Adaptive Ensemble Fault Diagnosis based on Online Learning of Personalized Decision Parameters

Data-driven fault diagnosis is increasingly prevalent in recent years for complex systems, but still suffers from two practical difficulties: one is the lack of enough historical examples of fault equipment and the other is the absence of personalized characteristics of the equipment to be diagnosed. This paper proposes an adaptive ensemble fault diagnosis method to resolve these two difficulties. The proposed method first extracts fault and normal patterns by constructing hidden Markov models (HMMs), respectively, from historical fault and normal examples. Then fault diagnosis decisions made by the supervised and unsupervised methods are intrinsically integrated through the hidden states identified by HMMs to overcome the difficulty of lacking enough historical fault examples. Finally, a learning process is designed and embedded in the proposed method to describe the impacts of personalized characteristics of the equipment on its fault diagnosis. Theoretical and experimental results both verify the effectiveness of the proposed method.