Adaptive Sensor Fault Detection and Identification Using Particle Filter Algorithms

• Published in: IEEE transactions on human-machine systems 2009-03, Vol.39 (2), p.201-213
• Main Authors: Tao Wei, Yufei Huang, Chen, C.L.P.
• Format: Article
• Language: English
• Subjects: Algorithmics. Computability. Computer arithmetics; Algorithms; Applied sciences; Communication switching; Computer science; control theory; systems; Computer simulation; Dynamical systems; Dynamics; Exact sciences and technology; Fault detection; Fault detection and isolation (FDI); Fault diagnosis; Markov processes; Mathematical models; Military aircraft; mixture Kalman filter (MKF); Monte Carlo technique; particle filter (PF); Particle filters; Redundancy; sensor failure; Sensor fusion; Sensor systems; sensor validation; Sensors; stochastic M -algorithm (SMA); Stochastic processes; Studies; Switches; Theoretical computing
• ISSN: 1094-6977; 2168-2291; 1558-2442; 2168-2305
• DOI: 10.1109/TSMCC.2008.2006759

Sensor fault detection and identification (FDI) is a process of detecting and validating sensor's fault status. Because FDI guarantees system reliable performance, it has received much attention recently. In this paper, we address the problem of online sensor fault identification and validation. For a physical sensor validation system, it contains transitions between sensor normal and faulty states, change of system parameters, and a fusion of noisy readings. A common dynamic state-space model with continuous state variables and observations cannot handle this problem. To circumvent this limitation, we adopt a Markov switch dynamic state-space model to simulate the system: we use discrete-state variables to model sensor states and continuous variables to track the change of the system parameters. Problems in Markov switch dynamic state-space model can be well solved by particle filters, which are popularly used in solving problems in digital communications. Among them, mixture Kalman filter (MKF) and stochastic M -algorithm (SMA) have very good performance, both in accuracy and efficiency. In this paper, we plan to incorporate these two algorithms into the sensor validation problem, and compare the effectiveness and complexity of MKF and SMA methods under different situations in the simulation with an existing algorithm - interactive multiple models.