Bond graph explained

A bond graph is a graphical representation of the energy flows though and between physical dynamical systems including those in the electrical, mechanical, hydraulic, thermal and chemical domains. It is used to model and analyse systems relevant to engineering and to systems biology.

Basic ideas

Because the concept of energy is common to all physical domains, the bond graph provides a unified description of all of these energy domains and can be thought of as a systematic use of the physical analogies introduced by the 19th century scientists James Clerk-Maxwell and Lord Kelvin.  The mechanical–electrical analogy is one example of a physical analogy.

Bond graphs use the concept of analogous power conjugate variables whose product is energy flow, or power; these variable pairs are called effort and flow and, for example, correspond to voltage and current in the electrical domain and force and velocity in the mechanical domain. These power conjugate variables are transmitted by bonds which connect bond graph components.

Bond graph  components are also based on analogies and, using the electrical and mechanical domains as examples,  include the  C component to represent  both mechanical spring and electrical capacitor, the  I component to represent both a mechanical inertia and  an electrical inductor and the R component to represent both mechanical damper and electrical resistor.

The electrical circuit notions of parallel and series connections are abstracted as 0-junctions and 1-junctions in bond graph terminology and again used as connection analogues for each physical domain.

The bond graph transformer  (TF) and gyrator (GY) components represent energy transformation within and between domains; thus an ideal gearbox in the rotational mechanical domain is represented by the  TF component  and an ideal DC motor transforming electrical into mechanical energy is represented by a GY component. Non-ideal transducers with flexibility, inertia and friction are modelled by including C, I and R components.

The concept of causality in the context of bond graphs is used not only to generate system equations in a number of forms including ordinary differential equation (ode), state-space and differential-algebraic equations (dae)  form suitable for simulation purposes but also to investigate dynamical system properties such as invertibility and zero dynamics. Causality can also be used to guide and correct modelling choices.

The bond graph use of energy flows leads to the systematic construction of hierarchical models of large multi-domain systems; thus the bond graph method provides a basis for constructing large computer models, or digital twins, of multi domain physical systems including systems relevant not only to engineering but also to systems biology and the life sciences.

The bond graph approach is related to the  behavioral modelling approach of Jan C Willems, and the port-Hamiltonian approach of Arjan van der Schaft and B. M. Maschke.

The bond graph method was originally proposed by Henry Paynter.

Analogies

The importance of analogies between physical domains was noted by Lord Kelvin and James Clerk-Maxwell.[1] [2] The bond graph can be thought of as a systematic approach to analogies. This section emphasises key features of bond graph analogies; more detail appears in a number of textbooks and tutorial papers.[3] [4] [5] [6]

As an introduction to the key features of bond graphs, the figure shows the bond graph of two analogous systems: one electrical and one mechanical. A brief description is given here and expanded in the following sections.

The bond graph uses three classes of analogy: analogies between variables, analogies between components and analogies between component connections; these are discussed in the following sections.

Analogies between variables

DomainEffort (units)Flow (units)
ElectricalVoltage (V)Current (A)
TranslationalForce (N)Velocity (m/s)
RotationalTorque (Nm)Angular velocity (rad/s)
HydraulicPressure (Pa)Volumetric flow (m^3/s)
ThermalTemperature (K)Entropy flow ((J/K)/s)
ChemicalChemical potential (J/mol)Molar flow (mol/s)
The bond graph use energy flow, or power, as the basis for abstracting analogies between different physical domains. Power conjugate variables are a pair of variables whose product is power and a list of these appears in the table for various physical domains. As indicated in the table, the bond graph uses the effort/flow analogy to categorise each of the two conjugate variables; the across/through analogy is also possible but not commonly used.[7] [8] The conventional bond graph symbol for effort is e, and that for flow is f.

As well as the two conjugate power variables e and f, the bond graph uses two integrated variables p and q where:

p=\intte(\tau)d\tau~~~and~~~q=\inttf(\tau)d\tau

equivalently:

•
p

=

dp
dt

=e~~~and~~~

•
q

=

dq
dt

=f

The harpoon symbol shown in the figure represents a power bond, or bond,  transferring energy; the harpoon direction corresponds to positive energy flow and corresponds to a sign convention. The conjugate variables are displayed on each bond for the purposes of illustration.

Analogies between components

The definitions relating p to e and q to f are indicated diagrammatically. Physical properties are encapsulated in constitutive equations relating the energy and power variables.  In the diagram, C represents the constitutive equation relating q and e,  I represents the constitutive equation relating p and f and  R represents the constitutive equation relating e and f. Instead of placing e, f, p and q on the four corners of a square as in the diagram, they can alternatively be placed at the four vertices of a tetrahedron; such a diagram is called the tetrahedron of state.

These three constitutive equations

\PhiC,\PhiI~and~\PhiR

correspond to three components each relating e and f: two dynamic components C and I incorporating an integrator and the R component. The C and I components store, but do not dissipate energy, R dissipates, but does not store energy. It is also possible to define a memristor component linking p and q, but this component is not commonly used.[9]

\PhiC,\PhiI~and~\PhiR

may be linear or nonlinear functions relating the variables as:

q=\PhiC(e),~~p=\PhiI(f)~~and~~e=\PhiR(f)

in the linear case:

q=Ce,~~p=If,~~e=Rf

where

C,~I,~and~R

are scalar constants representing generalised capacitance, inertia and resistance respectively.

Because e and f are conjugate variables, they are carried on a single bond and therefore the three components C, I and R are connected to a single bond and thus have a single energy port though which energy flows. The components, and impinging bond are shown in the figure. By convention, bonds point into these three components.

Notation

The colon (:) notation is sometimes used to refer to the components; thus, for example I:M, C:K and R:D could be used in the bond graph of the mass-spring-damper system to emphasise the link between the bond graph components and their physical analogues.

Analogies between connections

Electrical circuit diagrams have two sorts of connection: parallel and series or common voltage and common current; they distribute, but do not store or dissipate energy. The bond graph analogy of the common voltage  and common current connections are the 0-junction (common effort) and 1-junction (common flow) respectively; they both distribute, but do not store or dissipate energy.

All bonds impinging on a 0-junctions have the same effort. As energy is distributed, not dissipated, it follows that the sum of energy inflows (indicated by bonds pointing in) must equal the sum of energy outflows (indicated by bonds pointing out). Hence, if the common effort is e, the power flow constraint implies that the sum of the   

m

inflows
in
f
i
must equal the sum of the

n

 outflows
out
f
j
:
m
\sum
i=1
in
f
i

=

n
\sum
j=1
out
f
j

Using the same argument, the efforts impinging on a 1-junction are constrained by:

m
\sum
i=1
in
e
i

=

n
\sum
j=1
out
e
j

Analogies between external connections

Systems such as that the simple electrical and mechanical systems in the figure have no connection to the environment. Such connections may include external voltages and external forces (that is efforts) and external currents and external velocities (that is flows). Similarly, external measurements of efforts and flows are also important. For the purposes of building hierarchical system, it is convenient to define energy ports though which energy can flow between systems. These five possibilities correspond to five bond graph components:

Se

, an effort source analogous to applying external voltages and external forces

Sf

, a flow source analogous to applying external currents and external velocities

De

, an effort sensor (detector) analogous to measuring voltage and forces

Df

, a flow sensor (detector) analogous to measuring currents and velocities

SS

, a source/sensor component which acts both as

Se

-

Df

and as

Sf

-

De

pairs as well as an energy port for external connections.For example, the mass-spring-damper system can be augmented with an force

F

applied to the mass and a measurement of the mass velocity

v

using a single SS components. The colon notation has been included; the SS:io refers to the fact that the force and velocity pair could be considered as system input and output.

Analogies between energy transducers

The conjugate effort and flow variables have different units in each energy domain thus models in one domain cannot be directly be connected by bonds to a different energy domain. However, because power has the same units (J/s or W) in each domain, the two-port power transducing components TF and GY can be used to provide such connections. As the two components transmit, but do not store or dissipate power, it follows that the power associated with the conjugate variables of the left-hand and right-hand bonds must be the same:

e2f2=e1f1

Each component has a modulus

m

associated with it. In the case of the TF component:

e2=me1~~and~~f1=mf2

In the case of the GY component:

e2=mf1~~and~~e1=mf2

For example, a frictionless, massless piston of area

A

converts hydraulic power to mechanical power so that the hydraulic pressure

P

is related to mechanical force

F

by:

F=AP

and hydraulic flow

V

to piston velocity

v

by:

V=Av

Thus the bond graph analogy is the TF component with modulus

m=A

where

e1=P,~~f1=V

and

e2=F,~~f2=v

.

For example,  an ideal DC motor converts electrical power into rotational mechanical power so that the mechanical torque

T

is related to the electrical current

i

by:

T=ki

and back EMF (voltage)

E

to angular velocity

\Omega

by

E=k\Omega

Thus the bond graph analogy is the GY component with modulus

m=k

where

e1=E,~~f1=i

and

e2=T,~~f2=\Omega

.

In both cases, non-ideal transduction behaviour can be modelled by including C,I and R bond graph components in the model.

Bond graphs have been used to model systems relevant to the life sciences, including physiology and biology.[10] In particular, the use of bond graphs to model biophysical systems was introduced by Aharon Katchalsky, George Oster, and Alan Perelson in the early 1970s.[11] [12] More recently, these ideas were used in the context of Systems Biology to provide an energy-based approach to modelling the biochemical reaction systems of cellular biology[13] [14] [15] and to modelling the entire physiome.[16]

The bond graph approach has a number of features which make it a good basis for building large computational models of the physiome.

Variables

The bond graph variables for biochemical systems are:

x

(mol)

v

(mol/s)

\mu

(J/mol)[22]

Note that the product of effort and flow (μv) is, as always in the bond graph formulation, power (J/s).

Components

As detailed below, the main features of the components used to model biochemical systems are: the R and C components are nonlinear, there is no I component required and the R component is replaced by a two-port Re component.

Junction components

The bond graph zero (0) and one (1) components are no different in this context.

C component

The C component integrates the flow

v

to give the amount

x

of species:

x(t)=\inttv(\tau)d\tau

The effort, chemical potential

\mu

, is given by the formula:

\mu=\mu0+RTln

x
x0

where

\mu0

is the chemical potential corresponding to

x=x0

,

R

is the gas constant and

T

is the absolute temperature in degrees Kelvin.

The formula for

\mu

can be rewritten in a simplified form as:

\mu=RTlnKx

where

K=

1
x0

\exp

\mu0
RT

Because of the special form of this particular C component it is sometimes given a special name Ce analogously to the special Re component.

Re component

The Re (reaction) component has two energy ports corresponding to the left (forward) and right (reverse) sides of a chemical reaction. The forward

Af

and reverse

Ar

affinities are defined as the net chemical potential due to the species on the left and right sides of the reaction respectively. The Re component then gives the reaction flow

v

as:

v=\kappa\left(\exp

Af
RT

-\exp

Ar
RT

\right)

where

\kappa

(mol/s) is a rate constant.

Note that it is not possible to use the usual R component with 1-junction formulation as the flow depends on both the forward

Af

and reverse

Ar

affinities rather than the difference

Af-Ar

.

Modelling simple reactions

Reaction A <=> B

Sources:The simple reaction A <=> B is represented by three components:

\muA

; the flow is

-v

.

\muB

; the flow is

v

.

v

, forward affinity

Af=\muA

and reverse affinity

Ar=\muB

.

Using the above equations, the flow

v

is given by

v=\kappa\left(\exp

\muA
RT

-\exp

\muB
RT

\right)=\kappa\left(KAxA-KBxB\right)

where the subscripts correspond to the species. This is the simple mass-action equation:

v=k+xA-k-xB

where

k+=\kappaKAandk-=\kappaKB

.

Enzyme-catalysed reaction

As discussed in section 1.4 of Keener & Sneyd, an enzyme-catalysed reaction reversibly transforming species A to species B via enzyme E and enzyme complex C can be written as the pair of reactions:

A + E <=> C <=> B + E

The enzyme complex C is formed from A + E and decomposes into the species B and releases enzyme E. The bond graph shown in the figure shows how the enzyme is recycled.

The bond graph can be used to derive the properties of these reactions which are of generalised Michaelis-Menten form.

Energy transduction

The bond graph TF (transformer) component represents energy transduction either within or between energy domains. (Note that the TF component has been called the TD (transduction) component - TF is more widely used.) This section focuses on transduction between the chemical domain with effort

\mu

(J/mol) and flow

v

(mol/s) and a generic domain with effort

e

and flow

f

.

The key feature of the TF component is that it transmits energy without dissipation; hence, with reference to the figure:

ef=\muv

The transformer has a modulus m (with appropriate units) so that:

f=mv

the energy formula then implies that:

\mu=me

Stoichiometry

The stoichiometry of a chemical reaction determines how many of each chemical species occurs. Thus, for example, the reaction A <=> m B converts one mol of species A to m mol of species B.

The case where

m=1

corresponds to the simple reaction of the first example above. Using the same approach for general

m

, the reaction flow is:

v=\kappa\left(\exp

\muA
RT

-\exp

m\muB
RT

\right)=\kappa\left(KAxA-(KB

m
x
B)

\right)

Chemoelectrical transduction

This section looks at the case where the generic domain is the electrical domain so that effort is (electrical) voltage

l{V}

(

e=l{V}

) and the flow is current (

f=i

). Consider the flow

v

of charged ions where the charge on the molecule is

z\epsilon

(Coulomb) where

\epsilon

is the charge on the electron measured in Coulomb; the charge associated with a mole of ions is thus

z\epsilonNA

where

NA

is the Avogadro constant. The equivalent current is then

i=z\epsilonNAv=zl{F}v

where

l{F}=\epsilonNA

is the Faraday constant; thus the corresponding TF modulus is:

m=zl{F}

(C/mol)

Again, it follows that

\mu=me=zl{F}l{V}

In this context, the bond graph TF component can be used to model energy flows associated with action potential,[23] membrane transporters,[24] cardiac action potential,[25] and the mitochondrial electron transport chain.[26]

Chemomechanical transduction

Consider a long rigid molecule such as actin where a sub unit of length

\delta

(m) is added at a rate of

v

(mol/sec). Then the tip velocity

V

is given by:[27]

V=\deltaNAv

where

NA

is the Avogadro constant.

Thus the modulus

m=\deltaNA

(m/mol) and

\mu=mF=\deltaNAF

where

F

is the corresponding force at the tip.

These formulae have been used to generate force/velocity curves for actin filaments. The approach provides a useful alternative to the Brownian Ratchet approach[28] as the bond graph TF component can be potentially used with modular bond graph models of cellular systems.

Applications

A number of systems relevant to systems biology have been modelled using bond graphs. These include:

Causality

Causality is a word with many uses and connotations. In the context of bond graphs, however, it has a limited, precise but important meaning and allows the bond graph model of a system to be converted to various other forms including a (nonlinear) state-space representation. The causality concept can also be used to examine structural properties, including inversion, of the system represented by a bond graph as well as to expose modelling errors.

R, C and I components

The concept of causality  is visualised via the causal stroke notation. This notation is introduced in the three figures where the R, C and I components are connected to a bond augmented by the causal stoke: a short line perpendicular to the bond and located at either (but not both) end of the bond. (For clarity, the figures correspond to linear components; in the nonlinear case,

Rf

is replaced by

\PhiR(f)

and

e/R

by
-1
\Phi
R

(e)

and similarly for the C and I components. Whereas an acausal (without strokes) bond graph represents a set of equations (where the left and right sides of an equation can be swapped without change of meaning), a causal (with strokes on each bond) represents a set of assignment statements whereby the value of the left-hand side of the assignment statement (represented here by :=) becomes the value of the expression on the right-hand side of the assignment statement. Thus, for example, the constitutive equation of a linear resistor can be written as

e=Rf

and

f=e/R

without changing the meaning; but in contrast, the two assignment statements e:=Rf and f := e/R are different. In particular, in the first case, f must be known to compute e and in the second case, e must be known to compute f.

The assignment statement representation can be graphically visualised as a block diagram where each assignment statement is represented as a block with input representing the right-hand of the assignment statement and output representing the left-hand side of the assignment statement. The block diagrams for each causality are shown in the figures for each component. Note that each causality of a component leads to a different block diagram.

R component

The figure shows the causality of the R component with linear constitutive equation. (a) Flow is imposed on R and R imposes effort; this corresponds to the assignment statement e := Rf and the block diagram. (b) Effort is imposed on R and R imposes flow; this corresponds to the assignment statement f := e/R and the corresponding block diagram.

C component

The figure shows the causality of the C component with linear constitutive equation. (a) Flow is imposed on C and C imposes effort; this corresponds to the assignment statements

e:=q/C~~and~~q:=\inttf(\tau)d\tau

and the corresponding block diagram; this is called integral causality. (b) Effort is imposed on C and C imposes flow; this corresponds to the assignment statements

q:=Ce~~and~~f:=

dq
dt

and the corresponding block diagram. This is called derivative causality.

I component

The figure shows the causality of the I component with linear constitutive equation. (a) Effort is imposed on I and I imposes flow ; this corresponds to the assignment statements

f:=p/I~~and~~p:=\intte(\tau)d\tau

and the corresponding block diagram. This is called integral causality. (b) Flow is imposed on I and I imposes effort; this corresponds to the assignment statements

p:=If~~and~~e:=

dp
dt
and the corresponding block diagram. This is called derivative causality.

Source-sensor components

The SS (source sensor) component acts  as an effort source (

Se

), flow detector (

Df

) combination when the causal stroke is distant from the SS component and vice versa. The figure shows the causality of the SS (source/sensor) component. (a) The SS acts as an effort source (

Se

) flow detector (

Df

) combination. (b) The SS acts as a flow source (

Sf

), effort detector (

De

) combination.

Junctions

As, by definition, all efforts associated with bonds impinging on a 0-junction are the same, it follows that exactly one bond can impose effort causality. Similarly, all flows associated with bonds impinging on a 1-junction are the same, it follows that exactly one bond can impose flow causality. Thus if a bond imposes effort causality on a 0-junction, the junction imposes effort on the other bonds and if  a bond imposes flow causality on a 1-junction, the junction imposes flow on the other bonds.

Causal propagation

When one-port components (sources, C, I and R) are connected by a junction structure consisting of 0-junctions, 1-junctions, TF and GY, the causality assigned to each one port component propagates though the junction structure because of the causal constraints imposed by the junction structure components.This propagation can be applied systematically using the sequential causal assignment procedure (SCAP):

  1. Choose any source (SS, Se or Sf)  and assign the required causality. Immediately extend the causal implications though the bond graph as far as possible using the constraints on the junction elements (0 and 1) and the TF and GY elements.
  2. Repeat step 1 for all of the source components.
  3. Choose any storage element  (C or I)  and assign the integral causality. Immediately extend the causal implications though the bond graph as far as possible using the constraints on the junction elements (0 and 1) and the TF and GY elements.
  4. Repeat step 3 until all C and I elements have been assigned causality.

The figure shows the result of this procedure on the simple example. In particular, the I component imposes flow causality onto the 1-junction thus the one junction imposes flow causality on to the other three components: SS, C and R. This means that the C, R and SS components impose effort causality onto the one junction; thus the C component is in integral causality and the R component corresponds to the assignment

eR:=Rf

.

Although this procedure can be accomplished manually for small systems,  a computer-based approach is more generally useful.

There are three possible results of this procedure.

  1. It completes with all C and I components in integral causality and all bond causalities assigned; the resultant system is a causal bond graph and can be converted into a state-space system.
  2. All bonds have causality assigned but one or more C or I component has derivative causality.
  3. All C and I components have integral causality, but some bond do not have causality assigned.

Whereas case 1 leads to ordinary differential equations (ode) in state-space form, case 2 gives rise to differential-algebraic equations (dae). Depending on the context, it may be better to reconsider the underlying physical system to avoid daes, or it may be possible to reduce the dae to an ode[36] or the daes can be solved using an appropriate dae solver.

Alternative causality.

The sequential causal assignment procedure (SCAP) is designed to derive state-space equations, suitable for computation or control systems analysis. However, there are other forms of causal analysis designed to address other issues, including:

State-space equations

A bond graph system representation contains the constitutive equations of each component, embedded in the structure of bond and junctions.

The question of how to manipulate a set of equations into a form suitable for analogue computation was posed, and partially answered by Lord Kelvin.[44] This approach underlies the conversion of a bond graph model to a state-space representation suitable for digital computation.[45]

The bond graph uses the notion of (bond graph) causality to provide a systematic and constructive way to investigate whether a state-space representation exists, and, if so, what is that representation; this causality approach is well suited to computational implementation and has an intuitive representation on the bond graph itself using the causal stroke notation.

A causal bond graph can be put into state-space form if:

The sequential causal assignment procedure (SCAP) provides a graphical approach to determining whether  a system represented by a bond graph  has a state-space representation.

The bond graph is analogous to an number of different systems; for clarity, the mechanical analogue is used as an example, but exactly the same method applied to electrical and other physical domains.

It is convenient to choose the system states as the integrated variables corresponding to the components in integral causality; in the case of the simple example, there are two states:

p

corresponding to the mass M and

q

corresponding to the spring. The system inputs and outputs correspond to the Se, Sf and SS components. In this case, there is one input and one output corresponding to the SS component: the effort (applied force

F

) and the measured flow (velocity

v

). Guided by the causal strokes, the state derivatives (effort applied to I components, flow applied to C components) can be written in terms of the states. Thus:
•
\begin{align} q

&=f=

p\\
m
•
p

&=F-ec-er=F-kq-df=F-kq-d

p
m

\end{align}

This can be written in standard linear state space for as:

•
\begin{align} x

&=Ax+Bu\\ y&=Cx \end{align}

where the state

x

input

u

and output

y

are given by:

x=\begin{pmatrix}q\\p\end{pmatrix}, u=Fandy=v

and the matrices

A

,

B

and

C

are given by:

A=\begin{pmatrix}0&

1
m

\ -k&-

d
m

\end{pmatrix}, B=\begin{pmatrix}0\ 1\end{pmatrix},  C=\begin{pmatrix}0&

1
m

\end{pmatrix}

The state equation has three parameters: the mass

m

, the spring constant

k

and the damping constant

d

.

Using the usual Laplace transform notation of control systems theory, the system transfer function relating the input

u

to output

y

is:

{s\over{\left(k+ds+ms2\right)}}

Although it may be instructive to derive these formulae by hand, it is often convenient to make use of computer algebra to perform the same function.

System inversion

The behavior of a linear system is not only determined by the demominator of the system transfer function but also by the numerator. In the nonlinear case, the concept of the transfer function numerator is replaced by the concept of zero dynamics. Having a physical interpretation of zero dynamics is helpful in designing, and redesigning a dynamical system. For this reason, a bond graph approach to system inversion has been developed.[46] [47] [48]

This procedure is illustrated using the simple mass-spring-damper system bond graph which, although linear, exemplifies the key ideas. In particular, system inversion is performed by reversing the causality on the source-sensor component; thus the role of input and output is reversed. In this particular case, reversing the causality of the SS component implies that the causality of the I component must also be reversed as only one component can impose flow onto the 1-junction. Thus only one component, the C component, remains in integral causality. Thus the zero dynamics are first order which, in this simple case, corresponds to the fact that the transfer function of the original system has a first order numerator.

Following the causal strokes, the remaining state

q

can be written as in terms of the input

v

of the inverse system
•
q

=v

p

is no longer a state, but can be written as:

p=mv

the output

F

of the inverse system is:

F=kp+dv+

•
p

=kp+dv+m

•
v

This approach relies on the system input and output existing on the same SS component - the input and output are said to be colocated. If this is not so, the concept of bicausality must be used.

Bicausality

Although (standard) causality provides a powerful tool to investigate the inverse of a dynamical system described by a bond graph, it is restricted to co-located source-sensor pairs where the input and output reside on a single SS component. To remove this restriction, the concept of bicausality was introduced[49] and applied to various inversion problems.[50] [51] [52] Bicausality has also been used in the context of fault detection[53] [54], analysis of dynamical system structural properties[55] [56], control system design[57] [58] [59] [60] [61] and nonlinear analysis[62] .

Source sensor (SS) components

The SS (source-sensor) component has two possible causal configurations corresponding to Se/Df (effort source, flow sensor) and De/Sf (effort sensor, flow source). As illustrated in the figure, the SS component has two bicausal configurations corresponding to Se/Sf (effort source, flow source) and De/Df (effort sensor, flow sensor). This is represented graphically using causal half-strokes where the half-stroke on the harpoon side of the bond corresponds to flow and the half-stroke on the other side of the bond corresponds to effort.

The rules for the bicausality of junctions are the same as those for causality: only one bond imposes an effort onto a 0-junction and only one bond imposes a flow onto a 1-junction. In the context of inversion, the bonds connected to I, C and components cannot be bicausal.

These ideas are illustrated using an example with non-collocated source and sensor.

Example: inversion of non-collocated source-sensor system

The example is extended by adding a zero junction and a further SS component. In the mechanical case, this corresponds to inserting a force sensor and velocity source between the spring and the ground; in the electrical case it corresponds to adding a current source and voltage sensor in parallel to the capacitor. The example is simplified by setting the flow of the added flow source to zero as indicated on the bond graph.

As in the collocated example, the mechanical system is considered and system input is taken to be the force

F

acting on the mass M. However, the output is taken to be the force

Fs

of the spring which is not collocated with the applied force

F

.The corresponding state-space system is the same except that the output matrix C is given by:

C=\begin{pmatrix}k&0\end{pmatrix}

reflecting the change of the sensor. The transfer function relating input

F

to output

Fs

becomes:

{k\over{\left(k+ds+ms2\right)}}

The denominator remains unchanged, but the numerator is different reflecting the change of sensor.

The system is inverted by making the output

Fs

an input and the input

F

an output. Thus the left hand SS becomes both an effort and flow sensor and the right hand SS becomes both and effort and flow source with the corresponding causalities. As the I, C and R components must retain conventional causality, the bicausality propagates as shown. Both the I and C components are in derivative causality and thus the zero dynamics are zero order which, in this simple case, corresponds to the fact that the transfer function of the original system has a zero order numerator.

As in the collocated case, the transfer function of the inverse system is the reciprocal of the transfer function of the system; again, the bond graph causality method can also be used to examine the zero dynamics of nonlinear systems.

Deriving the bond graph of mechanical, electrical and electromechanical systems

Methods for deriving the bond graph of systems in various physical domains are explained in detail in the textbooks. A detailed derivation of a laboratory electromechanical system is given in a tutorial paper.

This section has methods and worked examples for some simple systems .

Electromagnetic

The steps for solving an Electromagnetic problem as a bond graph are as follows:

  1. Place an 0-junction at each node
  2. Insert Sources, R, I, C, TR, and GY bonds with 1 junctions
  3. Ground (both sides if a transformer or gyrator is present)
  4. Assign power flow direction
  5. Simplify

These steps are shown more clearly in the examples below.

Linear mechanical

The steps for solving a Linear Mechanical problem as a bond graph are as follows:

  1. Place 1-junctions for each distinct velocity (usually at a mass)
  2. Insert R and C bonds at their own 0-junctions between the 1 junctions where they act
  3. Insert Sources and I bonds on the 1 junctions where they act
  4. Assign power flow direction
  5. Simplify

These steps are shown more clearly in the examples below.

Simplifying

The simplifying step is the same regardless if the system was electromagnetic or linear mechanical. The steps are:

  1. Remove Bond of zero power (due to ground or zero velocity)
  2. Remove 0 and 1 junctions with less than three bonds
  3. Simplify parallel power
  4. Combine 0 junctions in series
  5. Combine 1 junctions in series

These steps are shown more clearly in the examples below.

Parallel power

Parallel power is when power runs in parallel in a bond graph. An example of parallel power is shown below.

Parallel power can be simplified, by recalling the relationship between effort and flow for 0 and 1-junctions. To solve parallel power you will first want to write down all of the equations for the junctions. For the example provided, the equations can be seen below. (Please make note of the number bond the effort/flow variable represents).\beginf_1=f_2=f_3 & & e_2=e_4=e_7 \\e_1=e_2+e_3 & & f_2=f_4+f_7 \\ & & \\e_3=e_5=e_6 & & f_7=f_6=f_8 \\f_3=f_5+f_6 & & e_7+e_6=e_8\end

By manipulating these equations you can arrange them such that you can find an equivalent set of 0 and 1-junctions to describe the parallel power.

For example, because e_3=e_6 and e_2=e_7 you can replace the variables in the equation e_1=e_2+e_3 resulting in e_1=e_6+e_7 and since e_6+e_7=e_8, we now know that

e1=e8

. This relationship of two effort variables equaling can be explained by an 0-junction. Manipulating other equations you can find that

f4=f5

which describes the relationship of a 1-junction. Once you have determined the relationships that you need you can redraw the parallel power section with the new junctions. The result for the example show is seen below.

Examples

Simple electrical system

A simple electrical circuit consisting of a voltage source, resistor, and capacitor in series.

The first step is to draw 0-junctions at all of the nodes:\begin & 0 & & 0 & \\ & & & & \\ & & & & \\ & 0 & & 0 & \end

The next step is to add all of the elements acting at their own 1-junction:\begin & & & & R & & & & \\ & & & & | & & & & \\ & & 0 & - & 1 & - & 0 & & \\ & & | & & & & | & & \\S_e & - & 1 & & & & 1 & - & C \\ & & | & & & & | & & \\ & & \underline & - & - & - & 0 & & \end

The next step is to pick a ground. The ground is simply an 0-junction that is going to be assumed to have no voltage. For this case, the ground will be chosen to be the lower left 0-junction, that is underlined above. The next step is to draw all of the arrows for the bond graph. The arrows on junctions should point towards ground (following a similar path to current). For resistance, inertance, and compliance elements, the arrows always point towards the elements. The result of drawing the arrows can be seen below, with the 0-junction marked with a star as the ground.

Now that we have the Bond graph, we can start the process of simplifying it. The first step is to remove all the ground nodes. Both of the bottom 0-junctions can be removed, because they are both grounded. The result is shown below.

Next, the junctions with less than three bonds can be removed. This is because flow and effort pass through these junctions without being modified, so they can be removed to allow us to draw less. The result can be seen below.

The final step is to apply causality to the bond graph. Applying causality was explained above. The final bond graph is shown below.

Advanced electrical system

A more advanced electrical system with a current source, resistors, capacitors, and a transformer

Following the steps with this circuit will result in the bond graph below, before it is simplified. The nodes marked with the star denote the ground.

Simplifying the bond graph will result in the image below.

Lastly, applying causality will result in the bond graph below. The bond with star denotes a causal conflict.

Simple linear mechanical

A simple linear mechanical system, consisting of a mass on a spring that is attached to a wall. The mass has some force being applied to it. An image of the system is shown below.

For a mechanical system, the first step is to place a 1-junction at each distinct velocity, in this case there are two distinct velocities, the mass and the wall. It is usually helpful to label the 1-junctions for reference. The result is below.\begin & & \\ & & \\1_\text & & \\ & & \\ & & \\ & & \\1_\text & & \end

The next step is to draw the R and C bonds at their own 0-junctions between the 1-junctions where they act. For this example there is only one of these bonds, the C bond for the spring. It acts between the 1-junction representing the mass and the 1-junction representing the wall. The result is below.\begin & & \\ & & \\1_\text & & \\| & & \\0 & - & C:\frac \\| & & \\1_\text & & \end

Next you want to add the sources and I bonds on the 1-junction where they act. There is one source, the source of effort (force) and one I bond, the mass of the mass both of which act on the 1-junction of the mass. The result is shown below.\beginS_e:F(t) & & \\| & & \\1_\text & - & I:m \\| & & \\0 & - & C:\frac \\| & & \\1_\text & & \end

Next power flow is to be assigned. Like the electrical examples, power should flow towards ground, in this case the 1-junction of the wall. Exceptions to this are R, C, or I bond, which always point towards the element. The resulting bond graph is below.

Now that the bond graph has been generated, it can be simplified. Because the wall is grounded (has zero velocity), you can remove that junction. As such the 0-junction the C bond is on, can also be removed because it will then have less than three bonds. The simplified bond graph can be seen below.

The last step is to apply causality, the final bond graph can be seen below.

Advanced linear mechanical

A more advanced linear mechanical system can be seen below.

Just like the above example, the first step is to make 1-junctions at each of the distant velocities. In this example there are three distant velocity, Mass 1, Mass 2, and the wall. Then you connect all of the bonds and assign power flow. The bond can be seen below.

Next you start the process of simplifying the bond graph, by removing the 1-junction of the wall, and removing junctions with less than three bonds. The bond graph can be seen below.

There is parallel power in the bond graph. Solving parallel power was explained above. The result of solving it can be seen below.

Lastly, apply causality, the final bond graph can be seen below.

State equations

Once a bond graph is complete, it can be utilized to generate the state-space representation equations of the system. State-space representation is especially powerful as it allows complex multi-order differential system to be solved as a system of first-order equations instead. The general form of the state equation is \dot (t) = \mathbf \mathbf(t) + \mathbf\mathbf(t) where \mathbf (t) is a column matrix of the state variables, or the unknowns of the system. \dot(t) is the time derivative of the state variables. \mathbf(t) is a column matrix of the inputs of the system. And \mathbf and \mathbf are matrices of constants based on the system. The state variables of a system are q(t) and p(t) values for each C and I bond without a causal conflict. Each I bond gets a p(t) while each C bond gets a q(t).

For example, if you have the following bond graph

you would have the following \dot(t), \mathbf (t), and \mathbf(t) matrices:

\dot(t) = \begin \dot_3(t) \\ \dot_6(t) \end\qquad \text \qquad\mathbf(t) = \begin p_3(t) \\ q_6(t) \end\qquad \text \qquad\mathbf(t) = \begin e_1(t) \end

The matrices of \mathbf and \mathbf are solved by determining the relationship of the state variables and their respective elements, as was described in the tetrahedron of state. The first step to solve the state equations is to list all of the governing equations for the bond graph. The table below shows the relationship between bonds and their governing equations.

Element typeBond NameBond with causalityGoverning equation(s)
Single-port elementsSource/ Sink, S

Se  \overset{style}{\underset{style}{---\rightharpoonup|}} 

input=e(t)

Sf  \overset{style}{\underset{style}{|---\rightharpoonup}} 

input=f(t)

Resistance, R:Dissipated Energy

  \overset{style}{\underset{style}{---\rightharpoonup|}} R

f(t)=

1
R

e(t)

  \overset{style}{\underset{style}{|---\rightharpoonup}} R

e(t)=Rf(t)

Inertance, I:Kinetic Energy♦

  \overset{style}{\underset{style}{---\rightharpoonup|}} I

f(t)=

1
I

\inte(t)dt

  \overset{style}{\underset{style}{|---\rightharpoonup}} I

e(t)=I

•
f

(t)

Compliance, C:Potential Energy

  \overset{style}{\underset{style}{---\rightharpoonup|}} C

f(t)=C

•
e

(t)

♦

  \overset{style}{\underset{style}{|---\rightharpoonup}} C

e(t)=

1
C

\intf(t)dt

Double-port elementsTransformer, TR

\begin{matrix} \overset{style}{\underset{style}{---\rightharpoonup}|} TR  \overset{style}{\underset{style}{---\rightharpoonup}}

\ \\ ^\end

f1=

1
r

f2

e2=

1
r

e1

\begin{matrix}

\ \overset\ TR\ \ \ \overset\ \\ ^\end

e1=re2

f2=rf1

Gyrator, GY

\begin{matrix}

\ \overset\ GY \overset\ \\ ^\end

e1=gf2

e2=gf1

\begin{matrix}  \overset{style}{\underset{style}{---\rightharpoonup}}

\ GY \overset\ \\ ^\end

f1=

1
g

e2

f2=

1
g

e1

Multi-port elements0 junctionOne and only onecausal bar at the junction

alle(t)=

\sumf(t)in=\sumf(t)out

1 junctionone and only one causalbar away from the junction

\sume(t)in=\sume(t)out

allf(t)=

"♦" denotes preferred causality.

For the example provided,

the governing equations are the following.

  1. e_1 = \text
  2. e_3 = e_1 - e_2 - e_4
  3. f_1 = f_2 = f_4 = f_3
  4. e_2 = R_2 f_2
  5. f_3 = \frac \int e_3 \, dt = \frac p_3
  6. f_5 = f_4 \cdot r
  7. e_4 = e_5 \cdot r
  8. e_5 = e_7 = e_6
  9. f_6 = f_5 - f_7
  10. e_6 = \frac \int f_6 \, dt = \frac q_6
  11. f_7 = \frac e_7

These equations can be manipulated to yield the state equations. For this example, you are trying to find equations that relate \dot_3(t) and \dot_6(t) in terms of p_3(t), q_6(t), and e_1(t).

To start you should recall from the tetrahedron of state that \dot_3(t) = e_3(t)starting with equation 2, you can rearrange it so that

e3=e1-e2-e4

.

e2

can be substituted for equation 4, while in equation 4,

f2

can be replaced by

f3

due to equation 3, which can then be replaced by equation 5.

e4

can likewise be replaced using equation 7, in which

e5

can be replaced with

e6

which can then be replaced with equation 10. Following these substituted yields the first state equation which is shown below.

\dot_3(t) = e_3(t) = e_1(t) - \frac p_3(t) - \frac q_6(t)

The second state equation can likewise be solved, by recalling that \dot_6(t) = f_6(t). The second state equation is shown below.

\dot_6(t) = f_6(t) = \frac p_3(t) - \frac q_6(t)

Both equations can further be rearranged into matrix form. The result of which is below.

\begin \dot_3(t) \\ \dot_6(t) \end =\begin- \frac & - \frac\\ \frac & -\frac\end\begin p_3(t) \\ q_6(t) \end +\begin 1 \\ 0 \end\begin e_1(t) \end

At this point the equations can be treated as any other state-space representation problem.

International conferences on bond graph modeling (ECMS and ICBGM)

A bibliography on bond graph modeling may be extracted from the following conferences :

See also

Systems for bond graph

Many systems can be expressed in terms used in bond graph. These terms are expressed in the table below.

Conventions for the table below:

P

is the active power;

\hat{X}

is a matrix object;

\vec{x}

is a vector object;

x\dagger

is the Hermitian conjugate of ; it is the complex conjugate of the transpose of . If is a scalar, then the Hermitian conjugate is the same as the complex conjugate;
n
D
t
is the Euler notation for differentiation, where: D_t^n f(t) = \begin\displaystyle\int_^t f(s)\, ds,& n=-1\\[2pt]f(t),& n=0\\[2pt]\dfrac, & n>0 \end

\begin{cases} \langlex\rangle\alpha:=|x|\alphasgn(x)\\ \langle{a}\rangle=k\langleb\rangle\beta\implies\langleb\rangle=\left(

1
k

\langlea\rangle\right)1/\beta\end{cases}

\textrm:\ \dfrac\\\textrm:\ \dfrac\\\textrm:\ \dfrac\end

Generalized flowGeneralized displacementGeneralized effortGeneralized momentumGeneralized power (in watts for power systems)Generalized energy (in joules for power systems)
Name

\vecf(t)

\vecq(t)

\vece(t)

\vecp(t)

P=\vecf(t)\dagger\vece(t)

E=\vecq(t)\dagger\vece(t)

DescriptionTime derivative of displacementA quality related to static behaviour.The energy per unit of displacementTime integral of effortTransformation of energy from one to another formConserved quantity in closed systems
Elements
NameHyperance

H

,
hyperrigitance

P=H-1

Compliance

C

,
rigitance

K=C-1

Resistance

R

Inertance

I

(or

L

)
Abrahance

A

Magnance

M

PropertiesPower dissipative elementCharge storage element(State variable: displacement)

(Costate variable: effort)

Power dissipative elementMomentum storage element(State variable: momentum)

(Costate variable: flow)

Power dissipative elementPower dissipative element
Quantitative behaviourFor 1-dimension systems (linear): P = H\cdot \left(D_t^0 q(t)\right)^2

For 1-dimension systems: e(t) = P \gamma\left[D^{-1}_t q(t)\right]

Impedance: Z(s) = \frac = \fracP

Potential energy for N-dimension systems: \beginV &=& \frac \vec q(t)^ \vec e(t) \\\vec q(t) &=& \hat \vec e(t)\end

Potential energy: V = \int_0^q e(q)\,dq

Potential coenergy: \overline=\int_0^e q(e)de

For 1-dimension systems: f(t) = C\cdot \frac + e\frac

Impedance: Z(s) = \frac = \frack

For 1-dimension systems (linear): P = R\cdot \left(D_t q(t)\right)^2Power for 1-dimension non-linear resistances (

e(f)

is the effort developed by the element): P=e(f)f

Rayleigh power: \mathfrak = \fracR\cdot f(t)^2

Rayleigh oower for non-linear resistances: \mathfrak=\int_0^f e(f) \, df

Rayleigh effort: e_=\frac = e(f)

For N-dimension systems: \beginP &=& \vec f(t)^ \vec e(t) \\\vec e(t) &=& \hat\vec f(t)\end

For 1-dimension systems: e(t) = R\cdot \gamma\left[D_t^1 q(t)\right]

Impedance: Z(s) = R

Kinetic energy for N-dimension systems: \beginT = \frac \vec \rho(t)^ \vec f(t) \\\vec \rho(t) = \hat \vec f(t)\end

Kinetic energy: T= \int_0^\rho f(\rho)\, d\rho

Kinetic coenergy: \overline=\int_0^f \rho(f) \, df

For 1-dimension systems: e(t) = L\cdot \frac + f\cdot \frac

Impedance: Z(s) = s L

For 1-dimension systems (linear): P = A\cdot \left(D_t^2 q(t)\right)^2

For 1-dimension systems: e(t) = A\cdot \gamma\left[D_t^3 q(t)\right]

Impedance Z(s) = s^2 A

For 1-dimension systems (linear) P = A\cdot \left(D_t^3 q(t)\right)^2

For 1-dimension systems e(t) = M\cdot\gamma \left[D_t^5 q(t)\right]

Impedance Z(s) = s^4 M

Generalized behaviourEnergy from active effort sources:

W=

q
\int
0

{esource

} \, dq Lagrangian

ak{L}=T-(V-W)

Hamiltonian

ak{H}=T+(V-W)

Hamiltonian effort:

eak{H

}=\frac

Lagrangian effort:

eak{L

}=\frac

Passive effort:

eak{LH

}=\frac\frac

Power equation:

dak{L
}+\frac = 0

Effort equation:

eak{L

}+e_=0

Lagrangian equation:

eak{R

}+e_=e_

Hamiltonian equation:

eak{R

}+e_=-e_

If

\overline{W}

is the coenergy,

W

is the energy,

S

is the state variable and

\overline{S}

is the costate variable,

\begin{aligned} \overline{W}+W=S ⋅ \overline{S}\\ \overline{W}=

\overline{S
\int
0
} S(\overline) \, d\overline\\W = \int_0^ \overline(S) dS\end

For linear elements:

\overline{W}=W=1
2

S ⋅ \overline{S}

Longitudinal mechanical power system[63]
Flow-related variables
6
D
t

x

pounce / pop

[m/s6]
5
D
t

x

flounce / crackle

[m/s5]
4
D
t

x

jounce / snap

[m/s4]
3
D
t

x

[m/s3]
2
D
t

x

(x2t

) [m/s2]
1
D
t

x

(xt) [m/s]

(flow)
0
D
t

x

displacement

(x) [m]

(displacement)
-1
D
t

x

[m ⋅ s]

-2
D
t

x

absity

[m ⋅ s2]
-3
D
t

x

abseleration

[m ⋅ s3]
-4
D
t

x

abserk

[m ⋅ s4]
Effort-related variables
1
D
t

F

[N/s]

0
D
t

F

t
(P
x) [N]
(effort)
-1
D
t

F

2t
(P
x) [kg ⋅ m/s]
(momentum)
Passive elements
Compliance (C)Resistance (R)Inertance (I)Abrahance (A)Magnance (M)
Spring \langle P^t_x\rangle = k \langle x\rangle where

k

is the spring stiffness
Damper \langle P^t_x\rangle = b \langle x_t\rangle where

b

is the damper parameter
Mass \langle P^t_x\rangle = m \langle x_\rangle where

m

is the mass
Abraham–Lorentz force\langle P^t_x\rangle = \frac \langle x_\rangle where

\mu0

is the permeability

q

is the electric charge

c

is the speed of light
Magnetic radiation reaction force\langle P^t_q\rangle =\frac \langle x_\rangle where

\mu0

is the permeability

q

is the electric charge

c

is the speed of light

R

is the radius of the magnetic moment
Cantilever \langle P^t_x\rangle =\frac \langle x\rangle

L

: cantilever length

E

: Young's modulus

I

: second moment of area
Cyclotron radiation resistance\langle P^t_x \rangle = \frac \langle x_t\rangle where

\sigmat

: Thompson cross-section

c

: speed of light

\mu0

: permeability

B

: magnetic field density
Prismatic floater in a wide waterbody\langle P^t_x\rangle = \rho_L g A \langle x\ranglewhere

\rhoL

: liquid density

A

: area

g

: gravitational acceleration
Viscous friction \langle P^t_x\rangle = b \langle x_t\rangle where

b

is the viscous friction parameter
Elastic rod \langle P^t_x\rangle = \frac \langle x\ranglewhere

E

: Young's modulus

A

: area

L

: rod length
Inverse of kinetic mobility \langle P^t_x\rangle = \frac \langle x_t\rangle where

\mu

is the kinetic mobility
Newton's law of gravitation\langle P^t_x\rangle = GMm \langle x\rangle^where

G

: gravitational constant

M

: mass of body 1

m

: mass of body 2
Geometry interacting with air (e.g. drag)\langle P^t_x \rangle = \fracc\rho A \langle x_t\rangle^ where

A

: contact area

c

: aerodynamic geometry coefficient

\rho

: fluid density
Coulomb's law\langle P^t_x\rangle = \frac \langle x\rangle^where

\varepsilon0

: permittivity

Q

: charge of body 1

q

: charge of body 2
Absquare damper \langle P^t_x\rangle = B\langle x_t\rangle^2 where

B

is the Absquare damper parameter
Casimir force\langle P^t_x\rangle = \frac \langle x\rangle ^where

A

: area of plate

\hbar

: Planck's reduced constant

c

: speed of light
Dry friction\langle P^t_x\rangle = \mu F_\langle x_t\rangle^0where

\mu

: kinetic friction coefficient

Fn

: normal force
Biot–Savart law\langle P^t_x \rangle = \frac \langle x\rangle^where

\mu0

: permeability

I1

: current in wire 1

I2

: current in wire 2

l

: length of wires
Piston pressing fluid inside an adiabatic chamberP^t_x = AP_0 \left(1+\frac\right)^ where

A

: area of piston

P0

: initial pressure inside

x0

: initial position of piston

\gamma

: poisson coefficient
Angular mechanical power system
Flow-related variables
3
D
t

\theta

[rad/s3]
2
D
t

\theta

(\theta2t

) [rad/s2]
1
D
t

\theta

(\thetat) [rad/s]

(flow)
0
D
t

\theta

(\theta) [rad]

(displacement)
Effort-related variables
1
D
t

\tau

[W/rad]

0
D
t

\tau

t
(P
\theta

) [J/rad]

(effort)
-1
D
t

\tau

2t
(P
\theta

) [J ⋅ s/rad]

(momentum)
Passive elements
Compliance (C)Resistance (R)Inertance (I)
Inverse of the angular spring constant \langle P^t_\rangle = k_r\langle \theta\rangle where

kr

is the angular spring constant
Angular damping \langle P^t_\theta\rangle = R \langle \theta_t\rangle where

R

is the damping constant
Mass moment of inertiaType

\langle

t
P
\theta

\rangle=I\langle\theta2t\rangle

where

I

is the mass moment of inertia
Rod torsion\langle P^t_\theta\rangle = \frac\langle \theta \ranglewhere

G

: shear modulus

J

: polar moment of area

L

: rod length
Governor (e.g. used in music boxes) \langle P^t_\theta \rangle = R \langle \theta_t\rangle^2 where

R

is the governor constant
Bending moment (cantilever)\langle P^t_\rangle = \frac\langle \theta\ranglewhere

E

: Young modulus

I

: second moment of area

L

: beam length
Parallel force field\langle P^t_\theta\rangle = \langle P^t_x\rangle \sin\left(\alpha-\theta\right) where

\alpha

: angle (counter-clockwise positive) between the polar axis and the force field

\theta

: current angle of the object (e.g. pendulum)
Electric power system
Flow-related variables
2
D
t

q

electric inertia

[A/s]

1
D
t

q

(qt) [A]

(flow)
0
D
t

q

(q) [C]

(displacement)
Effort-related variables
1
D
t

V

distension

[V/s]

0
D
t

V

t
(P
q) [V]
(effort)
-1
D
t

V

2t
(P
q) [V ⋅ s]

or[Wb ⋅ turn]

(momentum)
Elements
Hyperance (H)Compliance (C)Resistance (R)Inertance (I)Abrahance (A)
Frequency-dependent negative resistor (FDNR)\langle P^t_ \rangle = \frac \langle q^t\rangleLinear capacitor\langle P^t_q\rangle = \frac\phi_L \langle q\ranglewhere

\varepsilon

: Permittivity

\phiL

: vergent factor
Linear resistor\langle P^t_q\rangle = \rho \phi_L (1+\alpha (T-T_0)) \langle q_t\ranglewhere

\rho

: resistivity

\phiL

: vergent factor

\alpha

: thermal coefficient

T

: current temperature

T0

: reference temperature
Linear inductor (solenoid)\langle P^t_q \rangle = \frac \langle q_\rangle where

\mu0

: permeability

N

: number of turns

A

: area

L

: length
Frequency-dependent negative conductance (FDNC)Type \gamma^1 \iff V = R D^2_t i\quad [H\cdot s]
DiodeP^t_q = nV_T\ln\left(1+\frac\right)where

n

: ideality factor

VT

: thermal voltage

Is

: leakage current
Toroid\langle P^t_q\rangle = \frac \langle q_\ranglewhere

\mu

: permeability

N

: number of turns

A

: cross-sectional area

r

: toroid radius to centerline
Intra-gyratorInter-gyrator
Compliant gyratorResistive gyratorInertant gyrator
Hall effect device\begine_x = R_H\fraci_y\\e_y = R_H\fraci_x\endwhere

RH

: Hall coefficient

Bz

: vertical magnetic induction field

tz

: vertical thickness
Induction motor\begine_r^1 = M\cos(\theta)D_tf_s^1 + f_s^1D_t\left[M\cos(\theta)\right]\\e_s^1 = M\cos(\theta)D_tf_r^1 + f_r^1D_t\left[M\cos(\theta)\right]\endwhere

\theta

: electrical angle
1
e
r
: voltage at rotor bar
1
e
s
: voltage at stator bar
1
f
s
: flow through stator bar
1
f
r
: flow through rotor bar
DC motor\begin\tau_e=k_a\phi_a(i_e)i_a\\\omega_e = \frace_a\endwhere

\taue

: electromagnetic torque

\omegae

: axis angular velocity

ie

: field current

ia

: armature current
Faraday gyrator\beginF &= Bli\\V &= \fracv\endwhere

F

: force

v

: rod velocity

V

: rod voltage

B

: magnetic induction field

l

: rod length
Faraday disk\beginV &= \fracBr^2 \omega\\\tau &= \fracBr^2 i\endwhere

B

: magnetic induction field

r

: disk radius
Intra-transformerInter-transformer
Electrical transformer (only for AC signals)\beginV_2 &= \fracV_1\\f_2 &= \fracf_1 \end
Hydraulic / pneumatic power system
Flow-related variables
1
D
t

V

(V
t) [m3/s]
(flow)
0
D
t

V

(V) [m3]
(displacement)
Effort-related variables
0
D
t

P

t
(P
V) [Pa]
(effort)
-1
D
t

P

fluid momentum

2t
(P
V) [Pa ⋅ s]
(momentum)
Elements
Compliance (C)Resistance (R)Inertance (I)
Pipe elasticity\langle P^t_V\rangle = \left(\frac\right) \langle V\rangle where

r0

: nominal pipe radius

V0

: pipe volume (without stress)

tW

: wall thickness

E

: Young's modulus
Darcy sponge\langle P_V^t\rangle = \left(\frac \phi_L\right) \langle V_t\rangle where

\mu

: dynamic viscosity

k

: permeability

\phiL

: vergent factor
Fluid inertia in pipes\langle P^t_V \rangle = \left(\rho \phi_L\right) \langle V_\rangle where

\rho

: fluid density

\phiL

: vergent factor
Compressible fluid (approximation)\langle P^t_V\rangle = \left(\frac\right)\langle V\rangle = \underbrace_ \langle V\rangle where

V0

: pipe volume (without stress)

B

: bulk modulus

\rho0

: gas density at reference pressure

c

: speed of sound
Valve\langle P^t_V \rangle = \left(\frac\right) \langle V_t\rangle ^2

Cd

: discharge coefficient

\rho

: fluid density

A0

: smallest area for fluid passage
Tank with area A\frac\left(h+h_0\right)^ : P^t_V = \rho g h_0 \left[\left(1+\frac{nV}{A\frac{1}{[m]^}h_0^n }\right)^ -1\right] where

h

: height with respect to the ground

h0

: inlet height

\rho

: fluid density

A

: curve scaling (dimension of area)

[m]n-1

: correction of dimension

n

: curve parameter

g

: acceleration of gravity
  • Case

n=1

(prismatic case): P^t_V = \frac V
  • Case

n=0

(current diode case): P_V^t = \rho g h_0 \left(e^-1\right)
Poiseuille resistance for cylinders\langle P_V^t \rangle = \left(\frac\right) \langle V_t\rangle where

\mu

: dynamic viscosity

L

: pipe length

R

: pipe radius
Isothermal chamber\langle P^t_V\rangle = k \langle V_t\rangle ^ where

k

is chamber's constant
Turbulence resistance\langle P_V^t\rangle = a_t \langle V_t \rangle^ where

at

is an empirical parameter
Compressible fluidP^t_V = -B\ln V where

B

is the bulk modulus
Nozzle\langle P^t_V\rangle = \frac \langle V_t\rangle^2 where

\rho

: fluid density

Aout

: outlet area

Ain

: inlet area
Adiabatic bladderP^t_V = P^t_ \left(1-\frac\right)^ where
t
P
V,0

: reference pressure

V0

: reference volume

\gamma

: Poisson coefficient
Check valve P^t_V = k\ln\left(1+\frac\right) where

k

: empirical constant

Vt,0

: reverse bias absolute flow
Gyrator–capacitor power system
Flow-related variables
0
D
t

\varphi

(\varphi) [Wb]

(displacement)
Effort-related variables
0
D
t

l{F}

(l{F}) [A ⋅ turn]

(momentum)
Elements
Compliance (C)Resistance (R)Inertance (I)
Permeance (

l{P}

)\langle \mathcal\rangle = \frac \phi_L \langle \varphi \rangle\quad \mathrmwhere

\mu

: permeability

\phiL

: vergent factor
Magnetic complex impedance (

ZM

)\mathcal = Z_M \quad \mathrm
Magnetic complex inductance (

LM

)\mathcal = L_M \quad \mathrm
Gravitational power system
Flow-related variables
0
D
t

ig

gravitational current Loop

\left(ig=2\varepsilong

3\right) [kg/s]
v
orbit
(flow)
-1
D
t

ig

gravitational charge

(M) [kg]

(displacement)
Effort-related variables
0
D
t

Vg

gravitational voltage

\left(Vg=

1
2
4
v
orbit
c2
\right) [m2/s2]
(effort)
-1
D
t

Vg

gravitational momentum

\left(\phig=

\pi
3
v
orbit
r
c2
\right) [m2/s]
(momentum)
Elements
Compliance (C)Resistance (R)Inertance (I)
Gravitational capacitanceType

\gamma1\iffMg=CVg  

[kg/m2]

Cg=

2c2r2
GM

Gravitational orbital resistanceType

\gamma1\iffVg=Rgig  

[m2/kg ⋅ s]

Rg=

\mug
4

vorbit

Gravitational inductanceType

\gamma1\iff\phig=Iig  

[m2/kg ⋅ s2]

Lg=

2\pi2G
c2

r

Electric field volumetric power density system
Flow-related variables
0
D
t

J

(J) [A/m2]
(flow)
-1
D
t

J

(D) [C/m2]
(displacement)
Effort-related variables
0
D
t

E

(E) [V/m]

(effort)
-1
D
t

E

magnetic potential vector

(A) [V ⋅ s/m]

(momentum)
Elements
Compliance (C)Resistance (R)Volumetric power density inertance (

IV

)
Electrical permittivityType

\gamma1\iffD=\epsilon0E   [F/m]

Electrical resistivityType

\gamma1\iffE=\rhoJ   [\Omegam]

Magnetic permeability

\mu0 [H/m]

Magnetic field volumetric power density system
Flow-related variables
0
D
t

B

magnetic flux density

(B) [T]or

[Wb/m2]
(displacement)
Effort-related variables
0
D
t

H

magnetic field strength

(H) [A ⋅ {turn/m]}

(effort)
Elements
Compliance (C)Resistance (R)Volumetric power density inertance (

IV

)
Magnetic permeability for magnetic circuitsType

\gamma1\iffB=\muH   [H/(m ⋅ {turn2)]}

Gravitoelectric volumetric power density system
Flow-related variables
0
D
t

Jg

flux of mass

(J
g) [kg/m2s]
(flow)
-1
D
t

Jg

accumulated flux of mass

(D
g) [kg/m2]
(displacement)
Effort-related variables
0
D
t

g

(g) [m/s2]
(effort)
Elements
Compliance (C)Resistance (R)Volumetric power density inertance (

IV

)
Gravitational permittivityType

\gamma1\iffDg=\epsilongg  

[kg ⋅ s2/m3]

\varepsilong=

1
4\piG

Gravitational permeability

[m/kg]

I_V = \mu_g = \frac
Gravitomagnetic volumetric power density system
Flow-related variables
0
D
t

Bg

gravitomagnetic field

\left(Bg=

\omega
orbit\left(vorbit
c

\right)2\right) [Hz]

(displacement)
Effort-related variables
0
D
t

Hg

gravitomagnetic field strength

(Hg) [Pa ⋅ s]

(effort)
Elements
Compliance (C)Resistance (R)Volumetric power density inertance (

IV

)
Gravitational permeability
-4\piG
c2

 [m/kg]

Thermal power-temperature system
Flow-related variables
1
D
t

Q

heat rate

(\psit) [W]

(flow)
0
D
t

Q

total heat

(\psi) [J]

(displacement)
Effort-related variables
0
D
t

T

temperature

(T) [K]

(Effort)
Elements
Compliance (C)Resistance (R)Inertance (I)
Isobaric heatT = \frac \psi = \frac\psiwhere

\rho

: object mass density

V

: volume of object

cp

: pressure specific heat capacity
Conduction resistanceT = \frac \phi_L \psi_twhere

k

: thermal conductivity

\phiL

: vergent factor
Isocoric heatT = \frac\psiwhere

CV

is the constant-volume heat capacitance
Convection resistanceT = \frac \psi_twhere

h

: convection coefficient

A

: interface area
Isothermal heatT = \frac \psiwhere

R

: universal gas constant

n

: number of moles

Vf

: final volume

Vi

: initial volume
Stefan–Boltzmann law\langle T\rangle = \left(\frac\right)^\langle \psi_t\rangle^where

e

: emissivity

\sigma

: Stefan–Boltzmann constant

A

: interface area
Continuum mechanics volumetric power density system
Flow-related variables
1
D
t

\varepsilon

strain rate

•
(\varepsilon

) [Hz]

(flow)
0
D
t

\varepsilon

strain

(\varepsilon) [1]

(displacement)
Effort-related variables
0
D
t

\sigma

stress

(\sigma) [Pa]

(effort)
Elements
Compliance (C)Resistance (R)Volumetric power density inertance (

IV

)
Inverse of rigidityType

\gamma1\iff\varepsilon=C\sigma  

[Pa-1]
C=1
K
ViscosityType

\gamma1\iff\sigma=R

•
\varepsilon

  [Pa ⋅ s]

Power density inertance: density of material
\rho [kg/m3]

Other systems:

Further reading

External links

Notes and References

  1. Book: Thomson, William . Mathematical and Physical Papers . 2011-06-30 . Cambridge University Press . 978-1-108-02898-1 . 1 . 10.1017/cbo9780511996009.004.
  2. Clerk-Maxwell . J. . 1869 . Remarks on the Mathematical Classification of Physical Quantities . Proceedings of the London Mathematical Society . en . s1-3 . 1 . 224–233 . 10.1112/plms/s1-3.1.224.
  3. Book: Karnopp . Dean . System dynamics: modeling and simulation of mechatronic systems . Margolis . Donald L. . Rosenberg . Ronald C. . 2012 . Wiley . 978-0-470-88908-4 . 5th . Hoboken, NJ.
  4. Book: Bond Graphs for Modelling, Control and Fault Diagnosis of Engineering Systems . 2017 . Springer International Publishing . 978-3-319-47433-5 . Borutzky . Wolfgang . Cham . en . 10.1007/978-3-319-47434-2.
  5. Book: Gawthrop . Peter . Metamodelling: for bond graphs and dynamic systems . Smith . Lorcan . 1996 . Prentice Hall . 978-0-13-489824-7 . Prentice Hall international series in systems and control engineering . London . 10.5281/zenodo.6998395.
  6. 2007 . Bond-graph modeling . IEEE Control Systems . 27 . 2 . 24–45 . 10.1109/MCS.2007.338279 . 1066-033X.
  7. Book: Wellstead, P. E. . Introduction to physical system modelling . 1979 . Academic Press . 978-0-12-744380-5 . London ; New York.
  8. Book: Shearer . J. Lowen . Introduction to system dynamics . Murphy . Arthur T. . Richardson . Herbert H. . 1971 . Addison-Wesley Publ. Comp . 978-0-201-07017-0 . 2. print . Reading, Mass..
  9. Chua . L. . 1971 . Memristor-The missing circuit element . IEEE Transactions on Circuit Theory . 18 . 5 . 507–519 . 10.1109/TCT.1971.1083337 . 1971ITCT...18..507C . 0018-9324.
  10. Akbarpour Ghazani . Mehran . Pan . Michael . Tran . Kenneth . Rampadarath . Anand . Nickerson . David P. . 2024 . A review of the diverse applications of bond graphs in biology and physiology . Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences . en . 480 . 2294 . 2024RSPSA.48030807A . 10.1098/rspa.2023.0807 . 1471-2946 . free . 20230807.
  11. Oster . George . Perelson . Alan . Katchalsky . Aharon . 1971 . Network Thermodynamics . Nature . en . 234 . 5329 . 393–399 . 1971Natur.234..393O . 10.1038/234393a0 . 0028-0836 . subscription.
  12. Oster . George F. . Perelson . Alan S. . Katchalsky . Aharon . 1973 . Network thermodynamics: dynamic modelling of biophysical systems . Quarterly Reviews of Biophysics . en . 6 . 1 . 1–134 . 10.1017/S0033583500000081 . 4576440 . 0033-5835 . subscription.
  13. Gawthrop . Peter J. . Crampin . Edmund J. . 2014 . Energy-based analysis of biochemical cycles using bond graphs . Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences . en . 470 . 2171 . 1406.2447 . 2014RSPSA.47040459G . 10.1098/rspa.2014.0459 . 1364-5021 . 4197480 . 25383030.
  14. Rajagopal . Vijay . Arumugam . Senthil . Hunter . Peter J. . Khadangi . Afshin . Chung . Joshua . Pan . Michael . 2022 . The Cell Physiome: What Do We Need in a Computational Physiology Framework for Predicting Single-Cell Biology? . Annual Review of Biomedical Data Science . en . 5 . 1 . 341–366 . 10.1146/annurev-biodatasci-072018-021246 . 2574-3414 . 35576556 . subscription.
  15. Gawthrop . Peter J. . Pan . Michael . 2022 . Network thermodynamics of biological systems: A bond graph approach . Mathematical Biosciences . en . 352 . 10.1016/j.mbs.2022.108899 . 36057321 . free . 108899.
  16. Hunter . P. . de Bono . B. . Brooks . D. . Christie . R. . Hussan . J. . Lin . M. . Nickerson . D. . 2025 . The Physiome Project and Digital Twins . IEEE Reviews in Biomedical Engineering . 18 . 300–315 . 2025IRBE...18..300H . 10.1109/RBME.2024.3490455 . 1937-3333 . 39504298.
  17. Gawthrop . Peter J. . Cudmore . Peter . Crampin . Edmund J. . 2020 . Physically-plausible modelling of biomolecular systems: A simplified, energy-based model of the mitochondrial electron transport chain . Journal of Theoretical Biology . en . 493 . 1905.12958 . 2020JThBi.49310223G . 10.1016/j.jtbi.2020.110223 . 32119969 . 110223.
  18. Gawthrop . Peter J. . Cursons . Joseph . Crampin . Edmund J. . 2015 . Hierarchical bond graph modelling of biochemical networks . Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences . en . 471 . 2184 . 1503.01814 . 2015RSPSA.47150642G . 10.1098/rspa.2015.0642 . 1364-5021 . 20150642.
  19. Pan . Michael . Gawthrop . Peter J. . Cursons . Joseph . Crampin . Edmund J. . 2021-10-13 . Wallqvist . Anders . Modular assembly of dynamic models in systems biology . PLOS Computational Biology . en . 17 . 10 . 2021PLSCB..17E9513P . 10.1371/journal.pcbi.1009513 . 1553-7358 . 8544865 . 34644304 . free . e1009513.
  20. Gawthrop . Peter J. . Pan . Michael . Crampin . Edmund J. . 2021 . Modular dynamic biomolecular modelling with bond graphs: the unification of stoichiometry, thermodynamics, kinetics and data . Journal of the Royal Society Interface . en . 18 . 181 . 10.1098/rsif.2021.0478 . 1742-5662 . 8385351 . 34428949 . 20210478.
  21. Cudmore . Peter . Pan . Michael . Gawthrop . Peter J. . Crampin . Edmund J. . 2021 . Analysing and simulating energy-based models in biology using BondGraphTools . The European Physical Journal E . en . 44 . 12 . 10.1140/epje/s10189-021-00152-4 . 1292-8941 . 34904197 . subscription . 148.
  22. Book: Mathematical Physiology . 2009 . Springer New York . 978-0-387-75846-6 . Keener . James . Interdisciplinary Applied Mathematics . 8/1 . New York, NY . 10.1007/978-0-387-75847-3 . Sneyd . James.
  23. Gawthrop . Peter J. . Siekmann . Ivo . Kameneva . Tatiana . Saha . Susmita . Ibbotson . Michael R. . Crampin . Edmund J. . 2017 . Bond graph modelling of chemoelectrical energy transduction . IET Systems Biology . en . 11 . 5 . 127–138 . 10.1049/iet-syb.2017.0006 . 1751-8849 . 8687425.
  24. Pan . Michael . Gawthrop . Peter J. . Tran . Kenneth . Cursons . Joseph . Crampin . Edmund J. . 2019 . A thermodynamic framework for modelling membrane transporters . Journal of Theoretical Biology . en . 481 . 10–23 . 1806.04341 . 2019JThBi.481...10P . 10.1016/j.jtbi.2018.09.034 . 30273576.
  25. Pan . Michael . Gawthrop . Peter J. . Tran . Kenneth . Cursons . Joseph . Crampin . Edmund J. . 2018 . Bond graph modelling of the cardiac action potential: implications for drift and non-unique steady states . Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences . en . 474 . 2214 . 1802.04548 . 2018RSPSA.47480106P . 10.1098/rspa.2018.0106 . 1364-5021 . 6030650 . 29977132 . 20180106.
  26. Gawthrop . Peter J. . 2017 . Bond Graph Modeling of Chemiosmotic Biomolecular Energy Transduction . IEEE Transactions on NanoBioscience . 16 . 3 . 177–188 . 1611.04264 . 2017ITNB...16..177G . 10.1109/TNB.2017.2674683 . 1536-1241 . 28252411.
  27. Gawthrop . Peter J. . Pan . Michael . Rajagopal . Vijay . 2025 . Energy-based modelling of single actin filament polymerization using bond graphs . Journal of the Royal Society Interface . en . 22 . 222 . 10.1098/rsif.2024.0404 . 1742-5662 . 11780404 . 39881657 . 20240404.
  28. Peskin . C.S. . Odell . G.M. . Oster . G.F. . 1993 . Cellular motions and thermal fluctuations: the Brownian ratchet . Biophysical Journal . en . 65 . 1 . 316–324 . 1993BpJ....65..316P . 10.1016/S0006-3495(93)81035-X . 1225726 . 8369439.
  29. Hunter . Peter J. . Ai . Weiwei . Nickerson . David P. . 2025 . Energy-based bond graph models of glucose transport with SLC transporters . Biophysical Journal . 124 . 2 . 316–335 . 10.1016/j.bpj.2024.12.006 . 0006-3495 . 11788503 . 39644093 . 2025BpJ...124..316H .
  30. Su . Shan . Blanco . Pablo J. . Müller . Lucas O. . Hunter . Peter J. . Safaei . Soroush . 2020-08-26 . Bond Graph Model of Cerebral Circulation: Toward Clinically Feasible Systemic Blood Flow Simulations . Physiome . 10.36903/physiome.12859424 . 2744-6204. free .
  31. Pan . Michael . Gawthrop . Peter J. . Faria . Matthew . Johnston . Stuart T. . 2025 . Thermodynamically consistent, reduced models of gene regulatory networks . Royal Society Open Science . en . 12 . 7 . 241725 . 10.1098/rsos.241725 . free . 2054-5703 . 12312648 . 40746961.
  32. Gawthrop . Peter . Pan . Michael . 2025 . Energy-based analysis of biochemical oscillators using bond graphs and linear control theory . Royal Society Open Science . en . 12 . 4 . 241791 . 10.1098/rsos.241791 . free . 2054-5703 . 12040473 . 40309185 . 2025RSOS...1241791G .
  33. Gawthrop . Peter J. . Pan . Michael . 2022 . Energy-based advection modelling using bond graphs . Journal of the Royal Society Interface . en . 19 . 195 . 20220492 . 10.1098/rsif.2022.0492 . 1742-5662 . 9554522.
  34. Gawthrop . Peter J. . Pan . Michael . Crampin . Edmund J. . 2021 . Modular dynamic biomolecular modelling with bond graphs: the unification of stoichiometry, thermodynamics, kinetics and data . Journal of the Royal Society Interface . en . 18 . 181 . 20210478 . 10.1098/rsif.2021.0478 . 1742-5662 . 8385351 . 34428949.
  35. Gawthrop . Peter J. . Cudmore . Peter . Crampin . Edmund J. . 2020-05-21 . Physically-plausible modelling of biomolecular systems: A simplified, energy-based model of the mitochondrial electron transport chain . Journal of Theoretical Biology . 493 . 110223 . 10.1016/j.jtbi.2020.110223 . 32119969 . 1905.12958 . 2020JThBi.49310223G . 0022-5193.
  36. Cacho . R. . Felez . J. . Vera . C. . 2000-08-01 . Deriving simulation models from bond graphs with algebraic loops.: The extension to multibond graph systems . Journal of the Franklin Institute . 337 . 5 . 579–600 . 10.1016/S0016-0032(00)00025-9 . 0016-0032. subscription .
  37. Web site: Gawthrop . P. J. . 1995 . Bicausal Bond Graphs . 2026-02-08 . eprints.gla.ac.uk . en.
  38. Samantaray . A. K. . Ghoshal . S. K. . 2008-01-01 . Bicausal bond graphs for supervision: From fault detection and isolation to fault accommodation . Journal of the Franklin Institute . 345 . 1 . 1–28 . 10.1016/j.jfranklin.2007.05.009 . 0016-0032. subscription .
  39. Gawthrop . Peter J. . 2000-09-01 . Physical interpretation of inverse dynamics using bicausal bond graphs . Journal of the Franklin Institute . 337 . 6 . 743–769 . 10.1016/S0016-0032(00)00051-X . 0016-0032. subscription .
  40. Bideaux . Eric . Marquis-Favre . Wilfrid . Scavarda . Serge . 2000-09-01 . Formulation of the Constitutive Relations in the Bicausal Bond Graph: Symbolic and Numerical Issues . IFAC Proceedings Volumes . IFAC Conference on Mechatronic Systems, Darmstadt, Germany, 18-20 September 2000 . 33 . 26 . 655–660 . 10.1016/S1474-6670(17)39220-0 . 1474-6670. subscription .
  41. Dauphin-Tanguy . G. . Rahmani . A. . Sueur . C. . 1999-12-15 . Bond graph aided design of controlled systems . Simulation Practice and Theory . 7 . 5 . 493–513 . 10.1016/S0928-4869(99)00009-9 . 0928-4869. subscription .
  42. Sueur . C. . Dauphin-Tanguy . G. . 1991-01-01 . Bond-graph approach for structural analysis of MIMO linear systems . Journal of the Franklin Institute . 328 . 1 . 55–70 . 10.1016/0016-0032(91)90006-O . 0016-0032. subscription .
  43. Marquis-Favre . Wilfrid . Scavarda . Serge . 2002-09-01 . Alternative Causality Assignment Procedures in Bond Graph for Mechanical Systems . Journal of Dynamic Systems, Measurement, and Control . en . 124 . 3 . 457–463 . 10.1115/1.1481369 . 0022-0434. subscription .
  44. Murray-Smith . David J. . 2024 . Kelvin and Bush: Their Contributions to Today's System Modelling and Simulation Methods . SNE Simulation Notes Europe . 34 . 4 . 233–239 . 10.11128/sne.34.on.10716.
  45. Book: Kailath, Thomas . Linear systems . 1980 . Prentice-Hall . 978-0-13-536961-6 . Prentice-Hall information and system sciences series . Englewood Cliffs, NJ.
  46. Ngwompo . R. Fotsu . Scavarda . S. . Thomasset . D. . 1996-03-01 . Inversion of linear time-invariant SISO systems modelled by bond graph . Journal of the Franklin Institute . 333 . 2 . 157–174 . 10.1016/0016-0032(96)00025-7 . 0016-0032. subscription .
  47. Bera . Tarun K. . Samantaray . Arun K. . 2011 . BOND GRAPH MODEL-BASED INVERSION OF PLANAR PARALLEL MANIPULATOR SYSTEMS . International Journal of Modelling and Simulation . en . 31 . 4 . 10.2316/Journal.205.2011.4.205-5637 . 1925-7082. subscription .
  48. Gawthrop . Peter J. . 2000-09-01 . Physical interpretation of inverse dynamics using bicausal bond graphs . Journal of the Franklin Institute . 337 . 6 . 743–769 . 10.1016/S0016-0032(00)00051-X . 0016-0032. subscription .
  49. Web site: Gawthrop . P. J. . 1995 . Bicausal Bond Graphs . 2026-02-15 . eprints.gla.ac.uk . en.
  50. Ngwompo . R. Fotsu . Scavarda . S. . Thomasset . D. . 1996-03-01 . Inversion of linear time-invariant SISO systems modelled by bond graph . Journal of the Franklin Institute . 333 . 2 . 157–174 . 10.1016/0016-0032(96)00025-7 . 0016-0032. subscription .
  51. Gawthrop . Peter J. . 2000-09-01 . Physical interpretation of inverse dynamics using bicausal bond graphs . Journal of the Franklin Institute . 337 . 6 . 743–769 . 10.1016/S0016-0032(00)00051-X . 0016-0032. subscription .
  52. Ngwompo . R F . Ngwompo . R F . Scavarda . S . Thomasset . D . 2001-03-01 . Physical model-based inversion in control systems design using bond graph representation Part 1: Theory . Proceedings of the Institution of Mechanical Engineers, Part I: Journal of Systems and Control Engineering . EN . 215 . 2 . 95–104 . 10.1243/0959651011540888 . 0959-6518.
  53. Samantaray . A. K. . Ghoshal . S. K. . 2008-01-01 . Bicausal bond graphs for supervision: From fault detection and isolation to fault accommodation . Journal of the Franklin Institute . 345 . 1 . 1–28 . 10.1016/j.jfranklin.2007.05.009 . 0016-0032. subscription .
  54. Benmoussa . Samir . Bouamama . Belkacem Ould . Merzouki . Rochdi . April 2014 . Bond Graph Approach for Plant Fault Detection and Isolation: Application to Intelligent Autonomous Vehicle . IEEE Transactions on Automation Science and Engineering . 11 . 2 . 585–593 . 10.1109/TASE.2013.2252340 . 1558-3783. subscription .
  55. El Feki . Mariem . Jardin . Audrey . Marquis-Favre . Wilfrid . Krähenbühl . Laurent . Bideaux . Eric . Thomasset . Daniel . 2012-11-01 . Determination of Essential Orders From a Bond Graph Model . Journal of Dynamic Systems, Measurement, and Control . en . 134 . 6 . 10.1115/1.4004772 . 0022-0434. subscription .
  56. Bideaux . Eric . Marquis-Favre . Wilfrid . Scavarda . Serge . 2006 . Equilibrium set investigation using bicausality . Mathematical and Computer Modelling of Dynamical Systems . en . 12 . 2-3 . 127–140 . 10.1080/13873950500069003 . 1387-3954. subscription .
  57. Mouhib . Omar . Jardin . Audrey . Marquis-Favre . Wilfrid . Bideaux . Eric . Thomasset . Daniel . 2009-01-01 . Optimal control problem in bond graph formalism . Simulation Modelling Practice and Theory . Bond Graph Modelling . 17 . 1 . 240–256 . 10.1016/j.simpat.2008.04.011 . 1569-190X.
  58. Sanchez . Roberto . Dauphin-Tanguy . Geneviève . Guillaud . Xavier . Colas . Frédéric . 2010-10-01 . Bond graph based control of a three-phase inverter with LC filter – Connection to passive and active loads . Simulation Modelling Practice and Theory . 18 . 9 . 1185–1198 . 10.1016/j.simpat.2010.05.016 . 1569-190X. subscription .
  59. Vink . D. . Ballance . D. . Gawthrop . P. . 2006 . Bond graphs in model matching control . Mathematical and Computer Modelling of Dynamical Systems . en . 12 . 2-3 . 249–261 . 10.1080/13873950500068278 . 1387-3954. subscription .
  60. Ngwompo . R. F . Scavarda . S . 1999-12-15 . Dimensioning problems in system design using bicausal bond graphs . Simulation Practice and Theory . 7 . 5 . 577–587 . 10.1016/S0928-4869(99)00013-0 . 0928-4869. subscription .
  61. Ngwompo . Roger F. . Gawthrop . Peter J. . 1999-11-01 . Bond graph-based simulation of non-linear inverse systems using physical performance specifications . Journal of the Franklin Institute . 336 . 8 . 1225–1247 . 10.1016/S0016-0032(99)00032-0 . 0016-0032. subscription .
  62. Richard . P. Y. . Buisson . J. . Cormerais . H. . 2002-01-01 . ANALYSIS OF FLATNESS USING BOND GRAPHS AND BICAUSALITY . IFAC Proceedings Volumes . 15th IFAC World Congress . 35 . 1 . 25–30 . 10.3182/20020721-6-ES-1901.00254 . 1474-6670. subscription .
  63. Web site: Bond Graph Modelling of Engineering Systems .