A new Signal Processing-based Prognostic Approach applied to Turbofan Engines

For modern engineering industry, Prognostic has become a key feature in maintenance strategies since it enables to enhance system availability and safety while reducing operational costs and avoiding unscheduled maintenance. Prognostic can be seen as the prediction of the system's remaining useful life with the purpose of minimizing catastrophic failure events. Such task could be performed on the basis of an accurate physical representation of the system behavior and/or by using available historical data that have been collected. In this paper, a novel prognostic approach is proposed, based on data-driven category techniques. This approach uses mainly historical data, regardless of the underlying physical process, and it can be divided into two steps. First, an original signal processing technique is used to develop life prediction models. In the second step, the system's current health state is predicted and the RUL is estimated based on a proposed formula. This approach is validated by using four different data sets generated from the NASA's turbofan engine simulator (C-MAPSS) and the obtained results are compared with relevant existing approaches tested using the same collected data. The main outputs of our study attest that the proposed approach is robust, applicable and effective even in the presence of various fault modes and operating conditions.