ABSTRACT With the rapid expansion of deep‐sea development activities, the performance requirements for buoyancy materials have become increasingly stringent. Carbon fiber reinforced epoxy resin composites (CFRP) are a leading candidate due to their superior mechanical properties; however, the degradation mechanisms of the epoxy matrix and CFRP under high‐pressure water absorption remain poorly understood. In this study, standard epoxy and CFRP specimens were prepared and subjected to high‐pressure water absorption tests, followed by comprehensive mechanical characterization. Pearson correlation analysis revealed a strong association between water uptake and mechanical degradation. Eight machine learning algorithms—including Ridge Regression, Decision Tree, Support Vector Regression, Random Forest Regressor, Gradient Boosting Regressor, AdaBoost Regressor, XGBoost, and Neural Networks—were employed to construct predictive models; after data preprocessing and hyperparameter optimization, Gradient Boosting and AdaBoost achieved R 2 values above 0.89 and root mean square error (RMSE) below 1 for epoxy strength prediction, while ensemble methods for CFRP all exceeded R 2 of 0.90. Model forecasts indicate that the strength degradation of epoxy and CFRP stabilizes after approximately 36 and 40 days, respectively. These findings provide a theoretical basis for the design, performance optimization, and lifespan prediction of deep‐sea buoyancy materials and introduce a robust machine learning framework for material performance forecasting.
Meng et al. (Thu,) studied this question.