Beruflich Dokumente
Kultur Dokumente
DOI 10.1007/s13369-017-2915-2
Abstract In hydraulic system, the occurrence of fault is a ited methodology is modified to tackle with the situations
common phenomenon which mainly results due to abnormal where the generation of symbolic expression for estimating
behavior of its decisive components leading to sudden alarm- the suspected parameters is not trivial due to nonlinearity
ing rise in some of its critical parameters, and hence, the need in equation. Therefore, the estimation problem is resolved
for condition monitoring techniques arises. Residuals which numerically in Simulink. Due to same reason of the absence
happen to be an indicator of faulty signal are equivalent to of symbolic expression, the observer-based residual is gen-
the number of sensors set up in the plant. Hence, the need erated instead of ARR-based residual.
is to place in considerable number of sensors throughout the
plant to create residuals, a prime requisite required for isolat- Keywords Hydrostatic transmission system · Parameter
ing fault. But it appears a tedious task and seems technically estimation · Condition monitoring · Residual · Fault
improbable to isolate all potential faults with the instruments signature matrix
at our disposal, and it can turn out to be an expensive affair
to put inexact number of sensors in order to evaluate indi-
vidually every physical state. To tackle with such problem, 1 Introduction
an already-developed methodology has been implemented on
the test setup, i.e., a closed-loop hydrostatic transmission sys- The use of hydraulic system in heavy earth moving machiner-
tem consisting of a variable-displacement pump and a fixed ies (HEMM) can be felt widespread in automotive, agricul-
displacement motor. In this way, an experimental exposure tural as well as mining sectors. The type of transmission
is added to the revisited methodology which was lacking. system employed in such HEMM is continuous transmission
Moreover, the parameter estimation technique in the revis- system. Hydrostatic transmission system (HST) is one among
these transmissions which has found tremendous application
because of its special features. These include high power-
to-weight ratio, continuously varying output speed, ease of
B Santosh Kr. Mishra operation and remarkable control. It is quite evident that these
apsantoshmishra@gmail.com
HEMM are supposed to be operating for considerable hours
Jay Prakash Tripathi per day and that too in difficult terrains consisting of low-
jay@mece.ism.ac.in
lying areas. Under these circumstances, there is high risk of
J. Das breaking down of the HEMM due to faulty behavior of some
jayanta_mech_engg@yahoo.co.in
of its key or critical parameters affecting major or minor
Sanjoy K. Ghoshal components.
sanjoy.ghoshal@gmail.com
Literatures pertaining where critical parameters severely
1 Department of Mining Machinery Engineering, Indian affect the performance of the HST systems are discussed sub-
Institute of Technology (ISM), Dhanbad 826004, India sequently. Dasgupta [1] analyzed the effect of some critical
2 Department of Mechanical Engineering, Indian Institute of parameters on the functioning of a HST. The HST under study
Technology (ISM), Dhanbad 826004, India or investigation focused on steady-state response which con-
123
Arab J Sci Eng
tained an orbital rotor (low-speed high-torque hydrostatic FDI techniques basically consist of two phases: raising
motor) and pressure-compensated swash plate-controlled of alarm or sensing fault and isolating the defective part.
variable-displacement pump. He also examined the effect on The behavior of the system is examined constantly to check
the swash plate angle caused by the variation in the torque of whether fault has occurred or not during the alarm phase.
the output shaft and ultimately identified the leakage of the Immediately after the declaration of fault, the purpose of the
system. Fisher [2] studied the flow characteristics, restoring isolation stage is to find out the faulty part or component of
moment and load capacity related to the swash plate consid- the system. The defective part is being isolated on the basis
ering the parallel and tilted positions of the slipper. Bergada of the structural characteristics corresponding to the residual
et al. [3] proposed that the swash plate angle happens to be (errors between sensor and observer outputs). The residuals
the main contributing factor in the variation of cylinder pres- are tracked in real time in order to detect the fault expe-
sure ripple which means that by the reduction of the swash rienced by the overall system. They assume approximately
plate angle, pressure ripple would be reduced. They devised close to zero value during regular working conditions; other-
that the output pressure was mainly responsible for the pres- wise, under defective conditions, they have nonzero values.
sure ripple which in turn results due to variation of turning The residuals may be generated either through observers or
speed and swash plate angle. Hence, it is apparent from the through parity relations [6]. Once fault is indicated by resid-
above discussion that critical parameters greatly influence ual, a binary fault signature matrix (FSM) is employed to
the operation of HST and any fault or deviation in it can lead localize the fault between different components of the system
to adverse issues. In order to mitigate these kinds of situa- [4,7–9]. FSM is basically a table composed of residuals in
tions, the prime need is to put in place desirable measures the column side and vulnerable parameters in the row side. It
through online monitoring ideologies such as fault detection comprises values which are of Boolean nature, e.g., a value of
and isolation (FDI) and fault tolerant control (FTC) which 1 in the row representing parameters sensitivity corresponds
can provide a viable solution to deal with these kind of issues to the residual’s sensitiveness whereas a value of 0 represents
and to guarantee an effective, secure and smooth working of its insensitiveness.
the overall system [4,5].
123
Arab J Sci Eng
Table 1 List of components used in test setup fault [8–10]. Another method of fault localization among the
S/n Item description parameters sensitive to unstructured residuals without opti-
mization is presented in [11]. But, it is only valid with the
1 Electric motor assumption that only one component of the system can be
2 Variable-displacement pump faulty at a time (single fault hypothesis). In this method, non-
3 Charge pump isolable fault candidates (parameters) belonging to a specific
4 Pressure transducer (system) fault subspace are estimated from analytical redundancy rela-
5 LSHT motor tions (ARR) involving the corresponding parameters, and
6 Gear box (1:20) then, they are used in specifically designated fault models in
7 Loading pump which nominal parameter values are used for all other com-
8 Pressure relief valve ponents. Then separate models incorporating these estimated
9 Flow transducer (inlet) values are simulated parallely with the plant once the fault in
10 Flow transducer (outlet) the system is detected. Thereafter, the contrast linked with the
11 Pressure transducer (outlet) model behaviors guides to ascertain the faulty parameters.
12 Speed indicator (electric motor The motivation for carrying out this work is drawn from
13 Torque indicator (electric motor) the application of the HEMM in mining as well as in low-
14 Flow indicator (motor return) lying areas, where due to malfunctioning of some critical
15 Pressure indicator (system)
components, process failures are faced. Hence, there arises
the need of the utilization of FDI and FTC techniques which
16 Flow indicator (motor inlet)
can help in the detection and isolation of the defective com-
17 Speed indicator (motor)
ponents such that the normal working of the machinery is
18 Torque indicator (motor)
restored. The defective part is being localized on the basis
19 Computer with cRIO
of the structural characteristics corresponding to the resid-
20 Swash plate command signal
ual (errors between sensor and observer outputs) with the
21 Pressure relief valve command
help of FSM. The work of fault localization via multi-model
simulation as reported in this article is motivated by the fact
that it appears a tedious task and seems technically improb-
In order to have clear demarcation regarding the localiza- able to isolate all potential faults with the instruments at our
tion of fault, it is of prime importance to generate structured disposal or it can turn out to be an expensive affair to put
residuals (individual residual are considered to be sensitive in exact number of sensors in order to evaluate individually
to a single parameter) which can be accomplished only by every state.
installing many sensors in the whole plant/system. Else all The major contributions of this work are that it provides
the parameters which are sensitive to unstructured residuals the experimental validation of the methodology applied by
are expected to be evaluated through optimization and com- Samantaray and Ghosal [11] which was missing. Addition-
pared to their respective nominal values for localization of ally, the modification in the parameter estimation technique
123
Arab J Sci Eng
123
Arab J Sci Eng
123
Arab J Sci Eng
Fig. 9 Variation of system pressure with command voltage Fig. 10 Variation of flow rate with command voltage
of cRIO are used for configuring the analog input and output analog input blog by the analog input module, and the exe-
operations. The LabVIEW back panel program is shown in cuting signal is sent to the output module via output blog.
Fig. 12. Analog input and analog output block are the basic A ‘while loop’ shown in gray color line in Fig. 12 having a
constituents of the FPGA program. The signal is fed to the ‘stop’ button for stopping the loop confines the program. The
123
Arab J Sci Eng
Fig. 12 LabView back panel program for data acquisition and PRV control
123
Arab J Sci Eng
Pm 1 d Pm
−Dm ωm + Cd Ao (Ps − Pm ) − = (2)
Rlkgm m blkstf dt
Tld
ωm = (4)
Jl
Fig. 14 Variation of experimental motor speed with time
The outlet flow of the loading pump is restricted by varying
the port opening area of the proportional PRV by changing its
• There is prompt response of the unloading relief valve cracking pressure. As a result, the full pump flow will not be
and the check valves. able to rotate fully due to the resistance being developed by
• The discharge of fluid from the pump is steady; hence, the partial closure of flow path by the PRV. This will result in
for simplicity, the consequence of the flow ripples on the development of back pressure which is being applied by the
dynamics of the system is neglected. pump on motor which is mainly due to the viscous frictional
• Although the hydraulic system under study is a closed- coefficient Rv as the fluid in the loading pump is viscous
loop system, it is simplified into open-loop system for fluid. Finally, with the variation of load in PRV, the viscous
modeling. friction coefficient also varies.
The plant model is allocated zero initial conditions, and
then, only simulation is performed. It should be noted that
The modeling of hydraulic components has been done the record of number of samples is being maintained from
with the help of constraint analytical relations. In this respect, the point when the initial conditions corresponding to the
the variation in the control volume of fluid with respect to observer are found and which is further employed to derive
time at pump plenum is given by: the history of the process for comparing with the models
which are faulty. Figures 16 and 17 depict the response
obtained from the sensors established in the plant along with
Ps 1 d Ps those obtained from simulation. The detailed descriptions
Dp ωp − Cd Ao (Ps − Pm ) − = (1)
Rlkgp pblkstf dt regarding the specifications of the sensors have been pro-
vided in Table 4. The variations in response obtained from
the estimated model and the one obtained experimentally
The external leakage flow of the pump is RPlkgp
s
, where Rlkgp from plant approximately match with each other. Now, this
is the leakage resistance associated with the pump and pblkstf validated model can be used for further fault isolation anal-
represents the bulk stiffness of the fluid at pump plenum. ysis.
123
Arab J Sci Eng
Alv Port opening area of the proportional PRV Electric motor speed 152 rad/s
Ao Throttle port opening area of check valve Variable-displacement bidirectional pump
Cd Coefficient of discharge Maximum displacement 4.45 × 10−6 m3 /rad
Dp Maximum pump displacement rate used Rotational speed 152 rad/s
in the closed circuit Leakage resistance of pump 1013 Ns/m5
Dm Volume displacement rate of the motor
Bulk stiffness of the pump 5 × 1010 N/m5
Gr Gear ratio
Hydraulic motor
J Generalized inertia of the driving shaft
Maximum displacement 4.45 × 10−5 m3 /rad
and the connected load
Leakage resistance of motor 1011 Ns/m5
Mblkstf Bulk stiffness of the hydraulic motor
Mtrlkg Leakage resistance of the hydraulic motor Bulk stiffness of the motor 5 × 1010 N/m5
123
Arab J Sci Eng
Speed sensor
Model S T-60
RPM range 0–500
Accuracy Within 0.1% FSR ± 1 count
Flow sensor
Model TFM-1015
Flow range 2–50 lpm
Accuracy ± 0.5% over 10–100% flow range
Pressure sensor
Model S-10
Pressure range 0–100 Bar
Fig. 16 Comparison between predicted and the experimental pump Accuracy ≤ 0.25%
plenum pressure
r1 r2 r3
Mb Ib
ϕ1 1 0 0 1 1
ϕ2 0 1 0 1 1
ϕ3 U 0 0 1 1 0
Up1
ϕ4 0 0 1 1 0
ϕ5 Up
0 1 1 1 0
Up2
ϕ6 0 1 1 1 0
Fig. 17 Comparison between predicted and the experimental motor
speed
where algebraic expressions for generation of residual and Every single parameter φi ∈ U p (i = 1, . . . , m ) is being
parameter estimation are not available. In order to present a estimated with the idea that the remaining parameters ( j =
clear picture of the methodology, it is being again discussed 1, . . . , m and i = j) are considered to be nominal. This
here. ultimately heads to models m in number that needs to be
Suppose the number of sensors corresponds to n and solved at the same time.
apparently residuals will also be n in number. The residual A FSM shown in Table 5 further explains the definitions
set has been represented by rn . Let U = (ϕ1 , ϕ2 , . . . , ϕm ) discussed above, where ϕi = 1, . . . , 6 are independently
denotes the set of parameters and Ud stands for assigned X behaving parameters and r j , j = 1, . . . , 3 represent the
dependent parameters. So, the set or space containing inde- three considered residuals. A variable is monitored (Mb = 1)
pendent parameters is equal to; Ui = U − Ud . By taking the when it is present at nothing less than one residual. The
help of instruments at our disposal, l (l ≤ m − X ) parame- defective parts or components are isolated utilizing FSM
ters can be isolated from fault in FSM. Now, the parameters that accounts for the involvement of several constituents in
which cannot be isolated can be put in a subset or subspace every residual. The defect in any parameter associated with
U p (dimension m–X –p) which happens to be non-empty. the component is isolable (Ib = 1) under the condition that
Moreover, suppose n n ≤ n residuals, rn (rn ⊂ rn ), that firstly it should be monitored and its signature as seen in resid-
are suspected to be not behaving normally, as per decision uals is dissimilar with respect to the remaining variables and
process, are likely to be having multiple similarity in the parts/components.
FSM. So, U p ⊂ U p comprises the set (if m refers to their The coherence vector C = [1 0 0] can be assigned these
number) of suspect parameters that are included in every values in case residual 1 is faulty if at all it has a distinct
residual elements of rn . signature in the FSM and finally the element representing
123
Arab J Sci Eng
parameter ϕ1 is considered as isolable and is regarded as a which are suspected. The recursive parameter estimation is
faulty prospect. In the same way, ϕ2 can be localized under done in MATLAB–Simulink 2015 framework. These esti-
the condition that residual r2 is behaving abnormally. The mated parameters are incorporated in m parallel Simulink
faults associated with remaining parts cannot be isolated, and models and run parallely and the obtained responses are com-
hence, a set is constructed as U p = U p1 ∪ U p2 . In the case pared with that of the plant. The magnitude of the error among
of abnormality of residual r3 U p = U p1 , and under the con- ‘m’ models helps in identifying the faulty parameters. The
dition if residuals r2 and r3 both are showing abnormality, minimum error corresponding to a model implies that the sus-
U p = U p2 . Based on the concept of single fault hypoth- pected parameter which was incorporated after estimation is
esis which is being implemented in the FSM, the starting only having actual fault. The block diagram of the general
step involved in isolating fault furnishes a list of suspected condition monitoring methodology followed has shown in
prospects or candidates F ∈ [ϕ1 ϕ2 (ϕ3 , ϕ4 ) (ϕ5 , ϕ6 )], in Fig. 18.
which further refining is necessary for end two elements via The methodology proposed above is now being imple-
parameter evaluation. FSM alone is not capable to localize all mented in the hydraulic system under consideration to
the defective elements. The refining process is called second- examine its effectiveness in condition monitoring in a real-
level isolation technique. When required instruments are not istic situation.
available in the system, then only the need arises for higher-
level isolating techniques.
Eventually, in the second-level isolation, the fault isolation
methodology is linked with estimation of process parameters
123
Arab J Sci Eng
It is quite clear from Fig. 1 that two sensors are present in the
test setup. Consequently, it is possible to get two residuals.
The residuals deduced from the model are given as below:
Fig. 20 Variation of residual response with time corresponding to
r1 = Ps − Ps (5) motor speed
r2 = ωm − ωm (6)
Table 6 Fault signature matrix Residuals Mb Ib
The normalization of residuals r1 and r2 is done consid- (FSM)
ering the uncertainties linked with the process and is given r1 r2
below:
Parameters
ri |t = tk − i (tk )ηi (u k ) Dp 1 1 1 0
ri |t = tk = (7)
zΦi (u k )σi (u k ) Rlkgp 1 1 1 0
Dm 1 1 1 0
where ri represents the ith residual of kth sample, (tk ) and Rlkgm 1 1 1 0
φ(tk ) represents the time varying coefficients by changing Rv 1 1 1 0
the mean and the variance respectively, u k corresponds to Tld 1 1 1 0
the input, tk represents the time, ηi stands for residual mean
and z is a coefficient associated with the confidence level
[12].
The normalized residual curves corresponding to mea- From FSM, it can be deduced that the parameter subspace
surements shown in Figs. 13 and 14 have been shown in sensitive to residuals r1 and r2 are {Dp , Rlkgp , Dm , Rlkgm ,
Figs. 19 and 20, respectively. In healthy condition, the max- Rv , Tld }. Apparently, only a single parameter of this sub-
imum deviation in normalized residuals lies within ± 1. space can be faulty under single fault hypothesis. However,
Hence, ± 1 is considered as fixed threshold and shown by it is not possible to isolate them with the FSM only (first-
dash line. level isolation as shown in Fig. 18). Hence, the second-level
For isolation of faulty components, the FSM is constructed isolation process is being employed by incorporating the esti-
and is shown in Table 6. Model Builder software [13] can also mated value of the suspected parameters (Dm , Rv ) via two
be employed to build this. parallel operating multi-block Simulink models.
123
Arab J Sci Eng
Ps
Dp ωp − Cd Ao (Ps − Pm ) − =0 (8)
Rlkgp Fig. 23 Estimation of suspected parameter R̂v
Pm
−Dm ωm + Cd Ao (Ps − Pm ) − =0 (9)
Rlkgm of occurrence of fault has already been detected. From that
Dm Pm − Rv ωm − Tld = 0 (10) time, a time lag of 2 s is introduced before estimation of the
parameters. This is done in order to obtain the measurement
To solve Dm or Rv from Eq. (10), Pm need to be numeri- at steady state. The estimation done under steady-state con-
cally evaluated either from Eq. 8 or Eq. 9 and then substituted ditions are shown in Figs. 22 and 23, respectively.
in Eq. 10. The algebraic expression of Pm is comparatively
complex if obtained from Eq. 9 rather than Eq. 8. To avoid this 5.3 Fault Isolation Using Multi-Block Parallel
complexity, Pm is obtained numerically in Simulink which is Simulation Method
done quite simply even with Eq. 9 as shown in Fig. 21. It can
be seen that the Simulink model is containing an algebraic Until this juncture, the scenario is such that it is possible to
loop. To solve this loop, a loop solving routine is being called estimate the defective parameters on the basis of the response
by Simulink software during every time step. The solution got from the residual, but it is not feasible to clearly justify the
pertaining to the problem is ascertained with the loop solver right estimation or simply it can be concluded the localization
by performing iterations. of fault is not possible. The solution of the problem is pro-
Based on the estimation, the corresponding values of D̂m vided by parallel simulation technique [14] which has been
and R̂v corresponding to the model as obtained are given in applied here for isolating the fault. The isolation of faults is
Figs. 22 and 23, respectively. From residual plots, the time possible only by utilizing two models because of the exis-
123
Arab J Sci Eng
6 Conclusions
123
Arab J Sci Eng
a fixed displacement motor which provides an experimental 8. Samantaray, A.K.; Bouamama, B.O.: Isolation of structurally non-
validation of the proposed technique that can be quite benefi- isolatable faults. Model Based Process Superv. Bond Gr. Approach.
1, 347–372 (2008)
cial in understanding its aspects linked with physical system. 9. Samantaray, A.K.; Ghoshal, S.K.: Sensitivity bond graph approach
to multiple fault isolation through parameter estimation. Proc. Inst.
Mech. Eng. Part I J. Syst. Control Eng. 221(4), 577–587 (2007)
References 10. Borutzky, W.: Isolation of multiple parametric faults from a hybrid
model. In: Borutzky, W (ed.) Bond Graph Model-Based Fault Diag-
nosis of Hybrid Systems, pp. 123–148. Springer, New York (2015)
1. Dasgupta, K.: Analysis of a hydrostatic transmission system using
11. Samantaray, A.K.; Ghoshal, S.K.; Chakraborty, S.; Mukherjee, A.:
low speed high torque hydrostatic motor. Mech. Mach. Theory
Improvements to single-fault isolation using estimated parameters.
359(10), 1481–1499 (2000)
Simulation 81(12), 827–845 (2005)
2. Fischer, M.J.: A Theoretical Determination of Some Characteris-
12. Samantaray, A.K.; Ghoshal, S.K.: Sensitivity bond graph approach
tics of a Tilted Hydrostatic Slipper Bearing. British Hydromechan-
to multiple fault isolation through parameter estimation. Proc. Inst.
ics Research Association, Cambridge (1962)
Mech. Eng. Part I J. Syst. Control Eng. 221(4), 577–587 (2007)
3. Bergada, J.M.; Kumar, S.; Davies, D.L.; Watton, J.: A complete
13. Bouamama, B.O.; Samantaray, A.K.; Medjaher, K.; Staroswiecki,
analysis of axial pump leakage and output flow ripples. Appl. Math.
M.; Dauphin-Tanguy, G.: Model builder using functional and bond
Model. 36(4), 1731–1751 (2012)
graph tools for FDI design. Control Eng. Pract. 13(7), 875–891
4. Wang, D.; Yu, M.; Low, C.B.; Arogeti, S.: Model-Based Health
(2005)
Monitoring of Hybrid Systems. Springer, New York (2013)
14. Ghoshal, S.K.; Samantaray, A.K.; Mukherjee, A.: Improvements to
5. Blanke, M.; Kinnaert, M.; Lunze, J.; et al.: Diagnosis and Fault-
single fault isolation using estimated parameters. In: Proceedings
Tolerant Control. Springer, New York (2010)
of ICBGM’05, pp. 301–306 (2005)
6. Venkatasubramanian, V.; Rengaswamy, R.; Yin, K.; Kavuri, S.N.:
A review of process fault detection and diagnosis. Part 1: quanti-
tative model based methods. Comput. Chem. Eng. 27(3), 293–311
(2003)
7. Bouamama, B.O.; Medjaher, K.; Bayart, M.; Samantaray, A.K.;
Conrard, B.: Fault detection and isolation of smart actuators using
bond graphs and external models. Control Eng. Pract. 13(2), 159–
175 (2005)
123