Health Monitoring by Physical Modeling and Statistical Learning

Abstract

Health monitoring aims at preventing failures from happening. Our approach of health monitoring is to combine observations of the system with physical modeling. Information is extracted from complex observations by statistical learning. The study of this raw information can give first insights in the problem. To go further, the physical knowledge about the system is required. The modeling of the system components implements the physical knowledge we have of the system. Inversing this model enables the identification of the parameters responsible for the degradation. Model inversion is realized by bayesian inference. The evolution over time of the system parameters undergoing degradation enables precise diagnoses and the set up of predictive maintenance operations.