Ivo Punčochář and Miroslav Šimandl
Active fault detection, Controlled Markov chains, Input signal design
The paper deals with the problem of active fault detection with a given fault detector over an infinite time horizon. Systems that can be modeled using two interconnected discrete-time finite-state Markov chains are considered. The first Markov chain is unobservable and describes switching between fault-free and faulty modes. The second one is an observable controlled Markov chain that describes the system dynamics in fault-free and faulty modes. The maximum a posteriori probability fault detector is assumed, and an input signal generator that improves the decision quality is designed. The original problem is reformulated as a perfect state information problem and solved by dynamic programming. An infinite time horizon is considered to reduce off-line computational demands and a perceptron neural network is employed to lower memory requirements for on-line use. The results are illustrated in a numerical example.
Important Links:
Go Back
