Data-driven prediction of the fatigue performance of corroded high-strength steel wires

Corrosion damage determines the fatigue performance and intensifies the fatigue lifespan scatter of high-strength steel wires. Many experimental studies have been performed on corroded wires to quantify the correlation between fatigue life and corrosion. However, accurate analysis and fatigue performance prediction of various types of corrosion are still insufficient. This study proposes a predictive machine learning-based model to explore the quantitative influence of corrosion degree, stress ranges, loading ratio, loading frequency and other factors on the fatigue life of corroded high-strength steel wires. The proposed model is trained using the adaptive boosting algorithm on the basis of a collected database containing 323 samples from corrosion fatigue tests. A cross-validation method with the grid-search approach is applied to accelerate the optimisation of the hyperparameters. The performance of model is thoroughly compared with the empirical models and the common utilized machine learning algorithms. The prediction perfor-mance of the proposed model is considerably good, with a coefficient of determination of 0.96. The generalisation ability of the proposed model is superior to that of two common machine learning algorithms. Additionally, the physical meaning of the model is proven to be reasonable through a weight analysis of the input features. The stress range, corrosion rate and loading stress ratio dominate the fatigue performance of corroded high-strength steel wires.