340 Asian Journal of Control, Vol. 9, No. 3, pp. 340-344, September 2007 Manuscript received March 5, 2004; revised December 6, 2004; accepted September 25, 2006. Chee Pin Tan is with the School of Engineering, Monash University Malaysia, 46150 Petaling Jaya, Malaysia (e-mail: tan.chee.pin@eng.monash.edu.my). Christopher Edwards is with the Department of Engineering, University of Leicester, LE1 7RH Leicester, U.K. (e-mail: chris.edwards@leicester.ac.uk). －Brief Paper－ A ROBUST SENSOR FAULT TOLERANT CONTROL SCHEME IMPLEMENTED ON A CRANE Chee Pin Tan and Christopher Edwards ABSTRACT This paper presents a sensor fault tolerant control scheme applied to a crane sys- tem. Sensor faults affect the system’s performance in the closed-loop when the faulty sensor readings are used to generate the control input. In this paper, the measured outputs are separated into potentially faulty and nonfaulty components, and the latter are injected into a linear observer to reconstruct the faults. The reconstruction is sub- tracted from the faulty sensors to form a compensated ‘virtual sensor’ and this signal is then used to generate the control input. A design method for the observer is also presented in which the reconstruction signal is made as insensitive as possible to any uncertainties or nonlinearities present in the system. Good results have been obtained; with reduced performance degradation during faulty conditions. KeyWords: Fault tolerant control, fault reconstruction. I. INTRODUCTION Fault tolerant control (FTC) is an emerging area of re- search. Its objective is to minimize the degradation in per- formance of a system when a fault occurs. A reliable FTC scheme may help improve efficiency, productivity, reliabil- ity, generate financial savings or prevent catastrophic con- sequences such as loss of human life or serious environ- mental pollution. In the literature, FTC methodologies are separated into two categories: passive and active schemes. Passive schemes [1,2] are those whereby the controller is designed to work in all possible failure scenarios. They generally have fixed structure and parameters. Active schemes [3] are designed so that the control structure or parameters change or adapt in the event of different failure scenarios. The design for active schemes will therefore usually be less conservative than passive schemes, but more complicated. Most of the FTC schemes in the litera- ture usually do not make use of the wealth of work in the field of fault detection and isolation (FDI) [4]. This fact had been pointed out in [5], and only a few researchers have directly linked FDI with FTC [6-8]. This paper is concerned with sensor FTC that is reliant on good FDI. Sensor faults [9,10] are ones that occur in the sensors/transducers that measure the system variables, and do not directly affect the process dynamics (in open- loop). The source of these faults could be wear and tear of the sensor, prolonged use without calibration, or a total failure of the sensor. However, in the closed-loop, these faults will affect the process if the sensor measurements are used to generate the control signal. Therefore, inaccurate sensor readings will cause degradation in performance. In this paper, the outputs are firstly separated into nonfaulty and potentially faulty components. The control input and nonfaulty outputs are fed into a linear observer to generate an estimate of the internal states. A reconstruction of the sensor fault is obtained by subtracting a function of the estimated states from the measured outputs, and the result is multiplied by a scaling matrix. The reconstruction is subtracted from the faulty sensor to get a ‘virtual sensor’. In an ideal situation when the fault is estimated perfectly, the virtual sensor should give the output’s correct reading. The virtual sensor will then be used to generate the control signal, and the degradation in system performance should be eliminated. However, in a real system, there are system nonlinearities and uncertainties, which cannot be fully modelled. These elements will make the state estimate in- accurate, which in turn will corrupt the fault reconstruction as well as the output of the virtual sensor. Therefore, in this paper, an observer design method is presented to minimize the effect of the nonlinearities/uncertainties on the virtual sensor, using the Bounded Real Lemma [10]. In this paper,