Prediction of the compressive mechanical properties and reverse structural design of two-dimensional mesoscopic aluminum foam based on deep learning methods

This research introduces a novel approach to establishing a connection between the aluminum foam mesostructure features and mechanical properties. The proposed framework utilizes deep learning techniques to predict mechanical properties and enables a reverse structural design process for aluminum foam. Two-dimensional mesostructural images of aluminum foam are acquired using surface scanning and image recognition methods. Based on experiments and simulations, the compressive stress-strain response of aluminum foam is selected to establish the basic mechanical properties dataset. A two-dimensional convolutional neural network is utilized to predict aluminum foam mechanical properties, and a conditional generative adversarial network is used for reverse structural design. A two-way high-precision mapping relationship of aluminum foam is established, which links the mesostructure features with mechanical properties. The research results show that the established deep learning framework performs well in both forward prediction and reverse structural design. The error percentage of the forward mechanical properties prediction is within 3%, while the error percentage of the mechanical properties of the structure generated by the inverse structural design model is within 4%. The framework enables efficient and accurate mechanical property prediction and reverse design of aluminum foam structures. The deep learning framework has high engineering application potential and expansibility, extending the deep learning application scope in the porous materials field.