Damage monitoring and prognostics in composites via dynamic Bayesian networks

Abstract

This study presents a new structural health monitoring framework for complex degradation processes such as degradation of composites under fatigue loading. Since early detection and measurement of an observable damage marker in composite is very difficult, the proposed framework is established based on identifying and then monitoring “indirect damage indicators”. Dynamic Bayesian Network is utilized to integrate relevant damage models with any available monitoring data as well as other influential parameters. As the damage evolution process in composites is not fully explored, a technique consisting of extended Particle Filtering and Support Vector Regression is implemented to simultaneously estimate the damage model parameters as well as damage states in the presence of multiple measurements. The method is then applied to predict the time to failure of the component.