Indonesian J our nal of Electrical Engineering and Computer Science V ol. 17, No. 3, March 2020, pp. 1150 1156 ISSN: 2502-4752, DOI: 10.11591/ijeecs.v17i3.pp1150-1156 r 1150 Actuator fault detection and isolation f or r obot manipulator using higher order sliding mode obser v ers Khaoula Oulidi Omali, Mohammed Nabil Kab baj, Mohammed Benbrahim USMB A, LIST A LIST A, Sidi Mohamed Ben Abdellah Uni v ersity Fez, Morocco Article Inf o Article history: Recei v ed Jun 1, 2019 Re vised Aug 7, 2019 Accepted Oct 3, 2019 K eyw ords: F ault detection and isolation Higher order sliding mode unkno wn input observ ers Nonlinear system Robot manipulators ABSTRA CT F ault detection and isolation (FDI) method for actuator f aults for robot manipulator by applying a model-based FDI f ault, which may happen on a specific component of the system. T o detect and isolate actuator f aults, higher order sliding mode Unkno wn Input Observ ers (UIO) are proposed to mak e analyt ical redundanc y . The observ ers input la ws are designed according to the so-called Super -T wisting Second Order Sliding Mode Control (SOSMC) technique and the y are pro v ed to be able to guarantee the e xponential con v er gence of the f ault estimate to the actual f ault signal. The simulation results sho w the ef fecti v eness and rob ustness for the proposed approach. Copyright c 2020 Insitute of Advanced Engineeering and Science . All rights r eserved. Corresponding A uthor: Khaoula Oulidi Omali, USMB A, LIST A, LIST A, Sidi Mohamed Ben Abdellah Uni v ersity Fez, Morocco. Email: khaoula.oulidiomali@usmba.ac.ma 1. INTR ODUCTION A precise dynamic model of the system to be tested must be formula ted to design a high performance controller . Its parameters should be equally precisely identified. Furthermore applies to industrial robots. in that e v ent, the problem of identification becomes especially comple x because of the presence of nonlinearities and coupling ef fects, typical of robotic systems, by dint of the inertial, friction torques, gra vity , centripetal and friction torques [4]. The presence of a f ault can be modeled as an une xpected change in system parameters or by the presence of unkno wn signals in the industry . In a robot manipulator , a f ault be able to occur on a actuator , a sensor or a mechanical component of the system . specific Actuator and sensor f aults are more common thanks to the presence of electrical de vices, which possibly be subject for a lot of possible criticality . Diagnostic de vices generate on-line diagnostic signals in order to detect and isolate the presence of f ault. Specific methods considered to surmount this disadv antage, as lik e the use of Kalman filters [6], or gen- eralized moments, see [8]. These approaches, in the presence of typical uncert ainties of applied applications, can not ha v e the possibility e xactly to con v er ge from the state of the system to the s tate of the observ er . T o minimize this dra wback, sliding mode techniques are also frequently adopted to perform status observ ation [10, 11] o wing to their simplici ty of design and rob ustness. Usually , the FDI can be treat by incorporating v arious sliding mode observ ers [12, 13]. The aim of this paper is to study the performance in terms of the rob ustness and diagnostic capabil ities of sliding-mode input la w for the observ er . specifically , second-order sliding mode (SOSM) la w , the super - twisting la w [17] is considered. The diagnostic technique proposed in this w ork pro v es t o be capable to detect non-simultaneous sensor and actuator f aults, specially for actuator . J ournal homepage: http://ijeecs.iaescor e .com Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 r 1151 2. THE MANIPULA T OR MODEL The equations of motion of an n De gree of Fredeem (DOF) robot manipulators are described accord- ing to the Euler -Lagrange theory , as: = M ( q ) q + C ( q ; _ q ) _ q + G ( q ) + F ( _ q ) = M ( q ) q + n ( q ; _ q ) (1) where q ; _ q ; q 2 R n are the joint position, v elocity and ac celeration v ectors, respecti v ely , M ( q ) 2 R ( n n ) is the inertia matrix (symmetrical definite positi v e, thus, M ( q ) 1 al w ays e xists), C ( q ; _ q ) 2 R ( n n ) is the centrifug al and Coriolis matrix, G ( q ) 2 R n is the gra vitational v ector , F ( _ q ) 2 R n is the v ector of viscous friction torque at the joints. No w , introducing the v ariables x 1 ( t ) = q ( t ) , x 2 ( t ) = _ q ( t ) , model 1 can be re written in state space representation as: 8 > < > : _ x 1 ( t ) = x 2 ( t ) _ x 2 ( t ) = f ( ( t ) ; x 1 ( t ) ; x 2 ( t )) h ( t ) = x 1 ( t ) (2) Where the term f ( ( t ) ; x 1 ( t ) ; x 2 ( t )) is obtained after simple algebric manipulation of (1), i.e., f ( ( t ) ; x 1 ( t ) ; x 2 ( t )) = M 1 ( x 1 ( t ))( ( t ) n ( x 1 ( t ) ; x 2 ( t ))) (3) As pre viously mentioned, when f aults af fect the actua tors, the input torque for the mechanical system i s dif fer - ent from ( t ) . Then, in case of input f aults, (1) becomes: ( t ) + ( t ) = M ( x 1 ( t )) _ x 2 ( t ) + n ( x 1 ( t ) ; x 2 ( t )) (4) and, as a result, the state space representation is: 8 > < > : _ x 1 ( t ) = x 2 ( t ) _ x 2 ( t ) = f ( ( t ) + ( t ) ; x 1 ( t ) ; x 2 ( t )) q ( t ) = x 1 ( t ) (5) where f ( ( t ) + ( t ) ; x 1 ( t ) ; x 2 ( t )) is analogous to (3). In practice, model (3) is not e xactly kno wn and must be identified. Then, in case of f aults, the follo wing relationship holds. ( f ( ( t ) ; x 1 ( t ) ; x 2 ( t )) = M 1 ( x 1 ( t ))( ( t )+ ( t ) ^ n ( x 1 ( t ) ; x 2 ( t )) ( t )) (6) ( t ) = n ( x 1 ( t ) ; x 2 ( t )) ^ n ( x 1 ( t ) ; x 2 ( t )) (7) When ( t ) is uncertain and ^ n ( q ; _ q ) is the kno wn part of the model. Y et, by virtue of the particular application considered, ( t ) can be assumed to be bounded. Ob viously , to perform f ault diagnosis, one has to rely only on the kno wn part of model (3). Indeed, after a suitable identification procedure, such as the one proposed in [19], it is feasible (in absence of f aults) to determine only an approximated representation of f ( : ) , i.e. f ( ( t ) ; x 1 ( t ) ; x 2 ( t )) = M 1 ( x 1 ( t ))( ( t ) ^ n (( x 1 ( t ) ; x 2 ( t )) (8) in order that the actually usable model is: 8 > < > : _ x 1 ( t ) = x 2 ( t ) _ x 2 ( t ) = ^ f ( ( t ) ; x 1 ( t ) ; x 2 ( t )) q ( t ) = x 1 ( t ) (9) 3. THE PR OPOSED DIA GNOSTIC SCHEME By relying on the so-called Unkno wn Input Observ er (UIO) approach [11], ef ficient estimators of the input torques can be designed [21]. In this w ork, the UIOs of sliding mode type is proposed in order to detect the actuator f aults. The proposed UIOs can be jointly described as a multi-input-multi-state second order sliding mode observ ers. Actuator fault detection and isolation for r obot manipulator ... (Khaoula Oulidi Omali) Evaluation Warning : The document was created with Spire.PDF for Python.
1152 r ISSN: 2502-4752 3.1. Obser v er Design Let us consider the observ er: ( ^ x 1 ( t ) = ^ x 2 ( t ) + z 1 ( t ) ^ x 2 ( t ) = ^ f ( ( t ) ; x 1 ( t ) ; ^ x 2 ( t )) + z 2 ( t ) (10) where ^ x 1 ( t ) ; ^ x 2 ( t ) 2 R 2 n are the observ er states, and z ( t ) = [ z 1 ( t ) ; z 2 ( t )] T is an auxiliary input signal, which is designed relying on the sliding mode approach, as will be clarified. This signal is introduced so as to permit and guarantee the con v er gence of the observ er states to the actual state of the system. Each component of z ( t ) is an input la w of the observ er . 3.2. Dynamics of the Obser v er Err or The proposed f ault diagnostic scheme requires to steer to zero the signal e ( t ) = [ e 1 ( t ) ; e 2 ( t )] T 2 R 2 n , the components of which are gi v en by: ( e 1 ( t ) = x 1 ( t ) ^ x 1 ( t ) e 2 ( t ) = x 2 ( t ) ^ x 2 ( t ) (11) By steering to zero these quantit ies, it is possible to guarantee that the observ er (10) gi v es a good estimation of the unkno wn input, as it will be sho wn in the follo wing. The dynamics of the error v ariable e ( t ) is represented by a second order dynamical system: 8 > < > : _ e 1 ( t ) = e 2 ( t ) z 1 ( t ) _ e 2 ( t ) = f ( ( t ) ; x 1 ( t ) ; x 2 ( t )) ^ f ( ( t )+ ( t ) ; x 1 ( t ) ; ^ x 2 ( t )) z 2 ( t ) (12) which can be re written as: ( _ e 1 ( t ) = e 2 ( t ) z 1 ( t ) _ e 2 ( t ) = M 1 ( x 1 ( t )) ( ( t ) ( t )) z 2 ( t ) (13) No w , Second Order Sliding Mode approach is studied to design the multi-input-multi-state UIO input la w . This approach is the so-called Super -T wisting [17]. The proposal will be depicted in the ne xt subsections. 3.3. Super -T wisting based Obser v er The design of the observ er input la ws which are the components of z ( t ) = [ z 1 ( t ) ; z 2 ( t )] T using a Super -T wisting based approach (see [17]) is gi v en by: ( z 1 ( t ) = p j s 0j sig n ( s 0 ( t )) z 2 ( t ) = sing ( s 0 ( t )) (14) Where s 0 ( t ) = e 1 ( t ) = x 1 ( t ) ^ x 1 ( t ) . It can be pro v ed that a suitable choice of and e xists such that, starting from an y initial condition [ e 1 (0) ; e 2 (0)] T , the condition: ( e 1 ( t ) = 0 e 2 ( t ) = 0 (15) is guaranteed in finite time (the proof of this claim can be de v eloped as in [14]). T o implement the proposed method, the terms and ha v e been chosen after an e xperimental tuning procedure. Note that the term z 2 ( t ) is a discontinuous signal and, by virtue of the filt ering action considered in [22], the second equation of the system (13) can be re written as: z 2 eq ( t ) = M 1 ( x 2 ( t ))( ( t ) ( t )) (16) where z 2 eq ( t ) is the equi v alent input signal corresponding to the discontinuous signal z 2 ( t ) . Thus, theoret- ically , the equi v alent input signal is the result of an infinite switching frequenc y of the discontinuous term Indonesian J Elec Eng & Comp Sci, V ol. 17, No. 3, March 2020 : 1150 1156 Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 r 1153 sing ( s 0 ( t )) . In f act, the implementation of the observ er produces high switching frequenc y (since, in prac- tice, one can only implement z 2 ( t ) as in (14) and not z 2 eq ( t ) ) making necessary the application of a filter to obtain useful information from signal z 2 ( t ) . The filter has to eliminate the high frequenc y components of such a signal. It can be of the form: p _ z eq ( t ) + z eq ( t ) = z 2 ( t ) : (17) Indeed, in [25], it w as sho wn that : lim p ! 0 z eq ( t ) = z 2 eq ( t ) (18) Then, by taking a small p it is possible to assume that the equi v alent input l a w (16) is similar to the output of the filter . 4. THE CONSIDERED F A UL T SCEN ARIOS In this w ork, the occurrences of f aults on inputs of a robot manipulator is considered. In this sit uation, the real torque applie d by the actuators is unkno wn. That is, 2 R n being the nominal torque calculated by the robot controller , whil e 2 R n being the input f ault, the actual torque v ector which is the input of the robotic system, can be written as ( t ) = ( t ) + ( t ) as sho wn in Figure 1. Figure 1. The proposed FDI scheme for actuator f aults 5. RESIDU AL GENERA TION Error of the state estimation r ( t ) can be calculated as: r ( t ) = ( t ) ^ ( t ) (19) The residuals are supposed to dif fer from zero in the present of f aults ( r ( t ) 6 = 0 ) and to be zero when there are no f aults on the actuators ( r ( t ) = 0 ). So the residuals are defined as: ( r ( t ) = 0 if = ^ r ( t ) 6 = 0 if 6 = ^ (20) T able I represents the f ault signatures matrix for these residuals. W e find that the signatures for each of the f ailures are quite dif ferent. T able 1. Signature T able for Actuator F ault Isolation d i =r i r 1 r 2 r 3 r 4 r 5 d 1 1 0 0 0 0 d 2 0 1 0 0 0 d 3 0 0 1 0 0 d 4 0 0 0 1 0 d 5 0 0 0 0 1 6. SIMULA TION RESUL TS In this section, the performances of the proposed FDI scheme for robot manipulators are v erified, by simulating actuator f aults. T o carry out simulations, the model (5) has been simulated together with the Actuator fault detection and isolation for r obot manipulator ... (Khaoula Oulidi Omali) Evaluation Warning : The document was created with Spire.PDF for Python.
1154 r ISSN: 2502-4752 observ er (10) wi th the input la ws (14) rele v ant to the Super -T wisting approach. The presence of actuator f aults is simulated by introducing an abrut f ault signal on the dif ferent articulation of the robot (joint 1, 2, 3, 4, 5, respecti v ely). The simulation sho ws f ault detection and isolation for the v e articulations of robot manipulators. Detection and isolation of the f ault s for actuat o r s ( and ^ signals) by using the Super -T wisting input la w . Figures 2, 3, 4, 5, 6 pres ents tw o signals, one for the actual states and the other for the estimated states. Figures are simulated during the time of T = 10 s . The kind of f ault is ”Abrut”. the dif ference between the tw o signals gi v es a residual for each joint of the system. Figure (b) in figures 2, 3, 4, 5, 6 sho ws an observ ation er ror between the actual states and the estimated states. Residuals for all articulations are dif ferent from each other and react according to signature table for actuator f ault isolation. The f ault is appeared at the time t = 3 s . This methods gi v es a good results in comparison between the original state and the state es timate, therefore the state estimate con v er ges to the actual state rapidly , that’ s wh y we find the residuals equal to zero. So this proposed technique detect and isolate the actuator f aults in a robot manipulator . (a) F ault signal reconstruction (b) Residual signal for the first actuator Figure 2. Simulation of FDI on the first actuator ( and ^ signals). Detection and isolation of the f aults by using the Super -T wisting input la w . (a) F ault signal reconstruction (b) Residual signal for the second actuator Figure 3. Simulation of FDI on the second actuator ( and ^ signals). Detection and isolation of the f aults by using the Super -T wisting input la w . (a) F ault signal reconstruction (b) Residual signal for the third actuator Figure 4. Simulation of FDI on the third actuator ( and ^ signals). Detection and isolation of the f aults by using the Super -T wisting input la w . Indonesian J Elec Eng & Comp Sci, V ol. 17, No. 3, March 2020 : 1150 1156 Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 r 1155 (a) F ault signal reconstruction (b) Residual signal for the fourth actuator Figure 5. Simulation of FDI on the fourth actuator ( and ^ signals). Detection and isolation of the f aults by using the Super -T wisting input la w . (a) F ault signal reconstruction (b) Residual signal for the v e actuator Figure 6. Simulation of FDI on the v e actuator ( and ^ signals). Detection and isolation of the f aults by using the Super -T wisting input la w . 7. CONCLUSION The problem of f ault detection and isolation on a robot manipulator has been addressed. The pres ence of f ault detection is performed depending on higher order sliding mode Unkno wn Input Observ ers (UIOs). The observ er input la ws are designed by the so called Super -T wisting Second Order Sliding Mode Control (SOSMC). The proposed scheme allo ws one to detect and isolate f aults, e v en multiple and simultaneous, on the actuators of the robotic system. Simulations are presented pro v es The ef fecti v eness and the performance of the proposed technique. REFERENCES [1] L. M. Capisani, A. Ferrara, and L. Magnani, “Second order sliding mode motion control of rigid robot manipulators, Proc. 46th IEEE Conf. Decision Control, Ne w Orleans , LA, 2007. [2] C. Edw ards, S. K. Spur geon, and R. J. P atton, “Sliding mode observ ers for f ault detection and isolation, Automatica , v ol. 36, no. 4, pp. 541–553, Apr . 2000. [3] ——, An identification scheme for robot actuator f ault s, Proc. IEEE/RSJ International Conference on Intelligent Robots and Systems , Canada, Aug. 2005, pp. 1127–1131. [4] P . Pisu and A. Ferr ara, An observ er -based second order sliding mode v ehicle control strate gy , Proc. IEEE 4th Intelligent V ehicles Symposium , USA, pp. 180–185, 2000. [5] F . J. J. Hermans and M. B. Zarrop, “Sliding mode observ ers for rob ust sensor monitoring, Proc. 13th IF A C W orld Congress , California, USA, pp. 211–216, 1997. [6] C. Edw ards, S. K. Spur geon, and R. J. P atton, “Sliding mode observ ers for f ault detection and isolation, Automatica , v ol. 36, no. 4, pp. 541–553, Apr . 2000. [7] C. P . T an and C. Edw ards, “Sliding mode observ ers for detection a n d reconstruction of sensor f aults, Automatica , v ol. 38, no. 10, pp. 1815–1821, Oct. 2002. [8] J. A. Da vila, L. M. Fridman, and A. Le v ant, “Second-order sliding-mode observ er for mechanical sys- tems, IEEE T ransactions on Automatic Control , v ol. 50, no. 11, pp. 1785–1789, 2005. Actuator fault detection and isolation for r obot manipulator ... (Khaoula Oulidi Omali) Evaluation Warning : The document was created with Spire.PDF for Python.
1156 r ISSN: 2502-4752 [9] L. M. Capisani, A. Ferrara, and L. Magnani, “MIMO identification with optimal e xperiment design for rigid robot manipulators, Proc. IEEE/ASME International Conference on Adv anced Intelligent Mecha- tronics , Zurich, Switzerland, pp. 1–6, 2007. [10] R. Mattone and A. De Luca, “F ault detection and isolation in mechanical systems, Italy , May 2004. [11] W . Chen and M. Saif, “Rob ust f ault detection and isola tion in constrained nonlinear systems via a second order sliding mode observ er , Proc. 15th IF A C W orld Congress , Spain,, pp.1498–1500, 2002. [12] J. A. Da vila, L. M. Fridman, and A. S. Pozn yak, “Observ ation and identification of mechanical systems via second order sliding modes, International Journal of Control , v ol.79, no.10, pp. 1251–1262,2006. [13] L. M. Fridman, “The problem of chattering: an a v eraging approach, V ariable Structure , Sliding Mode and Nonlinear Control , K. Y oung and U. Ozguner , Eds. London, UK: Springer -V erlag, 1999, pp. 363–386. [14] khaoula. Oulidi Omali, Mohammed. Nabil Kabbaj, Mohammed. Benbrahim, Sensor F ault Detection and Isolation for a Robot Manipulator Based on High-Gain Observ ers. Lecture Notes in Real-T ime Intelligent Systems, Springer International Publishing A G, part of Springer Nature 2019 , R TIS 2017, AISC 756, pp. 426–435, 2019. [15] J. F . Bejarano, L. M. Fridman, and A. S. Pozn yak, “Exact state estimation for linear systems with unkno wn inputs based on hierarchical supertwisting algorithm, International Journal of Rob ust and Nonlinear Con- trol , v ol. 17, no. 18, pp. 1734–1753, Mar . 2007. [16] khaoula. Oulidi Omali, Mohammed. Nabil Kabbaj, Mohammed. Benbrahim, Nonlinear Observ er -Based F ault Detection and Isolation for a Manipulator Robot. Ne w Bioprocessing Strate gies: De v elopment and Manuf acturing of Recombinant Antibodies and Proteins . Springer Nature Sing apore Pte Ltd, 163-185, 2019. [17] Antonella. Ferrara, F ault Detection for Robot Manipulators via Second-Order Sliding Modes, IEEE T ransactions on Industrial Electronics , V OL. 55, NO. 11, 2008. [18] Raf f aella. Mattone, Alessandro. De Luca, Relax ed f ault detection and isolation: An application to a non- linear case study , Automatica , 42, 109-116, 2006. [19] khaoula. Oulidi Omali, Mohammed. Nabil Kabbaj, Mohammed. Benbrahim, F ault Detection and Isola- tion for Manipulator Robot, International Symposium on Adv anced Electrical and Communication T ech- nologies , 1-5, IEEE, 2018. [20] T rong-Thang Nguyen, Sliding mode control-based system for the tw o-link robot arm, International Jour - nal of Electrical and Computer Engineering (IJECE) , V ol. 9, No. 4, pp. 2771-2778, August 2019. [21] S. Nurmaini and B. T utuk o, Intelligent Robotics Na vig ation System: Problems, Methods, and Algorithm, International Journal of Electrical and Computer Engineering (IJECE) , v ol/issue: 7(6), pp. 3711-3726, 2017. [22] Zhang, X., Polycarpou, M. M., and P arisini, T ., A rob ust detection and isolation scheme for abrupt and incipient f aults in nonlinear systems, IEEE T ransactions on Automatic Control , 47(4), 576–593, 2002. [23] Luca. Massim iliano Capisani, Antonella. Ferrara, Alejandra. Ferreira de Loza, and Leonid M. Fridman, Manipulator F ault Diagnosis via Higher Order Sliding-Mode Observ ers, IEEE T ransactions on Industrial Electronics , V OL. 59, NO. 10, 2012. [24] Shahnaz. T ayebi-Haghighi, F arzin. Piltan, and Jong-Myon. Kim, Rob ust Composite High-Order Super - T wisting Sliding Mode Control of Robot Manipulators, MDPI Robotics , 2018. [25] Fraguela, L.; Fridman, L.; Ale xandro v , V .V . Position stabilization of a Ste w art platform: High-order sliding mode Observ ers-based approach, J. Frankl. Inst. , 349, 441–455, 2012. [26] Kamal, S., Bandyopadh yay , B., Higher Order Sliding Mode Control: A Control L yapuno v Function Based Approach, WSEAS T rans. Syst. Control , 9, 38–46, 2014. Indonesian J Elec Eng & Comp Sci, V ol. 17, No. 3, March 2020 : 1150 1156 Evaluation Warning : The document was created with Spire.PDF for Python.