Mechanical property prediction and configuration effect exploration of particulate reinforced metal matrix composites via an interpretable deep learning approach

Developing advanced particulate reinforced metal matrix composites (PRMMCs) with superior mechanical properties requires a deep understanding of their structure-properties relationships. However, the complexity and diversity in composite configurations of PRMMCs lead to the big difficulty in establishing their structure-properties relationships by traditional, time-consuming experiments and simulations. Herein, we propose an interpretable deep learning approach to accelerate the mechanical property prediction and configuration effect exploration of SiCp/Al composites. A spatial-temporal deep learning model is built to accurately predict the stress-strain relations of SiCp/Al composites across various configurations as well as to rapidly screen the configurations with superior strength-toughness matching. A 25 % improvement in strength-toughness matching of SiCp/Al composites is achieved by screening a million of composite configurations. Gradient-weighted regression activation mapping identifies the contributions of different configuration regions throughout tensile stages. Local configuration entropy is defined to characterize the configuration effects on mechanical properties, demonstrating a robust correlation with strength-toughness matching. © 2025 Elsevier B.V.