Prediction and Uncertainty Quantification of the Fatigue Life of Corroded Cable Steel Wires Using a Bayesian Physics-Informed Neural Network

Abstract

Cable integrity is critical for the serviceability of cable-supported bridges. Engineering practices show that the fatigue of corroded cable steel wires is one of the main reasons for cable failure. Predicting the remaining fatigue life (RFL) of corroded cable wires remains a significant challenge. Existing prediction methods are not only difficult to comprehensively consider various factors affecting fatigue but also challenging to quantify the uncertainty of prediction results. Therefore, a method based on a Bayesian physics-informed neural network (BPINN) is proposed for predicting the RFL of corroded cable wires. First, the fatigue mechanics of corroded wires and the Bayesian inference theory were introduced. Subsequently, the framework of the BPINN model was developed by integrating data, physics, and uncertainty, considering multiple factors that affect the fatigue of corroded wires. Finally, the sensitivity of input variables was analyzed, and the prediction accuracy and generalization performance of the BPINN model were compared with empirical models and other machine learning algorithms. The proposed method effectively predicts the RFL of corroded steel wires and quantifies the uncertainty of the prediction results. The results demonstrate that it has better accuracy and generalization capabilities than methods based on empirical formulas and purely data-driven models. Furthermore, the proposed method provides a new perspective and serves as a reference for assessing the fatigue condition in similar steel structures.