Microparameter calibration method for concrete DEM using metaheuristic-based explainable machine learning and multi-objective optimization

Yuting Zhang, Xiang Zhou, Guangcheng Long, Zhaofei Long, Jiwu Yang,
Microparameter calibration method for concrete DEM using metaheuristic-based explainable machine learning and multi-objective optimization,
Journal of Building Engineering,
Volume 112,
2025,
113903,
ISSN 2352-7102,
https://doi.org/10.1016/j.jobe.2025.113903.
(https://www.sciencedirect.com/science/article/pii/S2352710225021400)
Abstract: The discrete element model (DEM) is widely used for simulating and analyzing the mechanical behaviors of concrete. However, concrete DEM involve numerous parameters, and intricate nonlinear relationships exist between microparameters and macro performance. Establishing correlations between microparameters and macro performance for DEM microparameter calibration is a significant challenge. This study proposes a framework integrating machine learning and multi-objective optimization algorithms to predict macroparameter and calibrate microparameters in DEM modeling. First, an ensemble machine learning model incorporating metaheuristic algorithms was developed to predict DEM macroparameter, with microparameters as input features and macroparameters as output targets. Baseline machine learning models establish micro-macro parameter mappings, while metaheuristic algorithms enhance model accuracy through hyperparameter optimization. Subsequently, the best-performing macroparameter prediction model was selected as the objective function to develop an inversion model based on NSGA-III multi-objective optimization, enabling precise calibration of concrete DEM microparameters. Model explainability analysis was conducted using SHAP. Results demonstrate that the metaheuristic-enhanced ensemble machine learning model achieves high accuracy and generalization in macroparameter prediction, with the RIME-XGB outperforming other configurations. Furthermore, the developed microparameter calibration model effectively calibrates microparameters for concrete DEM. The validation across two case studies confirms its satisfactory calibration accuracy. This study presents an efficient and accurate technical approach for concrete DEM microparameter calibration, reducing resource consumption associated with experimental trials.
Keywords: Concrete discrete element model; Parameter calibration; Machine learning; Multi-objective optimization; Explainable analysis