Inter national J our nal of Electrical and Computer Engineering (IJECE) V ol. 8, No. 4, August 2018, pp. 2157 2171 ISSN: 2088-8708 2157       I ns t it u t e  o f  A d v a nce d  Eng ine e r i ng  a nd  S cie nce   w     w     w       i                       l       c       m     A New Hybrid Rob ust F ault Detection of Switching Systems by Combination of Obser v er and Bond Graph Method Mohammad Ghasem Kazemi and Mohsen Montazeri Department of Electrical and Computer Engineering Shahid Beheshti Uni v ersity , A.C., T ehran, Iran Article Inf o Article history: Recei v ed August 4, 2017 Re vised March 15, 2018 Accepted: April 5, 2018 K eyw ord: Switching system Rob ust f ault detection F ault sensiti vity Disturbance attenuation Bond Graph ABSTRA CT In this paper , the problem of rob ust F ault Detection (FD) for continuous time switched system is tackled using a h ybrid approach by combination of a switching observ er and Bond Graph (BG) method. The main criteria of an FD system including the f ault sen- siti vity and disturbance attenuation le v el in the presence of parametric uncertainties are considered in the proposed FD system. In the first stage, an optimal switching observ er based on state space representation of the BG model is designed in which simultaneous f ault sensiti vity and disturbance a ttenuation le v el are satisfied using H =H 1 inde x. In the second stage, the Global Analytical Redundanc y Relations (GAR Rs) of the switch- ing system are deri v e d based on the output estimation error of the observ er , which is called Error -based Global Analytical Redundanc y Relations (EGARRs). The paramet- ric uncertainties are included in the EGARRs, which define the adapti v e thresholds on the residuals. A constant term due to the ef fect of disturbance is also considered in the thresholds. In f act, a tw o-stage FD system is proposed wherein some criteria may be considered in each stage. The ef ficienc y of the proposed method is sho wn for a tw o-tank system. Copyright c 2018 Institute of Advanced Engineering and Science . All rights r eserved. Corresponding A uthor: M. G. Kazemi Shahid Beheshti Uni v ersity East V af adar Blvd., T ehranpars, T ehran, Iran +989177428946 Email mg kazemi@sb u.ac.ir 1. INTR ODUCTION F ault diagnosis has become an important topic in industrial applications, which further leads to more attention in research community . Model based f ault detection, isolation and identification has recei v ed much attention in the literature [1],[2],[3],[4],[5],[6]. Generally , obtaining a proper model of the system is crucial for f ault diagnosis design. Presenting a proper model by considering multi ph ysics nature and the interaction between continuous and discrete dynamics of the system is a challenging task in the model based FD system design for h ybrid systems. Bond Graph (BG) method may be considered as a con v enient tool for modeling and f ault diagnosis of dynamic systems. Dif ferent applications of the BG method can be found in the literature for modeling [7],[8],[9],[10],[11],[12] and f ault diag- nosis [12],[13],[14],[15],[16] applications. Because of its ef ficienc y in modeling and f ault diagnosis of continuous dynamic systems, the BG method is de v eloped to represent switching phenome n on in h ybrid dynamic systems, which leads to numerous applications of the method for modeli ng and f ault diagnosis of h ybrid systems in the literature. Due to the adv ancement of computer technology in control systems, switching systems ha v e dra wn much attention in recent years. Switching systems may be assumed as an important class of h ybrid systems wherein a finite number of dynamical subsystems and a logic rule define the o v erall dynamic of the system. The dynamic of the subsystems and thus, the o v erall system may be continuous or discrete, linear or nonlinear and so on. In this current paper , the linear continuous case of the switching system with a v erage dwell time is considered. F ault detection of the switching systems dra ws much attention in recent years. In [17], f ault detection J ournal Homepage: http://iaescor e .com/journals/inde x.php/IJECE       I ns t it u t e  o f  A d v a nce d  Eng ine e r i ng  a nd  S cie nce   w     w     w       i                       l       c       m     DOI:  10.11591/ijece.v8i4.pp2157-2171 Evaluation Warning : The document was created with Spire.PDF for Python.
2158 ISSN: 2088-8708 problem for a class of linear switched time v arying delay system is formulated as H 1 filtering problem with ADT in LMI form. A filter based f ault detection approach for continuous time switching systems with ADT is considered in [18]. Simultaneous F ault Detecti on and Control (SFDC) for continuous and discrete time linear switched systems with ADT based on dynamic filter is tackled in [19] whe rein a f ault sensiti vity is achie v ed for a gi v en disturbance attenuation le v el based on H =H 1 criteria. The rob ustness of an FD system, which is defined as the considering the ef fects of uncertainties, distur - bances and noises in the system, may be re g arded as the most important criteria in an FD system. Ho we v er , it is impossible to completely remo v al the ef fects of uncertainties, disturbances and noises in an FD system. Thus, man y studies are de v oted to attenuate the ef fects of disturbances, uncertainties and noises in an FD system. In other hand, the attenuation of disturbance ef fects may lead to reduce the f ault sensiti vity in an FD system. Therefore, the f ault sensiti vity must also be considered in an FD system design. The ef fects of uncertainties in the model may l ead to missed or f alse alarms in the FD system. T o w ard this end, tw o approaches are proposed in the literature including acti v e and passi v e approaches [20]. In acti v e approach, the rob ustness of the FD system is defined in the residual generation stage, while the rob ustness is considered in the residual e v aluation stage in the passi v e approach [20]. In [21], the rob ust f ault detection filter design of continuous time switched delay system s is considered. Design of observ er -based rob ust po wer system stabilizers by considering parametric uncertainties is noticed in [22]. The parametric uncertainty representation in the BG method in Linear Fractional T r ansformation (LFT) form is presented in [23] and noticed in thereafter studies for rob ust f ault diagnosis [24],[25]. Rob ust f ault diagnosis of ener getic system with parameter uncertainties is tackled in [24]. The nonlinear modeling, structural analysis, residual generation with adapti v e thresholds and sensiti vity analysis are done using the BG method. The proposed method is then used for a boiler system. The modeling and f ault diagnosis of a DC motor relying on the BG method is gi v en in [16]. The parameters of the BG method are obtained using the real data of the system. The a v erage v alues of the parameters and their standard de viations are assumed as the nominal v alues and uncertain parts of the parameters, respecti v ely . In [25], the residuals and thresholds generation in presence of parameter uncertainties in LFT form is considered and is applied for mechatronic systems. Rob ust f ault diagnosis and prognostic s of a hoisting mechanism based on the BG method considering the LFT form of the parametric uncertainties is addressed in [26]. Se v eral studies are concentrated on the f ault diagnosis of switched system and rob ust f ault diagnosis of continuous time systems as well, while there are fe w w orks on the rob ust f ault diagnosis of switched systems. Incremental BG approach is gi v en for h ybrid systems in [27], which is an e xtension of the proposed method in [28] for continuous time dynamic systems. The adapti v e thresholds are obtained considering the parametric uncertainties in the system, which are dependent on the modes of the h ybrid system as well. In [29], a rob ust f ault detection and isolation on the basis of pseudo BG model for h ybrid systems is presented. The parametric uncertainties are gi v en in the LFT form and are assumed 2% of their nominal v alues at maximum. Rob ust FD systems based on the BG method may be considered as the passi v e approach, since some thresholds are assumed in the FD system to detect or isolate the occurred f aults. In this paper , based on the BG method, a ne w acti v e rob ust FD system is proposed in which the disturbance is attenuated, the ef fects of uncertainties are considered and the f ault sensiti vity is enhanced as well. The proposed method has the benefits of both the BG and the observ er method. In summary , the main contrib utions of the paper may be stated as follo ws: A ne w rob ust acti v e f ault detection based on the combination of the observ er and BG method for linear switched system is presented, which simultaneously attenuates disturbance le v el and enhances f ault sensi- ti vity . Error form of the GARRs is used as the residuals which is based on the output estimation error of the observ er . The ef fects of disturbance and parametric uncertainties in the thresholds are considered as a fix ed threshold and an adapti v e threshold, respecti v ely . The remainder of the paper is or g anized as follo ws. The problem formulation and some lemmas and re- marks based on these lemmas are gi v en in section 2. The Research Method of the paper is gi v en in v e subsections of section 3 as disturbance attention le v el, enhanced f ault sensiti vity , simultaneous optimal f ault sensiti vity and disturbance attenuation le v el, the GARRs in error form based on the B G method and rob ustness of the FD system ag ainst parametric uncertainties. The simulation results and discussion for a tw o-tank system are gi v en in section 4, follo ws by conclusion in section 5. IJECE V ol. 8, No. 4, August 2018: 2157 2171 Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE ISSN: 2088-8708 2159 2. PR OBLEM FORMULA TION AND PRELIMIN ARIES Consider the continuous time linear switching system state space representation in the form of _ x ( t ) = A i x ( t ) + B i u ( t ) + B di d ( t ) + B f i f ( t ) (1) y ( t ) = C i x ( t ) + D di d ( t ) + D f i f ( t ) (2) which can be gi v en as Eq.(3) - Eq.(4) based on the BG model of the switching system. _ x ( t ) = A ( a; ) x ( t ) + B ( a; ) u ( t ) + B d ( a ) d ( t ) + B f ( a ) f ( t ) (3) y ( t ) = C ( a; ) x ( t ) + D d ( a ) d ( t ) + D f ( a ) f ( t ) (4) where x 2 R n is the states, y 2 R p is the outputs, u 2 R k is the inputs, d 2 R q is disturbances and f 2 R r is f aults in the switched system. A , B and C are respecti v ely system, input and output matr ices with appropriate dimensions denote the system , input and output matrices, respecti v ely . . B d , B f , D d and D f are the matrices of disturbance and f ault distrib ution on the states and on the outputs of the system, respecti v ely . The i; i 2 l = 1 ; 2 ; : : : ; N inde x is considered to define modes of the switched system. The a = [ a 1 a 2 : : : a z ] is the state of the controlled junctions in the BG model, which will define the acti v e mode of the switching system. The number of controlled junctions in the BG model is assumed as z . The system matrices are dependent on the parameters of the BG model ( R , C , L , . . . ) which represented as in the model. The switching Luenber ger observ er is assumed in the form of Eq.(5). _ ^ x ( t ) = A i ^ x ( t ) + L i ( y ( t ) ^ y ( t )) (5) ^ y ( t ) = C i ^ x ( t ) where L i is the observ er g ain and is dependent on the acti v e mode of the BG model. The observ er g ain is designed for each mode in such a w ay that some criteria are satisfied. The output estimation error , which is defined as Eq.(6), is used in the ne w form of the GARRs. e y ( t ) = y ( t ) ^ y ( t ) (6) The asymptotical stability of the observ er is in v estig ated on the error dynamic of the observ er as Eq.(7). _ e ( t ) = ( A i L i C i ) e ( t ) + ( B f i L i D f i ) f ( t ) + ( B di L i D di ) d ( t ) = A cl i e ( t ) + B cl f i f ( t ) + B cl di d ( t ) (7) The output estimation error is achie v ed as Eq.(8). e y ( t ) = C i e ( t ) + D f i f ( t ) + D di d ( t ) (8) In this paper , tw o criteria are considered for the output estimation error of the observ er for disturbance attenuation le v el and f ault sensiti vity , which is presented in Eq.(9) and Eq.(10), respecti v ely . sup jj e y ( t ) jj 2 jj d ( t ) jj 2 < ; > 0 (9) inf jj e y ( t ) jj 2 jj f ( t ) jj 2 > ; > 0 (10) The follo wing definitions and lemmas are used in the current paper . Definition . Let N ( t 1 ; t 2 ) stand for the s witching number of ( t ) on the interv al [ t 1 ; t 2 ) for a gi v en switching signal ( t ) and an y t 2 > t 1 > t 0 . If Eq.(11) is satisfied for N 0 0 and a > 0 , then a is called the ADT and N 0 is chatter bound [30]. N ( t 1 ; t 2 ) N 0 + ( t 2 t 1 ) = a (11) Lemma 1 : F or the continuous time switchi ng system Eq.(1) and Eq.(2), le t > 0 , 1 and i > 0 ; 8 i 2 l be constant scalars. Suppose there e xists positi v e definite C 1 function V ( t ) : R n ! R ; 2 l with V ( t 0 ) ( x ( t 0 )) = 0 such that [31]: _ V i ( x t ) V i ( x t ) y T t y t + 2 i u T t u ; 8 i 2 l (12) A Ne w Hybrid Rob ust F ault Detection of Switc hing Systems by ... (Mohammad Ghasem Kazemi) Evaluation Warning : The document was created with Spire.PDF for Python.
2160 ISSN: 2088-8708 V i ( x tk ) V j ( x tk ) ; 8 ( i; j ) 2 l l ; i 6 = j then the switching system is Globally Uniformly Asymptotically Stable (GU AS) and will satisfied Eq.(9), which is called H 1 performance, with inde x no greater than max( i ) for an y switching signal with ADT as Eq.(13). a > a = l n ( ) = (13) V is called the L yapuno v function. Remark 1 : In order to use the Lemma 1 in this paper , the u and y are substituted by d ( t ) and e y ( t ) , respecti v ely , which indicate the disturbance attenuation le v el on the output estimation error of the observ er . Lemma2 : F or the continuous time switching system Eq.(1) and Eq.(2), let > 0 , 1 and i > 0 ; 8 i 2 l be constant scalars. Suppose there e xists positi v e definite C 1 function V ( t ) : R n ! R ; 2 l with V ( t 0 ) ( x ( t 0 )) = 0 , which is called L yapuno v function, such that [32]: _ V i ( x t ) V i ( x t ) + y T t y t 2 i u T t u ; 8 i 2 l (14) V i ( x tk ) V j ( x tk ) ; 8 ( i; j ) 2 l l ; i 6 = j then the switching system is GU AS and will satisfied H performance as Eq.(10) with inde x no smaller than min( i ) for an y switching signal with ADT as Eq.(13). Remark 2 : In order to use the Lemma 2 in this paper , the u and y are substituted by f ( t ) and e y ( t ) , respecti v ely , which indicate the f ault sensiti vity on the output estimation error of the observ er . Lemma 3 [33]: F or a gi v en m m symmetric matrix Z 2 S m and tw o matrices U and V of column dimension m , there e xists matrix X that is unstructured and will satisfy Eq.(15). U T X V + V T X T U + Z < 0 (15) if and only if the follo wing inequalities as Eq.(16) and Eq.(17) with respect to X are satisfied. N T U Z N U < 0 (16) N T V Z N V < 0 (17) N U and N V are arbitrary matrices that their columns are a basis for null spaces of U and V , respecti v ely . 3. RESEARCH METHOD In this section, the LMI formulation for f ault detection problem of li near continuous time switched system Eq.(1)-Eq.(2) is gi v en. The o v erall results are gi v en in v e subsections. 3.1. Disturbance attenuation perf ormance in the output estimation err or of the switched obser v er The switching L yapuno v function for disturbance attenuati on criterion of the observ er is considered as Eq.(18), which its deri v ati v e must be ne g ati v e definite as Eq.(19). V ( t ) = e ( t ) T P i e ( t ) > 0 (18) _ V ( t ) = _ e ( t ) T P i e ( t ) + e ( t ) T P i _ e ( t ) < 0 (19) The switching L yapuno v function deri v ati v e may be obtained as Eq.(20) in f ault free case. _ V ( t ) = ( e ( t ) T A T cl i + d ( t ) T B T cl di ) P i e ( t ) + e ( t ) T P i ( A cl i e ( t ) + B cl di d ( t )) (20) which can be written as Eq.(21) for simplicity . _ V ( t ) = e ( t ) T ( A T cl i P i + P i A cl i ) e ( t ) + 2 e ( t ) T P i B cl di d ( t ) (21) According to Remark 1, we need e y ( t ) T e y ( t ) for using lemma 1, which can be gi v en as: e y ( t ) T e y ( t ) = e ( t ) T C T i C i e ( t ) + 2 e ( t ) T C T i D di d ( t ) + d ( t ) T D T di D di d ( t ) (22) IJECE V ol. 8, No. 4, August 2018: 2157 2171 Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE ISSN: 2088-8708 2161 Thus, Eq.(12) may be obtained as follo ws using Lemma 1. e ( t ) T ( A T cl i P i + P i A cl i ) e ( t ) + 2 e ( t ) T P i B cl di d ( t ) + V i ( t ) + e y ( t ) T e y ( t ) 2 d ( t ) T d ( t ) < 0 (23) Equation (23) can be gi v en as Eq.(24) by re g arding Eq.(18) and Eq.(22), which may be represented in matrix inequality form as Eq.(25). e ( t ) T ( A T cl i P i + P i A cl i + P i ) e ( t ) + 2 e ( t ) T P i B cl di d ( t ) + e ( t ) T C T i C i e ( t )+ 2 e ( t ) T C i D di d ( t ) + d ( t ) T D T di D di d ( t ) 2 d ( t ) T d ( t ) < 0 (24) A T cl i P i + P i A cl i + P i P i B cl di 0 + C T i C i C T i D di D T di D di 2 I < 0 (25) One may obtain Eq.(26) by some simplifications in Eq.(25). I A T cl i 0 B T cl di P i P i P i 0 I 0 A cl i B cl di + 0 C T i I D T di 2 I 0 0 I 0 I C i D di < 0 (26) By assuming Eq.(26) as N T U Z N U , we ha v e[32]: I A T cl i 0 0 B T cl di I 2 4 P i + C T i C i P i C T i D di 0 0 2 I + D T di D di 3 5 : 2 4 I 0 A cl i B cl di 0 I 3 5 < 0 (27) Thus, it can be concluded that: 2 4 P i + C T i C i P i C T i D di 0 0 2 I + D T di D di 3 5 < 0 (28) Therefore, U may be achie v ed as [ A cl i I B cl di ] taking into account that N U columns are basis for the null space of U . Furthermore, according to [32], N V and V are assumed as: N V = 2 4 I 0 0 0 0 I 3 5 V = [0 I 0] By using abo v ementioned results, Eq.(15) in Lemma 3 may be obtained as: 2 4 A T cl i I B T cl di 3 5 X i [0 I 0] + 2 4 0 I 0 3 5 X i [ A cl i I B cl di ] + 2 4 P i + C T i C i P i C T i D di 0 0 2 I + D T di D di 3 5 < 0 (29) which can be cast into one LMI as Eq.(30). LM I (1) : 2 4 11 12 13 22 23 33 3 5 < 0 (30) where 11 = P i + C T i C i + A T i X i + X T i A i N i C i C T i N T i 12 = P i X T i + A T i X i C T i N T i 13 = C T i D di + X T i B di N i D di 22 = X i X T i 23 = X T i B di N i D di 33 = D T di D di I = 2 ; N i = X T i L i A Ne w Hybrid Rob ust F ault Detection of Switc hing Systems by ... (Mohammad Ghasem Kazemi) Evaluation Warning : The document was created with Spire.PDF for Python.
2162 ISSN: 2088-8708 3.2. F ault sensiti vity perf ormance in the output estimation err or of the switched obser v er By considering the L yapuno v function as Eq.(31) for f aulty case, its deri v ati v e may be obtained as Eq.(32). V ( t ) = e ( t ) T P f i e ( t ) (31) _ V ( t ) = _ e ( t ) T P f i e ( t ) + e ( t ) T P f i _ e ( t ) = e ( t ) T ( A T cl i P f i + P f i A cl i ) e ( t ) + 2 e ( t ) T P f i B cl f i f ( t ) (32) According to Remark 2, we need e y ( t ) T e y ( t ) for using lemma 1, which can be gi v en as Eq.(33): e y ( t ) T e y ( t ) = e ( t ) T C T i C i e ( t ) + 2 e ( t ) T C T i D f i f ( t ) + f ( t ) T D T f i D f i f ( t ) (33) According to Remark 2 and Lemma2, one may obtain that: e ( t ) T ( A T cl i P f i + P f i A cl i ) e ( t ) + 2 e ( t ) T P f i B cl f i f ( t ) + e ( t ) T P f i e ( t ) e ( t ) T C T i C i e ( t ) 2 e ( t ) T C T i D f i f ( t ) f ( t ) T D T f i D f i f ( t ) + 2 f ( t ) T f ( t ) < 0 (34) which can be arranged into Eq.(35). e ( t ) T ( A T cl i P f i + P f i A cl i + P f i ) e ( t ) + 2 e ( t ) T P f i B cl f i f ( t ) e ( t ) T C T i C i e ( t ) 2 e ( t ) T C T i D f i f ( t ) + f ( t ) T ( D T f i D f i + 2 I ) f ( t ) < 0 (35) The matrix inequality representation of Eq.(35) is gi v en as Eq.(36). A T cl i P f i + P f i A cl i + P f i P f i B cl f i 0 + C T i C i C T i D f i 2 I D T f i D f i < 0 (36) Equation Eq.(36) is gi v en as Eq.(37) in form of N T U Z N U . I A T cl i 0 0 B T cl f i I 2 4 P f i C T i C i P f i C T i D f i 0 0 0 0 2 I D T f i D f i 3 5 2 4 I 0 A cl i B cl f i 0 I 3 5 < 0 (37) Thus, Z matrix may be obtained as follo ws. Z = 2 4 P f i C T i C i P f i C T i D f i 0 0 2 I D T f i D f i 3 5 (38) U = [ A cl i I B cl f i ] ; V = [0 I 0] Using Eq.(15) in Lemma 3, we ha v e: 2 4 A T cl i I B T cl f i 3 5 X i [0 I 0] + 2 4 0 I 0 3 5 X i [ A cl i I B cl f i ] + 2 4 P f i C T i C i P f i C T i D f i 0 0 2 I D T f i D f i 3 5 < 0 (39) Altogether , a linear matrix inequality as Eq.(40) is achie v ed by assuming = 2 . LM I (2) : 2 4 11 12 13 22 23 33 3 5 < 0 (40) where 11 = P f i C T i C i + A T i X i + X T i A i N i C i C T i N T i 12 = P f i X T i + A T i X i C T i N T i 13 = C T i D f i + X T i B f i N i D f i 22 = X i X T i 23 = X T i B f i N i D f i 33 = I D T f i D f i in which, , N i , P f i and X i are unkno wn v ariables of the problem. IJECE V ol. 8, No. 4, August 2018: 2157 2171 Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE ISSN: 2088-8708 2163 3.3. Simultaneous Optimal F ault sensiti vity and Disturbance Attenuation Le v el According to Eq.(30) and Eq.(40), the disturbance attenuation le v el and f ault sensiti vity are defined in tw o separate LMI problems, which must be considered simultaneously in the observ er design at a gi v en disturbance attenuation le v el or f ault sensiti vity . In this paper , by defining a weighted LMI optimization problem, the optimal v alues of the disturbance attenuation and f ault sensiti vity are obtained. The theorem is gi v en as follo ws. Theor em : Consider the continuous time switched system as Eq.(1)-Eq.(2) with ADT along with the switched observ er a s Eq.(5). The state estimation error dynamic Eq.(7) is asymptotically stable and satisfies the performance criteria as Eq.(9) and Eq.(10) for an y nonzero d ( t ) 2 l 2 [0 ; 1 ) , if there be the scalars v alues > 0 , > 0 and the matrices P i > 0 Q i > 0 X i N i such that the subsequent weighted LMI optimization problem has solution. min ( P i ;Q i ;X i ;N i ;; ) w (1 w ) s:t:LM I (1) < 0 ; LM I (2) < 0 P i < P j ; P f i < P f j where 0 w 1 is the weighting f actor for simultaneous f ault sensiti vity and disturbance attenuation perfor - mances and its v alue is defined by minimization of the follo wing cost function: min w Lastly , the g ains of the observ er for dif ferent modes of the switched system, optimal disturbance attenuation le v el and optimal f ault sensiti vity of the output estimation error of the observ er are achie v ed as: L i = X T i N i ; = p ; = p 3.4. Global Analytical Redundancy Relations in Err or F orm Using BG method The generic form of a GARR may be gi v en in Eq.(41) [34]. GAR R :   ( y ( n ) ; : : : ; y ; u ( m ) ; : : : ; u; ; a ) (41) where y , u and are denoted for outputs, inputs and parameters of the system, respecti v ely . As in Eq.(3)-Eq.(4), a = [ a 1 a 2 : : : a z ] is a binary v ector that determine the acti v e mode of the switched system.. The GARRs are dif ferential equations, which their orders are dependent on the sensor locations in the system. The lo wer order GARRs are more ef ficient for monitoring of a system. The generic form of the GARRs may be presented in matrix form by maximum second order assumption of deri v ati v es. GAR R :   ( y ; _ y ; y ; u; _ u; u; ; a ) = M 1 ( ; a ) y ( t ) + M 2 ( ; a ) _ y ( t ) + M 3 ( ; a ) y ( t )+ Z 1 ( ; a ) u ( t ) + Z 2 ( ; a ) _ u ( t ) + Z 3 ( ; a ) u ( t ) (42) The ef fects of disturbances, noises and parametric uncertainties on the GARRs may lead to f alse or missed alarms in the FD system. In this paper , a ne w method by combination of the BG method and observ er method is proposed to o v ercome these problems in the FD system based on the BG method, which further leads to some benefits in the FD system. In normal case i.e . in the case that there are no disturbances, f aults, noises and uncertainties in the system, the estimated states and outputs of the observ er con v er ge to the states and outputs of the system. Therefore, the GARRs by using estimated outputs may be gi v en as: GAR R h :   ( ^ y ; _ ^ y ; ^ y ; u; _ u; u; ; a ) = M 1 ( ; a ) ^ y ( t ) + M 2 ( ; a ) _ ^ y ( t ) + M 3 ( ; a ) ^ y ( t )+ Z 1 ( ; a ) u ( t ) + Z 2 ( ; a ) _ u ( t ) + Z 3 ( ; a ) u ( t ) (43) By subtracting Eq.(43) from Eq.(42), the GARRs in the error form, which is called EGARRs, are obtained. E GAR R : M 1 ( ; a ) e y ( t ) + M 2 ( ; a ) _ e y ( t ) + M 3 ( ; a ) e y ( t ) (44) The output est imation errors in Eq.(44) come from the observ er , which is designed in such a w ay that be more sensiti v e to f aults and simultaneously the ef fect of disturbance is attenuated. Therefore, the ef fects of the disturbances are reduced in the GARRs, while the f ault sensiti vity is enhanced. A Ne w Hybrid Rob ust F ault Detection of Switc hing Systems by ... (Mohammad Ghasem Kazemi) Evaluation Warning : The document was created with Spire.PDF for Python.
2164 ISSN: 2088-8708 Figure 1. The tw o-tank system 3.5. Rob ustness of the FD system against parametric uncertainties The f ault detection and isolation are based on the EGARRs, which are dependent on the parameters of the BG model. In order to obtain a rob ust FD system ag ainst parametric uncertainties, the proposed met hod in [23] is considered in this paper . In rob ust f ault detection based on the BG method, a GARR can be decoupled into nominal (GARRn) and uncertain part (UGARR) as follo ws. GAR R = GAR R n ( a; n ; u; y ) + U GAR R ( a; ; u; y ) (45) where n is the nominal part and is relati v e de viation compared to the nominal v alue of the parameter . The nominal part is the obtained GARRs, which contain the nominal v alues of the parameters and made the residuals. The uncertain part contains the uncertain parameters of the nominal GAR Rs, which further deter - mines the upper and lo wer bounds of the residuals. An EGARR may be gi v en as Eq.(46) re g arding this f act that the EGARRs are not directly dependent on the inputs and their distrib ution matrices. E GAR R = E GAR R n ( a; n ; e y ) + U E GAR R ( a; ; e y ) (46) The adapti v e thresholds on the residuals are defined as UEGARRs, which are dependent on the parametric uncertainty ( ), mode of the system ( a ) and the output estimation error of the observ er ( e y ). As mentioned abo v e, the proposed method is not directly dependent on the inputs and distrib ution matrices of the inputs ( Z 1 to Z 3 ). Thus, the rob ustness of the proposed method may be impro v ed. Finally , the residual e v aluation function and thresholds for f ault detection are assumed as Eq.(47) and Eq.(48), respecti v ely . J L ( t ) = jj r k jj 2 = ( l 0 + L X l 0 r T k r k ) 0 : 5 (47) J th = sup d ( t ) ;u ( t ) 2 l 2 ;f ( t )=0 J L ( t ) + jj k jj 2 (48) in which J L ( t ) is the residual e v aluation function, J th is the threshold on the residual e v aluation function, l 0 is the initial ti me, L is the windo w length for residual e v aluation function calculations and k is the ef fect of parametric uncertainties on the residual, which is obtained from UEGARR. Finally , the f aulty or f ault free conditions are declared by comparison between the residual e v aluation function and considered threshold. 4. RESUL TS AND DISCUSSION Consider the tw o-tank system in Fig.1, which in v olv e a flo w source, tw o output v alv es named as R 1 and R 2 and one interconnection v al v e ( R 3 ). The v alv es could be in open or closed state, which determined the acti v e dynamic of the switched system. Three configurations as T able I are assumed [35]. The BG models of the understudy tw o-tank system in inte gral and deri v ati v e causality in 20-sim softw are are sho wn as Fig.2 and Fig.3, respecti v ely . As mentioned earlier , the v alv es could be in open or closed state. The coef ficients a 1 , a 2 and a 3 are used to sho w the v alv e states of R 1 , R 2 and R 3 , res pecti v ely , which are set to ”0” in closed state and ”1” in open state. IJECE V ol. 8, No. 4, August 2018: 2157 2171 Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE ISSN: 2088-8708 2165 T able 1. Considered modes of the tw o-tank system Mode R 1 R 2 R 3 A = [ a 1 a 2 a 3 ] Mode 1 of f on of f [0 1 0] Mode 2 on on of f [1 1 0] Mode 3 of f on on [0 1 1] Figure 2. BG model of tw o-tank system in inte gral causality . In h ydraulic domain, the ef fort and flo w v ariables are re g arded as pressure and flo w , respecti v ely . The subscript c is considered to determine the c o nt rolled junctions of the model. Three controlled junctions are assumed for three v alv es. In BG theory of h ybrid systems, the flo w v ariable of the connected bonds to the i th 1-type controlled junction is multiplied by a i . T w o pressure sensors are used to measure pressures of the tw o tanks which is gi v en as P 1 and P 2 . The state space representation of a h ybrid system may be obtained in compact form using the BG model. The st ates of the system are considered as generalized momentum of inertia elements ( p I ) and generalized dis- placement of capaciti v e elements ( q C ). Hence, the state v ector of the tw o-tank system is achie v ed as [ q 2 q 9 ] T . The flo w through the v alv es is assumed laminar . This assumption leads to linear equations of the system [35]. Using the state space deri v ation method in [34], the go v erning equations in compact form are: _ q 2 = ( a 1 R 1 C T 1 a 2 R 2 C T 1 ) q 2 + a 2 R 2 C T 2 q 9 + q in (49) _ q 9 = a 2 R 2 C T 1 q 2 + ( a 2 R 2 C T 2 a 3 R 3 C T 3 ) q 9 y 1 = 1 C T 1 q 2 ; y 2 = 1 C T 2 q 9 The parameters definition and their v alues are gi v en in T able II. It is also noted that the tank capacity is defined as C T = A=g wherein g = 9 : 81 m=s 2 is the acceleration of gra vity . Thus, the system matrices for three considered modes of the tw o-tank system are: A 1 = 0 : 0007 0 : 0011 0 : 0007 0 : 0011 ; C 1 = 0 : 1962 0 0 0 : 3270 Figure 3. BG model of tw o-tank system in deri v ati v e causality . A Ne w Hybrid Rob ust F ault Detection of Switc hing Systems by ... (Mohammad Ghasem Kazemi) Evaluation Warning : The document was created with Spire.PDF for Python.
2166 ISSN: 2088-8708 T able 2. P arameters of tw o-tank system V ariable Definition V alue A 1 Cross sectional area of tank 1 50 m 2 A 2 Cross sectional area of tank 2 30 m 2 R 1 Output pipe of tank 1 resistance 300 s=m 2 R 2 Interconnection pipe resistance 300 s=m 2 R 3 Output pipe of tank 1 resistance 100 s=m 2 0 0.2 0.4 0.6 0.8 1 0.18 0.2 0.22 0.24 0.26 0.28 0.3 0.32 weight Disturbance attenuation level 0 0.2 0.4 0.6 0.8 1 0.34 0.36 0.38 0.4 0.42 0.44 weight Fault sensitivity Figure 4. Disturbance attenuation le v el and f ault sensiti vity v ersus weighting f actor A 2 = 0 : 0013 0 : 0011 0 : 0007 0 : 0011 ; C 2 = 0 : 1962 0 0 0 : 3270 A 3 = 0 : 0007 0 : 0011 0 : 0007 0 : 0044 ; C 3 = 0 : 1962 0 0 0 : 3270 B 1 = B 2 = B 3 = 1 0 The other matrices of the system as in Eq.(1) and Eq.(2) are assumed as follo ws. B d 1 = B d 2 = B d 3 = B 1 ; B f 1 = B f 2 = B f 3 = 1 0 0 1 D d 1 = D d 2 = D d 3 = 0 : 1 0 : 1 ; D f 1 = D f 2 = D f 3 = 0 : 3 0 0 : 1 0 : 5 The GARRs of the system may be obtained using the Causality In v ersion Method (CIM) in [34]. GAR R 1 = C T 1 dP 1 dt + a 2 R 2 ( P 1 P 2 ) + a 1 R 1 P 1 q in = 0 (50) GAR R 2 = C T 2 dP 2 dt a 2 R 2 ( P 1 P 2 ) + a 3 R 3 P 2 = 0 The matrices M i and Z i can be calculated as: M 1 = 0 ; M 2 = C T 1 0 0 C T 2 ; Z 1 = Z 2 = 0 M 3 = a 2 R 2 + a 1 R 1 a 2 R 2 a 2 R 2 a 2 R 2 + a 3 R 3 ; Z 3 = 1 0 The upper bound of parameter uncertainty is assumed as 2% of their nominal v alues [29]. Solving the weighted LMI optimization problem for = 0 : 08 , = 1 : 2 and = 0 : 9 , the disturbance attenuation le v el and f ault sensiti vity v ersus weighting f actor are gi v en in Fig.4. By minimization the cost function of the Theorem, we ha v e: w = 0 : 8 IJECE V ol. 8, No. 4, August 2018: 2157 2171 Evaluation Warning : The document was created with Spire.PDF for Python.