Failure mechanism analysis and machine learning-based residual strength prediction of CFRP with different ply orientations and loading types under six typical accelerated aging environments

Abstract

Carbon Fiber Reinforced Polymer (CFRP) is a pivotal engineering material valued for its high specific strength and stiffness, yet its long-term performance is susceptible to degradation under combined environmental exposures. To systematically investigate this issue, this study examined the performance evolution and failure mechanisms of CFRP under six distinct accelerated aging environments: low temperature and low humidity (LTLH), high temperature and low humidity (HTLH), high temperature and high humidity (HTHH), immersion at room temperature (IRT), thermal cycling (TC), and hygrothermal cycling (HC). Specimens with ply orientations of 0°, 45°, and 90° were fabricated for both tensile and compressive testing. Mechanical property evaluation was integrated with chemical and microstructural characterization via Fourier transform infrared spectroscopy (FTIR), thermogravimetric analysis (TG), and scanning electron microscopy (SEM) to elucidate underlying degradation mechanisms. The influence of aging environments, ply orientation, and loading type on residual strength was quantitatively assessed using analysis of variance (ANOVA), while four machine learning models—Back Propagation Neural Network (BPNN), Support Vector Regression (SVR), Random Forest (RF), and Extreme Gradient Boosting (XGBoost)—were employed to predict the residual strength of aged CFRP.

The results indicate that the residual strength of CFRP exhibits complex nonlinear degradation across all six aging environments, which can be attributed to the competition between performance gain from post-curing and performance degradation caused by hydrolysis and thermal oxidation. ANOVA further quantified the influence of individual factors (accelerated aging environment, ply orientation, load type) and multi-factor couplings on the residual strength of CFRP. Among the predictive models, BPNN demonstrated superior performance with a coefficient of determination (R²) of 0.985, enabling accurate prediction of CFRP residual strength across the multiple aging environments. This work provides a reliable technical basis for the long-term performance assessment and safe engineering application of CFRP.