Artificial Intelligence-Based Survival Analysis For Industrial Equipment Performance Management

This paper proposes an artificial intelligence-based survival analysis method to predict the time to occurrence of an abnormal event in industrial equipment as the main prescriptive maintenance decision criterion. The method exploits maintenance historical data comprising event data and continuous observations to build an updated set of survival models that reflect the ages and operating conditions of equipment. The event data is used to build the survival models while the continuous observations are labeled according to the equipment operating conditions to train an ensemble of different machine learning (ML) classifiers. Given a new observation collected from the monitored equipment, the trained ML ensemble predicts the probabilities of different health states and their corresponding survival curves, and then the overall equipment's survival curve is updated accordingly. The updated curve is then used to predict the time left for an abnormal event to occur in the monitored equipment. The proposed method is validated on NASA turbofan engine data, a common dataset in the realm of prognostics, to predict the remaining useful life (RUL) of a group of equipment working under varying operating conditions. The obtained results show the effectiveness of the proposed method and the ML ensemble used decreases the RUL prediction error by an average of 35.6% compared to the best-performing classifier in the ensemble.