Strain and Displacement Field Reconstruction of Metal Damage-Prone Structural Components Based on ResFCN

During the operation of major technical equipment such as tunnel boring machines, trains, and aircraft, fatigue represents one of the most common failure modes. Accurate acquisition and analysis of displacement and strain fields at critical structural locations constitute the key to understanding failure mechanisms. Based on this premise, this paper proposes a displacement and strain field reconstruction method for metal damage-prone structural components based on Residual Fully Connected Network (ResFCN). First, metal structural damage-prone characteristic specimens are designed based on similarity principles and feature extraction methods, with multi-level load spectra determined and datasets constructed through numerical simulation. Second, the reconstruction accuracy of the proposed ResFCN network is compared with traditional BP neural networks, deep fully connected neural networks, and the classical regression model Random Forest, with results demonstrating higher accuracy for the ResFCN network model. Finally, the proposed reconstruction model is utilized to predict displacement and strain of different features under various loads. Experimental results show that the ResFCN model achieves an average relative error of 0.75% for displacement prediction and 3% for strain prediction, validating the effectiveness of the proposed reconstruction method.