Mesoscale Thermomechanical Parameter Determination for Building Materials Based on Bioartificial Intelligence and the Finite-Element Method

Abstract The prediction of thermomechanical coupling behavior in brittle materials relies heavily on the accurate determination of their parameters. However, acquiring certain thermomechanical parameters of brittle materials, particularly at the mesoscale, poses significant challenges. Therefore, this research centers on establishing a computational method that synergistically combines bioinspired optimization [sand cat swarm algorithm (SCSO)], machine learning components [kernel extreme learning machine (KELM)], and numerical simulation techniques (finite-element method). The proposed method enables the systematic characterization of mesoscale thermomechanical properties of brittle materials using an inverse mapping from macroscopic deformation responses. To evaluate the effectiveness of the optimization strategy, the performance of KELM optimized by SCSO (SCSO-KELM) is systematically compared with KELM models optimized using genetic algorithm (GA), particle swarm optimization (PSO), and a cross-validation-based baseline KELM (CV–KELM). The results indicate that a stable and accurate relationship between macroscopic deformation and mesoscale thermomechanical parameters can be established under elevated temperature conditions. Compared with the CV–KELM model, all metaheuristic-optimized frameworks, including GA–KELM, PSO–KELM, and SCSO–KELM, achieve significantly reduced prediction errors, while the SCSO–KELM framework exhibits faster convergence and improved optimization stability in displacement prediction. Overall, the proposed framework provides an efficient and reliable approach for mesoscale thermomechanical parameter identification of brittle materials under high-temperature conditions.