Rail. Eng.

Science
https://doi.org/10.1007/s40534-020-00213-y

Real-time multibody modeling and simulation of a scaled bogie

test rig
Sundar Shrestha1,2 Æ Maksym Spiryagin1,2 Æ Qing Wu1,2

Received: 13 March 2020 / Revised: 9 May 2020 / Accepted: 12 May 2020

© The Author(s) 2020

Abstract In wheel–rail adhesion studies, most of the test equal or less than actual time, the real-time simulation
rigs used are simplified designs such as a single wheel or environment is prepared in two stages. To such end, the
wheelset, but the results may not be accurate. Alterna- computational time improved from 4 times slower than real
tively, representing the complex system by using a full time to 2 times faster than real time. Finally, the real-time
vehicle model provides accurate results but may incur scaled bogie model is also incorporated with the braking
complexity in design. To trade off accuracy over com- control system which slightly reduces the computational
plexity, a bogie model can be the optimum selection. performances without affecting real-time capability.
Furthermore, only a real-time model can replicate its
physical counterpart in the time domain. Developing such a Keywords Bogie modeling
Scaled bogie test rig
Real-
model requires broad expertise and appropriate software time simulation
Wheel–rail adhesion
Software in loop
and hardware. A few published works are available which
deal with real-time modeling. However, the influence of
the control system has not been included in those works. To
address these issues, a real-time scaled bogie test rig
including the control system is essential. Therefore, a 1:4
1 Introduction
scaled bogie roller rig is developed to study the adhesion
between wheel and roller contact. To compare the perfor-
Railway vehicles are the most efficient land-based trans-
mances obtained from the scaled bogie test rig and to
portation, generally having a higher carrying capacity at a
expand the test applications, a numerical simulation model
comparatively low energy cost. To progressively achieve
of that scaled bogie test rig is developed using Gensys
the requirements of power, speed, and complexity, the
multibody software. This model is the complete model of
modern railway vehicle requires more advances in different
the test rig which delivers more precise results. To exactly
domains of railway operation to test, validate, and optimize
represent the physical counterpart system in the time
performance. To perform such test and validation, the test
domain, a real-time scaled bogie test rig (RT-SBTR) is
rig has become more widespread because it provides a low
developed after four consecutive stages. Then, to simulate
cost, highly repeatable, low maintenance, and very safe
the RT-SBTR to solve the internal state equations and
modular testing environment.
functions representing the physical counterpart system in
In the development of mechatronics product or system,
for instance, a test rig integrates various stages such as
& Sundar Shrestha
design, verification of design (simulation), and physical
[email protected] implementation. It is important to understand the system
development process to avoid the huge cost incurred in
1
Centre for Railway Engineering, Central Queensland later modification. Additionally, the process focuses on the
University, Rockhampton, Australia
safety and reliability of the system. Figure 1 shows the
2
Rail Manufacturing Cooperative Research Centre (CRC), stages to develop the scaled test rig in the laboratory. The
Melbourne, VIC, Australia

123
 S. Shrestha et al.

authors as canvassed in [2, 3]. In [2], the authors have

Algorithm development
mentioned three different scaling strategies, namely DLR,
INRETS, and MMU. Furthermore, scholars [4, 5] have
compared three scaling methods proposed by Pascal,
Conceptual design Iwnicki, and Jaschinski. These investigations report two
Online simulation
fundamental strategies of scaling, namely dimension-based
Algorithm
scaling (i.e., geometrical scaling) and the equation of
motion-based scaling. However, neither of these strategies
works when dealing with some parameters which are not
Software-in-loop (detail design) scalable such as material elasticity and acceleration due to
gravity. The scaling strategies implemented in this project
Algorithm Test rig model RT-PC
are explained in [2].
Furthermore, these reduced-scale and full-scale test rigs
can both be further categorized into four subclasses
Hardware-in-loop (system integration) according to the type of the sub-system being used. They
are the single wheel, wheelset, bogie, and full vehicle. The
Algorithm Physical test rig RT-PC
presence of a higher level of a sub-system in the test rig
increases the accuracy of the test results; however, it also
Fig. 1 Development stages of the scaled bogie test rig raises the complexity. Most of the test rigs used are single
wheel and single wheelset which are simple in design, but
algorithm development phase and conceptual design phase there is room for development [6]. Conversely, the com-
have been accomplished and published in the literature [1]. plex system such as a full vehicle model delivers better
In this paper, the software-in-loop simulation stage of the results but may incur complexity in design. To trade off
scaled test rig development is explained. A simulation accuracy over complexity, the optimum selection can be a
model is the replication of the physical counterpart, which bogie model. While most of the published papers consider
is developed by using specialized multibody software single wheelset dynamics for adhesion studies [7–12], this
(MBS). It is essential for such a model to compare the paper will consider bogie dynamics for adhesion studies
performances obtained from the real vehicle and/or test rigs and contribute to the knowledge in that particular domain.
that are imitating the vehicle behavior. The application of
test rigs to the study of railway vehicle system dynamics 1.2 Research topic 2
has become more widespread because it provides a low
cost, highly repeatable, low maintenance, and very safe The second research question is “Is it possible to develop a
modular testing environment. real-time test rig model with braking control system inte-
grated?”. Currently, simulation modeling has been exten-
1.1 Research topic 1 sively used to study the different domains of railway
operation. The advanced simulation modeling to accurately
Before modeling and designing a test rig, it is necessary to characterize vehicle–track dynamics of the railway is per-
answer a vital question, “What is the optimal design of the formed in [13–15], while [16–23] show the simulation
test rig?”. The devices and validation technologies that are modeling of railway under traction and/or braking. More-
applied for adhesion studies can be classified into two types over, the exact representation of a physical system and its
by their size, full-scaled size and reduced-scaled size. The performance in the time domain is a challenging task. Only
full-scale test rigs are suitable to assess the dynamic per- a real-time model can replicate its physical counterpart in
formances of a prototype vehicle with high accuracy. Their the time domain. In a real-time simulation, the simulation
higher cost is the main concern that limits their extensive is performed in a discrete-time solver with a fixed time step
use. On the other hand, reduced-scale test rigs offer by solving the internal state equations and functions rep-
numerous advantages, namely that they occupy less space, resenting the physical counterpart system in less than that
and they are easier to maintain, easier to handle and require fixed step. It is useful in the design phase to investigate
less effort to change several vehicle parameters. Hence, the controller/actuators before implementing these into the
reduced-scale version of the test rig was selected for this physical model. This allows significant experimental cost
study. Nevertheless, the benefits of this are offset by some reduction and examines the new/potential designs in less
adverse factors. One of the major problematic factors is time. However, developing a real-time model and inte-
scaling. This problem has been investigated by many grating them into multidisciplinary models involve broad
expertise as well as appropriate software and hardware.

123 Rail. Eng. Science

Real-time multibody modeling and simulation of a scaled bogie test rig

Since the real-time model is the replication of the physical real-time environment is created in two consecutive stages.
system using specialized multibody system dynamics Then, a real-time model is developed by modifying the
software running on a computer, one of the major SBTR model in four stages. Finally, the braking control
requirements of such a software package is the computa- system is introduced in the model. The implementation of
tional speed which needs to be faster than real time. In the such advanced methodology will help to develop and
rail vehicle dynamics domain, no such software package verify the proposed solutions at initial design phases by
has been found so far. reducing probable design faults.
The possibilities of developing the model for real-time
application in MATLAB/Simulink have been shown by
Bosso et al. [24]. In [18], researchers developed a real-time 3 Scaled bogie test rig
model by transferring the model developed in SimMe-
chanics (toolbox of Simulink) into the real-time workshop A scaled bogie test rig as shown in Fig. 3 was developed in
in the dSPACE platform. MATLAB models integrally the Centre for Railway Engineering, Central Queensland
possess difficulties with modifying the parameters. Addi- University, Australia. The test rig is the 1:4 scaled version
tionally, the nonlinear components were characterized by of a freight bogie/wagon of 26.5 t axle load for 1067 mm
the comparative linear model, resulting in less accurate narrow gauge track and LW3 wheel profile. The scaling
simulation in the presence of disturbances. An advanced factor (Ϩ) followed in this project is Iwnicki similitude
approach to developing a real-time test rig model for a method which is explained in Ref. [26]. The snapshot of
heavy haul locomotive based on the Gensys MBS package the scale factor implemented for various parameters in this
is presented and verified through co-simulation in [25]. project is provided in Table 1. However, the scale param-
However, the influence of the control system calculation eter comparison between the actual bogie and SBTR is not
process was not included in that model which potentially the scope of this paper. The test rig consists of one bogie
decreases the real-time performance. To address that issue, frame, two wheelsets, and four rollers.
a real-time model integrating the braking control system is
developed in this study that can emulate the behavior of the 3.1 Scaled bogie test rig model
physical counterpart faster than the actual time.
This paper is organized as follows. In Sect. 1, the A scaled bogie test rig is modeled as shown in Fig. 4 to
practical objectives of the investigations are mentioned. In replicate the physical scaled test rig. The parameters for the
Sect. 2, the methodology to solve those objectives/prob- scaled bogie test rig (SBTR) model were derived from the
lems mentioned in Introduction section is formulated. In actual test rig model developed in Gensys [27]. Addition-
Sect. 3, a complete model of the bogie test rig is developed ally, the moment of inertia of the scaled model was derived
and validated with experimental results. In Sect. 4, various from the SolidWorks model of the SBTR. All the compo-
stages to prepare a real-time simulation environment and nents are modeled as rigid mass bodies with six degrees of
phases to develop the real-time scaled bogie test rig from freedom with some constraints as listed in Table 2. The
the complete model are explained. Simulations with dif- scaled test rig is shown in Fig. 3, and basic parameters are
ferent case studies and their results are presented in Sect. 5. presented in Table 3.
The simulation results and comparisons are further dis- The connection between masses is shown in Fig. 4 with
cussed and analyzed in Sect. 6. the number of DOFs indicated. The connection between the
car body and bolster includes 10 couplings made up of

2 Methodology ● two vertical coil springs with longitudinal, lateral, and

vertical stiffness;
The methodology to address the research questions men- ● one anti-roll stiffness element, two lateral bumpstops,
tioned in the introduction section is presented in Fig. 2. To one lateral damper, two vertical viscous dampers;
answer the first question, a scaled bogie test rig (SBTR) ● one damping element and one stiffness element work-
model of a railway freight wagon is developed. The model
ing in parallel in the longitudinal direction for traction
is first tested to ensure that the model code used is free of
rod.
errors and that the multibody model behaves as expected
when basic static and dynamic analyses are performed. The The connection between bolster and side frames
debugged model is then validated with the experimental includes 14 couplings made up of
results of its physical counterpart.
To address the second research gap, a real-time envi- ● six dampers and six stiffness elements for longitudinal,
ronment is essential to simulate the real-time model. The lateral, and vertical directions;

Rail. Eng. Science 123

 S. Shrestha et al.

Development scaled bogie Check and debug Solution of

Validate the model with
test rig (SBTR) model research topic 1
the model experimental findings

Develop real-time scaled Prepare real-time

Validate the RT-SBTR
bogie test rig (RT-SBTR) environment for MBS
model with SBTR
model [stages M1-M4] [stages E1-E2]
Solution of
research topic 2

Introduce braking control Real-time simulation

system on the RT-SBTR

Fig. 2 Methodology to solve the research questions

Fig. 3 Scaled bogie test rig in Centre for Railway Engineering, Centre Queensland University Australia

● two yaw dampers. response, and curve simulation. The difference in contact
patch formation and the distribution of normal and tan-
Each wheelset is modeled as a single mass with six
gential stresses are studied in [29]. In general, the vertical
DOFs. The connection between two side frames and two
curvature needs to be considered in wheel–roller contact,
wheelsets includes 24 couplings made up of
unlike wheel–rail contact where such curvature is huge.
● twenty-four stiffness and damping elements for longi- The finite curvature of the roller makes the contact surface
tudinal, lateral, and vertical directions; semi-axis shorter in the longitudinal direction which results
in the small size of the contact patch area. The standard
● four vertical and four lateral bumpstops. contact coupling wr coupl polach has been used in Gen-
Due to the finite vertical curvature of the roller as sys addressing the issues mentioned above [30].
compared to real track (i.e., rail), there exist differences
between wheel–rail and wheel–roller contact. In [28], the
authors report how those two types of contact differ with
regard to geometry, creep coefficient, stability, vibration

123 Rail. Eng. Science

Real-time multibody modeling and simulation of a scaled bogie test rig

Table 1 Scaling factor considered Gensys online documentation [32, 33] as well as proce-
Parameter Scale factor
dures provided in Table 1 of [34].
The bogie model was checked for syntax errors by using
Scale 1/Ϩ (1/4) Gensys program RUNF_INFO [32] to analyze model code.
Dimension 1/Ϩ (1/4) A visual check was accomplished using the GPLOT [33]
Area 1/Ϩ2 (1/16) utility by plotting the bogie model in three dimensions and
Volume 1/Ϩ3 (1/64) checking for faults such as inappropriate connection
Mass 1/Ϩ3 (1/64) mounts and other geometrical errors. The quasistatic
Velocity 1/Ϩ (1/4) analysis was performed to analyze suspension response due
Acceleration 1/Ϩ (1/4) to displacements of 1 cm in lateral (right) and vertical
Force 1/Ϩ4 (1/256) (downward) directions. In vertical displacements, the
Torque 1/Ϩ5 (1/1024) wheel loads increased evenly throughout all axles. Simi-
Stiffness 1/Ϩ3 (1/64) larly, with 1 cm lateral displacement, the bogie was yawed
Damping 1/Ϩ3 (1/64) with respect to the roller.
Moment of 1/Ϩ5 (1/1024) A time-stepping analysis was conducted to determine
inertia the critical speed at a normal time step of 1 ms. The initial
speed of 67 km/h was implemented with deceleration at
3 km/h per second. Initial excitation was applied to the car
3.2 Model-checking and debugging
body to introduce hunting. The bogie stops hunting at a
speed of approximately 35 km/h as shown in Fig. 5 which
The coefficient of friction in the wheel–rail contact patches
corresponds to 140 km/h full-scale speed.
is assumed to be 0.42 considering dry weather conditions
[31]. The procedure followed in this stage is based on the

Car body

6 10

Bolster

4 14
2

Side frame

24

Axle box 1 Wheelset 1 Axle box

Rollerset Rollerset
bearing 1 Roller set 1 bearing

Test rig base frame

Name Rigid body with label Force elements Y X

1 Z
Joint with number of DOFs

Fig. 4 Bogie test rig model with mass, force element, constraints, and joint with the degree of freedom. Adopted from [27]

Rail. Eng. Science 123

 S. Shrestha et al.

Table 2 Constraints on bodies

Degree of freedom Longitudinal (x) Lateral (y) Vertical (z) Roll (f) Pitch (k) Yaw (p)

Car body Yes, x=0 Yes Yes Yes, f=0 Yes, k=0 Yes
Bolster frame Yes Yes Yes Yes Yes Yes
Side frame Yes Yes Yes Yes Yes Yes
Axle Yes Yes Yes Yes Yes, k=0 Yes
Roller No No No No Yes No
x=0, f=0, and k=0 refer to longitudinal translation displacement, roll, and pitch rotations being fixed to be equal to zero

Table 3 Bogie mass and inertia parameters

Component Parameter
Mass (kg) Center of gravity, Moment of inertia (kg·m2)
vertical (m)
Roll Pitch Yaw

Car body 86.8 0.20 450 200 450

Bolster frame 4.45 0.11 0.14 0.015 0.14
Side frame 5.10 0.115 0.15 0.02 0.17
Axle 25.10 0.115 0.65 0.28 0.65
Roller 40.21 0.20 1.66 1.17 1.66

Fig. 5 Hunting from lateral displacement in critical speed

3.3 Validate with the previous finding conditions. In the first case, the bogie stabilized in all
conditions. For the second case, the bogie showed a
The model is compared with the experimental results from hunting motion when initial lateral displacement was
the physical scaled bogie test rig. The test rig was tested for applied. For the third case, the bogie showed self-hunting
six different test scenarios (see Fig. 6). First, it was tested motion after running for 5.5 s.
at a speed of 35 km/h with and without initial lateral dis- The second set of tests was accomplished to determine
placement. Second, it was tested at 54.6 km/h with and the maximum lateral displacement of the wheelsets when
without initial lateral displacement. Finally, it was tested at 850 N lateral force was applied on the wheelset. Three
67 km/h without initial lateral displacement. The initial different cases were compared as shown in Fig. 7. Forces
lateral displacement was developed on the front wheelset were applied on both wheelsets in the first case. In the
by applying an external force for 100 ms, with a 427 N load second and third cases, the force was applied on the rear
being applied in all cases to match the experimental wheelset and front wheelset, respectively. The maximum

123 Rail. Eng. Science

Real-time multibody modeling and simulation of a scaled bogie test rig

Fig. 6 Lateral displacement at speeds of: a 35 km/h, b 54.6 km/h, and c 67 km/h

Fig. 7 Lateral displacement on wheelset at 35 km/h with 850 N external lateral force applied on a both wheelsets, b rear wheelset, and c front
wheelset

lateral displacement of the front wheelset was found to be 4 Real-time simulation of the model
3.50 mm which was 0.25 mm more than the corresponding
experimental result. Furthermore, the maximum lateral The SBTR model is unable to replicate the physical test rig
displacement of the rear wheelset was found to be 2.5 in the time domain. To be able to imitate the characteristics
which is equal to the corresponding experimental result. of its physical counterpart, the RT-SBTR is modeled by
modifying the SBTR in four phases. For real-time simu-
lation, a dedicated simulator environment is also created.

Rail. Eng. Science 123

 S. Shrestha et al.

4.1 Calculation time in MBS collectively implemented to improve the calculation time
and satisfy the real-time simulation requirement [36].
The test rig model consists of one bogie frame, two In stage E1, the real-time kernel is implemented over a
wheelsets, and four rollers. All the components are mod- generic kernel. The Gensys MBS runs under the UNIX
eled as rigid mass bodies connected by force elements (i.e., environment. The real-time operating system (RTOS)
force couplings) and constraints. In every simulation step, kernel uses a preemptive-based scheduling algorithm
there are many other parameters and each of them has a which allows the scheduler to forcibly perform a context
unique task associated with it. Those parameters are: lsys switch to execute the desired high priority process without
defines the local coordinate system regarding the global waiting for kernel function to complete its execution. In
coordinate system, coupl determines the coupling forces, context switch, the state of a process or a thread is stored
func is the function in model script, mass creates mass which can be restored and resumed from the same stage
inertia in the model, cnstr evaluates constraints, integ is the later. It allows sharing the single CPU among multiple
numerical integrator, and ds represents the data storage processes. The symmetric multiprocessing (SMP) archi-
procedure. A special time estimator for each parameter has tecture of the kernel enables access to a single, shared main
been implemented in Gensys which are represented by the memory where two or more identical processors are con-
parameter subscript of t. The total time ttout in Gensys for nected. Since a CPU with multicore processors is used in
each output step is [25, 35]: this project, the SMP architecture applies to the cores,
ttout ¼ tlsys þ tcoupl þ tfunc þ tmass þ tcnstr þ tinteg þ tds ; ð1Þ considering them as separate processors.
In stage E2, the RAMdisk software is used. The time
where tlsys is the computational time spent on the position taken for data storage tds can be reduced by using RAM-
definition of local coordinate systems with reference to the disk. The computer’s RAM is still faster than even modern
global coordinate system, tcoupl the computational time solid-state drives. RAM can be used as a lightning-fast
spent on commands for coupling elements (coupling ele- virtual drive known as RAMdisk. This program would
ments are elements of various types that connect masses to reserve a section of RAM. All the files on the disk would
each other), tfunc the computational time required for the be stored in your RAM (see Fig. 9c). It could help to
calculation of defined functions in the model script, tmass optimize performance because the load times of the
the computational time spent on mass commands (a mass installed programs in a RAMdisk have near-zero latency
command creates inertia in the model, e.g., car body, bogie, because those data would already be stored in the fastest
wheelset, etc.), tcnstr the computational time spent on con- memory possible. Save a file also happens almost instantly
straint commands, tinteg the computational time required for as it would just be copied to another portion of RAM. This
calculation inside of the numerical integrator, and tds the would mean RAMdisk speeds up the application load times
computational time required for output data storage. and file read/write times for files saved in the RAMdisk. To
automatically create the RAMdisk at every boot, the auto-
4.2 Preparation of real-time environment for MBS mounting option was enabled. However, RAM is volatile
simulation memory and the content of the RAM can be lost if the
computer loses power. To deal with such nonpersistent
For a valid real-time simulation, the real-time simulator memory, a regular backup was set up by creating a bash
must accurately execute the output within the same time script to allow the periodic backup every 10 min in this
duration that its physical equivalent would. A dedicated work.
simulator environment is required to perform such simu- Another option that might be useful is RAM-based
lation. Two different environment preparation stages, solid-state drives (RAM-SSD). These are solid-state drives
namely E1 and E2, as shown in Fig. 8 have been that contain RAM instead of typical flash memory. They
are much faster to read and write as compared to RAM.
Such drives contain a battery so that they can maintain the
__
contents of the RAM even if the computer loses power.
E1 Implementation of a real-time kernel
Since RAM is more expensive than flash memory, RAM-
SSD is an expensive option.

E2 Storage process: RAMdisk or RAM-SSD tds

4.3 Preparation of real-time scaled bogie test rig
(RT-SBTR) model
Fig. 8 Stages to prepare a real-time environment for MBS simula-
tion. In each stage, the columns from first to third represent name, The SBTR model was further simplified in the following
task, and associated time of that stage, respectively stages, namely M1, M2, M3, and M4, as presented in

123 Rail. Eng. Science

Real-time multibody modeling and simulation of a scaled bogie test rig

Initialization of data and Initialization of data and

command input command input

Task Task parallelised Memory

lsys 1 Time=tn+1 lsys 1

func 1 func 1 Time=tn+1

Save data
Read and save data

Time integration
mass 1 mass 1

Time integration
Parallelised task
Memory

coupl 1 func 2

func 2 mass 2

mass 2
Time=tn
coupl 1 coupl 2
coupl 2 Time=tn

(a) (b) (c)

Fig. 9 Time-domain simulation in Gensys: a original serial computing-based numerical steps, b parallel computing-based numerical steps,
c parallel computing using RAMdisk

are presented in [25] for broader understanding. The SBTR

M1 Real-time numerical integrator tintag has large numbers of friction elements and small masses.
For accurate numerical simulation of such a model, it
requires a more dedicated numerical solver such as Runge–
M2 Simplified wheel-roller contact tfunc, tcoupl, tcnstr Kutta which is a four-step method with a fixed step-size
controller. For a simplified model in this study, the two-
step Runge–Kutta (Heun) numerical integrator has been
tfunc, tcoupl, tcnstr
chosen which ignores the backstepping.
M3 Reduction of the elements/components
In stage M2, a simplified contact model is considered.
The full model is capable of 3-point contacts. For a sim-
plified model, a single-point contact is considered. Another
M4 Parallel computing tcoupl reason for considering a single point is that 1 point enables
only one wheel–roller force coupling which can be
Fig. 10 Stages to model RT-SBTR. In each stage, the columns from assigned to 1 CPU core. Thus, wheel–roller force coupling
first to third represent name, task, and associated time of that stage, of a bogie can be assigned via 4 CPU cores simultaneously
respectively
to allow parallel computing. The contact subroutine can be
further simplified by using a lookup table which is not
Fig. 10 to obtain a model suitable for real-time application
considered in this project [38].
with reasonable trade-offs with the accuracy. The execu-
In stage M3, multiple stiffnesses and damping elements
tion time that is associated with tasks has been presented in
in couplings are replaced by equivalent stiffness and
the third column of Fig. 10. The significant reduction in the
damping elements. With the reduction in elements, the
associated time leads to reducing the total time required (i.
number of couplings and functions were reduced. Simi-
e., tout ) to execute in each step.
larly, graphical figures, saving variables, and the associated
In stage M1, an appropriate numerical solver is chosen.
post-processing section were removed after the initial
The list of potential numerical solvers for real-time simu-
check. Thus, the functions and storage were reduced. As a
lation in the Gensys MBS package is presented in [37]. The
result, the total calculation time was remarkably decreased.
real-time solvers in Gensys and its MATLAB equivalent

Rail. Eng. Science 123

 S. Shrestha et al.

In stage M4, parallel computing is implemented. Parallel

computing is not attempted to simulate the bogie model in
railway vehicle engineering. This attempt ensures faster
execution of the program. Parallel computing was enabled
by using the OpenMP [39]. In serial computing (see
Fig. 9a), the parameters are calculated in a series
arrangement. In parallel computing (see Fig. 9b), the nor-
mal components are executed in series and all the
remaining parallelized components are executed simulta-
neously at the end of every time step. In this project, the
wheel–roller coupling forces are parallelized. Thus, paral-
lel computing reduces the total time taken by reducing the
time taken to execute the coupling force (i.e., tcoupl ). Due to
the availability of only four CPU cores, parallel computing
was implemented on the wheel–roller force couplings
Fig. 12 Schematic diagram of the wagon model with brake controller
which are the most execution-time demanding force cou-
plings (see Fig. 11). Upon available of more CPU cores, The brake controller is a proportional–integral controller
other parallelizable force couplings can also be paral- which uses the slide error as the input signal. A first-order
lelized. Activating parallel computing in Gensys has to be lag filter is used to compensate for the associated delay of
compiled with fopenmp flag. the brake system. The slide is estimated from the following
relationship:
4.4 Brake control system xr
k¼1 : ð2Þ
v
The wagon model is then incorporated with a control
system. The brake controller uses a feedback control
approach. The detail of the brake controller is explained in
5 Simulation and results
[1]. The schematic of the braking control system imple-
In this section, the RT-SBTR is validated with SBTR.
mented in the model is presented in Fig. 12, where k
Then, the computational improvement in each stage during
denotes the calculated slide, ko the reference slide, Tb the
the real-time environment preparation and real-time model
brake torque, Ta the adhesion torque, r the nominal rolling
preparation is discussed.
radius of the wheel, Fa the adhesion force, v the linear
roller speed, and x the angular wheelset velocity.
5.1 Validation of the real-time SBTR model

Car bodyy The verification of the RT-SBTR with respect to SBTR has
Secondary been done by the critical speed test and vertical contact
suspension force test. For the critical speed test, the initial speed was
coupling
set to 67 km/h and reduced at the rate of 3 km/h per second.
Core 0

Bogie The critical speed of the RT-SBTR was found to be

Primary sus- approximately 31 km/h as compared to nearly 33 km/h for
pension cou-
pling
SBTR as can be seen from Fig. 13. The reason for the
slightly larger critical speed of the RT-SBTR is due to the
Wheelset 1 Wheelset 2
simplification of the model. The vertical contact force test
of the RT-SBTR was conducted to test the stability of the
Core 0-3

Wheel-roller
force coupling numerical solver. Forces from both SBTR and RT-SBTR
showed similar characteristics and stabilized in a short
time.

Roller-set 1
5.2 Case studies
Roller-set 2

In this section, three case studies are simulated and com-

pared. In Case 1, simulations based on the different stages
Fig. 11 Assignment of CPU core on the wheel–roller force coupling
elements followed to prepare a real-time environment as mentioned

123 Rail. Eng. Science

Real-time multibody modeling and simulation of a scaled bogie test rig

Fig. 13 a Critical speed, lateral displacement of wheelsets at critical speed for b SBTR, and c RT-SBTR

in Sect. 4.2 are performed. In Case 2, simulations based on real-time Linux environment. Since the simulated model is
the different stages followed to model the RT-SBTR as not memory intensive, not many differences were seen in
explained in Sect. 4.3 are shown. In Case 3, the simulation the second stage.
of RT-SBTR model incorporating the brake control system However, a clear difference in performance can be seen
is presented. if a complete railway vehicle is simulated. The total time
ttout in Gensys is presented in Fig. 15. In the RT operating
5.2.1 Case 1 environment, the output has less jitter as compared to the
non-RT environment. Overall, the computing time was
For all simulations, the SBTR model is simulated for 1 s. improved by 2.57 times.
The integration time step was set fixed at 0.25 ms, and the
results were saved every 1 ms. The first simulation was 5.2.2 Case 2
conducted with Gensys installed on the generic Linux
kernel. Then, the simulations were conducted at every This simulation is based on the different stages followed to
stage as explained in Sect. 4.2. Figure 14 presents the total develop RT-SBTR from SBTR mentioned in Sect. 4.3. For
time taken to execute the simulation of 1 s. all simulations, the RT-SBTR model is simulated for 1 s.
The computing time also indicates how many times the The integration time step was set fixed at 0.25 ms, and the
simulations are slower than real time. For example, the
original computing time is 3.99 times slower than real time. Generic
8 E1
In the first stage of improvement, the computing time is
improved by 2.55 times with the implementation of the E2

6
8
ttout (ms)

6
Computing time (s)

4
3.99
4

2
2 1.44 1.42

0 0
Generic E1 E2 Simulation time 1 s
Stages to prepare real-time simulation environment
Fig. 15 Total time ttout from the simulation in every stage with a
Fig. 14 Total computing time to execute the simulation of 1 s timestep of 1 ms for case 1

Rail. Eng. Science 123

 S. Shrestha et al.

results were saved every 1 ms. The simulation speed was 2

E1
improved by more than 600%, while preparing the real- M1
1.8 M2
time environment, however, it was not able to achieve the M3
real-time speed. 1.6 M4
To achieve real-time speed, the simulation needs to be
executed faster than 1 s in this work. Thus, in the second 1.4
stage, the scaled bogie test rig (SBTR) model is improved 1.2

ttout (ms)
in four stages to develop the RT-SBTR model. Figure 16
shows the total time taken to execute the simulation of 1 s. 1
The total time ttout in Gensys is presented in Fig. 17. Firstly, 0.8
a two-step Runge–Kutta (Heun) numerical integrator has
been chosen which ignores backstepping. It improved the 0.6
computing speed of 1.42 s by 42% to reach 0.82 s. Sec- 0.4
ondly, single-point wheel–roller contact was considered in
the RT-SBTR model which improved the speed by 17%. In 0.2

the third stage, multiple stiffnesses and damping elements 0

in couplings were replaced by single equivalent stiffness Simulation time 1 s
and damping elements. Similarly, graphical figures, saving
Fig. 17 Total time ttout from the simulation in every stage with a
variables, and associated post-processing section were also timestep of 1 ms for case 2
removed after the initial check. This improved the com-
putational speed by 29% to reach 0.48 s. Finally, parallel
computing has been applied to the wheel–roller force and 0.2 for wet conditions. The wet friction condition has
couplings which are the most execution-time demanding been chosen for the first 4 s, and the friction condition
force couplings. Since single-point wheel–roller contact is remains dry for the next 4 s. The brake is applied when the
considered in this work, the total 4 wheel–roller force speed is 5.56 km/h.
couplings were assigned to the 4 CPU cores available to The results obtained from the real-time simulation for
allow parallel computing. Thus, the computational speed of wheelset longitudinal slip, adhesion coefficient, speed, and
the final RT-SBTR model was found to be 2.5 times faster calculation time for wet and dry friction conditions can be
than the real time. seen in Fig. 18. Additionally, the wheelset slip is slightly
higher than the reference slip provided. The adhesion
5.2.3 Case 3 coefficient increased from 0.11 to 0.19 when the friction
condition shifts from wet to dry. Even with the brake
A braking control system is also incorporated in the model. control system implemented, the computational time is
Two different slip references have been chosen: 0.1 for dry faster than real time.

1.8 6 Discussion and conclusion

1.6
1.44 Most of the railway wheel–rail adhesion studies have been
1.4 conducted using simplified designs which may not be
Computing time (s)

1.2
accurate. As an alternative, the complex system of a full
vehicle model provides accurate results but may create
1 difficulty in preparing its design. Thus, a bogie model has
0.82
0.8 0.68 been selected in this project which has a slight trade-off
0.6
with accuracy to significantly reduce complexity in design.
0.48
0.39 Furthermore, to exactly represent the physical bogie test rig
0.4 in the time domain, the simulation model needs to run faster
0.2 than the physical test rig. In other words, the simulation
0
model needs to be capable of performing in a discrete time
E1 M1 M2 M3 M4 with fixed step by solving the internal state equations and
Stages to prepare RT-SBTR from SBTR functions representing the physical counterpart system in
less than actual time. In general, such a model is known as a
Fig. 16 Total computing time to execute the simulation of 1 s at
different stages of preparing RT-SBTR from SBTR
real-time model. An important question then arising is how

123 Rail. Eng. Science

Real-time multibody modeling and simulation of a scaled bogie test rig

(a)

Longitudinal
0.3
Reference slip

slip
0.2 Wheelset slip
0.1
0
(b)
0.2
coefficient
Adhesion

Wheelset
0.15

0.1
(c)
6
Speed (km/h)

Wheelset
4 Roller

0
(d)
1
ttout
Calculation
time (ms)

0.75
0. 5
0.25
0
0 1 2 3 4 5 6 7 8 9
Time (s)

Fig. 18 Parameters in time domain: a reference slip and an estimated longitudinal slip of wheelset, b adhesion coefficient, c speed of the
wheelset versus roller, and d total time for each output step of 1 ms

to achieve such a model which has higher computing speed, are more dedicated and powerful computers already
but also good modeling accuracy. To address this question, available in the market, which may help to achieve faster
the model has been prudently simplified, and then, an computational speed.
innovative computing scheme involving parallel computing Simulation modeling of a scaled test rig for an adhesion
has been implemented. study at the wheel–rail interface is described in this paper.
Three simulation cases were carried out for 1 s each for To progressively conduct the study of the complexity of the
comparison purposes. In case 1, the computational speed modern railway vehicle, hardware-in-loop simulation
while preparing a real-time simulation environment was integrating the physical test rig will be performed in future.
recorded. The computing speed was improved by 2.57
times by establishing the simulation environment without Acknowledgements The authors greatly appreciate the financial
support from the Rail Manufacturing Cooperative Research Centre
changing the model. However, the time taken to simulate (funded jointly by participating rail organizations and the Australian
was far from real time. In case 2, the model was further Federal Government’s Business Cooperative Research Centres Pro-
simplified, and parallel computing was implemented. This gram) through Project R1.7.1—“Estimation of adhesion conditions
reduced the computation time from 1.42 s to 0.39 s which between wheels and rails for the development of advanced braking
control systems.” Tim McSweeney, Adjunct Research Fellow, Centre
was 2.5 times faster than real time. In both cases, the for Railway Engineering is thankfully acknowledged for his assis-
control system was not incorporated into the model. In tance with proofreading.
general, the addition of the control system makes the model
more computably expensive and unstable. In this work, the Open Access This article is licensed under a Creative Commons
Attribution 4.0 International License, which permits use, sharing,
computational time of the model with the control system adaptation, distribution and reproduction in any medium or format, as
included is 0.62 s which is still faster than real time. Thus, long as you give appropriate credit to the original author(s) and the
the proposed technique can satisfy the requirements of the source, provide a link to the Creative Commons licence, and indicate
real-time simulation of the system. However, the brake if changes were made. The images or other third party material in this
article are included in the article’s Creative Commons licence, unless
control used in this work is simplified and a more precise indicated otherwise in a credit line to the material. If material is not
control scheme may lead to a higher computational com- included in the article’s Creative Commons licence and your intended
mitment. A computer equipped with Intel® Core™ i5-4570 use is not permitted by statutory regulation or exceeds the permitted
CPU @3.20 GHz with 8 GB of RAM has been used for the use, you will need to obtain permission directly from the copyright
holder. To view a copy of this licence, visit http://creativecommons.
simulation process with the Gensys MBS software. There org/licenses/by/4.0/.

Rail. Eng. Science 123

 S. Shrestha et al.

References 20. Onat A, Voltr P (2020) Particle swarm optimization based

parametrization of adhesion and creep force models for simula-
1. Shrestha S, Spiryagin M, Wu Q (2019) Friction condition char- tion and modelling of railway vehicle systems with traction.
acterization for rail vehicle advanced braking system. Mech Syst Simul Model Pract Theory 99:102026
Signal Process 134:106324 21. Park S, Kim J, Choi J et al (2008) Modeling and control of
2. Bosso N, Allen PD, Zampieri N (2019) Scale testing theory and adhesion force in railway rolling stocks. IEEE Control Syst Mag
approaches. In: Iwnicki S, Spiryagin M, Cole C, McSweeney T 28(5):44–58
(eds) Handbook of railway vehicle dynamics, 2nd edn. CRC 22. Pugi L, Malvezzi M, Papini S et al (2013) Design and preliminary
Press, Boca Raton validation of a tool for the simulation of train braking perfor-
3. Jaschinski A (1990) On the application of similarity laws to a mance. J Mod Transp 21(4):247–257
scaled railway bogie model. Delft University of Technology, 23. Chen ZG, Zhai WM, Wang KY (2018) Locomotive dynamic
Delft performance under traction/braking conditions considering effect
4. Allotta B, Pugi L, Malvezzi M et al (2010) A scaled roller test rig of gear transmissions. Veh Syst Dyn 56(7):1097–1117
for high-speed vehicles. Veh Syst Dyn 48(sup1):3–18 24. Bosso N, Spiryagin M, Gugliotta A et al (2013) Mechatronic
5. Bosso N, Gugliotta A, Somà A (2002) Comparison of different modeling of real-time wheel-rail contact. Springer, Berlin
scaling techniques for the dynamics of a bogie on roller rig. Veh 25. Spiryagin M, Sun YQ, Cole C et al (2013) Development of a real-
Syst Dyn 37(sup1):514–530 time bogie test rig model based on railway specialised multibody
6. Shrestha S, Wu Q, Spiryagin M (2019) Review of adhesion software. Veh Syst Dyn 51(2):236–250
estimation approaches for rail vehicles. Int J Rail Transp 7(2):79– 26. Iwnicki S, Spiryagin M, Cole C et al (2019) Handbook of railway
102 vehicle dynamics, 2nd edn. CRC Press, Boca Raton
7. Chang C, Chen B, Cai Y et al (2019) An experimental study of 27. Shrestha S, Wu Q, Spiryagin M (2018) Wheel-rail contact
high speed wheel-rail adhesion characteristics in wet condition on modelling for real-time adhesion estimation systems with con-
full scale roller rig. Wear 440–441:203092 sideration of bogie dynamics. In: Li Z, Núñez A (eds) 11th
8. Bruni S, Liu B, Sabbioni E et al (2018) Development of a MB international conference on contact mechanics and wear of rail/
model for designing a control strategy of a full scale roller rig. wheel systems, Delft, The Netherlands, pp 862–869
Dyn Veh Roads Tracks 2:885–891 28. Zhang W, Dai H, Shen Z, Rig R et al (2006) Handbook of railway
9. Bosso N, Zampieri N (2014) Experimental and numerical simu- vehicle dynamics. CRC Press, Boca Raton, pp 458–504
lation of wheel-rail adhesion and wear using a scaled roller rig 29. Liu B, Bruni S (2015) Analysis of wheel-roller contact and
and a real-time contact code. Shock Vib 2:1–14 comparison with the wheel-rail case. Urban Rail Transit 1
10. Bosso N, Zampieri N (2013) Real-time implementation of a (4):215–226
traction control algorithm on a scaled roller rig. Veh Syst Dyn 51 30. AB DEsolver, creep_polach_2. http://www.gensys.se/doc_html/
(4):517–541 calc_coupl.htm#jcreep_polach_2. Accessed 10 Apr 2020
11. Onat A, Voltr P (2020) Velocity measurement-based friction 31. Spiryagin M, Nielsen D, Wu W et al (2018) Advanced friction
estimation for railway vehicles running on adhesion limit: swarm measurement and their application for locomotive traction-track
intelligence-based multiple models approach. J Intell Transp Syst damage studies. In: CORE 2018: conference on railway excel-
Technol Plan Oper 24(1):93–107 lence, Railway Technical Society of Australasia (RTSA); Tech-
12. Zhao Y, Liang B, Iwnicki S (2014) Friction coefficient estimation nical Society of Engineers Australia, Sydney, Australia
using an unscented Kalman filter. Veh Syst Dyn 52(sup1):220– 32. AB DEsolver (n.d.) Users manual for program runf_info: the
234 gensys homepag. http://www.gensys.se/doc_html/misc_runf_
13. Zhai WM (2020) Numerical method and computer simulation for info.html. Accessed 28 May 2018
analysis of vehicle-track coupled dynamics. In: Zhai W (ed) 33. DEsolver (n.d.) Debugging a vehicle model: the gensys home-
Vehicle–track coupled dynamics. Springer, Singapore, pp 203– page. http://www.gensys.se/doc_html/analyse_check.
229 html#Mainmenu. Accessed 26 May 2018
14. Xu L, Zhai WM (2019) A three-dimensional dynamic model for 34. Spiryagin M, George A, Ahmad SSN et al (2012) Wagon model
train-track interactions. Appl Math Model 76:443–465 acceptance procedure using Australian standards. In: conference
15. Wu Q, Sun Y, Spiryagin M et al (2018) Parallel co-simulation on railway engineering, CORE2012, Brisbane, Australia
method for railway vehicle-track dynamics. J Comput Nonlinear 35. Spiryagin M, Cole C, Sun YQ et al (2014) Advanced simulation
Dyn 13(4):041004 methodologies, design and simulation of rail vehicles. CRC
16. Allotta B, Conti R, Meli E et al (2015) Modeling and control of a Press, Boca Raton, pp 277–311
full-scale roller-rig for the analysis of railway braking under 36. Belanger J, Venne P, Paquin JN (2010) The what, where and why
degraded adhesion conditions. IEEE Trans Control Syst Technol of real-time simulation. http://sites.ieee.org/pes-resource-center/
23(1):186–196 files/2013/10/11TP255E.pdf
17. Bosso N, Gugliotta A, Somà A (2006) Design and simulation of 37. AB DEsolver (n.d.) Users manual for program CALC. http://
railway vehicles braking operation using a scaled roller-rig. In: www.gensys.se/doc_html/calc.html. Accessed 19 Apr 2018
Brebbia CA, Tomii N, Tzieropoulos P, Mera JM (eds) Computers 38. AB DEsolver (n.d.) Users manual for program KPF. http://www.
in railways X. WIT Press, Southampton, UK, pp 869–883 gensys.se/doc_html/kpf.html. Accessed 3 Sept 2017
18. Kang CG, Kim HY, Kim MS et al (2009) Real-time simulations 39. OpenMP Architecture Review Board (2018) OpenMP application
of a railroad brake system using a dSPACE board. In: 2009 programming interface. https://www.openmp.org/wp-content/
ICCAS-SICE, Fukuoka, Japan, pp 4073–4078 uploads/OpenMP-API-Specification-5.0.pdf
19. Wu Q, Cole C, Spiryagin M (2019) Train braking simulation with
wheel-rail adhesion model. Veh Syst Dyn. https://doi.org/10.
1080/00423114.2019.1645342

123 Rail. Eng. Science