Chapter 12
Chapter 12
Chapter 12
Scope of this chapter is limited to arriving at transfer functions for linear time
invariant systems from their bond graph models via signal flow graphs. Once
transfer functions are obtained, further analysis may be routinely carried on by the
techniques well developed in classical control theory of these systems. One may refer
standard books [37, 52] on automatic control theory for this.
One of the most important steps in the analysis of a physical system is the
mathematical modeling of the system. In systems theory, signal-flow graphs (SFG)
and transfer function are valuable tools for analysis as well as design of linear
systems.
Here we write down the algorithm for the generation of the signal-flow graph of a
dynamic system from its bond graph representation. We consider the bond graph
shown in Fig. 12.1.
Fig. 12.1
Algorithm
(a) Each bond of the bond graph is associated with a pair of signals namely e
(effort) and f (flow), unless the bond is activated. If any of these signals is
suppressed or ignored due to activation (like in an ‘e’ activated bond ‘f’ is
APPROACHING CONTROL SYSTEMS 334
suppressed and in a ‘f’ activated bond signal ‘e’ is suppressed), then that
signal does not appear as a node on the signal-flow graph.
(b) All power bonds as well as flow activated bonds attached to a 1-junction
have same flow and would contribute only one flow node to the signal-flow
graph, bearing the bond numbers as suffixes. The 1-junction of Fig. 12.1
shown in Fig. 12.2(a) will contribute a flow node f1,2,3,4. Similarly the 1-
junction shown in Fig. 12.2(b) will contribute a flow node f6,7,9.
(c) All power bonds as well as effort activated bonds attached to a 0-junction
have same effort and would contribute only one effort node to the signal-
flow graph, bearing the bond numbers as suffixes. Thus the 0-junction of
Fig. 12.1 shown in Fig. 12.2(c) will contribute only one effort node e4,5,6. The
junction strong or equality laws are taken care of in steps (b) and (c).
(d) The remaining non-activated bonds (which have not yet contributed either
an effort node or a flow node or both) contribute individual effort nodes (in
the case of 1-junction) and individual flow nodes (in the case of 0-junction)
to the signal-flow graph. Thus in Fig. 12.1, the junction shown in Fig. 12.2(a)
contribute e1 and e2. The junction shown in Fig. 12.2(b) contribute e7 and e9
and the junction of Fig. 12.2(c) will contribute node f5.
2. Constitutive relations of the elements : The flow of a signal between the nodes in
a signal-flow graph is designated by directed lines. A signal travels only in the
direction of the arrow. Each directed line connecting two nodes is called a branch.
Now we determine the transference (or gain) associated with those branches which
relate to I, C, R, TF and GY elements. These transference may simply be numbers or
they may be transfer functions. The transfer functions may be determined from the
constitutive relations. The constitutive relations for the single port elements are as
follows.
335 MODELING AND SIMULATION OF ENGINEERING SYSTEMS THROUGH BOND GRAPHS
1 1 f s 1
(a) I: f
m e dt , or f s
ms
e s , or
e s ms
.
K e s K
C: e K
f , or e s f s , or
(b) dt .
s f s s
(c)
e s
R: R.
f s
(i.)
f s 1
R: .
e s
(ii.)
R
For transformers (TF) and gyrators (GY), the constitutive relations are their
respective moduli.
Thus from Fig. 12.1, the transference associated with the branch e2 f1,2,3,4 due to
1
the element I2 is .
m2 s
The transference associated with the branch f 5 e4,5,6 due to the element C5 is
K5
.
s
The transference associated with the branch f 6,7,9 e7 due to the element R7 is R7.
The transference associated with the branch e9 f 6,7,9 due to the element I9 is
1
.
m9 s
Next we determine the weak variables and the receptor for each junction.
(a) The weak variables of a junction are the power variables which are not
equated for all the bonds appended to that junction. The weak power
variables of a 1-junction are efforts of the bonds appended to it. Out of all
the weak variables of the bonds appended to a junction, one would be
called receptor variable. The receptor variable is the weak power variable of
APPROACHING CONTROL SYSTEMS 336
the particular bond which takes the information of this variable away from
the junction (as per causal orientation). In this example the receptor variable
of the 1-junction of Fig. 12.2(a) is e2 and likewise, the receptor variable for
the 1-junction of Fig. 12.2(b) is e9 .
(b) The weak variables of a 0-junction are flows. Thus the receptor variable for
the 0-junction shown in Fig. 12.2(c) is f5.
(a) At any junction the bonds which have the same direction of power as the
bond associated with the receptor variable are said to be co-oriented, while
those bonds that have power directions opposite to that of the weak
variable are said to be contra-oriented.
(b) At a 1-junction, the branches connecting the effort nodes of the co-oriented
bonds have gains –1 and those connecting the effort nodes of the contra-
oriented bonds have gains +1. For the part SFG of Fig. 12.2(a) and Fig. 12.2(b)
the gains are given in Fig. 12.3 and Fig. 12.4 respectively.
(c) At a 0-junction, the branches connecting the flow nodes of the co-oriented
bonds have gain -1 and those connecting the flow nodes of the contra-
oriented bonds have gains +1. From Fig. 12.1, for the junction of Fig. 12.2(c)
the gains are as shown in Fig. 12.5.
337 MODELING AND SIMULATION OF ENGINEERING SYSTEMS THROUGH BOND GRAPHS
Following the steps-1 through 4 the signal-flow graph of the bond graph given in the
Fig. 12.1 is as shown in Fig. 12.6.
Signal-Flow Graphs are manipulated and reduced in ways very similar to those used
with block diagrams. The input-output transfer function can be obtained by the
application of Mason's gain rule:
P i i
,
G s i
i = the for that part of the graph which does not touch the i th forward path.
Below, we define the terms used in connection with Mason's gain formula and SFG.
(a) Input node (source): A node which has only outgoing branches. In the SFG
given in Fig. 12.6, the node e1 is the input node.
(b) Output node (sink): It is a node having at least one incoming branch. In the
SFG of Fig. 12.6, e9 is an output node.
(e) Feedback path (loop): A path which originates and terminates on the same
node, along which no node is encountered more than once. In the SFG of
Fig. 12.6, e9 – f6,7,9 – e7 – e9 is a loop.
(f) Path gain: The product of the branch gains encountered in traversing the
path. For the path e1 – e2 – f1,2,3,4 – f5 – e4,5,6 of the SFG, the associated gain is
1 K K5
1 1 5 .
m2 s s m2 s 2
(g) Loop gain: The product of all the branch gains of the branches forming that
loop. For the loop e9 – f6,7,9 – e7 – e9 of the SFG of Fig. 12.6, the associated gain
is
1 R
R7 1 7 2 .
m2 s m2 s
In the SFG given in Fig. 12.6, there is only one forward path e1 – e2 – f1,2,3,4 – f5 – e4,5,6 – e9
between the nodes – e1and e9 with gain
1 K K5
p1 1 1 5 .
m2 s s m2 s 2
1 K5
m2 s 1 s 1 ,
L1 e2 f1, 2,3, 4 f 5 e4,5,6 e2
1 K5
1 m9 s 1 s ,
L2 e4,5,6 e9 f 6,7,9 f 5 e4,5,6
1
m9 s R7 1 ,
L3 e9 f 6,7,9 e7 e9
R7 K K 1 K 5 R7
1 5 5 and 1 1 .
2
m9 s m 2 m9 s m 2 m9 s 3
Therefore, transfer function between e1 and e9 (as given by Mason's gain formula) is
K5
1
P m2 s 2
G s 1 1
R7 m m9 K5 K 5 R7
1 2
2
mg s m 2 m9 s m2 m9 s 3
K5mg s
.
m2 m9 s 3 R7 m2 s 2 K 5 m2 m9 s K 5 R7
The dynamics of complex systems can be studied elegantly using bond graph models
for various control strategies. The designer can modify the control strategies or the
various parameters governing the system without much effort and gain valuable
insight into the behavior of the system. The designer can also get the transfer
functions governing the system. The critical parameters of the system can quickly be
established using the transfer function by applying the stability criteria of Routh,
Routh-Hurwitz or other graphical methods.
Three examples concerning position control and velocity control are given. The
control strategies used are as follows.
1. Proportional control.
2. Proportional-integral control.
APPROACHING CONTROL SYSTEMS 340
3. Proportional-integral-derivative control.
Problem statement: Figure 12.7 shows the system considered. By applying suitable
effort, the position of the mass is to be brought to the level marked Yref. A velocity
pickup measures the velocity of the mass. An integrator, integrates velocity to give
the current position. This is compared with the reference position. An effort
proportional to the positional error is applied on the mass. The bond graph of the
system is shown in Fig. 12.8.
Fig. 12.7
Fig. 12.8
341 MODELING AND SIMULATION OF ENGINEERING SYSTEMS THROUGH BOND GRAPHS
Fig. 12.9
Control strategy: Two different forces are applied on the mass to bring it to the
required position as shown in Fig. 12.10.
The integral control keeps track of the past history of errors and corrective action
(force) is decided by the same.
APPROACHING CONTROL SYSTEMS 342
Fig. 12.10
In Fig. 12.11 Q15 gives displacement at C15. The gyrator simply converts the effort
signal to flow signal and the flow signal is integrated at the activated capacitor C15.
SE5 corresponds to the effort applied on the mass which is proportional to the integral
of the error signal.
Fig. 12.11
Control Strategy: Three different forces are applied on the mass to bring it to the
required position as shown in Fig. 12.12.
The third corresponds to the derivative control. Here the force is generated by
projecting the future position based on the current rate. Derivative control dampens
out the oscillations around the set point.
Vref
Fig. 12.12
Fig. 12.13
In the system bond graph shown in Fig. 12.13 note the activation on bond 5 and 6.
The velocity pickup applies no force on the system while picking up velocity and
effort source does not pickup any velocity information.
APPROACHING CONTROL SYSTEMS 344
For all the three cases discussed so far the velocity pickup is physically present in the
system. They were not represented in the bond graph in the first two cases simply
because C3 was already giving the information on displacement of the mass and that
was all we needed. Now derivative control requires velocity information also and
bond 5 and 6 take care of the derivative action.
The signal flow graph of the system with only derivative feed back is shown in
Fig. 12.14.
Fig. 12.14
Fig. 12.15
1
f2 m2 s
e2 k R
1 3 2 4
m2 s m2 s m2 s
s
m2 s R4 – s k 3
2 .
The signal flow graph of Fig. 12.14 is now simplified as shown in Fig. 12.15.
Now the integral and proportional control action can be added to it as shown in the
signal flow graph of Fig. 12.16.
Fig. 12.16
The block diagram for the same has been shown in Fig. 12.17.
Fig. 12.17
yr e f
eref
s
Gp s G1 y ref
y2
m2 s R4 s k 3 Gp s G1
3 2 s
The order of the system has gone up to 3. It is because of the presence of the
integrator.
y 2 t t s y 2 s s 0 y ref
There is no steady state error. At t , the ultimate balance is supported by the
integrator alone. Using Routh criteria we can see the restrictions on the parameters.
Note that is opposing R4 when they are of the same sign. Both R4 and contribute
to the derivative control action.
s3 : m2 k 3 1 0
s2 : R4 – GI 0
s :
R4 – k 3 Gp – m2 G I
0 0
R4 –
s0 : GI 0 0
GI
R4 k3 Gp
For the system to be stable R4
m2
The parameters so selected can be used in simulation to study the dynamics of the
system near critical values of the parameters.
A more practical example concerning the velocity control of an electric motor driving
a load will now be considered. The bond graph approach to study the various control
strategies will be explored.
Problem statement:
In Fig. 12.18 a motor drives a load through a shaft. The objective is to maintain the
speed of the disk A at any desired value. A velocity pickup is used to measure the
velocity of the disk. This is compared with the reference velocity.
Fig. 12.18
The velocity pickup in Fig. 12.18 generates a voltage representative of the velocity of
the disk. This is compared with the reference signal representative of the desired
velocity. Voltage proportional to the error in velocity is applied to the motor.
The bond graph in Fig. 12.19, though representative of the model is found to be
loopy. All software may not be able to handle such loops. The loop in Fig. 12.19 can
APPROACHING CONTROL SYSTEMS 348
be avoided as shown in Fig. 12.20. Here the required state variable is fed back as a
15 is the rate
modulated source of flow or effort as the case may be. In Fig. 12.20 Q
of change of displacement of the activated C15 element and measures the speed of
the motor. By feeding back the state variable as a source of effort, the loops in the
system can be circumvented. The problem now is to tune the value of the
transformer modulus AMPL (proportional gain) to bring about the required control.
Fig. 12.19
349 MODELING AND SIMULATION OF ENGINEERING SYSTEMS THROUGH BOND GRAPHS
Fig. 12.20
Here the current in the armature is to be tuned to achieve the required velocity
control. The bond graph for the scheme is shown in Fig. 12.21. By adjusting the value
of AMPL (proportional gain), current control can be achieved.
Fig. 12.21
As shown in Fig. 12.22, the error in velocity is used to generate current proportional
to the error. The error signal is integrated and added to the original signal. The bond
graph for the system in Fig. 12.22 is shown in Fig. 12.23. The C24 element integrates
the current proportional to the error. The integrated error is fed back as a source of
effort on bond 23.
APPROACHING CONTROL SYSTEMS 350
Fig. 12.22
Fig. 12.23
Fig. 12.24 shows two cars are moving in the same direction. They are connected by
means of a spring. The motions of the cars are sustained by applying force on the
second car. It is proposed to control the velocity of the two cars by picking up the
velocity of the first car and applying an effort which is proportional to the error in
velocity to the second car. The bond graph for the system is shown in the Fig. 12.25
which can be reduced to the one shown in Fig. 12.26. The signal flow graph for
Fig. 12.26 is shown in Fig. 12.27.
Fig. 12.24
351 MODELING AND SIMULATION OF ENGINEERING SYSTEMS THROUGH BOND GRAPHS
Fig. 12.27
Using the signal flow graph of Fig. 12.27 and Mason's gain formula we can write
down the transfer function between f8 and e1 as
k
f8 s m2 m
e1 s
.
1 1 k
s3 k s
m2 m1 m1m2
1 1
s3 k s 0 .
m2 m1
1 1
or s 0 , i k .
m2 m1
APPROACHING CONTROL SYSTEMS 352
1 1 k
s3 k s 0
m
2 m1 m2 m1
1 1
s3 : 1 k 0
m2 m1
k
s2 : 0
m2 m1
k
s : 0 0
m2 m1
k
s0 : 0 0
m2 m1
Clearly the system is unstable and introducing velocity feed back in this case does
not solve the problems. The problem can be tackled by introducing additional
damper in between the two masses as shown in Fig. 12.28.
Fig. 12.28
353 MODELING AND SIMULATION OF ENGINEERING SYSTEMS THROUGH BOND GRAPHS
In this chapter the way to approach classical control theory starting with bond graph
models is presented. The very idea of causality has been enlarged by Gawthrop [22,
23, 24, 25] to address to the problems of inversion. Dauphin-Tanguy and her co-
workers [1, 61, 62] have used and extended the bond graph theory to handle
advanced ideas in analysis and design of control systems. Readers interested in these
aspects of bond graph may refer to their work. Linkens and Wang [40] have used
bond graph models for intelligent supervisory controls. This is a new application of
this tool.
Problems
P1. A rotor is used to draw viscoelastic fiber which is produced at certain rate
by the process of polymerization as shown in Fig. P12.1. The tension in the
fiber is being controlled by adjusting the armature voltage of a separately
excited field DC motor.
Fig. P12.1
Fig. P12.2
Air flows past the bluff body. The net influence of the air flow is that the
body experiences a force in vertical direction which is proportional to its
speed in vertical direction and in the same direction. This force then may be
treated as an influence of negative damper of coefficient of - 1.0. Now the
damper attached to the foundation Rf is to be designed such that the system
is not only stabilized but has maximum possible damping. Create root locus
diagram to arrive at the value (Hint: A damper attached to a massive body
and inertial frame may be treated as a feedback force applied to the body
which is proportional to its velocity).
Fig. P12.3
To control the velocity of the vehicle a velocity sensor picks up the velocity
of mass m2 and a feedback amplifier of gain is used to establish the
control. The nominal parameters are, to a suitable scale, m1 = m2 = 1, Kc =100
and Rc = Re = 0.2.
Fig. Pr 12.1
Create bond graph model of the main plant and compensator and obtain
their suitable transfer function through their signal flow graphs. Synthesize
the total transfer function to obtain (either by Routh's criterion or by root
locus diagram) a suitable value of R2 such that system becomes stable for
any positive value of feedback gain .
APPROACHING CONTROL SYSTEMS 356
Fig. Pr 12.2
The displacement sensor outputs voltage proportion to the displacement of load x1.
The constant of proportionality, = 1. Likewise the displacement command is also
set in voltage with the same proportionality constant. Other parameters are
K b 1 10 5 N / m, Rb 200 Ns / m, Ra 5, radius of the wheel R = 0.25m,
the crane mass Mc =1000 kg, load mass ML = 200 kg, hook stiffness Kh = 1 105 N/m,
hook damping coefficient Rh = 200 Ns/m, polar moment of inertia of the wheel = 2.33 kg
m2. The motor characteristic is 10 ia where is shaft torque and ia is
armature current.
The integral gain is set at 0.5. Obtain the critical value of proportional gain p by
creating a bond graph model of the main plant and then obtaining the transfer
function between input voltage to the motor and the displacement of the load. Then
append the control strategy to it. Derive the equations of motion of the system
equation taking every aspect into account. Simulate the response on the system.