Hybrid ANN-physical model for predicting residual stress and microhardness of metallic materials after laser shock peening

•A physics-informed ML model is proposed to predict the residual stress and microhardness of metallic materials after LSP.•The inputs for this model mainly include the characteristics of shock waves, elastic–plastic parameters of the materials.•This model demonstrates superior prediction accuracy compared to traditional ML models and empirical formula.•Due to its versatility across different materials, this model can be applied to various metallic materials subjected to LSP. Residual stress and microhardness formed through laser shock peening (LSP) are crucial for enhancing the mechanical properties of metallic materials in industries like aerospace, automotive, and biomedical engineering. Therefore, precise and efficient assessment of microhardness and residual stress is vital for the successful implementation of LSP in industrial applications. In this paper, we propose a physics-informed machine learning (ML) model to address these assessment challenges and accurately predict the residual stress and microhardness of metallic materials after LSP. Firstly, different physical parameters are determined according to their corresponding mechanisms, and ABAQUS software is used to resolve the attenuation characteristics of shock waves induced by lasers. Subsequently, these identified physical parameters are utilized as input features for the artificial neural networks (ANN) model in order to forecast residual stress and microhardness. The predicted results reveal that our model exhibits a high level of precision in predicting microhardness (correlation coefficient of 0.99935) and residual stress (correlation coefficient of 0.99562) for a wide range of materials subjected to LSP. By comparing our physics-informed ML model with the traditional ANN models and empirical formula, its superior performance is effectively demonstrated in terms of accuracy and effectiveness (lower error and higher precision). Its superiority lies in the effective integration of ML methods’ representational capabilities with the combination of domain knowledge and physical understanding. This approach not only establishes a robust theoretical foundation for predicting these behaviors but also holds great promise for practical applications in industries that utilize LSP due to the universality for various materials.