"A Method for Robust, Distributed Diagnosis using Dynamic Bayes Nets"

The proliferation of safety-critical embedded systems has created great demands for online fault diagnosis and fault-adaptive control techniques. A number of methodologies have been proposed, but the implementation of on-line schemes that integrate the fault detection, isolation, identification, and fault accommodation or reconfiguration tasks remains challenging. Moreover, model-based diagnosis methods can be computationally expensive, thus making online diagnosis schemes infeasible.

In this talk, I will present a distributed scheme using Dynamic Bayes Nets (DBNs) for robust diagnosis of dynamic systems. DBNs provide a systematic method for modeling the behavior of dynamic systems in uncertain environments that can include measurement noise and model uncertainties. The notion of structural observability applied to bond graph (BG) models of the physical system is exploited to derive DBN factors (DBN-Fs) that are independently observable, and together retain observability for the entire system. We have developed systematic methods for deriving the DBN-Fs from a BG model of the system, and we prove that these factors can be used as local diagnosers that generate globally correct results without a central coordinator, and without loss of accuracy. Running the independent factors significantly reduces the computational complexity of online diagnosis. Experimental results are presented to demonstrate the effectiveness of this scheme.