Fault monitoring and remaining useful life prediction framework for multiple fault modes in prognostics

Abstract The monitoring of fault evolution trend and the prediction of remaining useful life (RUL) are of great significance for complex engineering system since it can provide helpful decision support for maintenance. In general, it is difficult to distinguish the evolution tendency and the mode of multiple faults directly from original collected databases. To address this problem, a novel fault monitoring and RUL prediction framework under multiple fault modes is proposed in this paper, which can monitor the evolution tendency of fault, predict and identify the failure mode under multiple faults, and further accurately estimate the RUL. Firstly, gap metric is combined with deep belief network to extract the hidden degradation features from monitoring data. Following that, support vector data description is employed to establish a monitoring model to identify multiple fault patterns through a classification strategy. Afterwards, the RUL can be predicted through particle filter when the degradation characteristic falls into the fault feature described by support vector data. In the end, the application to a degradation engine dataset in multiple fault modes is given, and the experiment result indicates that the proposed framework achieved competitive results compared with the existed single fault prediction methods.