Predicting growth and interaction of multiple cracks in structural systems using Dynamic Bayesian Networks

Abstract

Digital twins have the potential to improve future health prognosis and operational safety of engineering structures. However, few studies have attempted to validate such twins with experimental data. In this work, two system-level digital twin models for structural capacity degradation are evaluated with experimental data from a four-crack structural fatigue specimen. Each twin model is based on a dynamic Bayesian network, but the two models differ in the complexity of the crack-to-crack interaction model. Using measured crack lengths from the experiment, the ability of each twin to predict the evolution of future crack size is assessed. For this specimen, a high level of crack-to-crack interaction must be captured for the twin to realistically trace the evolution of the system. The suitability of Bayesian network based twins for this type of problem is discussed.