Professional Documents
Culture Documents
2 Tank System Bondgraph
2 Tank System Bondgraph
2 Tank System Bondgraph
net/publication/257073197
CITATIONS READS
87 630
4 authors, including:
Geneviève Dauphin-Tanguy
École Centrale de Lille
238 PUBLICATIONS 1,897 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Modelling of Rolling Flexible Disc on Flexible Table System by Bond Graph Methods and Analysis View project
All content following this page was uploaded by Arun Kumar Samantaray on 21 October 2016.
Keywords: Bond graph, fault detection and isolation, FDI, on fault signatures deduced from Analytical Redundancy
analytical redundancy, residual Relations (ARRs) [1]. Modelling is an important and
difficult step in ARRs generation because of the complexity
Abstract of the monitored system along with its control equipment.
Any physical system in its normal operating condition obeys The bond graph methodology is widely used for modelling
certain mathematical constraints described by its behavioural purpose but only few works deal with monitoring of
model, which essentially is a set of algebro-differential complex systems using the bond graph tool. The structural
equations for the states of the system that can be derived control properties (controllability, observability) can be
directly from its bond graph model. However, in a fault deduced graphically from a bond graph model [2].
detection and identification paradigm, concurrent and initial Moreover, this approach allows the actuators and sensors
values of states are unknown and such constraints need to be placement for the model to be structurally observable or /
derived only in terms of known variables, i.e. the sources and and controllable, though only valid for linear or linearized
the measurements. Such constraint relations (also termed system models.
Analytical Redundancy Relations or ARRs) should always
be valid within a certain bound of error, while evaluated The causal properties of the bond graph model were initially
using measured data from the real system. Any discrepancy used for the determination of the origin of the faults for FDI
in holding one or more of these constraints is an indicator of and supervision [1]. In [3], the bond graph modelling is
fault(s) in some system component. Isolation of the faulty used not only to determine the causes of faults, but also to
component can be done based on the structural properties of quantify the effects of component fault using the
the ARRs. quantitative description of the system. However, this
method is applicable only to systems that operate at a steady
This paper deals with a new approach in derivation of the state (rate of states = 0.0). Published process monitoring
ARRs from the bond graph model of the system. This methodologies using bond graph [1] are constrained to
approach also generates prior indications of inversion of using 1-port elements and scalar bonds (modelling single
functional relations and redundancy of measurements, energy). Another approach for quantitative FDI based on
thereby providing a graphical means to optimise sensor temporal causal graphs [4] does not extend to non-linear
placements. This approach is illustrated using an example of systems. In contrast, the qualitative approach [5] while
a Two-Tank system. providing good monitorability indices is poor in fault
isolation due to the swarm of postulated possibilities that
INTRODUCTION need to be eliminated.
The operational safety of process engineering systems is
based essentially on fault detection and isolation (FDI) METHODOLOGY
procedures. These procedures consist of the comparison Analytical redundancy relations (ARRs) are derived in
between the actual behaviour of the system and reference terms of the parameters of the system, sources and the
behaviour. Different approaches for the FDI procedures have measurements. The number of redundancy relations
been developed, depending on the kind of knowledge used to derivable from any system model is equal to the number of
describe the process model (transfer function, state equation, sensors in the system [6]. The formal approach to the
structural model). Structural monitorability analysis is based derivation is illustrated using the example of a mass-spring-
damper system in Figure 1(a). The bond graph model is
shown in figure 1(b) with integral causalities; from which
behavioural equations can be derived.
MODIFIED METHOD
The proposed method is based on imposing the negative of
measured quantities on the system (as a pseudo source) and
equating the reactive factor to zero. This is equivalent to
writing the junction equations in an alternative form (e.g. for
above system, e5 = e1 – e3 – e4 – e2= 0). Imposition of the
measured quantities implies inversion of causality of the
corresponding port, wherever possible, without violating
other causality norms and using a preferred differential
causality in storage elements. Those sensors for which
causality inversion cannot not be effected are redundant Figure 1(c). Model in derivative causality
sensors [2], but they may still be required in the system to
isolate faults in some part of the system or sensors. The
inverted causality model is shown in Figure 2. The equations
for this model can be written as follows.
F(t) – 1/C2 1/s mf5 - I3 s mf5 – R4 mf5 = 0. Figure 2. Model with inverted detector causality
This relation can be differentiated to remove the integration The valve Vf1 and Vf2 can be used to simulate leakage
of mf5 and restructured to yield the same residual obtained respectively in tanks T1 and T2. The connection pipe
through the earlier approach. between tanks is placed at the bottom of the tanks. The
model variables and process parameters are given in the
Observations and Guidelines, Step –1 : table below. The output of all controllers and sources are
1. The deduction results in the residual directly by- measured signal values.
passing ARR formation.
2. The number of extra derivatives needed for integration Table 1. Model Parameters
free residuals is the maximum number of storage Symbol Description Value Units
elements in integral causality in all separate causal Hydraulic flow
Cvb 1.5938*10-4 m5/2/s
paths leading to the measurement. Conversely, the coefficient of valve Vb
derivative order in the residual for a measurement is Hydraulic flow
Cv0 1.5964*10-4 m5/2/s
the sum of the maximum of the number of storage coefficient of valve V0
elements in derivative and the maximum of the Cross-sectional area of
Ai (i=1,2) 1.54*10-2 m2
number of storage elements in integral causality; in all tanks Ti (i=1,2)
separate causal paths leading to that measurement. Height of water in tanks
hi (i=1,2) Variable m
3. If a second velocity pickup is used in the system, the Ti (i=1,2)
Df element corresponding to it cannot be assigned himax Maximum height of
inverse causality on the bond graph model. This shows 0.6 m
(i=1,2) water in tanks Ti (i=1,2)
up in the model bond graph as a redundant Maximum outflow from
measurement [2]. Qpmax 0.01 m3/s
pump P1
4. In presence of multiple sensors, if some residuals are Leakage flow from tanks
dependent (contain same variables), their linear Qfi 10-4 m3/s
Ti (i=1,2) in faulty mode
combinations can produce independent residuals. Set point of the PI level
h1c 0.5 m
controller
Application to Two-Tank System Proportional gain of PI
Kp 10-3 m-1
The example of a two-tank system illustrated in Figure 3 is level controller
considered next. The main aim of the two tanks is to provide Integral gain of PI level
a continuous water flow Q0 to a consumer. The process Ki 5*10-6 m-1/s
controller
consists of two tanks T1 and T2 connected by a pipe with a
valve. Tank T1 is filled by a pump P1 up to a nominal water The bond graph model for the system using inverted
level of h1=0.5m. The water level in this tank T1 is controlled causality in measurement ports is shown below. In the
by a PI level controller acting on the inlet flow Qp provided diagram, the controller signal measurements are not shown.
by the pump. The water flow between tanks T1 and T2 can be
controlled by a valve Vb using an “On-Off” controller in
order to keep the water level h2 in tank T2 at the medium level
(0.09m K2 11m). The quantity of water outflow Q0 to
the consumer is controlled by the valve V0.
Figure 3. Scheme of two-tank system Figure 4(b). Model using Symbols 2000 [7] capsules
Constitutive relations for the elements can be written as .. These residuals are
f1 = mQp, f2 = CT1 s e2, f8 = CT2 s e8, e10 = 0, mUp - ΦPI (me1) = 0, (R3)
f5 = ΦRvb (e5, mUb) = Cvb sign (e5) sqrt(|e5|) mUb, mQp - ΦC1 (me1) = 0, (R4)
f11 = ΦRv0 (e11, mU0) = Cv0 sign (e8) sqrt(|e8|) mU0, mUb - ΦC2 (me2) = 0. (R5)
where sign and sqrt are functions for sign and square root of All these residuals are automatically derived by FDIPad
variables; and CTi = Ai (i=i,2). The strong relations for the component module of Symbols 2000 [7].
junctions are
The above residuals resulting from redundant measurements
e1 = e2 = e3 = e4 = me1, simply relate output of some control device to its input. Bond
e6 = e7 = e8 = e9 = me2, graphically speaking, such devices are function blocks with
f4 = f5 = f6, fixed causality and the input signal is a known measured
f9 = f10 = f11; quantity. When the input signal is imposed on the model
through causality inversion, the causality of the measured
whereas the weak relations are signal on the output side cannot be inverted. This shows up
the redundancy of measurement [2] (also termed an algebraic
e5 = e4 – e6 = me1 – me2 , e11 = e9 – e10 = e9 = me2, linkage) in the model as illustrated in the figure 5, where a
f3 = f1 – f2 – f4 = 0, (a) bond graph realisation of the signal domain is used.
f7 = f6 – f8 – f9 = 0. (b)
{
cannot be evaluated for the other causal posture. An example
mQp = 0 if mUp ΦC1 (me1) of this is the field thermal resistance in conduction, where
mQpmax if mUp > mQpmax rate of change of entropy can be expressed in terms of effort
variables for temperature but not the vice versa. In certain
cases, though the inverse relation exists, it turns out to be
0 if me2 ≥ 11cm
{
singular in some operating regime. Let us consider a simple
mUb = 1 if me2 FP ΦC2 (me2) example of two tanks connected with a regulated valve and
Hysteretic state (Previous state) the flow through the valve is measured. The bond graph
If 9cm< me2 < 11cm model of such a system is shown in figure 6 and the model
with inverted measurement causality is shown in figure 7. It
Since these functions can be replaced in the two available can be verified from structural analysis [2] that one mode of
residuals as purely functions of me1 and me2, the controller the system is structurally unobservable because only one of
output measurements mQp, mUp and mUb are redundant in the storage elements can be assigned preferred derivative
the system. However, their presence leads to three more causality and the dualisation of the observer does not help.
residuals which can be used to isolate faults in the pump, the However, an additional effort sensor in any of the two tanks
PI controller and the On-Off controller, respectively. makes the entire system observable.
SIMULATION RESULTS
The two-tank system has been simulated with software
Symbols 2000 [7] under normal operating mode, with a tank
leakage fault in T1 and with valve blocking fault in Vb.
FDIPad program in Symbols 2000 has been used to enter the
object model of the system, derive the set of independent
residuals, the behavioural model in symbolic form and the
corresponding compound simulation model including faults.
REFERENCES
[1] Tagina M.; J.P.Cassar; G.Dauphin-Tanguy; M.
Staroswiecki. 1995. “Monitoring of Systems
Modelled by Bond Graph”, Proc. ICBGM’95 (Las
Vegas) 275-280.
The results shown in Figure 8 show perfect behaviour of the [7] Mukherjee A.; A.K.Samantaray. 2001. “System
system in normal operation. The residuals are noisy signals Modelling Through Bond Graph Objects on
and their mean values are approximately equal to 0.0 during SYMBOLS 2000”,Proc. ICBGM’01, SCS Publication,
Vol. 33, No. 1, 164-170.