Fatigue Crack Growth Prognostics by Particle Filtering and Ensemble Neural Networks

Abstract

Particle Filtering (PF) is a model-driven approach widely used in prognostics, which requires models of both the degradation process and the measurement acquisition system. In many practical cases, analytical models are not available, but a dataset containing a number of pairs component state - corresponding measurement may be available.In this work, a data-driven approach based on a bagged ensemble of Artificial Neural Networks (ANNs) is adopted to build an empirical measurement model of a Particle Filter for the prediction of the Residual Useful Life (RUL) of a structure whose degradation process is described by a stochastic fatigue crack growth model of literature. The work focuses on the investigation of the capability of theproposed approach to cope with the uncertainty affecting the RUL prediction.