Research on Composite Material Strength Prediction based on Deep Learning

Currently, composite materials stand as a leading choice for practical lightweight technology applications, favored in industrial designs for their lightweight, high strength, hardness, and significant specific modulus. Consequently, accurately predicting these materials’ mechanical properties is critical for their application. Traditional prediction methods primarily rely on theoretical models based on crystal structure to estimate metals’ yield strength, overlooking the potential of deep learning for precise material prediction. This study introduces innovative deep convolutional neural network models and attention mechanisms designed to autonomously extract features from raw material data, encompassing chemical composition, microstructural imagery, and historical mechanical property results. Our models demonstrate an ability to discern and leverage the intricate relationships between a material’s structure and its mechanical properties. Through comprehensive experimentation, we have verified that our approach achieves satisfactory prediction accuracy at manageable training and computational costs.