Analysis of Wooden Fracture Images and Yield Strength Estimation Using a Deep Learning Model with Grad-CAM

Abstract

In this study, a deep learning–based method was proposed to estimate the yield capacity from post-test images obtained from compressional tests of nail-timber. A convolutional neural network (CNN) was employed to learn image features, and it was confirmed that the yield capacity could be predicted directly from experimental images. In addition, by incorporating Grad-CAM (Gradient-weighted Class Activation Mapping), the regions of interest within the model were visualized, revealing that the network focused on critical areas such as the bearing zones around the nails. This approach enables visual interpretation of the AI’s decision-making process, thereby enhancing transparency in the application of artificial intelligence to structural engineering. Furthermore, the proposed method is expected to be applicable to the automatic identification of failure modes and the advancement of design guidelines.