Influence of voids structure parameters on frost resistance of phase change concrete: ICT non – destructive testing and machine learning

The investigation of frost durability in hydraulic concrete is of critical importance. In this study, microencapsulated phase change materials (mPCMs) with varying phase transition temperatures (5°C and – 5°C) and dosages (0 %, 5 %, and 10 %) were incorporated into hydraulic concrete to develop modified mPCM – concrete systems with distinct temperature gradient configurations. Experiments involving temperature regulation, rapid freeze – thaw cycles , and nondestructive ICT (Industrial Computed Tomography) testing were conducted. The results indicated that mPCMs effectively delayed internal temperature fluctuations, mitigated void structure degradation, and suppressed thermal cracking, albeit with a reduction in concrete strength . Increasing the dosage of mPCMs (5°C) enhanced frost resistance , whereas mPCMs (– 5°C) exhibited the opposite trend. Composite phase change concrete demonstrated optimal frost resistance. A dataset comprising 144 samples was established through ICT testing. To optimize predictive performance, the sparrow search algorithm (SSA) was employed to tune hyperparameters of the XGBoost model, while recursive feature elimination with cross – validation (RFECV) was applied to eliminate redundant features. Interpretability analysis via SHapley Additive exPlanations (SHAP) identified mass and maximum equivalent diameter as dominant factors influencing compressive strength . Compared to deep neural network (DNN) and random forest (RF) models, the DNN demonstrated superior prediction accuracy, enabling reliable estimation of compressive strength for mPCM – modified concrete. The findings of this study are expected to advance the integration of nondestructive testing technologies with material design, offering novel insights for addressing frost damage in hydraulic concrete structures. • A novel microencapsulated phase change concrete with different phase change temperatures was developed. • Composite PCM concrete incorporating 5 °C and −5 °C microcapsules exhibited superior freeze – thaw resistance. • ICT was employed to extract void structure features of phase change concrete, with redundant features eliminated via RFECV. • SHAP – based interpretability revealed key factors affecting compressive strength , with DNN yielding higher prediction accuracy.