Efficient online analysis of accidental fault localization for dynamic systems using hidden Markov model

This paper proposes a novel approach to do online analysis of accidental fault localization for dynamic systems by using Hidden Markov Model (HMM). By introducing reasonable and appropriate abstraction of complex system, HMM is used to represent the fault and no-fault states of system's components and system's behaviour. The HMM is parametrized to be statistically equivalent to real system's behaviour. Inspired by the principles of Fault Tree Analysis and maximum entropy in Bayesian probability theory, we propose the algorithms to estimate HMM's parameters, instead of learning, because in real systems the learning data for accidental fault is difficult to obtain. We design a specific test bed to generate large quantity of test cases, and give out the experimental results to assess the accuracy and efficiency. Meanwhile, we apply the approach to a simple helicopter control system case study, and give out convincing results.