Computational renormalization for thermal conductivity of porous asphalt concrete based on hybrid finite element-neural network method

• Published in: Construction & building materials 2024-11, Vol.450, p.138725, Article 138725
• Main Authors: Yu, Shujun, Chen, Jiaqi, Wang, Hao, Qu, Yichen
• Format: Article
• Language: English
• Subjects: 3-D heterogeneous model; Neural network; Porous asphalt concrete; Renormalization; Thermal conductivity
• ISSN: 0950-0618
• DOI: 10.1016/j.conbuildmat.2024.138725

Thermal conductivity of porous asphalt concrete (PAC) directly affects PAC pavement temperature field, which is closely related to the heat island effect in summer and pavement icing in winter. Due to the large and complex pores and the uneven surface in PAC, there is measurement uncertainty using traditional laboratory tests. In this study, a finite element-neural network (FE-NN) method was proposed to predict the thermal conductivity of PAC based on the three-dimensional (3-D) heterogeneous structure. Additionally, laboratory tests were conducted to measure the thermal conductivity of PAC, fine aggregate mixture (FAM), and basalt aggregate. The results served as input parameters for the FE-NN method and were used to verify its accuracy. Based on the validated method, various factors affecting the thermal conductivity of PAC were analyzed. The analysis results show that, compared to the transient plane source (TPS) method, the backcalculation method can achieve results similar to those obtained with the TPS method for basalt aggregate and FAM, but provides more stable measurements when applied to PAC. The accuracy of the FE-NN method is acceptable, with relative errors of 1.92∼4.34 % compared to the experimental results. Besides, 2-D structural models tend to underestimate the effective thermal conductivity of PAC, with values 83.2∼84.1 % lower than those calculated by 3-D models. When the representative volume element (RVE) sizes are larger than 256×256×256, the variances show little differences and are acceptable. Furthermore, typical aggregate thermal conductivity values were used as input, revealing a nonlinear increase in the effective thermal conductivity of PAC. When the void content increased from 18 % to 24 %, the effective thermal conductivity of PAC decreased by 19.83 %. The effective thermal conductivity of PAC changes more significantly with the thermal conductivity of the aggregate than that with the variation of porosity in an interval. The analysis findings provide insights for better evaluating the pavement temperature field and heat transfer to improve the PAC pavement material designs. •A hybrid finite element-neural network method was developed in this study.•A backcomputation method was proposed to evaluate the thermal conductivity of PAC.•Analyzed effects of various factors on thermal conductivity of PAC.