Journal Pre-proof

Evolutionary Optimisation of the Thrust Bearing Considering Multiple

Operating Conditions in Turbomachinery

Pavel Novotný , Martin Jonák , Jiřı́ Vacula

PII: S0020-7403(20)34345-9
DOI: https://doi.org/10.1016/j.ijmecsci.2020.106240
Reference: MS 106240

To appear in: International Journal of Mechanical Sciences

Received date: 24 September 2020

Revised date: 14 December 2020
Accepted date: 16 December 2020

Please cite this article as: Pavel Novotný , Martin Jonák , Jiřı́ Vacula , Evolutionary Optimisation of
the Thrust Bearing Considering Multiple Operating Conditions in Turbomachinery, International Journal
of Mechanical Sciences (2020), doi: https://doi.org/10.1016/j.ijmecsci.2020.106240

This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition
of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of
record. This version will undergo additional copyediting, typesetting and review before it is published
in its final form, but we are providing this version to give early visibility of the article. Please note that,
during the production process, errors may be discovered which could affect the content, and all legal
disclaimers that apply to the journal pertain.

© 2020 Published by Elsevier Ltd.

Highlights

 An effective approach to thrust bearing optimisation considering multiple operating conditions.

 Robust solution based on the principles of genetic algorithms.
 Balanced physical depth of computational models describing thrust bearing hydrodynamic and
thermal processes.
 Experimentally verified approach with the possibility of extension to other problems.

1
EVOLUTIONARY OPTIMISATION OF THE THRUST BEARING
CONSIDERING MULTIPLE OPERATING CONDITIONS IN
TURBOMACHINERY
Pavel Novotný1*, Martin Jonák1, Jiří Vacula1
1
Brno University of Technology, Technická 2896/2, 616 69 Brno, Czech Republic
*[email protected]

Abstract
Thrust bearings are an essential part of rotating machines, significantly impacting their overall efficiency.
The bearings must be designed to have optimal properties under the preferred operating conditions and be
able to operate under off-design operating conditions occurring exceptionally. The available literature
presents various possibilities for optimising bearing performance, but a method of including various
operating conditions has not been introduced. The presented solution uses an optimisation of the thrust
bearing that is based on genetic algorithms in combination with an efficient computational model of thrust
bearing lubrication and heat transfer. The presented evolutionary optimisation approach makes it possible to
design a bearing with exceptional performance under both specific and extreme operating conditions. The
optimisation approach is applied in the development of a prototype thrust bearing and bearing friction losses
decreased by approximately 30% compared to a serial bearing. At the same time, there is no significant
decrease in the bearing ability to operate under off-design operating conditions. The prototype bearing
designed using this approach reduced the mechanical losses of the entire turbocharger by approximately
20%.

Keywords
Thrust bearing; multiple operating conditions; hydrodynamic lubrication; genetic algorithms; turbocharger.

1 Introduction
Rotating machines are always designed to achieve maximum possible efficiency under the preferred
operating conditions. The designers of these machines must often take these operating conditions into
account and adapt the design accordingly. However, the machines can also be operated under off-design
operating conditions. The off-design operating conditions, although negligible in terms of machine operating
time, can cause serious damage. The result is that these machines are exposed to many load cases during
operation, these load cases can act independently of one another and have different probabilities of
occurrence.
Turbocharging is the boosting technology using rotating machines and is widely used in many applications,
including both transportation and power generation. The turbocharging is also the subject of extensive
research aimed at minimizing its disadvantages, such as turbo-lag [1, 2], compressor surge [3–5], low-end
torque [6, 7] or noise [8–10]. Even for the most experienced designer, it is difficult, if not impossible, to
assess the effects of various operating conditions on many design parameters. The situation is complicated
by specific conditions. For example, when designing a turbocharger (TC) lubrication system, designers must
maximise the load capacity of bearings while limiting lubricant flow rates in steady states or minimising
mechanical losses in fast transient states. Theoretically, the design of the bearing should satisfy all operating
conditions, according to the prescribed criteria. Unfortunately, a bearing design that meets all requirements is
difficult, if not impossible, to achieve. Reaching an optimal compromise, while meeting the prescribed
criteria, may be a solution.
Focusing on the TCs of internal combustion engines (ICE), mechanical efficiency is a quantity that is almost
always carefully evaluated, as it fundamentally influences the overall efficiency of the TC. The mechanical
efficiency of the TC is particularly crucial for transient operating states, under operating conditions. There
are applications where highly transient operation is needed with rapid ICE response to changing power
requirements, which leads to the transient operating conditions of the TC. TCs of passenger car ICEs or large
stationary ICEs, which compensate for power fluctuations in electrical networks, are examples.

2
If hydrodynamic bearings are used, the lubricating system has a fundamental influence on the mechanical
efficiency of the TC. Hydrodynamic thrust and journal bearings are used in most TC applications and are
therefore the main source of mechanical losses in the lubrication system. The TC must comply with the
ICE’s requirements and is thus operated under both steady and transient operating conditions. A typical
thrust bearing arrangement of exhaust gas turbochargers is presented in Fig. 1.
thrust ring on thrust ring on the segment of a
the thrust counter-thrust side bearing’s
side working surface
bearing
disc
shaft

Fig. 1 A cross section of the double-sided thrust bearing schematically showing the components and a
segment of a bearing’s working surface. The symbols and indicate the minimum thickness of the
lubricating gap on the thrust and counter-thrust side, respectively.

Many studies have analysed the effects of hydrodynamic bearings on mechanical losses and lubricant flow
rates. Hydrodynamic bearings have been studied, for example, by Deligant et al. [11] and they found that
thrust bearings have a greater impact on friction losses than journal bearings. Hoepke et al. [12] analysed the
friction losses of a passenger car TC and determined that the thrust bearing contributed to approximately
38% of the mechanical losses. However, the thrust bearing has a fundamental effect on the lubricant flow
rate of the TC, especially on its counter-thrust side, under conditions that correspond to the high axial load of
the rotor [13]. Novotný et al. [14] have shown that on a heavy-duty truck ICE’s TC, the lubricant flow rate
through a thrust bearing can be several times greater than through radial bearings.
TCs are a common source of ICE failures. Nahim et al. [15] reported up to 5% of diesel ICE failures due to
TC. Moreira [16] also presented a large share of TC on ICE failures of heavy-duty vehicles. The lubrication
system is the main source of technical failures of rotating machines, for example, Iliev [17] presented an
overview of possible types of failures and followed-up measures. Focusing on passenger cars, Dellis et al.
[18] found that more than 60% of serious TC failures recorded from a large group of samples were caused by
the lubrication system. The great influence of the lubrication system on TC failures was also reported by
Pastor et al. [19]. Frequent reasons for thrust bearing failure are the operation of the TC outside of design
conditions, such as operation of the compressor during surge. During surge, charge air recirculates to the
compressor inlet and axial force increases considerably and changes dynamically; this condition can cause
damage to the thrust bearing.
Thrust bearings, which mainly affect the axial movement of the rotor, can also have a great influence on the
lateral vibrations of the rotor. Some researchers [20, 21] have found that thrust bearings have a significant
effect on rotor dynamics. They found that thrust bearings can affect critical speeds, including resonance
fluctuations, and have a significant effect on rotor stability. Koutsovasilis [22] showed the effect of the
placement of individual pads on rotor vibration. Subsequently, he optimized the placement of pads and
reduced rotor vibration to a minimum.

3
It is therefore clear that designing thrust bearings with low friction losses and lubricant flow rates, while
meeting the requirements for load capacity, vibration, and operational reliability, is a considerable challenge.
Overcoming the defined challenges requires the formulation of a work objective. The work objective is to
develop an effective optimisation approach to finding the optimal design parameters of the hydrodynamic
thrust bearing. This effective approach will allow us to achieve excellent performance, with an emphasis on
reducing mechanical losses, limiting lubricant flow rates and preventing bearing damage under off-design
operating conditions.
This approach must include a combination of a suitable optimisation algorithm and a computational model to
describe the lubricant flow and heat transfer in the thrust bearing. When developing the approach, the
emphasis should be on the efficiency of the optimisation algorithm in finding the global minimum of the
objective function, depending on the design parameters, under many operating conditions. The speed of
computation, the ability to describe the flow of lubricant and the heat transfer in the thrust bearing should
also be monitored.

2 An Overview of Methods to Improve Bearing Performance

A commonly implemented method of bearing design is a computational approach that compares the impacts
of various design parameters with a subsequent qualified selection of the best design variants. Many different
design methods have emerged, mostly involving some of the computational methods. Analytical methods of
design have proved successful for designing the basic dimensions of bearings but no longer provide
sufficient physical depth to improve existing bearings. At present, numerically solved computational models
predominate. Examples are procedures presented in publications [23–25] that compare different thrust
bearing design variants using highly advanced three-dimensional (3-D) computational models based on
computational fluid dynamics (CFD). This approach allows the use of relatively detailed computational
models; however, in the case of many bearing input parameters, it has become difficult to find an optimal
bearing design in a reasonable time.
A more advanced method of improving bearing performance is the involvement of optimisation strategies.
Many optimization strategies have been also developed for design optimization of rotor-bearing systems.
Gradient based methods are the basis of most of these strategies. For example, Cheng and Chang [26] used
the conjugate gradient method in conjunction with a direct problem solver, which is based on the rarefied-
flow theory. An optimisation method using a gradient calculation was also presented by Van Ostayen [27]
who applied the Rayleigh step bearings method or similarly by Rajan et al. [28, 29]. Fesanghary et al. [30]
used a sequential quadratic method for the optimisation of the sectorial-shape thrust bearing, which is also
based on the calculation of objective function derivations. The integrated philosophy for designing and
optimising gas-bearing-supported rotors is presented by Saruhan et al. [31, 32]. These approaches are
suitable for the general design of the bearing lubricating gap and allow for the consideration of many design
parameters. However, due to their generality, these methods can lose efficiency in the targeted design of
thrust bearings for real applications.
Optimisation approaches using computational models that better describe the thrust bearings are presented,
for example, by Charitopoulos et al. [33]. The presented lubrication model uses the CFD-based approach for
the analysis of taper-land and pocket-thrust bearing designs. Parameter influences on both designs are
studied, and optimisations are performed by the in-house optimiser using genetic algorithms (GAs).
Regarding the taper-land bearing, two design parameters were chosen: a maximal height of the taper and a
ratio between the tapered and flat part of the working surface of the thrust bearing. A reduction of power loss
by more than 7% for a rotor speed of 200 thousand revolutions per minute (rpm) compared to the initial
design was the result. Significantly better results were obtained with the second design, the pocket-thrust
bearing, in which six design parameters were used. Power loss decreased by approximately 12%, and oil film
thickness increased by 40% in a comparison to the initial bearing design for design operating conditions.
Similarly, the CFD-based model with an in-house optimiser was used in a paper by Fouflias et al. [34] for
curved pocket thrust bearings, with the aim of maximising bearing load capacity and minimising friction
losses. The results show significant improvements when compared to other relevant literature, with a load
capacity improvement of up to 16% and a decrease in friction coefficient of 21%. A use of an evolutionary
optimization of micro-thrust bearings by GAs also successfully presented Papadopoulos et al. [35].

4
Turbocharging applications often require extending the description of bearing lubrication to include the
issues of heat transfer and changes in the properties of the related lubricant. Dadouche et al. [36] found that
inlet temperature of lube oil influences largely thrust bearing performance. Glavatskih et al. [37] investigated
thermal effects and heat transfer into the thrust bearing pads. They calculated displacements of pad surfaces
due to pressure and temperature fields and concluded, that effects of temperature must be included in the
computational model. An important issue is the ratio of heat transfer by conduction and convection
mechanisms. The results presented by Zhai et al. [38] showed only a very small part of heat is dissipated
through the pad and collar. This assumption is confirmed, for example, by Neal [39], who showed that the
heat transfer from the film to pads is not that significant, it represents 12% of the total power loss. Recent
successful application of optimisation approaches using CFD models, including the heat transfer phenomena,
is presented, for example by Chalkiopoulos et al. [40]. Temperature dependent lubricant properties can also
affect the bearing performance. Pavlioglou et al. [41] presented the optimisation of thrust bearings with their
spatially varying viscosity model. The maximisation of load capacity and the minimisation of friction
coefficient is studied in this Pareto optimisation problem using GAs. Improved bearing performance is
achieved in all the obtained solutions. Experiments and simulations [36–41] have proved that thermal effects
cannot be neglected for successful simulations of thrust bearings.
Elastic deformations of thrust bearing working surfaces can cause significant effects on lubrication. Thrust
bearing lubrication involving fluid-structure interactions between pads and collar through lubricant was
presented by Liang et al. [42]. They found a decrease of the minimum film thickness by approximately 11%
due to deformations compared with calculations without considering deformations. Similar conclusions can
be found in [43].
These approaches, which use optimisation methods in combination with physically detailed and
computationally demanding 3-D models, can describe the given physical problems (e.g. 3-D fluid flow,
thermal effects, deformations) very well. The disadvantage is the use of numerically demanding
computational models, making precise examination of many design parameters under many operating
conditions impossible.
The targeted modification of the working surface, using the texture of suitable profiles, is a widely published
issue of bearings. Previous studies, for example [44–48], have been limited to simple symmetrical profiles
and optimisations included only several variables. Recently, Lee et al. [49] presented the use of GAs for the
design of shape-targeted structures of working surfaces. Similarly, Schuh et al. [50] introduced design-driven
modelling of textured sliding surfaces. GAs can also be used for bearing design in combination with rotor
dynamics, as presented by Schiffmann and Favrat [51], or for crank mechanism bearing designs, as presented
by Zhang et al. [52]. A review clearly shows that GA principles, after some modifications, can be effectively
used for the solution of the optimisation problems involving thrust bearings in connection with other related
influences.
GAs are very popular optimisation algorithms that are used in a wide range of applications and are preferred
for their robustness. Another advantage is that GAs are often directly implemented in optimisation tools and
packages. An inherent disadvantage is the fact that with the consideration of a greater number of input
variables for optimisation, there is a sharp increase in the number of functional values that need to be
computed. It should be noted that GAs belong to a group of population-based metaheuristic methods that
work with more than one potential solution.
The main issue of the bearing optimisation approaches presented in the literature is the inability to assess
many types of design and off-design operating conditions in reasonable time. The approaches presented so
far do not allow for the inclusion of off-design operating conditions, which occur with only a small
probability and do not have a significant effect on important integral bearing parameters. However, these
operating conditions have a fundamental effect on the reliable operation of the thrust bearing.
This paper defines an effective optimisation approach for finding optimal thrust bearing parameters for many
operating conditions. This approach requires a proposal of the objective function, modifications to the GAs
and a computational model for the thrust bearing’s lubrication. The computational model of the bearing’s
lubrication must describe the physical processes taking place in the bearing in sufficient detail, but its
solution must take a maximum of a few seconds on commonly available computers. The duration of a single
computation is crucial because when using GAs, many individual computations of bearing characteristics
can be assumed.

5
3 Modelling of Thrust Bearing Lubricant Flow and Heat Transfer
A computational model that describes the behaviour of the thrust bearings of the TC rotor under different
operating conditions is the basic challenge of an optimisation approach. The thrust bearings of the TC rotors
are specific in certain aspects. They are double-sided and thus comprise of thrust and counter-thrust sides. In
addition, the design of the bearing includes lubrication channels for oil distribution to individual working
surfaces. Lubrication in thrust bearings is highly mechanically and thermally stressed. The relative speeds of
bearing working surfaces are extremely high and, in combination with the greater thickness of the lubricating
layer often occurring on the counter-thrust side, a significant effect of the turbulence and inertial forces in the
lubricating layer can be observed.
Modelling of thrust bearings requires to solve the Reynolds equation. The Finite Element Method [53, 54]
and Finite Difference Method [13, 14, 55, 56] together with the Newton–Raphson method are widely
adopted methods for the Reynolds equation solution. The current trend in modelling the lubricant flow and
heat transfer of hydrodynamic bearings is the use of the Finite Volume Method through CFD commercial
software. This approach makes it possible to model lubricant flow and heat transfer in a 3-D domain. The
commercially available models solve the essential physical processes described by the basic conservation
equations. Computational models are mostly formulated in 3-D and often include heat transfer; references
[23–25, 33, 34] can be used as examples. Generally, the CFD approaches represented by some commercial
software can be used in a very general way; however, they have the major disadvantage of computational
complexity. The computational complexity depends mainly on the requirements for the discretisation of the
solved domain, the lubricating gap, the lubricant supply channels and the number of physical problems
involved (mass and energy transfer, turbulence, etc.). Due to the nature of the bearing geometry, it is
necessary to perform a relatively detailed discretisation of the lubricating gap and the lubricant inlets and
outlets. The resulting CFD model can contain millions of cells per single thrust bearing segment.
The time required to numerically solve the steady state of the thrust bearing using the CFD approach can
then include units, as well as multiple hours of computational time on modern high-performance computers.
This does not allow to use this approach for multiple repetitive tasks involving the solution of hundreds or
thousands of operating conditions of the bearing. This approach, although certainly more physically accurate,
seems inapplicable for the work objective.
The suitable computational model must describe the problem of lubrication and heat transfer in the bearing
in sufficient detail, while being rapidly solvable on available computers; i.e. it must contain a balanced
combination of the physical depth of a problem description and its computational solving speed. Such a
model must make it possible to perform thousands of individual computations on computers in reasonable
time.
lubricating layer lubricant outlet

Lubricant inlet

Fig. 2 A schematic of the working surface of a typical thrust bearing segment. The schematic includes the
boundary conditions that are inlet pressure , outlet pressure , inlet temperature ( ) and outlet
temperature ( ). The lubricating layer is described with the distribution of lubricating gap thickness
, hydrodynamic pressure , lubricant viscosity and density and satisfying
the boundary conditions.

The model presented by Novotný and Hrabovský [55] is a suitable compromise that meets the defined
requirements and simultaneously provides a high-speed solution. This experimentally validated two-
dimensional model extends the generally known theory of thin lubricating layers by correcting the influence

6
of the lubricant temperature, inertia forces and turbulence. The lubricant is assumed to be a mixture of oil
and gas bubbles with constantly distributed properties throughout the lubricating gap volume. The
lubricating gap volume is described by coordinates , where is the radial coordinate, the
circumferential coordinate and the lubricating gap thickness. The model enables the computation of
hydrodynamic pressure , lubricant dynamic viscosity and lubricant density in
the lubricating layer, based on bearing geometry, operating conditions and lubricant properties. The
pressures in the inlet and outlet and the temperature in the inlet ( ) are the operating
conditions that must be included. A schematic of the working surface of a bearing segment is presented in
Fig. 2.
Working surface performance is quantified using integral characteristics, including the working surface load
capacity, the friction torque in the lubricating layer and the flow rate of the lubricant through the bearing.
The integral characteristics are calculated based on the hydrodynamic solution results. The load capacity of
the bearing working surface of all segments is defined by the relation
0 r1

Fb  np   p r d dr , (1)
0 r0

where represents a number of segments, the inner radius of the thrust bearing working surface, the
outer radius of the thrust bearing working surface and the angular extend of the segment. The outlet mass
flow rate of the thrust bearing ̇ can be determined with the integration of the unit mass flow through the
outer edge of the bearing working surface as
0
 1 mix h3 p 1 mix h3 
mr  np    mix r 2  rd . (2)
0 
12 mix r 40 mix 
Symbol represents the angular velocity of the rotor. Friction torque can be determined with equation
r 
1 0
 h p mix r  2
M f  np      r drd . (3)
r0 0  2 r  h 
The model introduced in reference [55] assumes heat transfer from the surrounding walls, but in the case of
the optimisation, this phenomenon can be neglected because it does not produce a significant effect. Then the
lubricant outlet temperature leaving the lubricating gap is set so that the total heat dissipated by the
lubricant flow balances the heat generated by the shear stresses, according to equation
Mf
Tout  Tin  . (4)
mr cp
Symbol represents the specific heat capacity of the lubricant.

4 Development of the Optimisation Process of Thrust Bearings

4.1 Bearing Geometry Parametrisation
A generalised geometry of the double-sided thrust bearing of the TC rotor is selected for optimisation
purposes. Both sides of the thrust bearing consist of several segments. Each segment contains a working
surface that is defined by design parameters. A parametrisation of the working surface of the thrust bearing
segment is presented in Fig. 3. The generalised geometry can also describe series-used bearing geometry.
In general, bearing lubricating gap geometry is defined by relative eccentricity , bearing clearance a ,
and the parameters of the working surface. The relative eccentricity of the thrust bearing expresses a relative
axial position of the bearing disk within the gap defined by the thrust rings. It can take values of 0, when the
bearing disc touches the thrust ring surface on the counter-thrust side, up to 1, when the bearing disc touches
the thrust ring surface on the thrust side. Relative eccentricity is not a bearing design parameter, but a

7
variable that depends on the operating conditions of the TC and must therefore be calculated from force
equilibrium.

groove land

taper

Fig. 3 A parametrisation of the working surface of a thrust bearing segment. Symbol represents the
number of segments, is the inner radius of the working surface, the outer radius of the working surface,
the inner radius of the taper part, the outer radius of the taper part, the taper wedge angle,
the limit wedge angle of the tapered part, the groove height, the taper part angle and the
angle of the lubricating groove.

Lubricating gap thickness relations on the thrust side ( and counter-thrust side can be obtained
with the following equations
hts  1    cax  hws,ts , (5)

hcts   cax  hws,cts , (6)

where and represent the lubricating gap thickness of the working surface on the
thrust side and the counter-thrust side, respectively. Symbol a ts ts indicates the axial clearance
of the thrust bearing – one of the main bearing design parameters, ts is the minimal thickness of the
lubricating gap on the thrust side and is the minimal thickness of the lubricating gap on the counter-
thrust side.
Tab. 1 An overview of the optimisation variables,
expressed in the dimensionless form necessary
for the definition of the lubricating gap of one
side of the thrust bearing only, with respect to
the basic design parameters.
Item Definition
Taper angle ratio [-]
Wedge taper angle ratio [-]
Outer radius ratio [-]
Taper inner radius ratio [-]
Taper outer radius ratio [-]
Groove angle ratio [-]
Axial clearance ratio [-] a

8
The thrust bearing design parameters can be divided into basic design parameters and free design parameters.
The basic design parameters cannot be changed during the optimisation and include the number of segments
, the inner diameter of the working surface , the height of the lubrication groove and the
maximal axial clearance . The free design parameters are optimisation variables. The optimisation
variables are for the purpose of parameterisation, normalised to the interval 〈 〉 and related to the basic
design parameters. Dimensionless expressions of the optimisation variables of the thrust bearing are
presented in Tab. 1. For a complete definition of the thrust bearing, it is necessary to define these variables
twice, with the exception of the axial clearance ratio for the thrust and counter-thrust side, respectively.

4.2 Turbocharger Operating Conditions used for the Optimisation

Choosing the operating conditions is a key part of the optimisation. By default, the components are
optimised with respect to the operating conditions that are most significant in terms of probability of use.
However, there may be situations where a malfunction occurs due to unlikely operating conditions or, in
other words, off-design operating conditions. The choice of multiple operating conditions, including those
that may occur exceptionally, is a measure against these situations. However, bearing performance must not
be significantly impacted by these off-design conditions. To avoid the negative effects of these conditions,
the significance of the individual operating conditions is expressed by the weighting factor .
In the case of high-speed rotating machines, it is always necessary to consider the issues of rotor dynamics
and stability as well. The load on the thrust bearing in the direction of the axis of rotation is usually caused
by the pressure conditions on the impellers, rapid changes in rotor speeds (transient states), aerodynamic
phenomena in the compressor or turbine, gas flow pulsation at the turbine inlet and rotor vibrations or
turbocharger vibrations.
The target turbocharger comprises of a rotor mounted on hydrodynamic journal bearings with a floating non-
rotating ring. The turbocharger is used in ICEs for combined heat and power production units and is
therefore primarily operated in several steady-state operating conditions. Rapid changes in rotor speed can
partially affect the response of the thrust bearing, but due to a use of the ICE and the corresponding operating
conditions, any time load variations can be neglected.
Pulsations in the gas flow at a turbine inlet also cause an increase in axial load, but in terms of dynamic
effects, these pulsations are at much lower frequencies than the rotor speed and the natural vibrational
frequencies of the rotor in the bearings. The axial load pulsations caused by rotor vibration are usually small
to negligible, but it is worth considering the results presented in references [20, 21]. Also, the thrust bearing
has only a small to negligible effect on the stability of the rotor because the stability of the rotor is mainly
influenced by the journal bearings. The result of these facts is that for a given area of application, the
operating condition can be considered a steady state, and it is not necessary to consider the additional
dynamic effects of the rotor caused by the thrust bearing. Thus, for simplicity, only a few operating
conditions are selected, and these are characterised by the rotor speeds, axial forces, load of the TC rotor,
parameters of the lubrication system and weighting factors. The specifications of the operating conditions are
presented in Tab. 2.
Tab. 2 Characteristics of the turbocharger operating conditions chosen for thrust bearing
optimisation.
Item Value
Operating condition no., [-] 1 2 3 4 5
Rotor speed, [rpm] 10000 20000 30000 30000 10000
Axial force, [N] 301 670 1068 2200 301
Absolute oil pressure at inlet, [MPa] 0.25 0.25 0.25 0.25 0.15
Absolute oil pressure at outlet, [MPa] 0.1 0.1 0.1 0.1 0.1
Oil temperature at inlet, [°C] 60 60 60 90 100
Weighting factor, [-] 0.4 0.3 0.2 0.05 0.05

The work objective is to minimise mechanical losses and thus increase the overall efficiency of the TC. It
can be concluded that the effect of mechanical losses has a greater impact on overall efficiency at lower TC

9
rotor speeds. Greater value of the weighting factor is therefore placed on operating conditions corresponding
to lower and medium rotor speeds. The characteristics of these operating conditions are given in Tab. 2
under numbers 1, 2 and 3. The characteristics of the operating condition were determined by a technical
experiment. Axial forces were measured with a specialised strain gauge system installed on a bearing disc of
the thrust bearing.
Surge is a significant aerodynamic phenomenon that occurs in centrifugal compressors during operation
under off-design conditions. Surge occurs when the compressor is operated at low mass flow rates and gas
backflow in the compressor dominates. This off-design operating condition is accompanied by increased
levels of vibration and noise and often leads to the failure of the thrust bearing or even the shaft. The surge of
a centrifugal compressor, including the impact on the axial load of the rotor, was investigated by Galindo et
al. [4]. They performed experiments on the TC of a heavy-duty vehicle and the shaft of the tested compressor
was destroyed. Brun et al. [57], who also measured the axial forces in a centrifugal compressor, noted
damage to the thrust bearing. Thus, the experience presented in references [4, 5, 57–59], as well as the
experience of the authors of this paper, shows that the surge operating condition should be taken into
account. In the case of this work, the surge operating condition is included in the optimisation and is given in
Tab. 2 under no. 4. Only the maximum value of the time-varying axial force, determined by the specialised
strain gauge system, is used for optimisation. Due to the low occurrence probability of this off-design
condition, a very low weighting factor is chosen.
An operating condition that combines a high oil inlet temperature, low oil inlet pressure and low rotor speeds
may sometimes occur. As a result of this combination, the bearing’s load capacity is relatively low, and the
risk of bearing damage is high. This off-design operating condition can significantly affect the operating
safety of the bearing, even assuming a low probability of occurrence. The characteristics of this operating
condition are presented in Tab. 2 under no. 5.

4.3 Computation of Thrust Bearing Operating States

During optimisation, it is required to determine the operating state of the thrust bearing as a function of the
TC operating condition characteristics, bearing design parameters and variables and lubricant properties. It is
thus necessary to determine the axial position of the bearing disk in the gap between the rotor thrust rings.
Finding the position of the disc therefore requires an equilibrium solution for the external rotor forces and the
internal forces in the lubricating layers on both sides of the thrust bearing. The evaluated operating states of
the bearing are assumed to be steady; therefore, the dynamic effects are neglected. All rotor forces are
summed in the nonlinear force residuum in the steady state of operating condition no. The force residuum
as a function of bearing relative eccentricity is defined as
f NL  Fax  Fbts ( )  Fbcts ( ) . (7)

represents total axial force on the turbocharger rotor with consideration of gas pressures on the
compressor and turbine wheels, represents the bearing load capacity of the thrust side and
represents the bearing load capacity of the counter-thrust side.
Finding the force equilibrium state of the bearing thus requires finding the bearing’s relative eccentricity so
that the nonlinear force residuum, defined by equation (7), is zero. From a mathematical point of view, it is a
search for a zero value with a non-linear equation with one variable. For such tasks, the Newton method can
be used in the following form
1
 f ( ) 
 s 1   s   NL s  f NL ( s ) , (8)
  
where is the number of equilibrium iterations in the given operating condition. The iterative solution
following eq. (8) proceeds until the relative residuum reaches the limit value defined by the following
relation
f NL ( s 1 )
 s 1    lim , (9)
Fax
where indicates the limit value of the relative residuum. The value is typically chosen.

10
The overall characteristics of the thrust bearing for the considered operating condition combine the
characteristics of both bearing sides with the following equations. The overall friction torque of the thrust
bearing is calculated with
M f,2s  M f,ts  M f,cts (10)

and the overall mass flow rate of the lubricant through the bearing is calculated with
mr,2s  mr,ts  mr,cts . (11)

The minimum thickness of the lubricating gap is defined as follows

hmin  min  h0,ts , h0,cts  , (12)

and the maximum outlet temperature of the worse side of the bearing is defined by

Tmax  max Tout,ts , Tout,cts  . (13)

The index containing the abbreviations ts and cts in the equations (10-13) defines the thrust and counter-
thrust bearing side, respectively.
The computation of the thrust bearing state is realised iteratively. On the basis of the operating condition
characteristics, lubricant properties and bearing design parameters, the solution proceeds according to the
flow chart presented in Fig. 4.
Inputs
Lubricating gap geometry:
Operating conditions:

Lubricant properties.
Initial conditions:
Force equilibrium cycle

Solve thrust bearing lubricant flow

and heat transfer and compute the
integral characteristics of both bearing
sides: and

Compute nonlinear Update

force using relative
eq. (7). eccentricity
using eq. (8).
and generate
lubricating
Check gaps
force
residual
no
using
eq. (9)
yes
Results
Compute overall thrust bearing
characteristics by combining both
bearing sides with equations (10–13):

Fig. 4 Flow chart representing an iterative computation of the overall integral characteristics of the thrust
bearing for operating condition of no. .

11
4.4 Adaptation of Genetic Algorithms
Many examples of optimization from various fields, such as structural mechanics, fluid flow, acoustics or
mechanical systems can be solved using GA [60]. The GA is a heuristic method that tries to find solutions to
problems by applying the principles of evolutionary biology, for which there is no usable exact algorithm.
GA problematics use standard terminology: an individual, a generation, a gene, an objective function or a
fitness function. Thrust bearing integral characteristics, as a result of the numerical solution of the lubricant
flow and heat transfer, are used to search for the optimal parameters of a thrust bearing. Each solution
(marked as an individual) is represented by a set of variables (marked as genes) and is evaluated using
integral characteristics (marked as the objective function). To create a new generation, the fitness function is
calculated for each individual, based on the objective function. The algorithm thus searches for the global
minimum of the objective function, depending on the genes. In this case, the fitness function is expressed as
a negative value of the objective function and expresses the quality of the thrust bearing design represented
by the individual.
Individuals are stochastically selected and modified by the genetic operators, including a selection, a
crossover and a mutation. After the application of these genetic operators, a new generation is created. The
proposed strategy includes a generation model of GA constrained by the limit values of genes and elitism
[61]. In this model, a new generation of individuals is created with each iteration, while a predetermined
proportion of the best individuals (referred to as elite individuals) from the previous generation is
maintained.
The optimisation process uses a comparison with the integral characteristics of the existing serial design of
the thrust bearing, denoted as ̇ The series-used geometry is hereinafter referred to as the
initial design, and, in the case of analytical notations, is referred to as the ‘init’ index.
The following objectives are set for the thrust bearing optimisation:
 Friction torque must be minimised.
 The outlet mass flow rate must not be significantly increased compared to the bearing’s initial
design and the reduction of the mass flow rate is not limited. The selected limit value of the
increase of mass flow rate is 10%.
 The minimal lubricating gap thickness must not be reduced below the thickness limit value, but its
increase is not restricted.
 The maximal outlet lubricant temperature must not be increased over the temperature limit value,
but its reduction is not restricted.
Reducing the mass flow rate or maximal outlet lubricant temperature and increasing the minimal lubricating
gap thickness are not restricted but are not the primary objectives of the optimisation. The optimisation is
aimed at minimising friction torque and at the same time penalising only the states leading to a reduction of
minimal lubricating film thickness or to an increase in the outlet temperature or an excessive increase in the
mass flow rate. Objective function , a value that will be minimised during the optimisation, is proposed
in the following form
kmax

f s s s wk
f,k m,k h,k t,k
f obj  x   k 1
kmax
, (14)
w
k 1
k

where [ ] represents the parent vector of size , reflects the number of the operating
condition and is the total number of operating conditions considered for the optimisation. The vectors

x ts  t,ts ,w,ts ,r,ts ,tmn,ts ,tmx,ts ,g,ts 

, (15)
xcts  t,cts ,w,cts ,r,cts ,tmn,cts ,tmx,cts ,g,cts 

correspond to the variables of the thrust side and counter-thrust side of the bearing, respectively.

12
Similarly, the initial bearing design is described by the initial parent vector . The friction torque ratio
is proposed relative to the friction torque of the initial bearing design under the considered
operating condition in the form
Mf
ff  , (16)
M f,init
and the mass flow rate factor that increases the value of the objective function in the case of an above-limit
lubricant flow rate increase is proposed as

  m  
sm  max 1,  r
 , (17)

   m mr,init 
  
where represents the factor that increases the tolerable mass flow rate and is a power exponent
greater than 1. The thickness factor, which increases the value of the objective function if it falls below the
limit value ( ), is defined as follows
  h  
sh  max 1,  lim   . (18)
  hmin  
 
The temperature factor, which increases the value of the objective function in case of temperature rise above
the temperature limit value ( ), is calculated as follows
  T  
st  max 1,  max   (19)
  Tlim  
 
The value of the power exponent determines how an unacceptable value of lubricating gap thickness or
outlet temperature increases the objective function value and thus reduces the fitness of the individual. For
example, is a reasonable value.
The optimisation problem is written in the form

min x  f obj (x)  (20)

and the values of the individual genes must be within the prescribed limit values
xmin,j  x j  xmax,j , (21)

where represents the order of the gene in the parent vector. The limit values are based primarily on the
compatibility conditions. The compatibility conditions allow only such combinations of variable values for
which the bearing can be unambiguously generated without collisions. The limit values are also influenced
by technological possibilities in the manufacturing of the bearing and by the TC design restrictions. Tab. 3
presents an overview of the limit values used for the optimisation.
Tab. 3 An overview of the limit values for
dimensionless variables defining the thrust
bearing.
Min. Max.
Item
value value
Taper angle ratio, [-] 0.05 0.95
Wedge taper angle ratio, [-] 0.05 1
Outer radius ratio, [-] 1.1 2
Taper inner radius ratio, [-] 0.05 0.45
Taper outer radius ratio, [-] 0.55 0.95
Groove angle ratio, [-] 0.05 1

13
Axial clearance ratio, [-] 0.5 1

The fitness function is used to in GEs to guide simulations towards optimal design solution and is typically
defined in a way that a higher value indicates better fitness; therefore, it is formally written in the form
ffit   f tot (x) . (22)

The distribution of the initial population of size is chosen uniformly within the limit values of the
individual genes. The fitness function is calculated for all individuals, and all individuals in the population
are sorted according to the value of their fitness function. From this population, a defined portion of best
individuals, with the highest fitness function values, called the elite individuals, is selected without any
further operations. These individuals are used to form a portion of the new generation. There is no possible
loss of a given percentage of the best solutions using this elitism [61].
To select the remaining portion of the new generation with share a selection operator is applied to
the previous population. Tournament selection [62] is applied as the selection operator, which does not create
a large pressure that might lead to a loss of diversity in the population and possible entrapment at the local
extremes. The selection of each parent begins with a random selection of the specified number of individuals
followed by selection with the tournament method.
Using the crossover operator, a portion of the new population is selected, with the share . The task of the
crossover operator is to create new individuals by combining the genes of the individuals selected with the
tournament method. It is assumed that the crossover of two individuals with high fitness function values
generates new and different individuals that contain quality genes inherited from their parents. The crossover
creates a child that lies on a line containing both parents and is at a small distance from the parent with the
better fitness value, in a direction away from the parent with the worse fitness value. The resulting child’s
gene can be written in the following form [63]

xchild,j  xp2,j   xp1,j  xp2,j  , (23)

where represents the gene of the parent with better fitness, represents the gene of the parent with
worse fitness and indicates the distance of the child from the parent with worse fitness.
The mutation operator is applied to create the remaining portion of individuals, with a share of in the
new population. The mutation is used to randomly modify the genes. However, because too frequent
mutations can lead to a loss of quality genes, this operator is used with only a small probability. The
mutation is applied with respect to the limit values adaptively and to the last successful or unsuccessful
generation. The mutation thus randomly selects a gene, and the magnitude of the gene value changes, but
only to the extent that satisfies the limits. The child’s gene is defined as follows
xchild,j  xp,j   N  0,   . (24)

Symbol represents the gene of the parent selected with the tournament method. Therefore, the mutation
adds a random number using the adaptive ratio taken from a Gaussian distribution with a mean value of
0 and a standard deviation taken from the previous generation for each entry of the parent vector. The
mutation finds a direction and ratio that satisfies the defined gene limits [63]. The process requires iterative
generation of trial points inside the gene limits and then applies a check of the feasibility of the resulting
steps to ensure that all limits are still valid.
The new generation is selected from the individuals with the elitism, crossover and mutation that satisfy the
following shares
e  c  m  1 . (25)

The optimisation process repeatedly creates new generations until the conditions for stopping are met. Two
independent criteria are used to stop the optimisation process. The first criterion is the completion of a
predetermined number of generations The second criterion stops the process if the relative change in

14
objective function values during steady generations is less than or equal to the tolerance for relative change
.
The optimisation involving the application of the genetic operators to the initial generation, and leading to
the creation of a new generation, is graphically shown in Fig. 5.
Evaluation and ranking based New generation
on fitness function values Elite
V1 Elite V1
selection
V2 V2
Tournament selection
V4
V3 V4 Crossover
V5 V4
V4 V4,3
V8
V5 V7 V7,8
V7

….
V6
….

V
V7
V6
V3
Mutation
V8 V3 V3
V4*
V7
….

….
V7
V8 V *

Fig. 5 Scheme of creating a new generation using genetic operators.

Typically, appropriate GA settings are a matter of experience, together with knowledge of the nature of the
problem to be optimised. The GA settings presented in Tab. 4 are determined by long-term testing of this
and other similar problems and lead to a relatively fast finding of results.
Tab. 4 An overview of the GA settings used for the
optimisation.
Step Item Value
Initial Population size, [ ] 200
generation
Selection Count ratio, [ ] 0.05
Tournament size, [ ] 4
Crossover Count ratio, [ ] 0.80
Heuristic ratio, [ ] 1.2
Mutation Count ratio, [ ] 0.15
Stop No. of generations, [ ] 50
criteria
Function rel. change, [ ]

5 Results and Discussion

5.1 Thrust Bearing Optimisation
This effective optimisation approach based on the GA is developed to be able to find the optimal thrust
bearing design, considering many design variables under many operating conditions. The approach optimises
both sides of the thrust bearing simultaneously, considering optimisation variables. However, in the case
15
of optimisation involving a large number of variables, the use of large initial populations is necessary to
search the entire multidimensional space in sufficient detail [64]. Sometimes, it is more efficient to optimise
both sides independently. The number of variables thus directly affects the required computing power of the
hardware.
5 5
mean
best
4 4

Objective function [-]

Objective function [-]

3 3

2 2

1 1

0 0
0 25 50 0 100 200
Generation [-] Individual [-]
Fig. 6 A development of the objective function of the individuals with the best fitness (best), the objective
function averaged from the whole population (mean) and the objective function distribution of all the
individuals in the last generation during the optimisation.

The search for the optimal bearing design progresses through the generations. The objective function of the
individuals of the corresponding generation is monitored and, subsequently, the criteria for stopping
optimisation are evaluated. By monitoring individual fitness over the generations, it can be stated that the
GA finds promising individuals relatively quickly. In the following generations, the GA only slightly
improves the fitness of the population. The development of the objective function of individuals with the best
fitness and the mean fitness of the whole population through the generations is shown in Fig. 6 on the left.
The fitness of an entire generation in the last generation is very similar; only a few individuals are
significantly worse. The distribution of objective function values for the entire population in the last step is
presented in Fig. 6 on the right. An overview of the found optimal values of the dimensionless variables
defining the thrust bearing is given in Tab. 5.
Tab. 5 An overview of the optimal values of the
dimensionless variables defining the thrust
bearing.
Thrust Counter-
Item
side thrust side
Taper angle ratio, [-] 0.90 0.79
Wedge taper angle ratio, [-] 0.58 1.95
Outer radius ratio, [-] 1.75 1.52
Taper inner radius ratio, [-] 0.15 0.24
Taper outer radius ratio, [-] 0.86 0.89
Groove angle ratio, [-] 0.98 0.97
Groove radius ratio, [-] 0.72 0.45
Groove width ratio, [-] 0.11 0.05
Axial clearance ratio, [-] 0.65

The numerical algorithm of the proposed effective optimisation approach was solved in parallel on eight
cores of an Intel XEON 3.6 GHz personal computer and the resulting optimisation time was approximately
24 hours. Such long optimisation times are acceptable, but it is necessary to consider that the calculation of
the objective function takes in the order of units of seconds. The computational complexity of the bearing
lubrication is very often a limiting factor and greatly limits the use of highly detailed 3D CFD-based
computational models of bearing lubrication. Sometimes the bearing lubrication under very simplified
16
conditions can be solved, as reported, for example, by Charitopoulos et al. [33] or Pavlioglou et al. [41], but
the situation is quite different in the case of a larger number of variables and when considering many
operating conditions.
The character of a variable space may also play a role, as it most likely does not contain many local
extremes. The optimisation of the thrust bearing thus very quickly finds a global maximum. However, this
property of the variable space cannot be generalised to any computational model describing hydrodynamic
lubrication.

5.2 Analysis of Thrust Bearing Performance

The thrust bearing designs under steady-state operating conditions of the TC with a gradual increase in rotor
speed are computationally analysed by the efficient computational model of bearing lubricant flow and heat
transfer [55]. The analysis is performed assuming lubricant absolute pressure of and
temperature of at the inlet.
From a tribological point of view, the minimum thickness and the maximum temperature of the lubricating
layer are strictly evaluated. The computational model of the thrust bearing considers only geometric states,
when the corresponding working surfaces defining the lubrication gap are parallel, and thus the minimum
gap thickness is calculated according to eq. (12). The maximum local temperature of the lubrication layer is
not directly calculated, so the maximum temperature is evaluated from temperatures of the lubricant at
outlets of both sides according to eq. (13). Fig. 7 shows the curves of the minimum gap thickness and the
maximum temperature of the lubricating layer as a function of the rotor speed and compares these values for
the serial and the new bearing designs.
0.10 72
Min. gap thickness [mm]

Max. temperature [ C]
0.08

68
0.06

0.04
64

0.02

0.00 60
10000 20000 30000
Rotor speed [rpm]
Fig. 7 Minimum lubricating gap thickness and maximum outlet temperature of the lubricant
comparing the serial (init) and the new bearing designs.

The new bearing design did not significantly affect the thickness of the lubricating gap but reduced the
lubricant temperature rise. At the same time the load capacity is maintained, and there are not greatly
increased the probability of roughness peak contacts of bearing surfaces or an overheating of lubricant in the
lubricating gap.
The integral characteristics presented in Fig. 8 show a significant reduction in friction losses in both
lubrication layers with a partial reduction in bearing lubricant flow rates at higher rotor speeds. On both sides
of the bearing, the algorithm found the bearing design with reduced outer diameters of the working surfaces.
This bearing design, in principle, could lead to a reduction in the load capacity of one layer. But for two-side
bearing the presence of two layers whose load capacity acts in opposite directions, these negative
consequences are partially eliminated. An important consequence is the recognition that both sides of the
bearing must be considered when optimizing.

17
 1600 1.00 0.16
init, ts init, ts init, ts
init, cts init, cts init, cts

Mass flow rate [kg.s-1]

new, ts new, ts new, ts

Friction torque [N.m]

Load capacity [N]

1200 0.75 0.12

new, cts new, cts new, cts

800 0.50 0.08

400 0.25 0.04

0 0.00 0.00
10000 20000 30000 10000 20000 30000 10000 20000 30000
Rotor speed [rpm] Rotor speed [rpm] Rotor speed [rpm]
Fig. 8 Characteristics of thrust bearing sides (ts and cts) including load capacity, friction torques and mass
flow rates and comparing the serial (init) and the new bearing design.

Fig. 9 shows that the proposed optimisation approach has found the new bearing design that reduces the
friction losses of the thrust bearing by approximately 40%, with a slight decrease in output mass flow rate
compared to the serial bearing design. The integral characteristics of the bearing presented in Fig. 9 are
expressed relative to the serial design as a function of a rotor speed. It is obvious that even though only a few
operating conditions were selected for the optimisation, the resulting reduction of friction losses is achieved
under all simulated operating conditions.
1.2

1.0

0.8
Ratio [-]

0.6
load capacity
0.4
friction torque
0.2 mass flow rate
temp. rise
0.0
10000 20000 30000
Rotor speed [rpm]
Fig. 9 Computed integral characteristics of the new thrust bearing design expressed relative to the serial
design.

5.3 Verification of Thrust Bearing Performance in a Turbocharger

A suitable way of evaluating the applicability of the optimisation approach is to experimentally verify the
thrust bearing under the typical operating conditions of the TC. Therefore, the new thrust bearing is
subjected to technical experiments on the TC being used in stationary ICEs. The TC is supplemented by a
temperature measuring system to measure lubricant temperatures at 16 places including the TC lubricating
system inlet and outlet, thrust bearing inlets and outlets or inlets and outlets of journal bearings. The TC is
externally driven by compressed air in so-called ‘cold conditions’ holding, with some small tolerances,
constant lubricant absolute pressure of and temperature of at the inlet.
Based on the measured average temperature rise and the lubricant volume flow rate, it is possible to estimate
the mechanical losses with reasonable accuracy. However, these mechanical losses include not only friction
in the bearing lubricating layer, but also other types of losses, such as ventilation losses. There is a proposed
strategy that mostly compensates for the inaccuracy of the measurement in this work. The strategy lies in a
comparison of the measured results containing the serial design and the new design of the thrust bearing.
Since all the other components and operating conditions of the technical experiments are identical, it can be
expected that the measured difference lies in the thrust bearing. This strategy also suits the optimisation
approach and thus allows a direct comparison of the suitability of this approach.
A direct impact of the new thrust bearing can be demonstrated by measuring lubricant temperature increase
during flow through bearing sidesError! Reference source not found., the thrust and counter-thrust Error!

18
Reference source not found.sides, respectively. The lubricant temperature increase is presented in Fig. 10
and the effect of the new bearing on the reduction of lubricant temperature increase is obvious.
12
init,ts
init,cts

Temperature increase [ C]
9 new,ts
new,cts

0
4000 10000 16000 22000 28000
Rotor speed [rpm]
Fig. 10 The experimentally determined lubricant temperature increase in the thrust side and the counter-
thrust side of the new thrust bearing and the serial thrust bearing.
120%

100%
Relative power losses [%]

80%

init
60%
new
40%

20%

0%
4000 10000 16000 22000 28000
Rotor speed [rpm]
Fig. 11 The experimentally determined relative power losses of the turbocharger, including the new thrust
bearing and the serial thrust bearing, the power losses are relative to the serial thrust bearing.

The evaluation of mechanical losses is based on the difference between the lubricant temperature at the
lubricating system inlet and at the outlet and can be derived from eq. (4). The results of comparing the power
loss differences of the TC, including the serial and new bearings, are presented in Fig. 11. The relative power
losses of the entire turbocharger measured during operation under several conditions show an approximately
21% reduction. It is obvious that the thrust bearing must achieve an even greater relative improvement, as
journal bearings and other subassemblies also contribute to the losses. The given relative reduction of losses
of the whole TC corresponds to the high influence of the thrust bearing on the total mechanical losses. The
fundamental influence of the bearing on energy losses finds support in the results presented in papers [11,
12].
Optimisation also includes the requirement of limiting the lubricant flow rate. Fig. 12 shows measured
relative lubricant flow rates through the TC and confirms restrictions included in the effective optimisation
approach.

19
 120%

Relative output flow rate [%]

100%

80%
init
60%
new
40%

20%

0%
4000 10000 16000 22000 28000
Rotor speed [rpm]
Fig. 12 The experimentally determined output lubricant flow rate through the TC lubricating system,
including the new thrust bearing and the serial thrust bearing, the output flow rates are relative to the serial
thrust bearing.

6 Conclusions
The developed optimisation approach combines the GA and an efficient computational model of the
lubrication and heat transfer in the thrust bearing. The approach shows high numerical stability, excellent
convergence and, in combination with the efficient model of the bearing, satisfies demands on the speed of
computation. An important feature is the fact that the optimisation approach searches for the global extreme
of the objective function from prescribed intervals of variable values. These intervals, in practical use,
represent some real limitations, for example of structural or technological nature.
The results confirm that this optimisation approach makes it possible to find the optimal values of the
variables defining the thrust bearing, considering many operating conditions. Preferred operating conditions
are more likely to have a formative effect on the bearing’s shape. In contrast, off-design operating
conditions, sometimes referred to as emergency operating conditions, make it possible to eliminate
combinations of variable values, leading to a reduction in bearing operability. Despite the known limitations,
the optimisation approach is applied in the form of a thrust bearing prototype and this prototype is verified
by technical experiments. The results of the experiments have demonstrated excellent bearing performance
under the typical operating conditions of the TC and show the usability of the approach in technical practice.
A computational model of the hydrodynamic thrust bearing is a key part of the effective optimisation
approach. The ways physical processes and their numerical solutions are described are very important
properties of all computational models. A physically detailed model is time-consuming to compute and,
conversely, a computationally fast model is insufficient in terms of physical description and does not allow
sufficient prediction of bearing behaviour. The presented approach is marked as effective and this
effectiveness lies in a balanced combination of the physical depth of the bearing process description,
reasonable computational complexity, appropriate choice of a bearing concept, selection of operating
conditions and robustness of the genetic algorithm. The approach also uses experience from the operation of
turbochargers, leading to the identification of the typical off-design operating conditions necessary for
analysis.
This optimisation approach also has disadvantages. The choice of a working-surface concept is a limitation.
This choice is related to the analytical description of the bearing working surface and the optimisation is
limited to this concept only. However, in principle, it is possible to choose any other working surface
concept to adapt the analytical description, including variables and their use in the optimisation approach.
Nevertheless, a more complex definition of the working surface concept often requires a larger number of
variables for a description and thus greater hardware complexity for a numerical solution.
GA itself includes a certain disadvantage because naturally contains specific properties and thus brings these
properties into the optimisation approach. GA-based optimisation may not be able to find the global
optimum with certain settings. However, the probability of finding the global maximum is greatly influenced
by the size of the initial population. A global maximum is more likely to be found with a larger population,
but it has higher computational demands.

20
This approach can be easily extended to other types of problems that occur in mechanical machines.
However, the effectiveness of the approach will always remain an important condition, regardless of the type
of problem.

Acknowledgement
The research leading to these results has received funding from the Specific research program at Brno
University of Technology, reg. No. FSI-S-20-6267.

Nomenclature
Abbreviations
ICE internal combustion engine
TC turbocharger
CFD computational fluid dynamics
GA genetic algorithm
no. number

Symbols
a axial clearance
maximal allowable axial clearance
lubricant specific heat capacity
axial force
load capacity
friction torque ratio
fitness function
objective function
force residuum
lubricating gap thickness
gene order
number of operating condition
friction torque
̇ outlet mass flow rate
number of generations
number of segments
size of initial population
number of variables
lubricant pressure
radial coordinate
thickness factor
mass flow rate factor
temperature factor
lubricant temperature
weighting factor
parent vector

21
 gene of child
gene of parent

factor
taper wedge angle
adaptive ratio
relative eccentricity
power exponent
lubricant viscosity
tolerance for relative change
lubricant density
circumferential coordinate
portion of individuals
heuristic ratio
rotor angular velocity
relative residuum
relative design variable

Subscripts
2s two sides
cts counter-thrust side
in inlet
init initial
lim limit value
max maximum
min minimum
mix mixture
out outlet
ts thrust side
g groove
t taper
Declaration of interests

☒ The authors declare that they have no known competing financial interests or personal relationships that
could have appeared to influence the work reported in this paper.

Author_statement
Pavel Novotný: Conceptualization, Methodology, Writing - Original Draft, Visualization.
Martin Jonák: Investigation, Software.
Jiří Vacula: Validation, Writing - Review & Editing.

22
References
[1] Pagot A, Duparchy A, Gautrot X, Leduc P, Monnier G. Combustion approach for downsizing: the IFP
concept. Oil and Gas Science and Technology 2006; 61(1): 139–153.
[2] Frei S, Guzzella L, Onder C, Nizzola C. Improved dynamic performance of turbocharged SI engine
powertrains using clutch actuation. Control Engineering Practice 2006; 14: 363–373.
[3] Marshall F, Sorokes J. A Review of Aerodynamically Induced Forces Acting on Centrifugal
Compressors, and Resulting Vibration Characteristics of Rotors. Proceedings of the 29th
turbomachinery symposium 2000, 263–280.
[4] Galindo J, Serrano J, Climent H, Tiseira AA. Experiments and modelling of surge in small centrifugal
compressor for automotive engines. Experimental Thermal and Fluid Science 2008; 32(3), 818–826.
[5] Andersen J, Lindström F, Westin F. Surge Definitions for Radial Compressors in Automotive
Turbochargers. SAE International Journal of Engines 2008; 1(1): 218–231.
[6] Galindo J, Tiseira A, Navarro R, Tarí D, Meano CM. Effect of the inlet geometry on performance,
surge margin and noise emission of an automotive turbocharger compressor. Applied Thermal
Engineering 2017; 110: 875–882.
[7] Podevin P, Clenci A, Descombes G. Influence of the lubricating oil pressure and temperature on the
performance at low speeds of a centrifugal compressor for an automotive engine. Applied Thermal
Engineering 2011; 31(2–3): 194–201.
[8] Raitor T, Neise W. Sound generation in centrifugal compressors. Journal of Sound and Vibration 2008;
314(3): 738–756.
[9] Li D, Gong R, Wang H, Wei X, Liu Z, Qin D. Analysis of Rotor-Stator Interaction in Turbine Mode of
a Pump-Turbine Model. Journal of Applied Fluid Mechanics 2016; 9(5): 2559–2568.
[10] Andersen J, Lindström F, Westin F. Surge Definitions for Radial Compressors in Automotive
Turbochargers. SAE International Journal of Engines 2008; 1(1): 218–231.
[11] Deligant M, Podevin P, Descombes G. Experimental identification of turbocharger mechanical friction
losses. Energy 2002; 39(1): 388-394.
[12] Hoepke B, Uhlmann T, Pischinger S, Lueddecke B, Filsinger D. Analysis of thrust bearing impact on
friction losses in automotive turbochargers. J Eng Gas Turbines Power 2015; 137(8).
[13] Novotný P, Hrabovský J, Juračka J, Klíma J, Hort V. Effective thrust bearing model for simulations of
transient rotor dynamics. International Journal of Mechanical Sciences 2019; 148: 374-383.
[14] Novotný P, Škara P, Hliník J. The effective computational model of the hydrodynamics journal floating
ring bearing for simulations of long transient regimes of turbocharger rotor dynamics. International
Journal of Mechanical Sciences 2018; 148: 611–619.
[15] Nahim HM, Younes R., Shraim H, Ouladsine M. Oriented review to potential simulator for faults
modelling in diesel engine. J. Mar. Sci. Technol 2016; 21: 533–551.
[16] Moreira MF. Failure analysis in aluminium turbocharger wheels. Engineering Failure Analysis 2016;
61: 108–118.
[17] Iliev H. Failure analysis of hydro-generator thrust bearing. Wear 1999; 225–229; 913–917.
[18] Dellis P, Evaggelos R, Alcibiades G, Elias G, Pesyridis A. Turbocharger Lubrication -Lubricant
Behavior and Factors that Cause Turbocharger Failure. International Journal of Automotive
Engineering and Technologies 2013; 2(1): 40-54.
[19] Pastor JV, Serrano JR, Dolz V, Lopez MA, Bouffaud F. Study of turbocharger shaft motion by means
of non-invasive optical techniques: Application to the behaviour analysis in turbocharger lubrication
failures. Mechanical Systems and Signal Processing 2012. 32: 292–305.
[20] Mittwollen N, Hegel T, Glienicke J. Effect of Hydrodynamic Thrust Bearings on Lateral Shaft
Vibrations. ASME. J. Tribol. 1991; 113(4): 811-818.
[21] Berger S, Bonneau O, Frene J. 2000. Influence of Axial Thrust Bearing on the Dynamic Behavior of an
Elastic Shaft: Coupling Between the Axial Dynamic Behavior and the Bending Vibrations of a Flexible
Shaft. ASME. J. Vib. Acoust. 2000; 123(2): 145-149.

23
[22] Koutsovasilis P. Impact of thrust bearing pad design and allocation on automotive turbocharger
rotordynamics. Journal of Sound and Vibration 2020; 485: 115546.
[23] Vassilios Z, Papadopoulos ChI. 3-D thermohydrodynamic analysis of textured, grooved, pocketed and
hydrophobic pivoted-pad thrust bearings. Tribology International 2017; 110: 426–440.
[24] Charitopoulos AG, Papadopoulos ChI, Kaiktsis L, Fillon M. Performance comparison between
textured, pocket, and tapered-land sector-pad thrust bearings using computational fluid dynamics
thermohydrodynamic analysis. Proceedings of the Institution of Mechanical Engineers, Part J: Journal
of Engineering Tribology 2014; 229(4): 376–397.
[25] Deligant M, Podevin P, Descombes G. CFD model for turbocharger journal bearing performances.
Applied Thermal Engineering 2011; 31(5): 811–819.
[26] Cheng ChH, Chang MH. The Optimization for the Shape Profile of the Slider Surface Under Ultra-Thin
Film Lubrication Conditions by the Rarefied-Flow Model. J. Mech. Des. 2009; 131(10): 101010.
[27] Van Ostayen RAJ. Film height optimization of dynamically loaded hydrodynamic slider bearings.
Tribology International 2010; 43(10): 1786-1793.
[28] Rajan M, Nelson HD, Chen WJ. Parameter sensitivity in the dynamics of rotor bearing system. ASME
Journal of Vibration, Acoustic, Stress, and Reliability in Design 1986;108:197–206.
[29] Rajan M, Rajan SD, Nelson HD, Chen WJ. Optimal placement of critical speeds in rotor-bearing
systems. ASME Journal of Vibration, Acoustic, Stress, and Reliability in Design 1987;109:152–7.
[30] Fesanghary M, Khonsari MM. Topological and shape optimization of thrust bearings for enhanced
load-carrying capacity. Tribology International 2012; 53: 12-21.
[31] Saruhan H, Rouch K, Roso C. Design Optimization of Tilting-Pad Journal Bearing Using a Genetic
Algorithm. The International Journal of Rotating Machinery 2004; 10(4): 301-307.
[32] Saruhan H. Optimum design of rotor-bearing system stability performance comparing an evolutionary
algorithm versus a conventional method. International Journal of Mechanical Sciences 2006; 48(12):
1341–1351.
[33] Charitopoulos AG, Visser R, Eling R, Papadopoulos Ch. Design Optimization of an Automotive
Turbocharger Thrust Bearing Using a CFD-Based THD Computational Approach. Lubricants 2018;
6(1): 21.
[34] Fouflias DG, Charitopoulos AG, Papadopoulos ChI, Kaiktsis L. Thermohydrodynamic analysis and
tribological optimization of a curved pocket thrust bearing. Tribology International 2017; 110: 291-306.
[35] Papadopoulos ChI, Nikolakopoulos PG, Kaiktsis L. Evolutionary Optimization of Micro-Thrust
Bearings with Periodic Partial Trapezoidal Surface Texturing. Journal of Engineering for Gas Turbines
and Power 2011; 133(1): 012301-1.
[36] Dadouche A, Fillon M, Bligoud JC. Experiments on thermal effects in a hydrodynamic thrust bearing.
Tribology International 2000; 33: 167–174.
[37] Glavatskih SB, Fillon M, Larsson R. The Significance of Oil Thermal Properties on the Performance of
a Tilting-Pad Thrust Bearing. J. Tribol. 2002; 124(2): 377–385.
[38] Zhai L., Luo Y., Liu X., Chen F., Xiao Y. and Wang Z. Numerical simulations for the fluid-thermal-
structural interaction lubrication in a tilting pad thrust bearing. Engineering Computations 2017[online].
Emerald Publishing Limited, 34(4), 1149-1165 [cit. 2019-04-09].
[39] Neil PB. Heat transfer in pad thrust bearings. Proc Instn. Mech. Engrs 1982; 28: 1960–216.
[40] Chalkiopoulos M, Charitopoulos A, Fillon M, Papadopoulos ChI. Effects of thermal and mechanical
deformations on textured thrust bearings optimally designed by a THD calculation method. Tribology
International 2020; 148: 106303.
[41] Pavlioglou SK, Mastrokalos ME, Papadopoulos CI. Kaiktsis L. Tribological Optimization of Thrust
Bearings Operated with Lubricants of Spatially Varying Viscosity. In: Proceedings of the ASME Turbo
Expo 2014: Turbine Technical Conference and Exposition 2014, Düsseldorf, Germany, V07BT32A004.
[42] Liang X, Xingping Y, Ouyang W, Wood RJK, Liu Z. Thermo-Elasto-Hydrodynamic analysis and
optimization of rubber-supported water-lubricated thrust bearings with polymer coated pads. Tribology
International 2019; 138: 365-379.

24
[43] Jiang X, Wang J, Fang, J. Thermal elastohydrodynamic lubrication analysis of tilting pad thrust
bearings. Proceedings of the Institution of Mechanical Engineers, Part J: Journal of Engineering
Tribology 2011; 225(2): 51–57.
[44] Etsion I. Improving Tribological Performance of Mechanical Components by Laser Surface Texturing.
Tribol. Lett. 2004; 17(4): 733–737.
[45] Ramesh A, Akram W, Mishra SP, Cannon AH, Polycarpou AA, King WP. Friction Characteristics of
Microtextured Surfaces Under Mixed and Hydrodynamic Lubrication. Tribol. Int. 2013; 57: 170–176.
[46] Johnston MT, King WP, Ewoldt, RH. Shear Stress Characteristics of Microtextured Surfaces in Gap-
Controlled Hydrodynamic Lubrication. Tribol. Int. 2015; 82: 123–132.
[47] Yu H, Wang X, Zhou F. Geometric Shape Effects of Surface Texture on the Generation of
Hydrodynamic Pressure Between Conformal Contacting Surfaces. Tribol. Lett. 2010; 37(2): 123–130.
[48] Hsu SM, Jing Y, Hua D, Zhang H. Friction Reduction Using Discrete Surface Textures: Principle and
Design. J. Phys. D: Appl. Phys. 2014; 47(33): 335307.
[49] Lee YH, Schuh JK, Ewoldt RH, Allison JT. Enhancing Full-Film Lubrication Performance Via
Arbitrary Surface Texture Design. J. Mech. Des. 2017; 139(5): 053401.
[50] Schuh JK, Lee YH, Allison JT, Ewoldt RH. Design-Driven Modeling of Surface-Textured Full-Film
Lubricated Sliding: Validation and Rationale of Non Standard Thrust Observations. Tribol. Lett. 2017;
65(2): 35.
[51] Schiffmann J, Favrat D. Integrated Design and Optimization of Gas Bearing Supported Rotors. J. Mech.
Des. 2010; 132(5): 051007.
[52] Zhang J, Qiu Y, Li M, Xu M. Sequential Multi-Objective Optimization for Lubrication System of
Gasoline Engines with Bilevel Optimization Structure. J. Mech. Des. 2017; 139: MD-16-1266.
[53] Markin D, McCarthy DMC, Glavatskih SB. A FEM approach to simulation of tilting-pad thrust bearing
assemblies. Tribology International 2003; 36: 807–814.
[54] Charki A, Diop K, Champmartin S, Ambari A. Numerical simulation and experimental study of thrust
air bearings with multipleorifices. International Journal of Mechanical Sciences 2013; 72: 28–38.
[55] Novotný P, Hrabovský J. Efficient computational modelling of low loaded bearings of turbocharger
rotors. International Journal of Mechanical Sciences 2020; 174: 105505.
[56] Jiang X, Wang J, Fang J. Thermal Elastohydrodynamic Lubrication Analysis of Tilting Pad Thrust
Bearings. Proceedings of the Institution of Mechanical Engineers, Part J: Journal of Engineering
Tribology 2011; 225(2): 51–57.
[57] Brun K, Simons S, Kurz R, Munari E, Morini M, Pinelli M. Measurement and Prediction of Centrifugal
Compressor Axial Forces During Surge—Part I: Surge Force Measurements. Journal of Engineering
for Gas Turbines and Power 2017; 140(1): 012601-1.
[58] Fink D, Cumpsty N, Greitzer E. Surge Dynamics in a Free-Spool Centrifugal Compressor System.
Journal of Turbomachinery 1992; 114(2): 321–332.
[59] Zheng X, Liu A. Experimental Investigation of Surge and Stall in a High-Speed Centrifugal
Compressor. Journal of Propulsion and Power 2015; 31(3): 1–11.
[60] Esping B. Design optimization as an engineering tool. Struct. Opt. 1995; 10: 137–152.
[61] Baluja S, Caruana R. Removing the Genetics from the Standard Genetic Algorithm. Pittsburgh, USA:
School of Computer Science Carnegie Mellon University; 1995. Technical Report CMU-CS-95-141.
[62] Gopi ES. Algorithm Collections for Digital Signal Processing Applications using Matlab. Dordrecht,
The Netherlands: Springer; 2007.
[63] Blum S, Puisa R, Riedel R, Wintermantel M. Adaptive Mutation Strategies for Evolutionary
Algorithms. In: 2nd Weiner Optimization and Stochastic Days; Weimar, Germany; 2005.
[64] Goldberg DE, Deb KA. Comparative Analysis of Selection Schemes Used in Genetic Algorithms.
Foundations of Genetic Algorithms 1991; 1: 69–93.

25
Graphical abstract

26