&VSPQFBO
$POUSPM
$POGFSFODF

&$$


1PSUP
1PSUVHBM

4FQUFNCFS


SIMULATION OF VARIABLE DYNAMIC DIMENSION

SYSTEMS: THE CLUTCH EXAMPLE
R. Zanasi, G. Sandoni, R. Morselli
DSI, Dipartimento di Scienze dell’Ingegneria,
University of Modena and Reggio Emilia, Via Vignolese 905, 41100 Modena, Italy,
Phone: +39 59 2056161; Fax: +39 59 2056126
e-mail: roberto.zanasi@unimo.it

Keywords: Modelling Control and Verifications, Automotive Different velocities ?   D ? 

Same velocity ?   ?   ?

system, Variable Structure Control, Non-Linear System, De-

sign Tools. 5

1
/

-
1 / 4

1
/

- -
/

- . - .

5 5

4 4

Abstract    

& & & &

Due to the presence of strong non-linearities (i.e. coulomb

 

   8   : ; = ?  + ?  

friction) some systems change their dynamic dimension while Second order model: First order model:
functioning. Such systems can be found in many application 

fields, such as in automotive and robotics. This paper proposes

 ?    + 8   : ; = ?  + ?  


 ,   ?   + 

& '

& & ' '

a simulation model for this type of systems that we will call as

 ?   8   : ; = ?  + ?   + 

& '

“Variable Dynamic Dimension Systems” (VDDS). Particularly,

a VDDS system composed by masses which slip and interact
together by means of the coulomb friction is analyzed. A con- Figure 1: Two different dynamic models for the clutch system.
gruent state space transformation is used to obtain a simple and
effective simulation model for the system. The performances
of the model are tested through simulation experiments applied is proposed. The paper is organized as follows. In Section 2
to an interesting automotive application: a clutch with torsional the simulation problem is stated and the typical schemes used
damper-spring. to simulate such problem are presented. The proposed sim-
ulation model is discussed in Section 3. A detailed example
explaining the features of the proposed model is presented and
1 Introduction simulated in Section 4. Finally, an interesting application to
The typical time-varying nonlinear systems are usually de- an automotive system, a clutch with torsional damper-spring,
scribed with the equation is described and simulated in Section 5.


where           

 

is the state vector,

      

the in-
      !   !       ! $ 

put vector and the time. The dynamic dimension of these 

2 Problem statement
systems is and usually it is constant. However when strong In this paper, systems having the framework shown in Fig. 2
non-linearities (i.e. coulomb friction) affect the system dynam- are considered. In the general case, the masses (or inertias)
ics, it can happen that the system behaves as a lower order sys- J

, ,... , move (or rotate) under the action of the ex-

 J  J 

tem. We will call such systems as “Variable Dynamic Dimen- ternal forces (or torques) , ,..., . Let , ,..., be M


M


M
      

sion Systems” (VDDS). A typical example of such systems is the positions (angles) of the masses (inertias). Due to the pres-
given by a clutch (see Fig. 1). When the clutch is slipping, the ence of the coulomb friction, the motion of a single mass de-
two inertias and move independently (one respect to the &


&


pends also on the motions of the neighborhood masses: ( N O N P



other) under the action of the torques and , and only the '


'


is the coulomb friction between the masses and . Let J

N
J

N P


coulomb friction is exchanged between them. Otherwise, , ..., denote the linear velocities of the
 

 

Q     Q    

when the clutch is locked, the two inertias rotate together. In masses. The system dynamics is given by the following differ-
this working condition the order of the model is equal to 1, and ential equations:
it is easily described by a first order model where the torque
for (1)


J Q  ,  +   X     

' acts on the inertia

 +

' . In many application fields,



&
 ,

&


N N M N ( N U O N ( N O N P V

such as in mechanics and in robotics, “variable dynamic dimen- where ( N O N P



is the amplitude of the coulomb friction given by:

sion systems” have to be handled. The simulation of such sys- YZ

tems is not easy due to the fact that the model changes its order. Z

Z
8

N O N P
 : ; = Q

N
+ Q

N P
  

if Q

N
 D Q

N P


This leads to models that are or exact but huge and very com- [

,   J  +  +    J

plex, or simple but approximated. In this paper a new model for ( N O N P

 

Z\
8

N O N P
 : ` a b
M N ( N U O N N P M N P ( N P O N P N

(2)
8  J , J   d

the exact simulation of “variable dynamic dimension systems” N O N P N N P

if Q

N
 Q

N P


Proceedings of the European Control Conference 2001 3149

*4#/


Authorized licensed use limited to: BME OMIKK. Downloaded on October 07,2020 at 10:17:18 UTC from IEEE Xplore. Restrictions apply.
 

Model 1


 

 

 



Model 2 

In Out
 

Model 3

     

..
) )

 

       
.


Figure 2: Example of a variable dynamic dimension system



Model 

where is the coulomb friction coefficient (that is the 

 

Selection Module
maximum absolute value of the coulomb friction). When


, the coulomb friction is equal to the force

 




 


that keeps the relative velocity equal to zero. The 

 




Figure 3: Framework of the multi-subsystem simulation model

force has a limited amplitude: . When this ampli-

 
 

 


tude is exceeded, the two masses and start to move 



at different velocities. Otherwise, when the



 



 


from the old model to the new one.

value of the force is the following, see eq. (2),:
 


c) Inputs and outputs of the simulator have to be properly con-

           
nected to the new model (see Fig. 3).
d) At each instant of the simulation time, the “selection mod-

    
  

 

 

 

ule” has to choose which one, among all the possible models,
In fact, substituting this result in equation (1) one obtains: is the correct model to be use at that time.

For all these reasons, the multi-subsystem simulation model
shown in Fig. 3 is generally huge and very complex to be used
                

 
   
  

when the number of masses is greater than 3.



                 

  
   
  

that is, the two masses and move together (as a single 

2.2 Sliding mode simulation model



mass ) at the same velocity under the

 

     

 

action of the resulting force . 








  

 
 




 
 

To cope with the all these problems, a different approach,

In this condition, the dynamic dimension of the system reduces mainly based on the sliding mode theory, is commonly used.
to . If the coulomb friction between the masses is able
$  '

The sliding mode simulation model of the considered system

to keep relative velocities to zero, the dynamic dimension of

can be obtained from (1)-(2) by replacing the second part of

the system reduces to . The coulomb friction between $ 

equation (2) with equation

,  
  

 
 + - /  

 


 

the masses influences and is influenced by the dynamics of the which is exactly equal to the former part of equation (2). In
whole system. Due to the particular form of the coulomb fric- this case the coulomb friction is seen as a sliding input variable
tion given in (2), the exact simulation of the “variable dynamic for the system. When the previous equation is used and the
dimension systems” is particularly difficult. relative velocities are not zero, the dynamic behaviour of the
system remains the same as when equation (2) is used. On the
2.1 Multi-subsystem simulation model contrary, when the relative velocity becomes zero, the 

 




corresponding coulomb friction starts switching at infi- 

The typical scheme used to simulate the “variable dynamic

 

nite frequency between the two values trying to keep to  

dimension systems” is strictly based on the definition of the

  

zero the relative velocity (see Fig. 2). The continuous-

   

coulomb friction given in (2). When the system is composed



time equivalent value of the sliding variable can be eas- 

by masses, there are relative velocities that have to

 

$ $  '

ily computed as shown in [6], and the corresponding dynamic

be checked to choose which one of the relations of equation ( 

behaviour is fully equivalent to the second relation of equa-

(2) has to be used, consequently there are differ-


tion (2). In this case, just one model is used to simulate the
ent possible configurations (see [1]). The block scheme used
behaviour of the whole system in every functional condition.
in this case in shown in Fig. 3: at each instant the “selection
Unfortunately, this sliding model can not be “exactly” simu-
module” chooses which model has to be used. With this type
lated. In fact, all the simulators can not handle an ideal infinite
of simulation model, the following problems arise:
switching frequency. Therefore a “finite” switching frequency
a) The number of different models increases exponentially ( 

is really applied to the model and this introduces a strong ap-

with the number of masses: .


$  

proximation, as shown later in Section 4. Moreover, to improve

b) When the system switches from a model to another, the up- the simulation accuracy a very small integration step is neces-
dating of the initial conditions of the new model is required, sary and therefore simulation time increases dramatically. To
that is, the state variables of the system must be converted $

avoid the very high switching frequency of the sliding variables

Proceedings of the European Control Conference 2001 3150

Authorized licensed use limited to: BME OMIKK. Downloaded on October 07,2020 at 10:17:18 UTC from IEEE Xplore. Restrictions apply.
In Out slipping conditions are:

Main Dyn.

T

01

  
,

 +   %  '       ' 

  ,

 '  ' + ' 7 %  '       '   % '     '    

(3)
#

Rel. Dyn. 1
,

13

    +  7 % '     '      %        

  + 7 %        

 

Rel. Dyn. 2

Let us now define the following vectors and matrices:

 

           

..
 "    " 

       

#


. 

+
#
'

 : +
" '

  +
 '

+
"
,

'

 "    " 

       

Rel.Dyn.   

"  "
,

Selection
#



  

 

   
 

     %  '

 '

Figure 4: Variable dynamic dimension systems: block scheme. 

 
  +


    


  +


% ' 

 

     % 

System (3) can be rewritten in matrix form as follows:

and to reduce the simulation time, the sign function “ ” is   

 ,

T
 ,

 +         "  +  #

(4)
sometimes substituted by the saturation function “ ”, but    

this solution is even more approximated than the previous. Vectors $

+  

and # + 

T      

are:
  

3 Proposed simulation model for variable dy-

%  '       ' 

 
 

    '

% '     '      %  '       ' 

namic dimension systems $

+


 '   


 # +


%          % '     '    

The complexity of the variable dynamic dimension systems    

 %        

lies in the definition of the coulomb friction given by equation

(2). The basic idea of the proposed simulation model, shown System (3) can be simplified if the following state space con-
in Fig. 4, is the introduction of a proper congruent state space gruent transformation is considered:
transformation that allows to see the system dynamics from a 


A


A .


A 0 A .


A 0 A 0

 

new perspective that decouples the main system dynamics from

 

1 1 1

A A . A 0 A 0

 

all the other relative dynamics. The transformation shows

1 1 1

 ' ,
 +  + * +

A A A A 0

C C B

  

that the coulomb friction acts only on the relative dynam-

 

1 1 1

A A A A A A .

    

C C B C B

  

   

ics given by the relative velocity . Consequently,

1 1 1

   

  

decouples the dynamics of the original system into indepen- 

where 2
+   7  ' 7   7 

. The new state vector has the

)

dent dynamics. The Main Dynamics shown in Fig. 4 describes following physical meaning:
the lower dimensional dynamics of the whole system, namely 


 

the dynamics of the system when all the bodies move together.
    7  '  ' 7     7   

 

The remainder submodels describe the Relative Dy-

    '

     

-1
) )

+  " +

namics among the masses. The Main Dynamics and the

 '   

 

  

Relative Dynamics are simply obtained by applying the state

   

space transformation . In the transformed space, the task of In this case, the Main Dynamics is described by the trans-
the Selection Module is just to add or remove the relative dy- formed velocity (the mean velocity weighted by the masses), 4


namics when the dynamic dimension of the system increases or while the Relative Dynamics are described by the other three
decreases: the -th Relative Dynamics is added when the mass


relative velocities , and . By using , system (4) 4

'

4


4
 +
)

slides on mass
 , and it is removed when and 

 
 




 


transforms as :
 7
)
,

+
7

 #
7

move together. The use of the transformation is very simple

and leads to a simulation model much easier than the solutions < = 9 :

T T T
 ,

(5)
) )

shown in the previous section. Let us now again consider the

+        C 

9 : < = 9 : < =

9 : < =

system shown in Fig. 2. Four masses , , move 

   
 
  

> ? @ ?

D ?

in the direction subjected to forces . The masses interact

"

# 

The new vectors and

7

have now a very simple

)

+  #

by means of the coulomb friction which, in this case, is com- $

pletely described by the friction coefficients , and . %  ' % '  % 

structure:
The positions of the four masses are given by the variables .
  

"

 
 

'

Let us denote with , , and the velocities of the four    '   

+
4


 #
7

+
%  '  

4
'

 

masses: . The differential

, , , ,

4
$
  + "    ' + " '    + "    + "

% '    

 

equations describing the dynamics of the considered system in 4

%   

Proceedings of the European Control Conference 2001 3151

Authorized licensed use limited to: BME OMIKK. Downloaded on October 07,2020 at 10:17:18 UTC from IEEE Xplore. Restrictions apply.
 

The benefit of using transformation is that now matrices where is the matrix obtained by selecting the rows and the 

T
and -1 are block diagonal matrices. columns of matrix
    

corresponding to the sliding variables

$



; is the sub-matrix of matrix



obtained by se-
$

 
  )

  

lecting the rows corresponding to the sliding variables

 

 

  

-1


  

  


-1


 















-1

and the columns corresponding the other variables . The  
0 

main functions of this algorithm are the following:



  


 

 


 

  



 

a) it checks if the relative velocities are equal to zero;  

Let vectors , b) it computes the possible equivalent control related to

 

and


be partitioned as follows:


"  $ % 
























the transformed variables ;  

  

c) it verifies the sliding mode conditions on

 

  


 

"  $ %   $ % 

  

  


variables ;
 
 

  

 

  

     

d) it keeps at zero all the variables that satisfy the sliding mode
 

 

  

conditions; for these variables it imposes ;



   

Since
 

is a block diagonal matrix, system (5) can be rewrit- e) it computes the proper input for all the variables not



 

ten as: satisfying the sliding mode conditions.

 

 

-1
Note: the proposed algorithm works correctly also when the
  
 

    

(6)
   

-1
parameters are time-varying. For understanding the
          
 

 

       

) $ % )

computational method used for solving system (7), let us con-

that is:







sider the following two cases:


I) The case when all the three variables are equal to zero:


 

. In this condition a sliding mode arises

 


 
  

 


  
 

(7)
    

  

  


in the system iff:

  

 

 


 

 

   



  


     

    


 


         


  

(8)


   

 
 



   

 



 

  

The state space transformation decouples the original



The functions start switching between the two values



   

 )

system in two independent parallel systems: variable is not 



at infinity frequency with an average value that, at each




) $


% )

influenced by variables , and and viceversa. The sec- instant, is equal to the external transformed force


. If condi-

  

ond part of system (7) describes the relative dynamics and can tions (8) are satisfied, variables , , and are kept to zero. 

 


be interpreted as a three dimensional Variable Structure Sys-   

II) The case when and . Solving system 
0 

 



tem. When , if relation is satisfied,  



"  $


% 


 $


(7) with respect to the equivalent controls and "

  

a sliding mode arises in the system: is the equivalent "  $



" one obtains: 




 




control associated with the switching term , that 

 $


% 
  

 



 




    

is the time mean value of the term , and it is 

 $


% 
  

 
















equal to the force that at each instant the two masses


 


 

& 

 "


 

 



 

and exchange due to the presence of the coulomb friction.

 

&

 "

  

When all the conditions are satisfied, in a "  $



% 


 $


that is:
finite time the system (7) converges towards the sliding mani- 







-1 $


fold . When one of the sliding surfaces





  


"  

  

.       


 

is reached, a sliding mode can arise in the system, that is



  
   



 

" 

 )

the term can start to switch at infinite frequency 

 $


% 
  

 


keeping the variable at zero. In this condition the dynamic  

If signals and

satisfy the sliding mode conditions  
"


"




dimension of the model decreases of one unit. Since matrix and " , the variables

and are kept 

"



 


to zero and the dynamics of variable is described by the fol-



is not diagonal, when one of the sliding variables is equal

$

to zero, its equivalent control influences immediately all the lowing equation:
 

other relative dynamics. To correctly simulate the system (7),



    

 


the following simulation algorithm has been designed:

 

 


 

  

 

  
 "


 

 "

loop 

 
 

If the sliding mode conditions are not satisfied, a different con-

  % %

 

            . 

figuration for the sliding variables must be considered (for in-

"  $ %  )  ) $ % ) ) $ % )  )   $ % 

  

if   

stance , and ). All the other cases must be

"  $ %   $ % 

0  0  

  

   

treated in a similar way.

 

else 


            

   ) $ % ) ) $ % )  )

4 Simulation results


end if
end loop The parameters used in simulation are: &  & &

&

Kg; N, N, 
8    8  8  


8 

N and ini-

Proceedings of the European Control Conference 2001 3152

Authorized licensed use limited to: BME OMIKK. Downloaded on October 07,2020 at 10:17:18 UTC from IEEE Xplore. Restrictions apply.
 Velocities v1(r−−), v2(b−.), v3(g−), v4(m:) Equivalent Control: force exchange between the masses
2


10

T12 (N)
1
0
m/sec

0 −10


0 0.5 1 1.5 2 2.5 3 3.5 4

−1 

10
−2
0 0.5 1 1.5 2 2.5 3 3.5 4

T23 (N)
Time (sec) 0

Figure 5: Velocities  of the masses,   






. −10

0 0.5 1 1.5 2 2.5 3 3.5 4

Transformed Velocities z1(r−−), z2(b−.), z3(g−), z4(m:)
2
10


T34 (N)

0
m/sec

−10

0 0.5 1 1.5 2 2.5 3 3.5 4

−1

Time (sec)

−2
0 0.5 1 1.5 2 2.5 3 3.5 4
Figure 7: Equivalent controls #  



  

,   





.
Time (sec)
Exact velocities v1(r−−), v2(b−.), v3(g−), v4(m:)


Figure 6: Transformed velocities

 ,   





. 0.1 

m/sec 0.05

0


tial conditions m/s. The forces applied    



 

 

 

−0.05

to the masses are sinusoidal functions  




      


 

−0.1
with amplitudes N and frequencies 



 
 
 
  




1.75 1.8 1.85 1.9 1.95 2 2.05

Time (sec)

 

Hz. Note that the amplitudes



are lower
 

  

 


Sliding velocities v1(r−−), v2(b−.), v3(g−), v4(m:)

than the friction coefficients . This ensures that in finite 

 


0.1 


time the four masses collapse to an unique mass. The veloci- 0.05
ties obtained in simulations are shown in Fig. 5. Note that for
m/sec

0


s it happens that


, but the exchanged torque

    

#



−0.05
is greater than the friction coefficient and therefore masses  

−0.1
$

and continue to slip. For



s it happens that $

  
    

1.75 1.8 1.85 1.9 1.95 2 2.05

, but in this case
 

, so the two masses col- #

  

Time (sec)

lapse together. This condition persists until about s,   

Figure 8: Comparison between the proposed simulation model

when the external forces and the inertia dynamic action over- and the sliding approximated model: the velocities . 

come the coefficient , so the two masses start slipping again.  

The transformed velocities are shown in Fig. 6. The “Main



Dynamics” is described by velocity while the “Relative Dy-

5 Clutch with Torsional Damper-Spring

namics” are described by the other variables , and . Note

Let us now consider the automotive transmission system shown

that for s and 0.8 s 1 s, when , the corre-     





 

in Fig. 10. The torsional damper-spring is an elastic element

sponding transformed variable is equal to zero. The equiv-

that often is inserted between the clutch disk and the primary
alent controls are shown in Fig. 7. They represent the #  


of the gear shaft with the objective of filtering the torque spikes
torques that at each instant the masses reciprocally exchange. generated by the engine. Let and denote the inertias *


*


When the relative angular velocity is equal to zero, the cor-



of the engine shaft, the torsional damper disk and the transmis-
responding equivalent control is not saturated, and it be- #  


sion shaft, respectively. The system simulation is critical due to

longs to range . 
 

 






 





the presence of coulomb frictions between the three inertias , *



* and . The coulomb friction



is assumed to be constant
*
 

4.1 Approximated simulations: the sliding mode model while the coulomb friction is supposed to be modulated by   

A simulation model for the considered system is the sliding an external normal force acting on the disk. 
,

model described by equation (3). A comparison between the

simulation results obtained with this sliding model and the 5.1 Dynamic Model of the clutch with the damper-spring
model proposed in section (2.2) is shown in Fig. 8. The match- The differential equations describing the system are:
ing is good except for obvious chattering phenomena on the -.

output velocities of the sliding approximated model. In the

.

 2       2       '   2   2  

* 4

sliding model, the mean value of the switching action repre-

/

 2      2  ;     '   2   2       '   2   2        

sents the torque that at each time is exchanged by two consecu-

.0

2    2 ;    '   2   2  ;      

* 4

tive masses: . This fact is clearly shown

   2   2

     '   

#      


in the lower part of Fig. 9 where the equivalent control ob- #



tained with the proposed model is compared with the switching where 4


, 4 are the engine torque and the resistant external

action filtered by a first order Butterworth filter. torque; 2 
2 

and are the engine angular velocity, the tor-

2

Proceedings of the European Control Conference 2001 3153

Authorized licensed use limited to: BME OMIKK. Downloaded on October 07,2020 at 10:17:18 UTC from IEEE Xplore. Restrictions apply.
 Sliding actions exchange between m2 and m3 W1 engine angular velocity(r−−), W2 dumper angular velocity(b),W3 transmission angular velocity(g−.)
15
15
10

5
10
Tslid23 (N)

rad/sec
−5 5
−10  

−15 0
0 0.5 1 1.5 2 2.5 3 3.5 4
Time (sec)

Filtered sliding action

15 −5
0 0.5 1 1.5 2 2.5 3 3.5 4
10 Time (sec)

5
Figure 12: Angular velocities , and .
T23 (N)

              

−5

−10

−15
0 0.5 1 1.5 2 2.5 3 3.5 4
to increase. At time s all the inertias move together at
   ! 

Time (sec)
the same angular velocity . Note that                 

Figure 9: Equivalent control '

 

compared with the mean value 

is smoother that
   

due to the presence of the torsional     

of the sliding actions. damper-spring. When the coefficient decreases to zero the   

angular velocity starts to differ again from and .

 

     

 








6 Conclusions



In this paper, the problem of finding a simple and effective sim-

ulation model for a class of Variable Dynamic Dimension Sys-





%
%
%

tems (VDDS) composed by masses that reciprocally slip and %

interact by means of the coulomb friction, has been presented.


 


This type of systems is very common in mechanics. For sim-
ulating this type of systems, in literature one can found huge


Figure 10: The clutch with the torsional damper-spring. and complex models, or approximated models that give rough
results. In this paper, a particular space state transformation
that puts in evidence the main dynamics and the relative %  

sional damper disk angular velocity and the transmission shaft dynamics of the considered system, has been presented. Due to
angular velocity, respectively; and are the viscous fric-    

this new point of view, the exact simulation of the considered

tion coefficients; and are the coulomb friction coeffi-      

VDDS is easier and faster. Finally, the proposed simulation

cients; is the elastic torque of the torsional damper-
       

model has been applied to a VDDS system of interest in auto-

spring which is a function of the relative position        

motive applications: the clutch with “torsional damper”.

between the two inertias and . 





References
5.2 Simulation results


[1] J.H. Taylor. Rigorous Handling of State Events in MAT-

The parameters used in simulation are: Kg m ,
 

  


 

LAB, 1995 IEEE International Conference on Control



Kg m , Kg m , N m s/rad, N m s/rad, 
      
  

Application, pp. 156–161, (1995).



N m s/rad, 

N m and

. The time      


  

behaviours of the coulomb friction and the torque are   



[2] A. Haj-Fraj, F. Pfeiffer. “Dynamics modelling and anal-
shown in Fig. 11. The angular velocities , and               
isis of automatic transmissions”, IEEE/ASME Advance
Engine torque C1(r), Columbian friction coefficient K12(b−.)
Intelligence Mechatronics, pp. 1026–1031 (1999).
600  

[3] A. Haj-Fraj, F. Pfeiffer. “Dynamics of Gear Shift Opera-

500

400
tions in Automatic Transmission”, DYMAC’99, Manch-
ester, UK (1999).

300
Nm

200 [4] J.-J. Slotine. “Sliding controller design for non-linear sys-
100 tems”, Int. Jornal of Control, 40, pp. 421–434, (1984).
0

−100
[5] C. Edwards, S.K. Spurgeon. Sliding Mode Control: The-
0 0.5 1 1.5 2
Time (sec)
2.5 3 3.5 4
ory and Application, Taylor and Francis, London (1998).
[6] V.I. Utkin. Sliding Mode in Control Optimization,
Figure 11: Time behaviours of the coulomb friction   

and
Springer-Verlag, New York (1992).
torque . 


are shown in Fig. 12. At time s the velocity starts    !   

to increase because the torque N m is applied; the 

    

velocities and remain at zero because the coefficient       

is zero. At time s, when the coefficient starts to in-      

crease, the velocity decreases and velocities and start      

Proceedings of the European Control Conference 2001 3154

Authorized licensed use limited to: BME OMIKK. Downloaded on October 07,2020 at 10:17:18 UTC from IEEE Xplore. Restrictions apply.