In-situ residual strength prediction of composites subjected to fatigue loading

A novel approach is introduced for in-situ residual strength prediction of glass epoxy composites subjected to fatigue loading, by integrating piezo-resistivity-based structural health monitoring with machine learning techniques. In this process, composite samples made conductive with carbon nanotubes are subjected to fatigue loading while their electrical resistance (ER) is monitored. The ER features most closely related to the residual strength are identified and used to train various machine learning algorithms. Ridge regression, K-nearest neighbor (KNN), Decision Tree (DT), Random Forest, Extreme Gradient Boosting, and Support Vector Regressor (SVR) are implemented in two different approaches: as standalone predictors, and in an ensemble learning approach to predict the residual strength. The analysis shows that the KNN meta -model within an ensemble framework, integrating DT, SVR, and KNN as base models, demonstrates superior performance, with a mean absolute percentage error of 4.7% in predicting residual strength.