Prognostics and Health Monitoring in the Presence of Heterogeneous Information

Diagnosis and prognosis methodologies have used dynamic Bayesian networks (DBNs) to fuse many types of information. These methodologies, however, fuse problem- specific information and focus only on a subset of information types. By using only a subset of information, the interactions between or individual behaviors of subsystems, components, and faults may not be fully realized. In this paper, a general framework for system level diagnosis and prognosis of a mechanical system in the presence of heterogeneous information using dynamic Bayesian network (DBN) is proposed. Due to their ability to fuse heterogeneous information — information in a variety of formats from various sources — and give a probabilistic representation of a system, DBNs provide a platform naturally suited for diagnosis and uncertainty quantification. In the proposed methodology, a DBN is first constructed via an established machine learning algorithm from heterogeneous information. The DBN is then used to track the system and detect faults by monitoring the Bayes’ factor of the system state estimate. When a fault occurs, the underlying system model changes, and the Bayes’ factor of the DBN system model decreases. The state estimate provided by tracking indicates the most likely fault scenario and quantifies the diagnosis uncertainty. Estimation of remaining useful life and quantification of uncertainty in prognosis can then proceed based upon the diagnosis results. The proposed methodology is then demonstrated using a cantilever beam example with a possible loose bolt at the connection or a crack in the middle of the span.