You are on page 1of 48

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/333619295

SIMULATION OF RAIL VEHICLE DYNAMICS WITH UNIVERSAL MECHANISM


SOFTWARE

Chapter · January 2005

CITATIONS READS

18 530

1 author:

Dmitry Pogorelov
Bryansk State Technical University
78 PUBLICATIONS   747 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Simulation of drillstring dynamics View project

Numerical methods in multibody system dynamics View project

All content following this page was uploaded by Dmitry Pogorelov on 17 September 2019.

The user has requested enhancement of the downloaded file.


10
D. Yu. POGORELOV1)

SIMULATION OF RAIL VEHICLE DYNAMICS WITH UNIVERSAL


MECHANISM SOFTWARE

Summary. Some algorithms and numerical methods implemented in the program package
Universal Mechanism are considered. The program package is intended for simulation of
multi-body system dynamics, especially for computer-aided 3D modelling rail vehicles.

1. INTRODUCTION

Modelling rail vehicle dynamics by means of representation the vehicles by system of rigid
and/or elastic bodies (multi-body system, MBS) gives the most efficient method of running
gear analysis for designed carriages and locomotives, their parametric and structural
optimization, comparison of dynamic properties of various vehicles, analysis of vehicle
behavior in non-standard situations, e.g. the derailment process.
Mathematical model is the basis for simulation of rail vehicles. As a rule, the model is a
system of ordinary differential or differential-algebraic equations. Due to large size of
equations and strong non-linear nature of force elements, the equation should be solved by a
computer. Computer-aided modelling of rail vehicle dynamics started in 1960s. At that time
researchers programmed the equations of motion manually derived. A considerable progress
in computer-aided simulation of rail vehicle dynamics was achieved due to development of
program packages, which allow the user to automate the process of derivation of motion
equations. The equations are derived according to inertia, geometric, kinematic parameters
and force elements described by the user. At present this field in the market of software is
presented by the program packages ADAMS/RAIL, GENSYS, MEDYNA, NUCARS,
SIMPACK, VAMPIRE etc. [1]. The main way to simulate the vehicle dynamics in these
programs is a numeric integration of motion equations (i.e. analysis in the time domain).
Some programs have been developed to solve special problems; for instance, the TRAIN
program is intended for simulation of longitudinal dynamics of trains [2].
In this paper some methods and algorithms implemented in the program package Universal
Mechanism (UM, www.umlab.ru) are discussed. The UM software has been developed in
Bryansk State Technical University since the end of 1980s. One of the main directions for the
program practical usage is simulation of 3D dynamics of rail vehicles and tractive
connections. With the help of UM Software some aspects of dynamics of diesel locomotives
TE116, TEP70, TEP80, TA15 and TA35 (TA15 and TA35 are the models of main-line
locomotives designed in RLS Bryansk machine-building plant), TEM21, diesel-train DR1A,
electric locomotives VL10u, EP10, high-speed coach 61-4170, freight cars with three-piece
bogies of different design, ballast cleaner SHOM1200, railcar AS4 etc., have been analyzed.
______
1)
Bryansk State Technical University, Russia

11
One of the latest developments is the 3D model of three freight cars connection. Main
goals of research were the optimization of parameters of existing and designed vehicles
directed on improving the running gear dynamics, decreasing rail/wheel wear, increasing
safety etc. A considerable part of research was devoted to optimizing parameters of wheelset
self-steering mechanisms for a number of models of electric and diesel locomotives. The
model of the freight car with three-piece bogies and the model of the tractive connection of
cars are used for analysis of derailment process.
In the paper we present algorithms, which allow a very fast and reliable simulation of rail
vehicle with UM. First of all, formalisms for deriving equations of motion are considered
(Sect.2). These algorithms can be used for generating equations both in symbolic and in
numeric-iterative forms. One of the most important features of UM is the implementation of
both forms of equations. This fact joins advantages of quickness, symbolic equations and
flexibility of numeric-iterative ones.
In a considerable degree, the universality of MBS software depends on a reliable
description of various interactions between bodies based on mathematical models of forces
and force elements. In Sect.3 we consider a number of mathematical models of forces, which
are used for modelling springs, dampers as well as contact interactions. Sect.4 is devoted to
description of fast algorithms for a rail-wheel interaction. In particular, a semi-analytic
modification of the well-known Kalker’s FastSim algorithm is discussed.
A specific structuring of rail vehicles, tractive connections, and trains models called
subsystem technique is developed in UM (Sect.5). This technique serves not only to the
purpose of reduction user’s efforts in development of models but also gives bases for
connections of sub-models of different nature such as rigid body systems, flexible parts,
electric systems.
Numeric methods for integration of equations of motion are discussed in Sect.6. A special
consideration in this section is given to a problem of stiff equations as well as to computation
of approximate Jacobian matrices. Analytic expressions for local Jacobian matrices of some
forces are derived.
In the final section we present some models of rail vehicle developed by means of UM.
Manchester benchmark vehicles are used to illustrate effectiveness of algorithms described in
the paper. A freight car with three-piece bogie is considered as an example of complicate
models.

2. EQUATIONS OF MOTION

2.1. Kinematics of multi-body systems


Let us content ourselves with models of railway vehicles consisting of rigid bodies
connected with ideal joints and arbitrary force elements. Let the MBS be holonomic and
scleronomic.
Often equations of motion of rail vehicles are derived in a geometrically linear form, it is
r0001

supposed that generalized coordinates are of small relativity to a prescribed stationary motion.
At the same time, 3D simulation of tractive connections and trains, as well as simulation of
in-board mechanisms (for instance, tools of track packing machines) requires deriving the
equations in a full non-linear form. Consider the so-called Newton-Euler method, which is
some of the most general formalism for generation necessary kinematic and dynamic
relations. This formalism usually uses relative coordinates in joints as a set of coordinates.

12
Suppose the system contains n bodies and m  n joints. If the system has closed loops
( m  n ), then m  n joints must be cut. Introduce an ordered numeration 1..n of bodies and
joints, which are not cut. Consider a chain of the system tree, which begins at body 0 (inertial
frame SC0). Indices of bodies increase along the chain, i.e. the minimal index has the body
connected with body 0. Indices of joints in the chain are equal to those of bodies, namely joint
j connects bodies i and j, j  i  j  . Suppose cut joints have greater indices n+1,..,m.
Introduce sets of indices: J k  is the set of indices of joints included in the path from body
k to body 0; Bk  is the set of indices of bodies, which paths to body 0 contain joint k. Let q
be coordinates of the system, q  q1T    qnT 
T
, where q j is the n j  1 matrix of local
coordinates in joint j, which set the position of body j relative to body j  .
Suppose rj , v j , a j , A0 j ,  j ,  j be the radius vector, velocity and acceleration of the origin
as well as the direction cosine matrix, angular velocity and angular acceleration of a body-
fixed frame of body j. The following recursive relations are valid:
   
r j q   ri q   A0i q riji q j , A0 j q   A0i q Aij q j ,
v v  ~ r v ,    ,
j i i ij ij j i ij
a j  ai  ~ ~
i rij   ~ ~
i i rij  2i vij  aij ,
~   ,
 j  i  i ij ij
where rij , Aij , vij , ij , aij , ij specify the local joint kinematics, i.e. the position and motion of
body j relative to body i.
vij  d j q j , ij  b j q j ,
aij  d j q
 j   j , ij  b j q
 j   j .
Here and all over the paper we use the matrix notations introduced in [3]. In particular, an
index over a matrix-vector denotes a number of a body-fixed system of coordinates (SC), in
which the vector is presented; if an index is not presented, the vector is resolved in SC0 (the
inertial frame); the direct cosine matrix Ai 0 is inverse to A0i , and the sign ~ above a vector
corresponds to the skew-symmetric matrix
 bx   0  bz b y 
  ~  
b   b y   b   bz 0  bx  .
b   b 0 
 z  y bx

zi zj zi zj
b yj
yi , y j
yi
xj
φ y
xi xi xj
a

Fig. 1. Rotational and prismatic joints

Local joint kinematics depends on a joint type. The corresponding relations for joints with
any number of rotational and translational degrees of freedom can be easily obtained by

13
means of a recursive procedure. As an example consider local kinematic relations for the
rotational and prismatic joints shown in Fig. 1.
a) Rotational joint.
 0  1 0 0   0    
       
rij   a  b cos  , Aij   0 cos   sin  , vij    b sin  , ij   0  ,
 b sin    0 sin  cos    b cos   0
       
 0   0   1
     
aij    b cos  , ij  0, d j    b sin  , b j   0  .
2

  b 2 sin    b cos    0


     
b) Prismatic joint.
0 0  0
     
rij   y , Aij  I 3 , vij   y , ij  0, aij  0, ij  0, d j   1 , b j  0.
0 0  0
     
Here I 3 is the 3 3 identity matrix.

Introducing the global velocities and accelerations



V j  v Tj Tj ,
T
W j  a Tj  Tj 
T

yields the recursive formulas


V j  CijVi  S j q j ,
W j  CijWi  S j q  j  jr ,
~  j 
 I  rij  d j 
Cij   3  , S j    ,  j    ,
0 I3   bj   j 
and the explicit expressions
V j   j q, W j   j q    j , (1)
   C Sˆ  C  Sˆ  .
j kj k 0j k
(2)
k J  j  k J  j 

Here we introduced the notation for 6  n j matrices X k : X k  C0k1 X k , as well as for 6  m


matrices (m is total the number of coordinates):
Xˆ k  0   0 X k 0   0 .
In the case of MBS with closed kinematic loops, the coordinates q are redundant, and
constraint equations corresponding to closure conditions for cut joints should be derived. If
joint s with any number of rotational and translational degrees of freedom is cut, the
corresponding constraint equations can be written as
 rv q   ru q   A0u q ruv
u
qs  (3)
 3
hs q, qs    1 ~  0.
 2  ek A0v q Avu qs Au 0 q ek
 k 1
Here u,v are indices of bodies connected by the joint, and ek , k  1,2,3 are the bases vectors
of SC0. These equations depend both on coordinates q and on auxiliary local coordinates q s
in the cut joint.
Deriving equations of motion requires Jacobian matrices of constraint equations. In the
above case the Jacobian matrix looks like this:

14

H s  H s qT 
H s qsT   v  Cuv u  Sˆs (4)

2.2. Formalisms for generation of equations of motion


Most of the known formalisms for computer-aided generation of equations of motion are
based on the Newton-Euler equations written for each body of the system. If the body-fixed
coordinate systems have body center of masses as origins, the Newton-Euler equations for
body j are
M jW j  k j  F j  R j , (5)
where M j , k j are the mass matrix of the body and the matrix of inertia forces
m j I3 0   0 
M j   , k j   ~


 j J j j  ,
 0 Jj  
m j , J j are the mass and the inertia tensor, the 61 matrices F j , R j contain applied and
reaction forces and moments reduced to the center of mass of body j.
The Newton-Euler formalism for generation of equations of motion of MBS consists of
multiplying Eq.(5) from the left by the matrix  Tj , Eq.(2), substituting expression (1) for
accelerations W j , and summarizing over the set of bodies in the system. The formalism leads
to the following differential-algebraic equations:
M q q
  k q, q   Qq, q, t   H T q, p  H , (6)
hq, p   0 ,
S sT q, qs  Hs  0, s  n  1,..., m
where M is the mass matrix, k and Q are the columns of generalized inertia and applied forces,
the vector-function h is a combination of constraint equations (3) for separate cut joints, the
matrix H unites the corresponding Jacobian matrices of constraint equations (4),

 H  THn 1 ... THm 
T
is the matrix-column of reaction forces in cut joints
 Hs , s  n  1,..., m . The vector of auxiliary variables p  qnT1 ... qmT 
contains local
T

coordinates in cut joints.


Auxiliary coordinates in cut joints can be excluded from the constraint equations (3), so
that the constraint equations for cut joints and the corresponding Jacobian matrices become
functions of coordinates q
g s q   0, Gs q   g s qT ,
and Eq.(6) takes the well-known descriptor form
M q q
  k q, q   Qq, q, t   G T q  , (7)
g q   0, G  g qT .
Matrices M, k, Q in Eq. (6), (7) are computed according to the following relations:
Q  k    Tj F j  M j  j  k j     Tj G j .
n n n (8)
M    Tj M j  j ,
j 1 j 1 j 1
Computation of elements of equations by multiplying the corresponding matrices
according to Eq.(8) coincides with the direct method of deriving the equations. In case of a
 
chain of n bodies it requires O n 3 floating point operation for evaluation of the mass matrix,

15
 
and O n 2 operations for evaluation of the generalized forces. So, the direct method is not
optimal for MBS with long kinematic chains, but it is quite good for rail vehicles, where
kinematic chains are normally short.
The composite body method is much more effective for MBS with long kinematic chains
[4]. It can be easily obtained from Eq.(8) if the structure (2) of the matrices  j as well as the
following rearrangement of summation formulas
n n n
    
n n
    ,
j 1k J  j  k 1 j  B  k  j 1k J  j  mJ  j  k 1m 1 jB max k , m 

are taken into account. Really, a chain of transformations for the mass matrix should be
considered.
n n n
M    SˆkT C0T j M j C0 j Sˆk     SˆkT M j Sˆm 
j 1k J  j  mJ  j  k 1m 1 jB max k , m 

n n   n n
  SˆkT   M j  Sˆm    SˆkT M max k , m Sˆm
k 1m 1  jB max k , m   k 1m 1

This formula gives a very simple expression for the nk  nm block M km of the mass matrix
M
M km  SkT Mmaxk , m Sm .
The similar chain of transformation for generalized forces gives
n n n n
Q  k    SˆkT C0T j G j    SˆkT C0T j G j   SˆkT  G j   SˆkT G k ,
j 1k J  j  k 1 j  B  k  k 1 j  B k  k 1

so that nk  1 block of the matrix Q  k is computed according to the formula


Q  k k
 S kT G k .
If we consider a chain of n bodies, the evaluation of the composite matrices
M k   M j , G k   G j
j  B k  j  B k 

as well as the generalized forces requires O n  operations, and the evaluation of the mass
 
matrix is as costly as O n 2 . This is the reason why the composite body method is very
popular.
The third formalism for generation of equations implemented in UM is the articulated
rigid body method. It is effective for systems with very long kinematic chains and without
closed loops. Moreover, its use for stiff equations is not trivial. So, we do not consider the
articulated body method in this paper.

2.3. Implementation of formalisms


UM, like all MBS software, generates equations of motion automatically. Two different
approaches are used for this purpose. The first one consists in deriving equations in a
symbolic form. Generation of equation in the symbolic form means that kinematic relations as
well as elements of equations of motion (6) are obtained as expressions either in Object Pascal
or in C with the help of a built-in specialized computer-algebra system (CAS) [5]. The
equation codes should be compiled with an external compiler. The simulation module loads
the resulting dynamic linked library. The CAS provides a very fast generation of equations. A

16
built-in optimizer of symbolic expressions reduces the number of floating point operations in
the code and allows generating equations for large MBS with hundreds of degrees of freedom.
The alternative method for deriving equations is a numeric-iterative algorithm. It provides
kinematic relations and elements of the equations at each step of numeric integration.
Deriving the equations of motion in the symbolic form has advantages in simulation speed,
whereas the numeric-iterative algorithms make the simulation more flexible with respect to
changing the object structure during the simulation.
A very important feature of UM consists in a full parameterization of vehicle models. This
means that description of graphic, geometric, inertia parameters as well as parameters
defining force element models can be set by symbolic identifiers and expressions. For
instance, spring ratios, their static deformations, coordinates of attachment points can be
parameterized.
Parameterization of models builds the basis for analysis of the vehicle behavior depending
on parameters, comparative analysis of similar models as well as parametric optimization.
Consider a model of the frictional wedge of the three-piece bogie as an example, Sect.7.2,
Fig. 39. The parameterization of the image and coordinates of contact points with the same
identifiers and usage of macroinstructions for identifiers (i.e. simultaneous assignment of
specific values to the set of identifiers with the help of a single action) allows changing the
wedge type by means of the “Type of wedge” macroinstruction.

3. MODELS OF APPLIED FORCES

In this section we consider only small part of mathematical models, which are necessary
for simulation of rail vehicles. In particular, rail-wheel contact forces are considered in the
next section.

3.1. Bipolar force elements

Oi
i Oj
-G j G
A B
r

Fig. 2. Bipolar force element

A bipolar force element connects two chosen fixed points A, B of a pair of bodies,
Fig. 2. The force acts along the straight line between the attachment points and may depend
on distance r between the points, its time derivative v  r , and time t.
G  f (t, r, v)e .

17
where e is the unit vector along the element.
If the distance r equals zero, the degeneration of the force element occurs due to the
uncertainty of the force direction, and the force is assumed to be zero in this case.
The distance r depends on the body positions and on the local coordinates of the
attachment points
r  r  rT r , (9)
AB AB AB

rAB  rB  rA  r j  A0 j  jj  ri  A0i ii .


Differentiation of this formula with respect to time gives the expression for v

 
r T r ~  v  ~ (10)
v  r  AB AB  eT v j   j j i i i
e
The element of this type is often used in models of rail vehicles for modelling dampers,
rods etc. Consider a number of mathematical models of the force.
The first example is a linear viscous-elastic force element with c and d parameters as
stiffness and damping coefficients. Let the force be equal to f when the length of the element
is x0 and the velocities vanish. The analytic expression for the force looks like
f  f 0  c  x  x0   dv

The next example is a bipolar friction force. This type of force is mainly used for
modelling frictional dampers. The force description includes two modes: sliding and sticking.
In the sliding mode the force satisfies the formula
f   Fsignv 
analogously to the Coulomb friction with a constant friction force F, and v as a sliding
velocity. In the sticking mode, the force model looks like
f  f 0  c  x  x0   d  v ,
i.e. it presents a linear viscous-elastic force with the c and d parameters as stiffness and
damping coefficients.
The sliding-sticking transition occurs at the moment when the velocity v changes its sign.
At this moment the force f and the coordinate x values are stored (the f0 and x0 parameters in
the formula for the force in the sticking mode).
The sticking-sliding transition occurs when the force reaches its maximal value
f  F0 ,
where F0 is the maximal value of static friction force.

C2
C1

x2 x1
x

Fig. 3. Load dependent friction force

18
Scheme of a load dependent friction force is shown in Fig. 3. The force element of this
type can by used for modelling leaf springs, internal friction in rubber elements etc. The
element includes a spring ( c1 ) in series with a parallel combination of the second spring ( c2 )
and a Coulomb friction element. In contrary to the previous model of the frictional element,
here the friction force is not constant. Its value depends on a deformation of the spring in
parallel c2 .
Consider the mathematical model of the element. Let x1 , x2 be the length of the springs,
and x2 0  0, x1 0  x0 be the initial values of these variables. This means that at the
beginning of a simulation the second spring c2 has zero length and this position corresponds
to its undeformed state. Let L0 be the length of the spring c1 in an undeformed state. Then the
forces produced by the springs can be computed from the expressions
f1  c1x1  c1 x1  L0   c1 x  L0  x2   c1 x  x2 ,
f 2  c2 x2  c2 x2 .
Here x  x  L0 .
As usual, the friction has two models. At sticking we accept a proportionality of the
friction force to the force produced by the spring c2
F fr   sgnx2  f 2   sgnx2  c2 x2   sgnx2 sgnx2 c2 x2 .
where  is the dynamic coefficient of friction. The x2 deformation can be computed from
equality of two forces: the force in the spring c1 and the sum of the friction force and the
force produced by the spring c2
f  f1  f 2  F fr ,
i.e.
 c1 x  x2   c2 x2   sgnx 2  sgnx2 c2 x2 ,
and finally
c1x c1x
x2   ,
c1  c2 1   sgnx 2  sgnx2  c1  c2 1   
x1  x  x2 .
From the expression for x2 we obtain that at sliding the variables x and x2 have equal
signs at least for   1 .
At sticking the deformation x2 is constant and the resultant force produced by the element
is computed from the formula
f  f1  c2 x  x2  ,
Finally, the slip-stick transition occurs when the velocity x 2 changes its sign. The sign of the
velocity is estimated on the difference x2  x2 , where x 2 is the value of the coordinate at the
previous integration step. The stick-slip transition occurs when
F fr  f1  c2 x2   0 c2 x2 ,
where  0 is the static coefficient of friction.
A typical hysteresis (force vs. coordinate) as well as the dependence force vs. time is
shown in Fig. 4. The amplitude of vibrations decreases exponentially like a viscous damper,
but the element realizes a frequency independent damping.

19
Fig. 4. Force vs. coordinate x; force vs. time

Fig. 5. Stiffness and damping in series

Stiffness and damping in series is shown in Fig. 5. Such elements are used for modelling of
dampers, as a part of models of rubber, elastomer, etc.
Mathematical model of the element is obtained from equality of elastic and damping forces
and includes the following differential equation
x1  cx  x1  .
Thus, the element adds a new variable x1 to the model and the corresponding differential
equation. If the time constant T   c is small, the differential equation is stiff. It is worth to
note that if T is small, and the analyzed object motion is slow, the element is equivalent to a
simple linear damping with the same damping ratio.

3.2. Generalized linear force element


When modelling railway vehicles, force elements linearly depending on the relative
displacements and velocities of bodies are so often to be obtained whereas the influence of
their own dynamics is insignificant. A spring usually belongs to this type of elements. To
model such forces automatically UM uses the generalized linear and damping force elements.
An element of this type connects two different bodies i,j. Body i is considered the first, body j
- the second. The element connects fixed points of the bodies and is set by the coordinates of
the points of its attachment to the bodies, the stiffness 6x6-matrix C (the elastic element) or
the matrix of dissipation D (the dissipative element) of the same dimension and a stationary

20
value of the force and moment (only for the elastic element). Here the action of the element is
reduced to the appearance of the force G and the couple with the moment M, which being
reduced to the attachment point of the element to the second body in the SC fixed to the first
for the elastic element are as follows:

 Gi   G   r i 
F i   i  =  0 -CR, R   i 
M   0    
The analogous values for the dissipative element are
 vi 
F i  -DV , V   i 
 
Here  r is the vector of the relative displacement of the second attachment point and  is
that of the relative rotation of the second body (its deviation from the stationary position); v,
 - the corresponding linear and angular velocities, G0 is the stationary value of the force.

r Bj i
Oj

ij Bi

Oi
j A

Fig. 6. To the notion of linear force element

Let us consider the model of a linear elastic force element in more details (see Fig.6). Two
bodies are shown. The A and Bj points of the bodies are connected by the linear elastic force
element. The body-fixed systems of coordinates SCI, SCJ have Oi and Oj as origins. The
element attachment points are set by two vectors ii ,  jj , each of them should be set in the SC
of the corresponding body.
An additional point Bi is assigned to the first body and corresponds to the position of the
second end of the force element when  r vanishes ( r  0 ), in this case it coincides with
point Bj. The iij vector specifies its position in SCI. Point Bi is the origin of an additional
system of coordinate (SCBI), which axes are parallel to those of SCI. Another additional
SCBJ is fixed in body j, but its axes are arbitrary oriented relative to SCJ. A constant rotation
matrix ABj sets the position of SCBJ relative to SCJ. The relative rotation  is equal to zero
when the axes of SCBI and SCBJ are parallel.
Thus, the force is equal to the stationary value, and the moment vanishes when SCBI and
SCBJ coincide. Displacement of BJ relatively to Bi sets the  r vector, whereas the rotation of
SCВJ relatively to SCВI sets the  vector. Both  r and  are assumed to be small and
define the force and the moment values. The following formulas take place:
r  r  A  j  r  A i (11)
j 0j j i 0i ij

21
1 3 ~i 1 3 ~i
i   k i 0 0 j jB k
e A A A e i
  ek AiBeki
2 k 1 2 k 1
where eki are the bases vectors of SCI, and the direct cosine matrix AiB  Ai 0 A0 j A jB specifies
the orientation of SCBJ relative to SCBI. If   i is small then AiB  I 3 .
The following stiffness matrix is the most frequently used for modelling vertical springs:
 cs 0 0 0  cs H 2 0 
 
 0 cs 0 cs H 2 0 0
 0 0 cl 0 0 0
C  
 0 cs H 2 0 c 0 0
 c H 2 0 0 0 c 0
 s 
 
 0 0 0 0 0 c a 
Here c s , cl , c , ca are shear, longitudinal, bending and torsion stiffness of the spring, H is
the height of the spring. As a rule the length H for vertical springs takes into account the
static load.

3.3. Contact forces

O1
n

1
A
O2 2
C
O

Fig. 7. Point-plane contact geometry

Contact with friction is an important type of interaction of bodies. For example, the model
of a freight car with three-piece bogie includes more than 200 contacts. One of the possible
contact descriptions implemented in UM gives a points-plane contact force element, Fig. 7. A
set of body-fixed points is assigned to the first body (body 1, one point C is shown in Fig. 7).
A body-fixed plane is assigned to the second body (body 2). The plane is specified by a body
fixed point (A) and by an external normal n. The contact points, point A and normal n should
be given in SC of the corresponding bodies. Number of contact points is unlimited. Consider
a mathematical model of the contact for a separate point.

22
Fig. 8. Penetration of point into plane

If the distance  between the contact point and the plane is positive, the contact force is
equal to zero (no contact, Fig. 8 left). If <0 (Fig. 8 right) the force appears. The following
formula gives the value of:
  n  rc  nT rc ,
rc  r2  A0222  r1  A0111

Fig. 9. Contact forces

The contact force has two components: the normal force N directed along the normal n,
and the friction force F f situated in the contact plane (Fig. 9).
Fc  Nn  F f
A linear viscous-elastic model is used for the normal force
N  c    0 ,
The penetration velocity is computed from the formula
  n T v2  
~    v  
2 2 1
~   r   n T v
1 1 c c12
If the normal force is negative, the force is equal to zero (no adhesion). Parameters c,  are
constant stiffness and damping coefficients.
There exist two modes for the Coulomb friction low: sliding and sticking. The force model
in the sliding mode is

23
F f   fN vs vs ,
where v s is the sliding velocity (projection on the contact plane of velocity of point C relative
to body 2), f is the dynamic friction coefficient. The following formula is valid for the sliding
velocity:
c12 c12 3
 n  v  nnT v  I  nnT v ,
vs  vc12    c12
Here I 3  nnT is the projector on the contact plane.
The sliding-sticking transition occurs when the sliding velocity changes its direction to an
opposite one. The program realization compares the scalar product of the velocity on the
current integration step and the velocity on the previous step. If the product is negative, the
sticking occurs, and the friction force on the previous step Fg and the rg 0 =AD vector are
stored, where D is the projection of point C on the contact plane (Fig. 8).
The friction force model in the sticking mode looks like this:
 
F f  Fg  c rg  rg 0  vs
where rg is the current value of the vector AD. Thus, the point D is connected with the
contact plane by a linear viscous-elastic element at the sticking mode. The force Fg in the
formula ensures the continuity of the friction force at the sliding-sticking transition.
The sticking mode is over when the friction force exceeds its maximal value
Ff  f0 N ,
where f 0 is the static friction coefficient, f 0  f .

Note. The points-plane contact model can be easily generalized to a point - Z-surface
contact element, which is mathematically similar to the above one. In the case of this force
model, the contact plane is replaced by a surface, which is described by the functional
relation z  f x, y  .

Contact stiffness and damping are important parameters characterizing the contact. It is not
quite evident how their values should be chosen in simulation of a definite system. Let us
consider some methods for estimation numeric values of these parameters.
The real contact stiffness due to elastic deformations in contacts is usually very high (say,
11
10 N/m). It is clear, that for reasonable values of contacting body masses such stiffness
introduces a very high frequency in the model (about c m , where c is the stiffness, and а m
is the mass). This makes the model oscillatory stiff and increases CPU expenses considerably
due to decreasing the integration step size. In practice much less values the stiffness can be set
in the model. Consider the main reasons for that.
Applied theory of ordinary differential equations states (a strict proof can be obtained with
the help of the theory of singular degenerate equations) that if a system has two groups of
frequencies of different order, e.g. 1–10 Hz in the first group and 200–1000 Hz in the second
one, the high frequency processes do not affect practically the low frequency processes.
Moreover, changing parameters, which determine high frequency processes, do not lead to
considerable changes of the low frequency processes. That means, if the stiffness c is ‘large
enough’, its further increase does not affect analyzed processes, if they are ‘slow’. The
stiffness is ‘large enough’ if the introduced local frequencies are at least by order of
magnitude greater than the frequencies of analyzed object. The frequency k introduced by the
contact stiffness may be estimated by the formula k  c m , where m is the lower mass of

24
interacting bodies. Thus, if the user choose a ‘large enough’ local contact frequency
f  k 2 Hz, the corresponding contact stiffness can be computed according to the formula
c  4 2 f 2 m .
Consider an example. Let m  20 kg, f  200 Hz (i.e. the main frequencies of the object
are 1–10 Hz). Then c  3.16  107 N/m.
A stiffness coefficient computed according to this methodology should be verified and
corrected by the user. To do this, simulations should be run for different values of the stiffness
and plots of object performance variables should be compared. If 2÷10 times increasing the
coefficient does not affect the results, the stiffness is ‘large enough’. In this case it is
recommended to try decreasing its value. This may reduce CPU expenses. The boundary of
the parameter is its value when the plots of performances are changed.

Let us discuss now a methodology for choosing a dissipation coefficient  . If this


coefficient is ‘too small’, high frequency undamped oscillations may appear in the model,
which introduce large accelerations. Moreover, frequencies of these oscillations according to
the methodology above have nothing in common with the reality. If the damping is ‘too
large’, the equations of motion become stiff, and CPU expenses increase. A correct choice of
the damping coefficient is especially important for systems with unilateral constraints, i.e. for
systems with gaps and impacts, because the damping is responsible for a value of coefficient
of restitution. Finally, the value of the damping must correlate with the value of the contact
stiffness.
To get a justified value of the contact damping, a very important notion of damping ratio of
critical  and a damping factor   is used. To clarify this notion consider the equation of free
linear damped oscillations
mx  x  cx  0 ,
As it is well known, the solution of this equation depends on ratio of two parameters: the
frequency of free undamped oscillations k  c m and the damping coefficient n   2m  .
If n  k , the motion is a damped oscillation with the frequency k   k 2  n 2 . If n  k , an
aperiodic solution takes place. The boundary value n  k corresponds to a critical damping.
The damping ratio and the damping factor satisfy the following relations
n
 ,
k
n n
    [0, ],

k k 2  n2
Oscillations are undamped if     0 , and by   1 (    ) the damping is critical. The
parameters can be expressed in terms of each other,
 
   ,  
1  2 1  2
If   1, a decrement factor is computed as



D  e    e 1 .
2

The decrement factor determines an amplitude fall on a half of the period of oscillations. If
the damping is small,    and

25
D  e   .
Values of the decrement factor for different degrees of damping are presented in Tab. 1.

Table 1
Decrement factor versus damping ratio
 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
D 1 0.729 0.527 0.372 0.254 0.163 0.095 0.046 0.015 0.0015 0

Fig. 10. Shows solutions for different  .

Fig. 10. Damped vibrations for various damping ratios

If   0.1, the damping is often considered as a small one. The value   0,1 can be
recommended for an approximate in rubber-metal elements. The value   0,3 is
recommended for dampers of transport vehicles. If   1 , the system is strongly overdamped.
The above analysis clarifies the notions ‘too small’ or ‘too large’ damping coefficient. We
recommend the damping   0,1  0,4 for the contact forces. Thus, if you choose some values
for stiffness and damping ratio, the coefficient of damping is computed according to the
following formula
  2 mc .
For the above example c  3.16  107 N/m with the damping ratio   0,2 we get
  104 Ns/m.

4. WHEEL-RAIL CONTACT FORCES

Three main parts of the wheel-rail contact computation can be pointed out
 contact geometry, i.e. computing locations of contact points for a definite position
of a wheelset;
 computation of kinematic characteristics at contacts (creepages and spins, rail
velocities);
 computation of normal forces and creep forces at contacts according to actual
geometrical and kinematic parameters.

26
As it is shown below, these problems are not solved independently; a general iterative
procedure is necessary.
Dr. V.S.Kossov is a co-author of algorithms presented in Sect.4.1, 4.2 [6].

4.1. Algorithms for wheel-rail contact geometry


To make computation of contact points faster and more reliable, two main ideas are useful.
Firstly, if a rail profile does not change along the track, computation of contact points is
executed once for a given pair of rail and wheel profiles before the simulation start. UM
creates tables of contact coordinates for different relative positions of the wheel profile
relative to the rail profile (lateral displacement and rotation about the longitudinal direction).
By simulation the coordinates of contacts are interpolated with the help of these data.
Secondly, computation of contact points on the profiles is based on a procedure, which
computes a nearest point between two curves. These algorithms proved to be very fast and
reliable, and do not depend on smoothness of the curves.

Fig. 11. Relative position of profiles

Let us consider an arbitrary position of the wheel profile relative to system of coordinates
of the rail profile YrZr (SCR), Fig. 11. Introduce a new system of coordinates Y0Z0 (SCR0),
which origin coincides with the origin of SCR. The Z0-axis is perpendicular to the track plane
as if the inclination angle  r 0 is zero. The SCR YrZr is inclined on the angle  r 0 relative to
SCR0. Position of SCW (system of coordinates of the wheel profile) YwZw relative to SCR0 is
defined by coordinates of the origin y ,  z , and the angle  . Coordinates in Fig. 11 are
positive. All angles are considered to be small.
It is necessary to find a pair of points on the profiles, which have the same Y-coordinate in
SCR0 and the minimal difference in Z-coordinate in SCR0
min  z  minz w0  z r 0 
Obviously, the solution does not depend on  z .
Let us start with the algorithm for computing the value
z  z w0  zr 0
for two points having the same Y-coordinates in SCR0. The coordinate yr of a point on the
rail profile in SCR (YrZr) has a given value. We should compute the coordinates y w , z w of the
corresponding point on the wheel profile in SCW as well as z .

27
Firstly, here are the coordinates of the point on the rail profile in SCR0
y r 0  y r  zr  y r  r 0 ,
zr 0  zr  y r  r 0 ,
where zr is the coordinate of the point in SCR.
Then a point on the wheel profile with coordinates y w and z w in SCW should be found,
which has the following abscissa in SCR0: y w0  y r 0  y , Fig. 11. Since the angle  is
small, the following nonlinear relation takes place:
y w0  y w  z w  y w 
or
y w  y w0  z w  y w  .
To solve this nonlinear equation relative to unknown value y w , the direct iterations
0
yw  y w0 ,
i 1
yw  i
 y w0  z w y w , i  0,1,2,...
could be used.
As it is known, direct iterations converge if the condition
dzw
  1
dyw
takes place. Since   1 , this condition is always valid for real profiles.
Finally, the coordinate z w 0 is computed from the formula
z w 0   z  y w   z w ,
and the value of minimized function z  yr  is evaluated.
In this way the algorithm allows us to compute a pair of points and the distance between
them. The main advantage of the algorithms is its simplicity, reliability and independence on
smoothness of profiles.

Now consider computation of tables of contact points. As it is already mentioned,


computation of tables of contact geometry information is executed before the start of the
simulation, if the rail profile does not change along the track. This information is obtained in
dependence on the wheel profile position relative to the rail profile. UM allows two kinds of
pairs of profiles: profiles with one-point contact only and profiles with a possible two-point
contact (Fig. 12).

Fig. 12. Pairs of profiles allowing two-point contact (left) and one-point contact only (right)

In a one-point contact mode the contact point position depends on the lateral shift y and
the angle  , Fig. 11. Coordinates of the contact point correspond to the profile point with
the minimal distance in the vertical direction. Thus, the table contains coordinates of the

28
contact point in SCR and SCW for a discrete set of variables yi ,  j , i=1..Ny, j=1…N
with a proper step size of discretization.
Computation of one-point contact is discussed later. Here we consider some features of
computing a two-point contact.

Fig. 13. Types of contact for a pair of profiles allowing two-point contact: one-point
contact, two-point contact and creeping up

Fix the value  and set the maximal lateral displacement y (Fig. 13, left). In this
position of the profile we have obviously a one-point contact. Now decrease y without
changing  , i.e. shift the wheel profile to the left. There exists some critical value of the
lateral coordinate y  for which a two-point contact occurs, if the profiles allow it (Fig. 13,
center). Further decreasing y does not change coordinates of contact points, the rail profile
moves to the left together with the wheel profile due to elastic deformations. This mode of the
two-point contact can disappear in two different ways. Firstly, the wheel will move to the
right and the flange contact disappears (Fig. 13, left). Secondly, the contact on the running
surface disappears (the corresponding normal force becomes zero) and the wheel goes to the
creeping up state (Fig. 13, right). This mental experiment is the base of an algorithm for
computing the two-point contact. Let us start discussions of the algorithm.
Choose intervals for possible values of the  , y coordinates with a definite reserve,
  [  min ,  max ] , y  [ y min , y max ]
and introduce a homogeneous discretization of the intervals, N and Ny are numbers of
subintervals. Now for each fixed value  j , j=1.. N+1 compute coordinates of contact point
on the running surface successively decreasing the lateral shift yi , i=Ny+1,…,1. If the
profiles allow the two-point contact, the position of the contact point for some yi will
change by a large enough jump. This fact means that the flange contact occurs inside the latest
change of y . Denote this value of y as y  . Two conditions formalize the notion of a
‘large enough jump’ and give a criterion of passing through the two-point contact:
y r ,i  y r ,i 1   y , nr , y / nr , z  n
where yr,i , yr,i 1 are the successive coordinates of the contact point on the rail (in SCR),
nr, y , nr, z are the projections on the normal to the rail profile at the contact point at the latest
step,  y , n are some empirical criterion numbers, e.g.  y  20mm, n  0.5 . Thus, the two-
point contact is considered to be found, if the contact coordinate on the rail profile changes by
a large jump, and the new coordinate of the contact point lies on the side of the rail.
After confirmation of the fact, that the state of the two-point contact is crossed, the interval
[ yr,i 1, yr,i ] is discretized on small subintervals (about 0.0001 mm), and the critical value
y j  is defined more exactly.
To model the process of creeping up the wheel, the computation is continued
for y  y j . Then the computations are repeated for other values of the angle  j . The

29
contact tables contain coordinates of contact points for various values  j , yi as well as the
y j 
critical values of the lateral shift for each value of the angle  j .

Computation of two-point contact geometry assumes evaluation of a wheel ‘overswing’ xc,


i.e. the longitudinal coordinate of the flange contact for nonzero value of the angle of attack .
UM uses an approximate analytic expression for the value of this parameter. To get it,
consider a simplified geometrical model of the flange-rail contact, Fig. 14. In this model a
plane with the same normal replaces the side surface of the rail near the contact, and the
flange is replaced by a circle with a radius equal to the wheel radius at the contact.

Fig. 14. Simplified model of the flange contact for computing an overswing

Let  be the angle of attack,  is the angle defining the flange contact with the overswing.
The  value can be found from the condition that the tangent to the flange circle at contact is
perpendicular to the plane normal n  (0,nry , nrz )T . The unit tangent vector is set by the
following expression
  (rw cos,0, rw sin )T ,
where rw is the wheel radius at the flange contact.
Introducing the direct cosine matrix corresponding to the angle of attack,
 cos   sin  0 
 
A01   sin  cos  0  ,
 0 1 
 0
yields the following orthogonal condition
nT A01  0 ,
or, taking into account that angles and are small,
 nry   nrz   0 .
For small angles the overswing can be found as xc  rw , which results in the final
formula
xc  k, k  rwnry / nrz .

30
4.2. Rail-wheel contact forces
A rail is considered in UM as a massless force element. This means, both stiffness and
damping of the rail is taken into account, but not the inertia properties. Generalized
coordinates are not introduced for the rail and its lateral and vertical deformations must be
computed from the equilibrium equations.
The following assumptions take place:
- rail deformations for different wheelsets are independent and can be computed
separately;
- deformations of the left and right rails are independent;
- rail deformations include independent lateral y r , vertical z r deformations (Fig.
15), which are parallel to the corresponding system of coordinates of the track;
- rail roll is not considered.
- the rail as a linear force element both in the lateral and vertical directions; lateral
dissipation is taken into account for two-point contact mode only.

yr
zr

Fig. 15. Rail as massless force element

Let c y , c z be the lateral and the vertical stiffness of the rail, d y , d z be the corresponding
damping constants. Forces acting on the rail due to the deformations are the following:
R y  c y yr  d y y r , (12)
Rz  cz zr  d z zr .
Because the rail has no mass, these forces must be balanced by contact forces acting on the
rail from the wheel. The contact forces acting on the wheel for one- and two-point contacts
are shown in Fig. 16. Longitudinal forces are not shown in this figure.

31
N1 
1 N1

1
F1
F1 F2

2

N2

Fig. 16. Forces acting on wheel at one- and two-point contacts

Equilibrium equations for one-point contact written in SC of the track are


R y  F1 cos1  N1 sin 1  0, (13)
Rz  N1 cos1  F1 sin 1  0.
Analogous equations are valid for a two-point contact
R y  F1 cos1  N1 sin 1  F2 cos2  N 2 sin 2  0, (14)
Rz  N1 cos1  F1 sin 1  N 2 cos2  F2 sin 2  0.
Here 1, 2 are the angles between the normal to the rail at contact and the axis perpendicular
to the track.
Eq. (13) for a one-point contact and (14) for a two-point contact are complicated systems
of nonlinear algebraic equations relative to unknown deformations of the rail and normal
reactions N1, N 2 . Consider the main ideas for their solving not going into details.

1. Position of the wheel as well as the rail shift due to irregularities, gauge widening and cant
are known by the computing the contact problem; the only unknown are deformations
yr , zr and their time derivatives. Note that the vertical deformation z r is not an
independent variable because when the lateral y r deformation is known, the value of
z r can be obtained from the geometry of the contact. This value is equal to the value
z , Sect. 4.1.
2. Calculation of the rail deformations and contact forces is an iterative process. Iterations
include two cycles: internal end external. The internal iterations are used for solving
equations (13), (14) for known values of tangential forces (lateral creep forces). External
iterations calculate creep forces. Values of creep forces on the previous integration step
are used as initial approximations. Thus, the contact computation looks like this: internal
cycle of iterations computes the lateral deformation of the rail and the normal force/forces
in contact/contacts, the lateral creep forces are taken from the previous step. After that the
new values of creep forces are computed. When the new values differ from the previous
ones more than an error tolerance, the external iterations start, and equations (13), (14) are
solved for corrected values of creep forces.

32
Fig. 17. Creepages, spin and creep forces

Modern models of tangential forces in a wheel-rail contact are based on nonlinear


dependencies of the general form:
Fx  Fx ( N ,  x ,  y , , p), Fy  Fy ( N ,  x ,  y , , p)
Here the following notations are used:
Fx , Fy are the longitudinal and lateral creep forces lying in the tangential plane of the rail;
N is the normal force in the contact;
 x ,  y are the longitudinal and lateral creepages;
 is the spin;
p is a set of geometrical parameters characterizing rail and wheel profiles, e.g. curvatures
of contact surfaces in the case of the FASTSIM algorithm.
As it is known, the creepages and the spin satisfy the following relations:
 x  v x / v0 ,  y  v y / v0 ,   n / v0 ,
where v x , v y are the corresponding component of sliding velocity at the contact point on the
wheel relative to the rail; v0 is the longitudinal velocity of the wheelset; n is the projection
of the wheel angular velocity on the normal to the rail at the contact point.

The following algorithms for computing the creep forces are available with UM.
1. Mueller’s method is the simplest one for computation of the creep forces according to
the following analytic expressions [7]
   2x   2y , P = 0.001N , k c  P( 235  P( 2.4  0 .01P ) )
1000k c
Fxy   ,
1 + (kc  / f / P)m 1m
Fx    x Fxy , Fy    y Fxy
Thus, the model is very simple. In particular, forces do not depend on the spin. If the two-
point contact is presented, forces at the flange contact are computed as simple friction forces
in the sliding mode.
Some advantage of the algorithm consists in its simplicity. The disadvantage is its lower
accuracy especially for a one-point flange contact where the spin is not small.

33
2. FASTSIM is the known algorithm for creep forces by Kalker [8]. It takes into account
both the spin and the geometry of the contact surfaces. The algorithm is used for calculation
of creep forces both for one- and two-point contacts.
dx y
dy

Rolling direction

Fig. 18. Discretization of the contact ellipse

According to this data FASTSIM solves a system of differential equations (in the adhesion
area of the contact patch) or a system of differential-algebraic equations (in the sliding area of
the contact patch) relative to tangential stresses. For this purpose the contact ellipse is divided
into a number of narrow slices of the same width. In turn, each slice is divided into n elements
of equal length within one slice (Fig. 18).
To compute the creep forces and to obtain adhesion and sliding areas of the patch,
FASTSIM solves the above equations for each of the slice successively. In fact, the
discretization on elements gives the constant step size for numeric solving the differential
equations by the explicit Euler method. Thus, CPU expenses are of order m n operations,
and depend on the discretization level.

3. FASTSIM_A is a semi-analytic modification of the classical FASTSIM algorithm. For a


slice it uses an exact solution of the FASTSIM governing differential equations in the
adhesion area of the contact patch and an approximate analytic solution for differential-
algebraic equations in the sliding area. The solution was implemented in UM as FASTSIM_A
(FASTSIM - Analytic) procedure. The number of operation for computing creep forces is
proportional to the number of slices m.

4. Modified Kik and Piotrowsky algorithm for calculation of non-elliptic multipoint


contacts [9, 10].

4.3. Semi-analytic FastSim algorithm


J.J. Kalker proposed the FastSim algorithm in 70th as a fast solver of the tangential contact
problem. Now this algorithm and its modifications for non-Hertzian contacts are the most
often used methods for computation of creep forces in many railway vehicle dynamics
simulation software. Here we consider some modifications of the algorithm, which allow us
to make the computation of creep forces several times faster in comparison with the
traditional computation scheme.
The FastSim algorithm is based on a simplified theory of rolling contact [11]. The main
assumption of this theory consists in introduction of a contact layer, for which the linear

34
relation between the tangential displacements u = {ux, uy}T and the tangential stresses
p = {px, py}T takes place
u  Lp, (15)
where L is the constant compliance.
At a stationary rolling, the relative sliding velocity s = {sx, sy}T within the contact patch
satisfies the equations
s x   x  y  u,
s y   y  x  u,
where  x , y are the longitudinal and lateral creepages, and  is the spin, '   x .
Taking into account assumption (15), the equations for the tangential stresses can be
obtained
Lp x   x  s x , (16)
Lp y   y  s y ,
 x   x   y,  y   y  x.
In the adhesion area of the contact patch the sliding velocity vanishes, and the equations
for tangential stresses become simpler
Lpx   x , (17)
Lpy   y .
Within the adhesion area the Coulomb’s law takes place
p x2  p 2y  fpz , (18)

where p z is the normal pressure.


Numeric integration of Eq. (17) starts from the leading edge along a slice, and the
Coulomb condition (18) is verified for each the element. In case of violation of the inequality,
the tangential stresses are corrected according to the formulas
px (19)
p x : fpz ,
px  p y
2 2

py
p y : fpz ,
p x2  p 2y
where the normal stress distribution is set by the paraboloid equation
2N  x2 y 2 
p z ( x, y )  1   ,
 a b  a 2 b2 
Here a, b are the semiaxes of the Hertzian contact ellipse.
Summarizing the tangential forces over the whole set of elements yields the required
tangential (creep) forces.

35
sliding area y adhesion area

dy xs(y)
x

xl(y)
separatrix

Fig. 19. Contact patch

In fact, FastSim solves ordinary differential equations (17) within each of the slices with a
finite difference formula similar to the explicit Euler method, which has a low accuracy and
requires a sufficiently small step size (i.e. a fine discretization of the contact patch). Though
the algorithm is fast, it requires a considerable number of floating point operations, and
development of alternative methods for solving equations (17), (18), (19) is desirable.

Eq. (17) has a quite evident explicit analytic solution for fixed y at the adhesion area.
Taking into account, that the tangential stresses are zeroes at the leading edge, we obtain
1 x (   y )( x  xl ) (20)
p x   x l (  x   y ) dx   x ,
L L
1 xl  ( x  x)
p y   x ( y  x ) dx  ( y  ( xl  x )) l ,
L 2 L
where xl  xl ( y )  a 1  y 2 b 2 is the abscissa of the entering edge for the given y, Fig. 19.
To compute the tangential forces within the adhesion area of a slice, a boundary of this
area is necessary. It can be obtained from the condition
  2
p x2  p 2y  fpz 0 xl2  x 2 ,  (21)

 
where p z 0  2 N  a 3 b .
Substitution of relations (20) in Eq. (21) gives the quadratic equation
 
x  xl 2  f 2 pzo2 L2       y x  xl    x  y 2  2y  0
2

 4 
with the following solution for the boundary between the adhesion and sliding regions
x s ( y )   xl  
1
2a
b  b 2  4ac( y ) . 
Here
2
a  f 2 p zo
2 2
L  , b    y , c( y )   x   y 2   y2 .
4
Now the tangential forces at the adhesion area of the slice with the width dy is computed
according to Eq. (17)
 xl 
   p x dx dy   x   y  xl  x s 2
dy
dFxa ( y ) ,
x  2L
 s 

36
 xl    x3  2x    dy
dFya ( y )    p y dx dy    y ( xl  x s ) 2    s  xl2  l  x s    .
x      2L
 s    3 3  
This exact solution does not require discretization of the slice, and usually can be computed
much faster then in the FastSim algorithm.
Consider the sliding area. Here the sliding velocity is directed opposite the vector of the
tangential stresses, so that
s x  p x , s y  p y ,

where λ is the coefficient of proportionality, and Eqs. (16) are transformed to the following
equations:
Lp x   x  p x , (22)
Lp y   y  p y .
Joining the Coulomb equation
p x2  p 2y  p x   fpz 0 ( xl2  x 2 ). (23)

leads to the closure of the equations. In fact, we have differential-algebraic equations of index
2, where Eq. (23) is the algebraic constraint equation, and  is the Lagrangian multiplier.
We did not find exact solution of these equations, but a good approximation of it can be
proposed. Consider the following change of variables:
p x  p e x , p y  p e y ,
where ex , e y are the new variables. They are components of the unit vector directed along the
stress vector or opposite the sliding. The differential-algebraic equations in the new variables
are
Lp  ex   x  p  p e x , (24)
Lp  ey   y  p  p e y ,
ex2  e 2y  1.
To transform differential-algebraic equations (24) into differential equations, i.e. to
eliminate the Lagrangian multiplier, the constraint equation should be differentiated with
respect to x
ex ex  ey e y  0 ,
and expressions for derivatives ex , ey are substituted in this identity
p  p   x ex   y e y .
So, the new variables satisfy the following ordinary differential equations
Lp  ex    x e y   y ex e y , (25)
Lp  ey    x e y   y ex ex .
Now we consider the approximate solution of Eqs.(25):
~  x  y (26)
e x0 x  
px
 ,
~
p x2  ~
p 2y 
 x   y 2   y  ( xl  x ) 2 2 
~
py  y   ( xl  x ) 2
e y x  
0
 .
~
p x2  ~
p 2y 
 x   y 2   y   ( xl  x ) 2 2 
Functions ~ px ~
p y are obtained from solution (20) for stresses, which is valid at the adhesion
regions of the slice. In fact, approximate solution (26) corresponds to the assumption that the

37
direction of the stress vector in the sliding area is close to that obtained from the equation
valid for the adhesion area.
Let us prove some important properties of the approximate solution, which allow us to
consider it as a good approximation of the exact one.

1. Relations (26) satisfy Eqs. (25) for   0 , i.e. our approximation is exact for zero spin.
Really, if   0 , formulas (26) give the constant values
x y
ex0   const, e0y   const
x   y
2 2
 x  2y
2

satisfying Eqs. (25) because of the identity

  yx   yx
  x e 0y   y e x0   0.
2x  2y

2. Functions (26) satisfy the following conditions at the begin of the sliding area:
e x0 e x
e x0 x s   e x  x s ,  ,
x xs
x xs

e 0y e y
e 0y xs   e y xs ,  ,
x x xs
xs
i.e. at the begin point of the sliding interval of the slice the approximation functions and their
first derivatives are equal to the exact values. The first statement is trivial because of the
continuity. To prove the second statement, consider differential equations, which are satisfied
by functions (26)
e x0

~
p x

~
px ~
p x  ~
py ~
p y ~
p 
  x e 0y   y e x0 e 0y,

x  
~ x
p x2  ~
3
p 2y ~ 2 ~
px  p y 2 2 ~
L px  p y
2 ~ 22

e 0y ~
p y
 ~2 ~2 
~
px ~
p x  ~
py ~
p y ~
py  
  x e 0y   y e x0 e x0  
x px  p y ~ 2 ~
px  p y
3
2 2
 L ~ 
p x2  ~
p 2y
or
e 0
 
(27)
L ~ p x2  ~
p 2y x    x e 0y   y e x0 e 0y ,
x
e 0y
L ~ p x2  ~
p 2y
x

    x e 0y   y e x0 e x0 . 
Comparing Eq. (27) with Eq. (25) and taking into account the relation (cf. Eq. (21))
~
p2  ~ p 2  p x 
x y  s
xs
we prove the statement.
The above statements show that the approximation is very good at least for small values of
the spin. Now consider other properties of the solution, which show restrictions of the
approximation.

38
3. Functions (26) satisfy the following conditions at the trailing edge of the contact patch if
0:
ex0  xl   ex  xl , e0y  xl   e y  xl  .
To obtain the exact value of the variable at the trailing edge, consider the transformation
ex  cos , e y  sin  .
The first equation in system (25) becomes
 Lp sin     x sin    y cos sin 
or
 2x   2y (28)
   sin   x 
Lp 
where
 y  x
sin   , cos  .
 2x   2y  2x   2y
If we take into account that p  xl   0 , the following relations can be stated:

e x  xl   cos  x
  e x0
 x   x l
x   y
2 2
 x   xl
   xx  
  y
l

 e y  xl   sin  x   x l   e 0y
 x   y
2 2
 x   xl

The proven statement shows that the approximate solution for stresses satisfies the following
conditions
p x0  xl   p x  xl   0, p 0y  xl   p y  xl   0,

p x0 p x p 0y p y
 ,  .
x x   xl
x x   xl x x x   xl
x   xl

4. Approximate solution cannot give adhesion regions inside the contact patch, which does
not start at the leading edge and can be found by the FastSim algorithm for large spins.

a) b)

39
c)

Fig. 20. Tangential stresses in a strip: a) φ = 0,1; b) φ = 0,3; c) φ = 0,5

Comparison of the exact and approximate solutions for tangential stresses in a strip for
different spin values is given in
Fig. 20. The following parameter values are chosen for this example: ξx=0.005, ξy=-0.002,
y=0, a= 8.428 mm, b= 3.041 mm, N=21.1 kN. In the considered case the approximation for
the longitudinal stress is much better than for the lateral one.
Now we can obtain an expression for approximate values of tangential forces within the
sliding area of a strip
 xs 
dFxs 0 ( y )    p x  ex0  x  dx dy,
x 
 l 
 xs 
dFys 0 ( y )    p x  e 0y  x  dx dy.
x 
 l 
Values of these integrals can be easily computed as explicit expressions, and we obtain the
analytic approximation for tangential forces in a strip
dFx0  y   dFxa  dFxs 0 , dFy0  y   dFya  dFys 0 .
Summation over the set of strips gives the total creep forces Fx0 , Fy0 according to the analytic
algorithm FastSim_A.

Fx, kN
30

25 
20 

15 

10

0
x
0 0.005 0.01 0.015 0.02
a)

40
Fy, kN
30

25 
20 
 
15

10

0
0 0.005 0.01 0.015 y 0.02

b)
F, kN
30

25 
20 
 
15

10

0
0 0.005 0.01 0.015
 0.02

c)
Fig. 21. Comparison of total creep forces versus creepages for FastSim (solid lines) and
FastSim_A (markers)

Comparison of total creep forces computed with FastSim and FastSim_A versus creepages
for different values of the spin is given in Fig. 21. Fig. 21a corresponds to the case
 x  0,  y  0 , Fig. 21b -  y  0,  x  0 , and Fig. 21c -  x   y ,   2x  2y . The force F
in Fig. 21c is the module of the creep force. All results in Fig. 21 are computed for
N  100kN .
These results show that the use of the FastSim_A procedure gives quite coincidence of
results with the FastSim algorithm for profiles allowing two-point contacts, because at the
first contact point the spin is usually small, and at the flange contact a pure sliding with large
creepages takes place. It is surprising that even for curving in case of one-point profiles the
comparison results are very close as well. Fig. 22 shows simulation results for a freight car -
hopper running in a curve with R = 300 m and 14 cm cant. Irregularities correspond to the
track of a satisfactory condition. Simulation was executed for the R65 rail profile and the
DMetI wheel profile (Fig. 12, right). This pair of profiles allows one-point contact only.

41
Fig. 22. Comparison of longitudinal and lateral creep forces for the outer wheel of the
leading wheelset: Fx , Fy – FastSim; Fx0 , Fy0 –FastSim_A

Tf 12
/Ta
10
8
6
4
2
0
0 10 20 30 40 50 60 70 80 90 100
n

Fig. 23. Comparison of CPU expenses for FastSim (Tf) and FastSim_A (Ta) versus number
of elements in a strip

As it was expected the FastSim_A algorithm requires less floating-point operations than
FastSim (Fig. 23). Even if the analytic solution is applied for the adhesion area of the contact
patch only (the FastSim_SA algorithm), a reduction of CPU expenses takes place. Fig. 24
shows a comparison of the FastSim and FastSim_SA algorithms for a typical contact on the
running surface. What about effectiveness of the FastSim_A algorithm, it makes about 90000
computations of the creep forces per second on Celeron 1,32 GHz.

42
5
Tf /Tsa

1
0 20 40 60 80 100
n

Fig. 24. Comparison of CPU expenses for FastSim (Tf) and FastSim_SA (Ta) versus
number of elements in a strip

Finally, consider a way to improvement of the approximate solution in the sliding area,
which is similar to the next iteration of the Picard’s method of solving differential equations.
The improved solution e1x , e1y satisfies the differential equations
e1y
Lp 
x
 
   x e 0y   y e x0 e 0y ,

e1y
Lp 
x

    x e 0y   y e x0 e x0 , 
where the previous approximation is substituted in the right hand side of the equations. This
approximation has the exact second derivative at the beginning of sliding, and allows about
double reduction of the error in creep forces. Comparison of FastSim solution and both
approximations is shown in Fig. 25 in the case corresponding to Fig. 22 for   0.5 .

Fig. 25. Tangential stresses in a strip: F – FastSim, Fa0 – the first variant of approximation,
Fa1 – the improved approximation

43
5. SUBSYSTEM TECHNIQUE

Bushing Bushing
shaft - wheelset motor assembly-shaft Anchor

Reducer
Cardan shaft

Fig. 26. “Wheel-motor assembly” subsystem

Creation of computer models of complex technical systems such as a railway vehicle or a


tractive connection of several vehicles is a very time consuming procedure. Its considerable
simplification and acceleration can be achieved with the help of subsystem technique [12].
Let us consider the model of the electric locomotive EP200 to illustrate the subsystem
method. The model creation consists of four stages. Firstly, a “Wheel-motor assembly” is
created as a separate multi-body system, Fig.4. This assembly as a subsystem contains a
wheelset (the standard UM subsystem), a motor, a reducer, a hollow cardan shaft, and force
elements: a tractive characteristic of the motor, bushings and gearing forces, Fig. 26.

Fig. 27. Primary connection of wheelsets

Secondly, the new model “Wheelset Group” is created. This subsystem includes the
“Wheel-motor assembly” two times. In addition, two bodies (balance beams) should be added
as well as force elements: springs and dampers of the primary suspension. Force elements
connecting bodies of the subsystem with the frame, which are not included in the subsystem,
are denoted as external, and visually marked in Fig. 27 by the letters L (linear element) and B
(bipolar element).

44
Fig. 28. Bogie

Then the model “Bogie” is created. Model of the bogie includes the “Wheelset Group”
subsystem two times, adds one body (frame) as well as external force elements of the
secondary suspension, Fig. 28. External elements of the two-wheelset groups are connected
with the frame by assignment of attachment points (connection points) on the frame.

Fig. 29. Electric locomotive EP200

Finally, the locomotive model is developed. This model includes two “Bogie” subsystems,
and a car body. Connection points on the car body are assigned to the external force elements
of the secondary suspension.
In fact the model of the locomotive EP200 is more complicated because it includes the full
mathematical description of electric part developed in co-operation with South-Russian
Technical University [13].
This example concerns the representation of a vehicle by a tree of included subsystems. In
this case the head object (model of the locomotive) is the owner of all elements described in
subsystems. Equations of motion of included subsystems are generated as parts of equations
of motion of the head object.

45
Fig. 30. Model of a train

External subsystems, unlike the included ones, do not belong to the head object. It contains
references to the external subsystems only. In addition, it can include kinematic and force
connections between separate subsystems. Equations of motion of the external subsystems are
compiled as dll and linked to the UM before start the simulation process.
A typical example of an object, which contains several external subsystems, is a train, Fig.
30. Each vehicle in the train model is presented by an external subsystem. Usually different
vehicles are referred to a one and the same external object.

Fig. 31. Model of a tractive connection of freight cars

The model of the connection of three freight cars is build by three references to a single
external object “Freight car”. The head object assigns connections between automatic
couplers, which are described in the car model and contains external contact force elements.
All changes in the “Freight car” model are automatically accepted by all models, which refer
to the “Freight car” as an external subsystem.
Subsystem technique gives some additional useful tools. Firstly, all elements, which are
defined in included and external subsystems, obtain long names, which contain names of
subsystems, for instance, “Bogie1.WMB2.Damper1Left”. As a result, elements are easy
identified. Secondly, identifiers of the same name for different subsystems generated by
identical objects can have different values. Therefore, all three cars in the model of the
tractive connection can differ in inertia and geometric parameters, though the single model of
the freight car produces them. As a result, one car in the connection can be empty, another
one - loaded, one could be a hopper, another one – an open wagon etc. The subsystem
technique is the tool for creation of bases of typical subassemblies.

46
Screens

Fig. 32. Hybrid model of ballast cleaner SHOM 1200

The subsystem technique is used for modelling objects containing subsystems of different
nature, e.g. a rigid body system and a flexible body [14].
Finally, modelling a train or a tractive connection of vehicles as sets of external
subsystems gives an ideal structuring of the models for parallel computation using local
computer nets. Really, equations of motion of such the models have the following form:
M i qi  qi  ki qi , qi   Qi qi , qi , t    Qij qi , qi , q j , q j , t   GiT qi  i ,
j

g i qi   0,
where i=1,…,N is the index of the subsystem. It is clear that almost all computations for a
subsystem including evaluation of equations of motion and their solving can be executed on
separate processors. Communications between processors are necessary for obtaining external
forces acting between the neighbor vehicles, which define generalized forces Qij , as well as
for analysis of results of a single simulation step for the purpose of acceptance or rejection of
the step and choice the size of the next step. Thus, data of a minimal size should be sent on
each step of the integration.

6. NUMERIC METHODS

From the point of view of reduction of CPU expenses, efficiency of simulation process
depends mainly on the solver, which is used for integration of differential-algebraic equations
of motion (6), (7). The implicit Park method, which is a half-sum of the 2nd and 3rd order
backward differentiation formulas, is proved to be one of the fastest and reliable solvers in
simulation of rail vehicles [15, 16]. The implementation of the Park method in UM includes
predictor formulas for coordinates (the interpolation polynomial)
qip1  4qi  6qi 1  4qi  2  qi  3
and for velocities (the Park formula)
vip1  qip1 
1
6h

10qip1  15qi  6qi 1  qi  2 

47
Here i is the index of the integration step, and h is the step size. The corrector relations are
qi 1  qip1  qi 1 , (29)

vi 1  vip1  qi 1  ,   0.6h,

ai 1  qi 1 
1
10vi 1  15vi  6vi 1  vi  2   aip1  qi 1  2 .
6h
Substitution of the corrector formulas (29) in Eq. (7) leads to nonlinear algebraic equations
relative to unknown qi 1 ,  i 1 (indices are omitted)
 
M ( q p  q) 2 a p  q  (30)

 2 Q q p  q, v p  q   k q p  q, v p  q   G T q p  q  ,
g ( q p  q)  0.
Equations of motion of rail vehicles are often stiff, i.e. a Jacobian matrix (JM) of Eq.(30) is
necessary. Evaluation of the JM by finite differences is a very CPU time-consuming
operation, even if it can be computed not on every simulation step. An alternative method is
proposed in [16, 17]. The method consists in evaluation of approximate Jacobian matrix based
on simplified analytic expressions for local JM of stiff forces.
To introduce an analytic expression for JM of stiff forces, consider a variation of position
and velocity of body i. Let i be the vector of a small rotation
~ i  A0i Ai 0 ,
then the following formula takes place:
 r 
Ri   i    i q .
 i 
The variation of velocities is
 v  1  i q  
δVi   i    i q   i q    i  q.
 i   qT 
If we take into account that the integration step is usually small ( h ~ 1  0.3 ms), the following
estimation is valid:
 i q 
 i   ~ Vi h ,
qT
and we obtain the approximate relation for the variation of velocities
1
δVi   i q.

Now consider an applied force F acting on body j from body i. Let the force depend on the
positions and velocities of bodies i, j as
F  F ri , rj , vi , v j , A0i , A0 j , i ,  j  .
The corresponding force acting on body i is
 I3 0
  ~ ~  F  CijT F

 rj  ri I 3 

According to Eq.(8), the contribution of the force in the generalized force is the following:

   
QF   Tj   Ti CijT F    SˆkT   SˆmT  C0T j F
 k J  j  mJ i  
The variation of this expression includes two terms

48
  
QF    Tj  Ti CijT F   Tj  Ti CijT F 
The next simplifying assumption consists in neglecting the first term in this expression, i.e.
in neglecting the force variation due to change of its direction in comparison with the force
variation due to change of its dependence on relative displacement and velocities of bodies.

QF  Tj  Ti CijT F
Now, the variation of the force should be evaluated.
F  FRi Ri  FVi Vi  FRj R j  FVj V j  FRi  FVi   i  FRj  FVj  j
   
Q    SˆkT   SˆmT  C0T j  FRi  FVi  C0 j  SˆkT  FRj  FVj  C0i  Sˆm  
 k J  j  mJ i    k J  j  mJ i  


 
  
   SˆkT   SˆmT   FRi  FVi   SˆuT  FRj  FVj   Sˆv 

 k J  j  mJ i   uJ  j  vJ i  

This formula gives the approximate expression for contribution of the force in the JM
 T 

 
    (31)
J F    SˆkT   Sˆm   FRi  FVi   Sˆu  FRj 
 FVj   Sˆv 
 k J  j  
mJ i    uJ  j 

vJ i  

We introduced the important local JM of the force
FRi  C0T j FRiC0i , FVi  C0T j FVi C0i , FRj  C0T j FRjC0 j , FVj  C0T j FVj C0 j
If the force depends on relative position and velocities of bodies i, j, the following relations
are valid:
FRi   FRj , FVi   FVj .
To simplify transformations consider a force depending on relative positions only
 
F i  F i riji , Aij . Let us derive the expression for the variation of the force.
A
F   0i
0  i  A0i 0  i  A0i 0  i
 F   F    
F  Fr A0i  riji  F  j  i  
 0 A0i   0 A0i   0 A0i 
 Fr ~
rij i  rj  ri   F  j  i   Fr F R j  Fr F Cij Ri
Therefore
FRi  C0T j Fr F Cij C0i  C0T j Fr F C0 j   FRj
and the expression for the JM (31) becomes more compact

J F    SˆkT   Sˆm
 k J  j 
T 


 FRi  

 FVi    Sˆk   Sˆm

 


 mJ i    k J  j  mJ i  
It should be noted that in many cases the relation FRj  FVj with a scalar multiplier 
takes place. Moreover, often both the approximate local and the global JM of force elements
are symmetric.

Now consider derivation of approximate local JM for the force elements described in
Sect.3.
1. Bipolar force element. To simplify transformations consider a force, which depends on
the element length x only, Eq. (9). The variation of the length is

49
rAB
   
T
rAB ~   r  
~  
r   rAB
T
rAB   eT r j  A0 j  jj  ri  A0i ii  eT r j   j j i i i
r

 
 I3  ~ rj   ~ ~ ~
j  1  I 3  ri  i  1 
 e T 
0   C0 j R j   C0i Ri .
 0 I 3   0 I 3  
T ~ ~ T ~ ~ T~
Note that e ( ri  i )  e ( r j   j )  e rB , therefore

       
~
 I 3  rB  1
r  e T
0   C0 j R j  C0i1Ri  eT 0 C0 B C01j R j  C0i1Ri .
0 I3 
Now we can consider the variation of the force
 G   I 3 0  e  I 0  e 
F   ~   f r  ~    f r ~3


  j I 3  0 r
  jG    j I 3  0    
and the final expression for the corresponding approximate local JM is
 
FRj   FRi  f rJ
where the symmetric matrix
 P  Pe ~
rB 
J   ~ e ~ ~ 
 rB Pe  rB Pe rB 
and the projector Pe  eeT are introduced.
The local JM for the velocities are as follows:
FVj   FVi  f vJ .

2. Generalized linear force element. Consider variation of expressions (11).


r i  ~  ~   r  
r i ii  Ai 0 r j   j i i ij j i0 
~   A r  
j j
~   r  
i i
~ 
ij j 
i 
1 3 ~i
2 k 1
  1 3
 
 ek Ai 0 A0 j A jB  Ai 0 A0 j A jB eki   e~ki  ~ii  ~ij AiBeki  ij  ii
2 k 1
Here we have taken into account that the vector  r is small, the estimation AiB  I 3 as well
as the identity
1 3 ~i ~i i
 ek b ek  bi
2 k 1
which is valid for every vector b.
Substituting these relations into the variation of the force gives the formula
 I  ~ ~i 
r ji    ~i ~i 
j C 1R   I 3  ri  i C 1R 

F i  CR  C   3  0  .
 0 0 j j 0 i i 
 I3   I3  
If we note that r ji  ij  r ji  ij  r  r ji  ij  rBi , we obtain the following final
expression for the approximate local JM of the generalized linear elastic force element:
 
FRj   FRi  C0TB CC0 B ,
~ ~
 I  rj   j 
C0 B   3 .
0 I 3 
Similar transformations lead to the analogous result for a generalized linear dissipative
force element
 
FRj   FRi  C0TB DC0 B ,

50
3. Point-plane contact force. Here we consider only the local JM for the normal force
FRj   FRi  N  J ,
FVj   FVi  N  J
where the symmetric matrix
 P  Pn ~
rC 
J   ~ n  , rC  r1  1
 rC Pn
~ rC 
 rC Pn ~
and the projector Pn  nnT are introduced. These expressions are derived quite similar to the
analogous ones for a bipolar force element.

7. SOME MODELS OF RAIL VEHICLES

7.1. Manchester benchmarks


Manchester Benchmarks have been developed in 1998 to let researchers compare different
software for simulation of railway vehicle dynamics. Detailed information about the
benchmarks can be found on the site www.sci-eng.mmu.ac.uk/rtu/research01.htm.
Full description of the benchmark models, their parameters, track cases and list of
evaluated variables can be downloaded from the same site. The benchmark results for
ADAMS/Rail, MEDYNA, GENSYS, NUCARS, SIMPACK and VAMPIRE are published in
[1]. Description of UM models of the benchmark vehicles as well as full information about
the UM test results are included in the free UM Demo3.0 and available for download and full
dynamic analysis (www.umlab.ru).

Fig. 33. UM model of benchmark Vehicle 1 (left) and Vehicle 2 (right)

The Manchester Benchmarks contain two models of rail vehicles, Fig. 33:
Vehicle 1 – a simplified model of a passenger coach with 42 degrees of freedom in the UM
model;
Vehicle 2 – a model of a two-axle freight vehicle with a load dependent friction with 18
degrees of freedom in the UM model.
We will use here the models of the vehicles to compare different approaches to simulation
and algorithms described above.

51
Tab. 2 presents the comparison of CPU expenses for simulation of the both vehicles with
use of symbolic or numeric-iterational generation of equations of motion (Sect.2.3), and
FastSim or FastSim_A with 10 or 20 elements in a strip (Sect.4.3). Both symbolic deriving
the equations and the use of the semi-analytic modification of the FastSim algorithm allow a
considerable reduction of computational costs. For instance, the variant Symbolic+FastSim_A
is almost two times faster than the variant Numeric+FastSim with 20 elements in a strip.
Anyway, the simulation of the benchmark models with UM is much faster than the real time
one.

Table 2
CPU time for simulation of benchmark models with UM (Pentium IV, 2.4GHz)
CPU time, sec
Simulation
Cases Symbolic + Numeric + Symbolic +
time, sec
FastSim_A FastSim_A FastSim 10/20
Vehicle 1
Track Case 1 42 22.5 32.2 26.2/31.1
Track Case 2 4 0.90 1.4 1.1/1.25
Track Case 3 5.7 2.2 3.3 2.9/3.5
Vehicle 2
Track Case 2 5.5 0.8 1.1 0.9/1.1
Track Case 3 6.5 1.7 2.2 2.1/2.2
Track Case 4 15 5.9 7.45 6.7/7.8

Consider some simulation results for Vehicle 1, Track Case 1 to compare the procedures
FastSim and FastSim_A. Track Case 1 corresponds to the vehicle motion in a right curve of
150 m radius and 100 mm of cant with a constant 4.4 m/s speed. Track irregularities are
presented by a single linear dip of 20 mm in the outer rail at the end of the leaving transition.
Fig. 34 - Fig. 36 show total forces for the outer and inner wheels of Wheelset 1 as well as a
difference in results for these forces between FastSim and FastSim_A. The following two
main conclusions can be drawn. First, in the considered case the FastSim_A procedure gives
very close results to FastSim. The difference normally does not exceed 5%, and for the total
vertical force the difference is less than 0.2%. Second, the difference in results decreases with
growth of discretization of strips, but of course does not tent to zero.

Outer

FastSim 40

Inner FastSim 20

Fig. 34. Total longitudinal forces for Wheelset 1 (left) and the corresponding difference in
force value for the outer wheel between FastSim 20/40 and FastSim_A (right)

52
Inner FastSim 20

Outer
FastSim 40

Fig. 35. Total lateral forces for Wheelset 1 (left) and the corresponding difference in force
value for the outer wheel between FastSim 20/40 and FastSim_A (right)

Outer Outer

Inner Inner

Fig. 36. Total vertical forces for Wheelset 1 (left) and the corresponding difference in
values between FastSim 20 and FastSim_A (right)

Outer rail

Displacement

Inner rail

Fig. 37. Lateral displacement of Wheelset 1(left); difference in the displacement between
FastSim 20 and FastSim_A (right)

The next example (Fig. 37) compares lateral displacement of the leading wheelset of
Vehicle 2, Track Case 3. The freight vehicle runs in a straight track section with a constant
22.5 m/s speed. The track is widened by 25.4 mm. The sinusoidal lateral track irregularities
are shown in Fig. 37 left. The plots of the rail irregularities in the figure are separated by the

53
flange clearance, so that touching the displacement and irregularity plots corresponds to the
flange contact. The example confirms again a good approximation of creep forces by the
FastSim_A.

7.2. Freight car with three-piece bogies

Fig. 38. UM model of a freight car (open wagon)

A freight car with three-piece bogies is one of the most difficult models of rail vehicles
developed with UM. Its difficulty is caused by the fact that almost all interactions between
bodies in the model are contacts with friction. Interactions frictional wedge – bolster, wedge –
side frame, bolster – car body, frame – axle box are contacts. More than 250 point-plane
contact forces (Sect.3.3) are used in the model.
To make the model more universal and exact, frictional wedges are presented by rigid
bodies with six degrees of freedom each. 16 points describe contacts of a wedge with the
bolster and the side frame. Parameterizing coordinates of contact points allows usage of the
model for wedges with different shapes, Fig. 39. Thus, the model contains 19 bodies and has
114 degrees of freedom. Taking into account a mechanism of auto-coupling for creation of
tractive connections (Fig. 31) adds 3 bodies, 8 degrees of freedom and about a dozen of
contact points.

Fig. 39. Shapes of frictional wedges

Numerous contacts make the equations of motion of the model stiff. For example, the
contacts of a wedge introduce frequencies about 1000 Hz in the model. Use the implicit Park
method accompanied with computation of the approximate Jacobian matrices for contact
forces (Sect.6) make simulation of the freight car model fast enough, Tab. 3.

54
Table 3
Some performances of simulation process
Simulation Average CPU time for
Track Speed m/s CPU time
time, sec step size, ms a step, ms
Straight 27.78 21.6 222 0.194
1.6
Curve 20 30 306 0.196

Full parameterization of the model allows the user to analyze the dynamics a variety of freight
cars, which differ in geometric and inertia parameters (open wagons, tank cars, hoppers etc. in
loaded and empty states), Fig. 39, Fig. 40.

Fig. 40. UM models of tank car and hopper

Acknowledgements
The author is grateful to Russian Foundation for Basic Research (RFBR), grant 02-01-
00364, and to the scientific program “Universities of Russia – Basic Research”, grant
UR.04.01.002 for the financial support of this research.

LITERATURE

1. Iwnicki S. D.: The Manchester benchmarks for rail vehicle simulation / ed. by S. Iwnicki.
- Lisse: Swets & Zeitlinger, 1999.
2. Hauptmann D. et al.: Eisenbahntechnische Rundschau. 1998, nr 5, p. 306 – 309.
3. Wittenburg J.: Dynamics of systems of rigid bodies, Teubner, Stuttgart, 1977.
4. Featherstone R.: Robot Dynamics Algorithms, Kluwer Academic Publishers, Boston,
1987.
5. Pogorelov D.: Introduction in simulation of multibody system dynamics, Bryansk, BSTU,
1997. (in Rissian).
6. Kossov V.S.: Reduction of loading the locomotive running gear and track. Doct. techn.
sc. thesis, Kolomna, 2000. (in Russian).
7. Ushkalov V.F. at al.: Mathematical modeling vibrations of railway vehicles. V.F.
Ushkalov (Ed), Kiev, Nauk. dumka, 1989. (in Russian).
8. Kalker J.J., Piotrowski J.: Some New Results in Rolling Contact, Vehicle System
Dynamics, 1989, nr 14, p. 223-242.

55
9. Kik W., Piotrowski: J.: A fast approximate method to calculate normal load at contact
between wheel and rail and creep forces during rolling, Proceedings of 2nd mini. conf.
Contact Mechanics and Wear of Rail/Wheel Systems, 1996, p. 52-61.
10. Kovalev R., Yazykov V.N., Mikhalchenko G.S., Pogorelov D. Yu.: Railway Vehicle
Dynamics: Some Aspects of Wheel-Rail Contact Modelling and Optimization of Running
Gears, Mechanics Based Design of Structures and Machines, 2003, v. 31, nr 3, p. 315-
335.
11. Kalker J.J.: Simplified theory of rolling contact. Delft progress report. Series C:
Mechanical and aeronautical engineering and shipbuilding. 1973, nr 1, p. 1-10.
12. Pogorelov D.Y.: On numerical methods of modeling large multibody systems,
Mechanism and machine theory, 1999, nr 4, p. 791-800.
13. Kolpakhchyan P, Pogorelov D.: Simulation of electric locomotives as mechatronic
systems, EUROMECH 452, Advances in Simulation Techniques for Applied Dynamics.
Martin-Luther-University Halle-Wittenberg, Germany, March 1-4, 2004.
(http://cantor1.mathematik.uni-halle.de/euromech452/papers/pogorelov.pdf)
14. Mikheev G.V.: Usage of hybrid models for simulation of railway vehicle dynamics,
Visnik Skhidnoukrainskogo Nat. Univ. im. V.Dalya. - Lugansk, 2002, p. 32-38. (in
Russian).
15. Garg V.K., Dukkipatti R.V.: Dynamics of Railway Vehicle Systems, Academic Press,
1984.
16. Pogorelov D.: Differential-algebraic equations in multibody system modeling, Numerical
algorithms, 1998, p. 183-194.
17. Pogorelov D.: On Calculation of Jacobian Matrices in Simulation of Multibody Systems,
Preprints of the NATO Advanced Study Institute on Virtual Nonlinear Multibody
Systems. v.1. Edited by W. Schiehlen and M. Valasek, Czech Technical University in
Prague, Prague, June 23 – July 3, 2002, p. 159-164.

56

View publication stats

You might also like