MOGA-subspace k -means and LightGBM–SHAP framework for acoustic emission damage diagnosis in CFRP cables

This paper studies the tensile damage development of carbon fibre reinforced polymer (CFRP) cables. Their light weight, high strength, and corrosion resistance make them widely used in bridges, cableways, etc. Extensive experiments were conducted to perform tensile tests on CFRP cables. To address the issue of accurately identifying damage modes, a multi-objective genetic algorithm (MOGA) subspace k-means algorithm was proposed, and a damage identification model for CFRP cables integrating LightGBM and Shapley Additive explanations (SHAP) technology was established. The results demonstrate that CFRP cables have excellent tensile properties (with an ultimate failure load reaching 378 kN). The proposed clustering algorithm adaptively determines the optimal clustering number based on the joint optimisation of multiple indicators, and combines with VMD to successfully cluster the AE signal of the CFRP cable into three clusters, representing matrix cracking, fibre/matrix debonding, and fibre fracture, respectively. K-fold cross-validation confirmed the LightGBM optimised by the Kepler optimisation algorithm (KOA) algorithm model’s classification accuracy exceeding 99.43%, demonstrating excellent performance and strong generalisation capability. Through feature contribution ranking and interaction effect analysis based on the SHAP interpretability framework, the criteria for identifying the three major damage modes were clarified.