Sensor Fault Tolerant Control for AUVs Based on Replace Control

Sensors & Transducers, Vol. 58, Issue, November 03, pp. 408-43 Sensors & Transducers 03 by IFSA http://www.sensorsportal.com Sensor Fault Tolerant Control for AUVs Based on Replace Control Xao LIANG, Jundong ZHANG, 3 We LI College of Traffc Equpment and Ocean Engneerng, Dalan Martme Unversty, Chna College of Marne Engneerng, Dalan Martme Unversty, Chna 3 College of Envronmental Scence and Engneerng, Dalan Martme Unversty, Chna Tel.: +86-04-8477985 E-mal: langxao9800@6.com Receved: 8 August 03 /Accepted: 5 October 03 /Publshed: 30 November 03 Abstract: To mprove the relablty of the moton control system for an autonomous underwater vehcle (AUV), sensor fault tolerant control based on replace control was talked about. Ths paper analyzed the decouplng controller and the sensor system frstly, and proposed a fault tolerant control strategy based on the deas of replacng the fault sensor outputs by the sldng-mode observer outputs (also called replace control). For the rrecoverable sensor faults, the strategy mnmzed the damage to the AUV. For the jumpng faults of sensors, replace control s used durng jumpng. The smulaton results proved the valdty of the strategy. Copyrght 03 IFSA. Keywords: Autonomous underwater vehcle (AUV), Decouplng control, Sensor fault tolerant control, Replace control, Sldng-mode observer.. Introducton Autonomous Underwater Vehcle (AUV), because of the mportance n busness and mltary and many challenges n technology, more and more scentsts and techncans take t serously, and have carred out a lot of research work. Fg. shows the whole flow chart of the AUV control system. Plannng system produces the AUV locaton nformaton durng the next tme, Controller calculates the force and moment by comparng plannng nformaton of plannng system and nformaton of sensor after measurng, and then obtans the voltage of each thruster by the thrust allocaton. Fnally, sensors send the AUV message durng the next tme to the controller, thus formng a complete feedback control system. plannng system control algorthm data processng thrust allocaton Fg.. Moton control system of the AUV. thruster AUV sensor system In the control system, sensors provde the actual nformaton for the AUV. If the sensors break down durng the use, ts outputs wll no longer reflect the real value, t may lead to paralyss of the entre 408 Artcle number P_50

Sensors & Transducers, Vol. 58, Issue, November 03, pp. 408-43 system. In order to guarantee the AUV successfully complete ts msson, there s need to buld a sensor fault-tolerant control system for montorng and dagnoss and treatment. Fault-tolerant control system manly ncludes three parts: fault detecton, fault locaton and fault handlng (fault-tolerant control). At present, there are many fault processng methods, such as the method based on the analytcal model or based on artfcal ntellgence. In ths paper, consderng the hardware redundancy of sensors n actual work s not large, output observer s proposed to nstead the fault sensor outputs of fault-tolerant control method and fault handlng strategy. Ths strategy doesn't need to ncrease addtonal sensors, t also can be called replace control strategy.. Decouplng Control Decouplng control s of great sgnfcance n ndustral process control. Industral processes usually don't want the change of a setted value to cause bg swngs n other controlled varables. The system after decouplng s better than the general multvarable systems, no matter from settng or relablty. Therefore, when desgnng the controller, we should avod the mutual couplng. For AUVs, the moton for all degrees of freedom (DOF) s coupled to each other. It s dffcult to qualtatvely descrbe and express the couplng. In order to avod the dffcultes of mutual couplng, we consder decouplng for each DOF, and desgn a controller (the controller for each DOF s the nonlnear functon of the devaton and the devaton change rate) on each DOF. Here, we adopt a novel, smple and practcal control method S surface control. The structure of the controller s smple and the quantty of nput s lttle, t s sutable for nonlnear system. The controller combnes the deas of fuzzy control wth PID control, and t s not only smplfed the controller desgn, but also can guarantee the effect of control, whch has been adopted successfully n real sea test of AUV. Let`s desgn a S surface controller on each DOF for the AUV: ( ( k ) ) e k e u =.0 /.0 + e.0 + u f () = K u ( =,,6 ), () where e and e are the nput nformaton of DOF (devaton and the devaton change rate, normalzed processed), u s the control output of DOF, k and k correspondng to the control parameters of devaton and rate of change respectvely, and the changng velocty of the correspondng DOF can be changed. f s the magntude of the requred force, K s proporton coeffcent, =,,, 6, represents sx DOF for the AUV moton: surge, sway, heave, yaw, ptch, roll. u s magntude of the nterference force obtaned by the adaptve ways (normalzaton). Adaptve ways are as follows: a) Determne whether the rate of the AUV moton on DOF s less than a set threshold, If t s, turn to (b), f not, to (c). b) Gve the devaton value of ths DOF a setted of array, at the same tme, add to the setted count value. When the count value reaches a predetermned value, turn to (d). c) Put the value of array forward one bt, and the count value mnus, turn to (a). d) Weghted average the value of the array, use the mean value of the offset to calculate the devaton of control output. The output of adaptve adjustment controller, In order to elmnate the fxed devaton, and set the count value and array to zero, then carry out the next cycle. Then we have desgned a sx DOF controller. The controller of each DOF presents the control output, based on the current devaton and the rate of devaton change. But the nput of the controller depends on the dfferences and change rate between the output of plannng devce and the measurement of sensors. 3. Sensor System In ths paper, a certan knd of AUV developed by Harbn engneerng unversty led as research object. The nstallaton of the sensors s shown n Table below. The sound and lght vsual sensors (60 magng sonar, 3D magng sonar and TV) of the AUV manly are used for the trackng and recognton of underwater target, the mpact on AUV moton control s not bg. GPS sgnals are used for postonng correcton when the AUV s longendurance. For ths paper, the replace control research of AUV s manly amed at the sensors whch can affect the robot s poston, such as fber optc gyro, Doppler velocmeter (DVL), postonng sonar, altmeter sonar and depth gauge. There are three forms when the AUV sensors break down, that s, () the outputs nformaton of sensors reman unchanged; () the outputs nformaton of sensors jump wthn a certan tme or perod; (3) the outputs nformaton of sensors shock on the tmelne. Generally speakng, the frst two knds of falures may happen on fber optc gyro, Doppler velocmeter, altmeter sonar and depth gauge; all three knds of falures are lkely to happen on postonng sonar. For the thrd knd of falure of postonng sonar, we adopt lnear smoothng method for processng and succeed on sea trals. For the frst knd of falure, we montor t by adoptng the method of cumulatng the amount of samplng perod durng whch the change rate of data keeps contnuously constant; for the second knd of falure, we montor t 409

Sensors & Transducers, Vol. 58, Issue, November 03, pp. 408-43 by adoptng the method of extractng of mutatons n the sgnal through wavelet analyss; Replace control method s used for troubleshootng. As can be seen from the table, the AUV s equpped wth all knds of sensors wthout hardware redundancy. Therefore, t s dffcult to realze fault repar by usng the method of nformaton fuson. Table. The confguraton of the sensor AUV. Type of State parameters of Sensors sensors measurng Headng angleψ Inerta Fber optc ptch angleθ gyroscope rollng angleφ Longtudnal velocty Doppler u transverse velocty velocmeters v vertcal velocty w Sonar altmeter Dstance from the bottom of the sea Plane coordnates of Ultrasonc Short base lne AUV n the sonar wave postonng sonar array The dstance d and 60 magng drecton Ω of sonar underwater targets or obstacle 3D magng sonar Acoustc mage of underwater targets Pressure Depthometer Submerged depth Rado GPS Longtude, lattude Optcs TV Optcal mage of underwater targets 4... The Moton Equaton of the AUV In the condton where the gravty and buoyancy has been balanced and external dsturbances have been gnored, the space moton equaton of the AUV n fve DOF can be expressed as: MX = F ( X ) + F, (3) vs where M s the mass matrx, ncludng the addtonal mass and addtonal moment of nerta; X, X and X represent the dsplacement matrx, velocty matrx and (angle) acceleraton matrx respectvely; F t s the force matrx of thruster; F vs s the vscous hydrodynamc matrx. For easy analyss, type (3) nto the realzaton form of correspondng space state. Make x = X, x = X. Then the system can be expressed as the followng space state equaton: where x = x, (4) x = M [ Fvs ( x) + Ft ] M s the nverse matrx of M. 4... The Desgn of Sldng-Mode Observer To the state space equatons of the AUV represented n (4), we desgn a nonlnear sldng mode observer as follows: t 4. Replace Control of Sensors x ˆ ˆ ˆ = x + ηtanh( x x) x ˆ = M [ F ( x ) + F] + η tanh( x xˆ ) vs t (5) 4.. The Fault Assumpton of Sensors The replace control of sensors s based on the followng assumptons: a) Only one sensor malfunctons; b) Thruster does not malfuncton; c) Replace control s carred out based on whch knd of fault happened on whch sensor that has been dagnosed. 4.. Sldng-mode Observer When the AUV woks n complcated marne envronment of hgh pressure and poor vsblty, there are many uncertan factors. Sldng-mode observer has good robustness for the system uncertanty. Thus, we ntroduce the concept of sldng mode to the fault processng of control system. where, ˆx and ˆx respectvely represent the estmates of state vector x and x, that s the AUV velocty and the estmates of the acceleraton; η = dag( λ, λ,, λ5) s postvely steady opposte angle gan matrx that to be desgned; η = dag( λ, λ,, λ5) s postvely steady feedback gan matrce that to be desgned. In (5), we use tangent functon tanh( ) nstead of symbolc functon sgn( ) that used n conventonal sldng mode observer. The purpose of dong ths s to elmnate the chatterng of sldng mode observer Defne the error vector of observer: e = x ˆ x e ˆ = x x (6) Accordng to the equaton () and (3), we can calculate the error output equaton of observer: 40

Sensors & Transducers, Vol. 58, Issue, November 03, pp. 408-43 e = e η tanh( e) e = η tanh( e) (7) We can know that by reasonable selecton of gan T matrx η and η, t can make ee< 0, that s sldng mode observer meets the attractve condtons. Ths means that the sldng resdual error e can converge to the vcnty of sldng surface and has zero mean value. Thus e 0, e 0. There are 0 outputs for the sldng mode observer. û, ˆv, ŵ, ˆq, ˆr can be obtaned drectly from the observer. The other 5 observed quanttes ( ˆx, ŷ, ẑ, ˆ θ, ˆϕ ) can be obtaned through the tme ntegral of velocty ganed by observer. The acceleraton requred n the study s obtaned through velocty dfference. Thus, we set up an sldng mode observer for the AUV. Then we can acheve the replace control about Doppler velocmeter, fberoptc gyro (DVL), sonar altmeter, postonng sonar and depthometer. 4.3. Replace Control The frst knd of fault of sensors s a knd of fault that can't be repared for the actual confguraton of the AUV. If the outputs are used n feedback control, t can lead to the control out of the AUV and even paralyss of the system. The second knd of fault of sensors can lead to nstablty of AUV moton. But when both knds of faults happened, the outputs of the observer are stll vald n a certan tme. Therefore, we can replace the outputs of sensors n the feedback control. The prncple of replace control s shown n Fg. below. As shown n Fg. 3, the fault montorng module checks whether the sensor malfunctons, the module of level analyss of fault judges the type of fault, the extent of the fault and the handlng measures. For the thrd type of fault of postonng sonar, we perform a sngle process. If the poston sensor has jump fault n a certan tme n the second type of fault, then we need to use replace control durng the perod of jump fault, n order to mantan the stablty of control system. If the poston sensor has the frst type of fault, the module of level analyss of fault wll assess the current task. We can know whether completon tme of automatc buoyancy task s n the range of tolerance (t depends on the maxmum tme that the error accumulaton determned by test can wthstand) of replace control. If wthn the range, we can use replace control, then take approprate task; f not. We can judge that the AUV has serous fault that s unable to resst. At ths pont, the module of level analyss of fault wll assess the rsk level of current task. f the rsk level s hgh, t wll take buffer tasks wthn the range of tolerance of replace control. For example, keep away from obstacles task, then the AUV can dscards excess load and starts the buoyancy devce (Referred to as the task of savng tself ). If the rsk level s low, t wll execute the task of savng tself drectly. Sensor fault montorng Fault level analyss Replace control Abandon load, start the buoyancy devce Buffer tasks Automatc buoyancy Contnue workng No AUV Postonng sonar fault handlng controller fault dagnose Sensors fault YES sldngmode observer Fg.. The prncple of replace control. 5. Fault Dagnoss for Sensors and Fault Tolerant Control System In ths paper, we manly demonstrate the replace control model and method of the AUV sensors. To mantan the ntegrty and systemc, t s necessary to descrbe the operatng process of fault dagnoss for sensors and fault tolerant control system. Fg. 3. Sensor fault dagnoss and fault tolerant control system. It should be noted that buffer task s a short-tme movng task whch s carred out when the fault unable to resst s detected. It can lead the state of AUV moton to the stop, even to be far away from the danger. In ths way, t can avod loss caused by sudden falure of sensors when the AUV s n dangerous envronment. At the same tme, for the rsk level and executon tme of task, the msson plannng module wll allocate task n real-tme when the AUV s movng underwater. It wll also estmate the executon tme of current task accordng to the state of the current task and nformaton of path planner and set up rsk level for the current task. For example, for the long-endurance msson of AUV, f t met obstacles, then the path planner wll dvde the 4

Sensors & Transducers, Vol. 58, Issue, November 03, pp. 408-43 long-endurance msson nto task of before avodng obstacles, task of avodng obstacles and task of after avodng obstacles. It wll also estmate the executon tme of current task and set the rsk level of task of avodng obstacles to be the hghest. Partcularly, for local task of avodng obstacles, because we can't accurately estmate the tme t takes, once sensors have the frst type of fault, we need to take the buffer operaton that s performng replace control whle tmng. If keepng away from the obstacles s realzed wthn the scope of permssble tme of replace control, AUV wll carry out save tself task n the safest stuaton. If the AUV s stopped when t s not away from obstacles wthn the tme scape, at ths tme, the state of AUV moton has nearly stopped and t s also a relatvely safe tme to save tself. 6. Smulaton Results 6.. Handlng Smulaton of the Frst Type of Fault The smulaton s dvded nto two parts. We take an example that the AUV completes the task of avodng obstacles met n the long-endurance msson by usng Doppler velocmeter and optcal fber gyro control, and Doppler velocmeter has the frst type of card fault (the outputs of Doppler velocmeter are suddenly constant n a certan pont). Smulaton one s to determne the maxmum wthstandng tme of error accumulaton of replace control n smulaton test. Smulaton two s the processng strategy of fault when the AUV s dong the task of avodng obstacles. Smulaton one s shown n Fg. 4, set the AUV to be from startng pont (0,0,0) passng pont (5,5,π/4), pont (0,5,π/), pont (0,0,π), to the target pont (0,0,3π/) (Three coordnates respectvely s the x coordnate, y coordnate and headng angle ψ). The smulaton tme s nearly 0 mnutes. The smulaton s n 0.5 s for a beat and the fault happened n the 30th beat. The tme detected s 35 beats and replace control lasted for nearly 7 mnutes. Judgng from the results of smulaton, the effects of replace control are better n ths perod of tme. Judgng from the results of smulaton, the effects of replace control are better n ths perod of tme. We have done many tmes smulaton tests to determne the durng tme of replace control. Here the tme lmt we use s 5 mnutes. Smulaton two s shown n Fg. 5. The executon tme of task of avodng obstacles sets for 5 mnutes n replace control. The 60 collson avodance sonar s nstalled n front of the AUV and the scannng range only can be 90 degrees. Therefore, buffer task s setted to stop after the task of avodng obstacles s carred for 5 mnutes under replace control. Fg. 4. Handlng smulaton of the frst type of fault. Fg. 5. Smulaton of obstacle avodance strategy. In Fg. 5, pont A s the place where fault happened; pont G s the termnal pont of local plannng when there are no faults; pont G s the place where the AUV stopped after buffer task when faults happened; pont G3 s the place where the AUV ht obstacles wthout buffer task; pont O s startng pont. As can be seen, f we ordered the AUV to stop controllng and save tself nstead of takng replace control when the Doppler velocmeter had the frst type of fault, the AUV wll ht the obstacles due to the AUV large nerta. However, f we took replace control, the AUV would stop after takng buffer task to avod obstacles for 5 mnutes. After the buffer task, the AUV state was nearly statc, and now the AUV can save tself safely. In addton, for the settngs of buffer task, f the scan range of the AUV sonar nstalled s wde, the methods of buffer task can be more and more reasonable. 4

Sensors & Transducers, Vol. 58, Issue, November 03, pp. 408-43 6.. Handlng Smulaton of the Second Type of Fault Wavelet analyss technology can detect the sngularty and rregular jump of sgnal, and replace control manly processes the random jump parts of sensor faults. We analyss the effects of replace control accordng to the data of fber optc gyroscope n a sea test. From Fg. 6, we can see that outputs of headng angle of fber optc gyroscope jumps n several beats. However, durng the tme of jump fault, we replaced the jump data of headng angle by the outputs of observer n order to mantan the stablty of the AUV moton. that the AUV can complete the tasks successfully or securely save tself when the sensors malfuncton. Acknowledgements Ths work s supported by the Natonal Natural Scence Foundaton of Chna (Grant No. 50905) and the Chna Postdoctoral Scence Foundaton (Grant No. 03T607, 004959) and Fundamental Research Funds for the Central Unverstes of Chna (Grant No. 0760, 330338). yaw/ rad -3,0-3,5-4,0-4,5-5,0 0 00 00 300 400 500 t/ 0.5s Fg. 6. Handlng smulaton of the second type of fault. 7. Conclusons For the frst type of fault of sensors, because of the AUV sensor confguraton, we are unable to repar the deadly fault. But we can mnmze the dangers of the fatal fault by takng the strategy of replace control, and even can guarantee that the AUV completes the tasks successfully. For the rregular jump fault n a perod of the second type of fault, we can take replace control wthn the tme of jump fault. Therefore, when the exstng AUV sensor confguraton s not under the redundant stuaton, we don't need addtonal sensors. We can take some fault-tolerant control strateges, n order to guarantee References []. D. R. Bldberg, Autonomous underwater vehcles: a tool for the ocean, Unmanned Systems, Vol. 9, Issue, 99, pp. 0-5. []. Yuru Xu, Y. J. Pang, Y. Gan, Y. S. Sun, AUV-stateof-the-art and prospect, CAAI Transactons on Intellgent Systems, Vol., Issue, 006, pp. 9-6. [3]. Yuru Xu, Kun Xao, Technology development of autonomous ocean vehcle, Journal of Automaton, Vol. 33, Issue 5, 007, pp. 58-5. [4]. Xuemn Lu, Yuru Xu, S control of automatc underwater vehcles, Ocean Engneerng, Vol. 9, Issue 3, 00, pp. 8-84. [5]. Xao Lang, Jundong Zhang, We L, Sensor fault dagnoss for autonomous underwater vehcles, Sensor Letters, Vol. 9, Issue 5, 0, pp. 06-066. [6]. M. Taka, T. Ura, Development of a system to dagnose autonomous underwater vehcle, Internatonal Journal of Systems Scence, Vol. 30, Issue 9, 999, pp. 98-988. [7]. Ye L, Jancheng Lu, Mngxue Shen, Dynamcs model of underwater robot moton control n 6 degrees of freedom, Journal of Harbn Insttute of Technology, Vol., Issue 4, 005, pp. 456-459. [8]. A. Alessandr, G. Bruzzone, M. Cacca, et al., Fault detecton through dynamcs montorng for unmanned underwater vehcles, n Proceedngs of the 4 th IFAC Symposum on Fault Detecton, Supervson and Safety for Techncal Processes, Kdlngton, UK, 000, pp. 95-956. 03 Copyrght, Internatonal Frequency Sensor Assocaton (IFSA). All rghts reserved. (http://www.sensorsportal.com) 43