2016 Conference of Basic Sciences and Engineering Studies (SGCAC)

Bond Graph Models for Human Behavior

Abdelrhman Mahamadi and Shivakumar Sastry
Department of Electrical and Computer Engineering
The University of Akron, Akron, OH 44325 USA

Abstract—Advances in technology have led to rapid increase

in the number and the complexity of engineered systems. Conse- Element Description Notation
quently, the need for effective tools and techniques for designing,
implementing and analyzing such systems has increased. Bond Se Sources, e.g. electric mains (voltage Se
Graphs were proposed as domain independent approach for Sf source), gravity (force source), pump (flow
modeling dynamic systems in 1960. This approach is a unifying Sf
methodology to represent and analyze systems in which there is source).
energy exchange and, hence, one can represent, validate, analyze
and generate models for the behavior of electrical, mechanical, R Resistor dissipating free energy, e.g. R
chemical, fluid or hydraulic system. In this short paper we electric resistor, mechanical friction.
describe how to create bond graphs for a system with a focus
on modeling Human Behavior. We describe how a model can R
be used to study the energy transfers in the system and how to
obtain a mathematical model for the dynamic behavior of the
C Storage element for a effort-type variable,
system from the bond graph model. e.g. capacitor (stores charge), spring C
I. I NTRODUCTION (stores displacement).
Bond graphs are a domain independent graphical models
that describe the dynamic behavior of physical systems [1]. I Storage element for a flow-type variable,
These models were introduced by [2] as a unifying notation e.g. inductor (stores flux linkage), mass I
for systems involving mechanical, electrical and hydraulic
components that involve energy exchange. These models are (stores momentum).
constructed by first considering energy flows between the
ports of the components of a system. The physical effects TF Transformer, e.g. an electric transformer, n
and their interactions are considered initially in a qualitative TF
toothed wheels, lever.
manner and the process details are specified in later stages of n
the model development process [3]. The physical components TF
of systems are represented as vertices of the bond graph.
GY Gyrator, e.g. electromotor, centrifugal r
Every component is said to have specific ports through which
energy is exchanged. Components are connected using bonds pump. GY
— and every bond represents a bilateral exchange of effort r
and flow [4]. GY
II. B OND G RAPH N OTATIONS
0-Junction 0– and 1–junctions, for ideal connecting
A domain neutral, common, notation for the components 1 0
1-Junction two or more sub-models.
of a system is an important foundation for bond graphs. This
notation allows systems involving mechanical, electrical and
hydraulic components to be analyzed in a common frame-
work [5]. Figure 1 shows the commonly used elements in a
bond graph. Fig. 1: Commonly used Bond Graph Elements
III. C REATING B OND G RAPHS
The topology of the system, i.e., the connectivity between 2) Identify distinct efforts and represent each by a 0-
the physical components of the system, guides the construction junction; a distinct effort corresponds to a point in the
of a bond graph model [3]. The following procedure can be system where there is difference in the level of the effort.
used 3) Connect sources, storage elements, dissipator elements,
1) Enumerate the physical components in the system and transformers or gyrators using a 1-junction between the
identify a corresponding bond graph element illustrated two corresponding 0-junctions.
in Figure 1. 4) Add half arrows to all bonds; the direction of the arrow
978-1-5090-1812-3/16/$31.00 2016
c IEEE must identify the direction of energy transfer in the bond.

110
 5) In bond graphs of electrical systems, choose a potential
as a reference; eliminate its corresponding 0-junction
along with all incident bonds. If two sub-circuits of
an electrical network are connected via an isolating
transformer, a reference potential must be chosen in each
sub-circuit.
6) Simplify the bond graph.
One of the main utilities of a bond graph model is that
it enables the designer to systematically analyze the energy
transfers in the system. This process is called causality anal-
ysis.
IV. C AUSALITY A NALYSIS
The objective of causality analysis is to determine whether
the flow of energy in the system is consistent [1]. The bond
graph methodology admits four types of causality and by
following these rules, one must assign causality to the bonds.
Causality is denoted by a stroke (“|”) on one end of the bond.
The flow of energy in a bond is from the component on the
bond without the stroke to the element that has the stroke.
Note that half arrows only denote whether the corresponding
physical component is generating or consuming energy. The
details of assigning half arrows and causal strokes will be
illustrated with a specific example after describing the four
types of causality that must be considered.
1) Fixed Causality: The energy flow is constrained in one
specific direction. For example, when the element at one
end of a bond is a Source of Effort, the causal stroke
must be at the other end, i.e., the flow is assumed to be
out of the source of effort and into the element at the
other end of the bond.
2) Constrained Causality: In bond graph elements such
the T F , GY , 0−Junction and 1−Junction, there is
a physical relation between the different ports of the
element because the causality of one port imposes the
causality on the other ports.
3) Preferred Causality: The storage elements C and I
could be represented in either of integration or differen-
tiation forms. The integration form is preferred because
it forces the designer to specify an initial condition for
the elements.
4) Indifferent Causality: In this case, the element does
not impose any constraints on the causality. For example,
in the case of a Resistor element, the flow can either be
into the element or away from the element.
When carrying our causality analysis, it is necessary to
first start with the components that have fixed causalities.
Usually, these are the sources. Next, it is necessary to
check if any of the assigned causalities affect the elements
with constrained causality, i.e., T F , GY or 0−Junction and
1−Junction. If such connections exist, the assigned causalities
must be propagated through these elements. The next step is
to consider the components with preferred causality. If the
preferred causalities can be assigned without violating any
of the assigned causalities; any changed causalities must be
111
propagated through the Junctions. Finally, it is necessary to
check if there are unassigned causalities and the strokes are
assigned to the indifferent causality elements.
Causality analysis can serve as a diagnosis tool that can
check whether or not the systems are modeled properly. When
the causalities assigned are inconsistent with the rules, it
usually denotes that a model change may be warranted.
V. DYNAMIC E QUATIONS
The dynamic equations of an energy system can be derived
from the bond graph of that system. These equations are
written in terms of the states of the system. Generally, there are
two kinds of states P and Q for each bond that supplies energy
to a storage element; the P state represents the integration of
the effort variable and is usually used with the I type storage
elements; the Q state is the integration of the flow variable
and it is used with the C type storage elements.
The concept of a strong bond is important and is related to
the Junction elements. Only one bond in a 1−Junction must
supply the flow, i.e., only one bond should have the causality
stroke outwards. Similarly, only one bond in a 0−Junction
must supply the effort and that bond should have the causality
stroke inwards. These unique causality bonds at the junctions
are called strong bonds.
The procedure of deriving the dynamic equation from the
bond graph is described below.
1) Start with the Sources, Inductors, Capacitors and Resis-
tor elements and write their equations according to the
causalities.
2) Resolve the unknown values using the values across the
strong bonds.
3) Write the equations of the two ports elements, i.e., T F
and GY , and the Junction elements for the strong bonds
variables.
4) Resolve the variables in the right hand side by using the
states variables and system parameters.
5) Check to see if all the equations are completely reduced.
If not, it means there is some loop in the system and this
can be resolved by solving these equations as a system
of linear equations.
VI. I LLUSTRATIVE E XAMPLE FOR M ODELING H UMAN
B EHAVIOR
We believe that human behavior involves complex en-
ergy transfers across multiple domains. In the context of
our ongoing investigation into decision-support systems for
wellness management [6], there is a keen interest to develop
actionable models for human behavior that can guide the
decision-support. Recently models for human behavior have
been developed using a fluid analogy paradigm [7]. To explore
the potential of bond graphs for this domain, we present a
model for the interaction between a Client and a Therapist
that is inspired by the work in [8].
A. Client Therapist Interaction
Figure 2 illustrates a fluid analogy model that represents
the interaction between a client and a therapist. Following the
ideas in [8], we represent the valence or affect of the client
and the therapist as the levels of the two tanks, i.e., I1 and I2 .
The valence or affect of the therapist is a function of his or
her training and is represented by the valve N1 . We assume
that a better trained therapist, i.e., more flow in N1 , would
have higher valence. The valence of the client is affected by
the environmental conditions as represented by N2 . Through
the interaction, the valence of the client and the therapist are Fig. 3: Distinct Efforts in the Client Therapy Interaction Model
affected as indicted by the valves labeled Therapy (R1 ) and
Feedback (R2 ).

Fig. 4: Step 2
Fig. 2: Fluid Analogy Model for the interaction between a
Therapist and a Client.
5) The simplified bond graph is shown in Figure 7. Here,
junctions that had only one input and one output were
B. Bond Graph for Human Behavior Model removed.
Applying the procedure mentioned above for bond graphs
creating step by step as follows. C. Causalities Assignment
Using the causality assignment procedure described in Sec-
TABLE I: Components of the System and Corresponding Bond tion IV, the bond graph shown in Figure 8 was obtained.
Graph Elements D. Dynamic Equations
Physical Element Type Value The dynamic equations were derived for this system us-
Training source Se S1
Environment source Se S2 ing the procedure described in Section V. As illustrated in
Training valve TF N1 Figure 9, a unique number is assigned to each bond in the
Environment valve TF N2 system.
Therapist valence I I1
The dynamic equations were derived using these assign-
Client valence I I2
Therapy R R1 ments. In the following ei is the effort variable associated
Feedback R R2 with bond i and fi is the flow variable associated with bond
i.

1) The components of the system and corresponding bond

graph elements were identified as shown in Table I.
2) All the distinct efforts in the client therapist interaction
model were identified as shown in Figure 3. Each effort
was represented with a unique 0−Junction as illustrated
in Figure 4.
3) Add all the other components in the system us-
ing 1−Junctions between the two corresponding
0−Junctions as illustrated in Figure 5.
4) Add half arrows to the bonds to indicate whether the
element is generating or consuming energy. This is Fig. 5: Step 3
illustrated in Figure 6.

112
 2) Resolve unknown values using the values across the
strong bonds in the Junction elements
e2 = N1 × e1 = N1 × S1 (7)
f1 = N1 × f2 (8)
P3
f2 = f3 = f4 = f12 = (9)
M3
e3 = e2 − e4 + e12 = N1 × S1 − e4 + e12 (10)
e4 = e5 = e6 = R5 × f5 (11)
P3
Fig. 6: Step 4 f5 = f4 − f6 = − f6 (12)
M3
P7
f6 = f7 = f8 = f10 = (13)
M7
e7 = e6 + e8 − e10 = R5 × f5 + e8 − e10 (14)
e8 = N2 × e9 = N2 × S2 (15)
P7
f9 = N1 × f8 = N2 × (16)
M7
e10 = e11 = e12 = R11 × f11 (17)
P7 P3
f11 = f10 − f12 = − (18)
M7 M3
Fig. 7: Step 5
3) Write equations for the two port elements and Junction
elements.
1) Precisely specify the effort and flow values related to e1 = S1 (19)
the Sources, Inductors, Capacitors and Resistors. P3
f3 = (20)
M3
e1 = S1 (1)
P3 P7
P3 e5 = R5 × f5 = R5 × ( − ) (21)
f3 = (2) M3 M7
M3
P7
e 5 = R5 × f5 (3) f7 = (22)
P7 M7
f7 = (4) e9 = S2 (23)
M7
P7 P3
e9 = S2 (5) e11 = R11 × f11 = R11 × ( − ) (24)
M7 M3
e11 = R11 × f11 (6) e2 = N1 × e1 = N1 × S1 (25)
P3
f1 = N1 × f2 = N1 × (26)
M3
P3
f2 = f3 = f4 = f12 = (27)
M3
(28)

e3 =e2 − e4 + e12 =
P3 P7 (29)
Fig. 8: Causality Assignment N1 × S1 − (R5 + R11 ) × ( − ).
M3 M7
P3 P7
e4 = e5 = e6 = R5 × ( − ). (30)
M3 M7
P3 P7
f5 = f4 − f6 = − . (31)
M3 M7
P7
f6 = f7 = f8 = f10 = . (32)
M7
e7 =e6 + e8 − e10 =
Fig. 9: Assigning numbers to the bonds P3 P7 (33)
N2 × S2 + (R5 + R11 ) × ( − ).
M3 M7

113
 e8 = N2 × e9 = N2 × S2 (34)
P7
f9 = N1 × f8 = N2 × . (35)
M7
P7 P3
e10 = e11 = e12 = R11 × ( − ). (36)
M7 M3
P7 P3
f11 = f10 − f12 = − . (37)
M7 M3
4) Express all the variables in terms of the state variables
and system parameters.
d
(P3 ) = e3 =
dt (38)
P3 P7
N1 × S1 − (R5 + R11 ) × ( − ).
M3 M7
d
(P7 ) = e7 =
dt (39)
P3 P7
N2 × S2 + (R5 + R11 ) × ( − ).
M3 M7
Equations 39 and 38 describe the dynamics of how the
Fig. 11: Client valence after going to an untrained therapist
client’s and the therapist’s valence changes over time. These
dynamic equations can be used to simulate the interaction in a
stand alone system or a decision-support system such as [6].
VII. R ESULTS
In order to test the bond graph mechanism that we are
using to model humans behavior, we will choose some values
to all the variables of the therapist-client model and check
how the client’s valence changes in a response to the therapy.
We consider three cases, first e assume that, the environment
decreases the valence of the client. In the first case we choose
equal values for the two valves to indicate that the same
amount of received therapy equals to the negative feedback
from client. The second case consider a well trained therapist,
with a flow of therapy higher than the negative feedback. The
last case depicts an untrained therapist, in which the therapy
rate is lower than the negative feedback. For these three cases
we maintain a constants to all the other variables.

Fig. 12: Client valence after going to a well trained therapist

We can conclude form Figures 10, 11 and 12 that the

Fig. 10: The valences of the therapist and the client under
equal flow exchange of therapy and feedback.
results are as expected. Notice that when the client visits a
neutral therapist, his valence decreases linearly as expected
just under the environment effects. Further, when the client
visits an untrained therapist, his valence decreases rapidly. On
the other hand if the client visits a well trained therapist, the
therapy will overcome the negative effects of the environment,
hence the valence of the client will increase.

114
 VIII. C ONCLUSION AND F UTURE W ORK
Bond Graphs are powerful models for representing and
reasoning about energy transfers in systems. We described a
method for constructing a bond graph using the example of
human interaction. The model was used as a basis to carry
out causality analysis and also derive dynamic equations that
described the behavior of the system.
In the future, we plan to extend these models to represent
and reason about more complex behaviors involving indi-
viduals and groups of individuals. This approach can also
be extended to incorporate well-known psychology theories
such as the Theory of Planned Behavior and Social Cognitive
Theory for human behavior.
R EFERENCES
[1] J. F. Broenink, “Introduction to physical systems modeling with bond
graphs,” in in the SiE whitebook on Simulation Methodologies, 1999.
[2] H. M. Paynter, Analysis and design of engineering systems. M.I.T. Press,
Cambridge, 1961.
[3] W. Broutzky, “Bond graph modelling and simulation of multidisciplinary
systems - an introduction,” Simulation Modelling Practice and Theory,
vol. 17, pp. 3–21, 2009.
[4] P. C. Breedveld, “Modeling and simulation of dynamic systems using
bond graphs,” in Control Systems, Robotics and Automation - Modeling
and System Identification I, EOLSS Publishers Co. Ltd./UNESCO, 2008.
[5] P. Gawthrop, “Bond graphs: A representation for mechatronic systems,”
Mechatronics, vol. 1, pp. 127–156, 1991.
[6] M. k. Chippa, S. M. Whalen, F. L. Douglas, and S. Sastry, “Goal-
seeking formulation for empowering personalized wellness management,”
in Medical Cyber Physical Systems Workshop, 2014.
[7] C. A. Martin, D. E. Rivera, W. T. Riley, E. B. Hekler, M. P. Buman,
M. A. Adams, and A. C. King, “A dynamical systems model of social
cognitive theory,” in American Control Conference, 2014.
[8] L. Liebovitch, P. Peluso, M. Norman, J. Su, and G. J.M., “Mathematical
model of the dynamics of psychotherapy.,” Cognitive Neurodynamics.,
vol. 3, pp. 265–275, 2011.

115