Comprehensive Diagnosis of Continuous Systems Using Dynamic Bayes Nets

Fault diagnosis is essential for guaranteeing safe and reliable operation of complex engineering systems. Our work focuses on diagnosis of parametric faults in the components of dynamic systems, whose temporal profile can be categorized as incipient (slow) or abrupt (fast). The diagnosis of abrupt and incipient faults using qualitative approaches is challenging, since in many situations, these faults produce similar qualitative effects. Quantitative estimation methods may provide more discriminatory power, but these approaches can be computationally infeasible for large systems with nonlinearities and complex dynamics. In this paper, we combine a qualitative fault isolation scheme with an Dynamic Bayes net-based particle filtering approach for the comprehensive diagnosis of incipient and abrupt faults in continuous systems. We also present experimental results to demonstrate the effectiveness of our approach when applied to a two-tank system.