Interfacial stress prediction and load inversion of a coat-substrate system via Green’s functions-based AI

This paper proposes a novel data-driven method based on Green’s function-generated data. The Green’s function solution under concentrated loads of the coating structure is obtained in this paper, which is the foundation for data generation. The response was constructed under arbitrarily distributed loads using the superposition principle, thereby fully characterizing mechanical field information under any loading condition. Relying on the theoretical model of Green’s functions, this approach enabled the efficient generation of large-scale, high-precision training datasets, which were subsequently used to train a neural network model capable of mapping the relationship between arbitrary load and stress response. Intelligent models for stress identification and load inversion in layered structures serve as critical foundations for safety design and tactile sensor design. However, the core challenge in developing high-performance intelligent models is determining how to obtain sufficient and high-quality training data. The advancement of many existing intelligent models has been constrained by this limitation. The method proposed in this paper breaks through the bottleneck of insufficient quality and scale of training data. Additionally, because stress distributions under arbitrary loads could be obtained by superimposing Green’s function solutions for concentrated loads, the neural network trained with the data generated by this method inherently carried complete stress field information and possessed broad generalizability. Once trained, the model could achieve high-precision stress identification and load inversion under any loading condition. Numerical examples demonstrated that the model exhibited excellent performance in both identification accuracy and computational efficiency.