ 1 pˆ max (to be speciﬁed in text)

Multi-stepped bearings (Rule #1)

Multi-sloped bearings (Rule #2)

5.3. Multi-stepped slider bearings

5.4. Multi-sloped slider bearings

6. Stability and sensitivity analysis

6.1.1. Analytical model of bearings with known proﬁle

A linear ﬁlm height variation is assumed inside the slice j. Therefore:

9.1 aj ≡ 6μUmj , bj ≡ mj , cj ≡ hj − mj xj , dj ≡ 6μU(hj − mj xj )

9.4 Tj ≡ Rj (xj + 1 ) − Rj (xj )

6.1.2. Testing the analytical model

Quantity Boundary conditions Lower limit Multi-stepped Multi-sloped

Upper limit Lower limit Upper limit

The optimal proﬁles are shown in Fig. 7a and 7b, respectively.

6.1.3. Stability of optimal proﬁles

6.2. Sensitivity analysis

Parameter Range of variation

Optimal value Dimensionless Dimensional (mm) Fraction of optimal value (%)

Dimensionless Dimensional (mm) minimum maximum minimum maximum minimum maximum

Multi-stepped slider bearing