Data-Driven Structure Displacement Response Surrogate Model Using Convolutional Neural Network

Accurate prediction of structure’s displacement response is imperative to evaluate structure performance. With the aid of the recent advances in deep learning, a data-driven structure displacement response model based on encoder-decoder convolutional neural network (CNN) is proposed for the efficient and accurate displacement response prediction. A multi-channel input–output data organization is developed, empowering CNN to learn the full-field mapping relationships among varied static physical features, including geometry, boundary conditions, etc. Available physics constraints are formulated by approximation distance functions (ADFs) based on the theory of R-functions, and are then encapsulated in input variables, which improve the accuracy and robustness of the trained model for complex boundary conditions. The surrogate model is evaluated on several common structures, including the simply supported plate and shear wall with opening. The performance of the model is demonstrated through the numerical experiments, and the significance of ADFs is validated by the comparison between its absence and presence.