Enhancing the thermal efficiency to decrease the energy consumption of structures has been the topic of much research. In this study, a graphite-enabled microencapsulated phase change material (GE-MEPCM) was used in the production of a novel thermal energy storage engineered cementitious composite feathering high heat storage capacity and enhanced thermal conductivity. The surface morphology and particle size of the microencapsulated phase change material (MEPCM) were investigated by scanning electron microscopy (SEM). Thermal properties of MEPCM was determined using differential scanning calorimetry (DSC). In addition, thermal and mechanical properties of the cementitious mortar with different admixtures were explored and compared with those of a cementitious composite. It was shown that the latent heat of MEPCM was 162 J/g, offering much better thermal energy storage capacity to the cementitious composite. However, MEPCM was found to decrease the thermal conductivity of the composite, which can be effectively solved by adding natural graphite (NG). Moreover, the incorporation of MEPCM has a certain decrease in the compressive strength, mainly due to the weak interfaces between MEPCM and cement matrix.

Maintaining load-bearing capacity while accommodating deformation and absorbing energy remains a challenge for metallic roadway supports in rockburst-prone

Ground-penetrating radar (GPR) is among the primary techniques for detecting concealed road cavities; however, radargram interpretation is highly dependent on manual expertise, which may result in delayed or incorrect judgments and hinder timely structural and safety risk mitigation. This paper proposes RacNet, a lightweight model that integrates finite-difference methods with deep learning for rapid and automated road cavity detection. The model establishes a morphology-based theoretical formulation derived from real road cavity characteristics and constructs a multi-source cavity radargram dataset including engineering, experimental, and simulated data. An adaptive feature extraction and efficient feature fusion strategy is adopted for improved detection performance while reducing computational complexity. Additionally, a diversified batch simulation system was developed to encode and quantify real data characteristics, enabling rapid and diverse road cavity radargram generation. Experimental results show that RacNet achieves a detection accuracy of 98.4%, reduces computational complexity by 29.3%, and increases inference speed by 34.1%. Furthermore, RacNet outperformed baseline models and was successfully applied in real-world road inspections. The detected cavities were validated via borehole imaging, confirming the effectiveness and generalization capability of the proposed method. These findings support the application of deep learning in engineering measurement and provide technical support for risk assessment and safety management in road infrastructure.