Improving the Accuracy of Maintenance Decision-Making via Deep Forest-Based Failure Prognostics

Accurate system failure prediction plays a vital role in the prognostics and health management of a system. It can avoid sudden shutdowns and ensure the system safety through making proper maintenance decisions including inventory decision. Aimed at improving the accuracy of maintenance decision-making, this paper deals with the system failure prediction by using a deep forest model. Firstly, crucial sensory data are selected by using the Fisher’s discriminant ratio. Secondly, according to the operation planner’s requirements, the use of deep forest is proposed to provide the system failure probabilities in different time windows in future. With obtained prognostic information, optimal maintenance and inventory decisions are made by quickly evaluating the costs of different maintenance options. Verification results on aero-engine degradation data from NASA reveal the feasibility and advantages of the proposed method.