Data-driven characterization of plastic deformation and mechanical properties in hot-pressed Ti/Al/Ti laminates

Abstract

This study investigates the prediction and evaluation of mechanical characteristics in hot-pressed Ti/Al/Ti laminates using a Random Forest (RF) machine learning model. The training dataset was generated through numerical simulations, encapsulating the laminates’ complex mechanical behavior under diverse conditions. To optimize model performance, hyperparameter tuning techniques, including Grid Search (GS), Particle Swarm Optimization (PSO), and Genetic Algorithms (GA), were applied. Among these, the GA-tuned RF model exhibited the highest predictive accuracy, achieving R² values of 0.947 for yield stress, 0.937 for yield strength, and 0.928 for Poisson’s ratio. The superior performance of the GA-tuned model is attributed to its effective feature selection and optimization capabilities, surpassing GS and PSO by identifying the most relevant input features. Relevance score analysis also revealed a balanced contribution of material geometry (e.g., thickness) and pressing parameters for predicting yield stress and ultimate strength, while induced strain played a significant role in predicting Poisson’s ratio. A case study using the GA-RF model further unveiled intricate relationships between input variables and mechanical properties, providing valuable guidance for optimizing hot-pressing parameters to enhance laminate performance.