CNN-based prediction of compressive strength in porous Ti6Al4V alloy from micro-CT images

Recent studies have increasingly attempted to apply deep learning for predicting the mechanical properties of porous materials. In this study, a convolutional neural network (CNN) was applied to predict the mechanical properties of Ti6Al4V alloy foam using micro-CT images. For model training, data on compressive yield strength (YS) corresponding to porosities ranging from 40% to 70% were prepared. After combining micro-CT images with porosity and compressive YS data, the porosity and compressive YS were trained separately. The developed model achieved accurate prediction of porosity and compressive YS with high determination coefficients (R 2 ) of 0.9707 and 0.9818, respectively. For validation, Ti6Al4V alloy foam with porosities of 44%, 55%, and 66% were fabricated, followed by micro-CT scanning and compression testing. The developed model predicted the porosity and compressive YS of the fabricated specimens based on the micro-CT images, showing high prediction accuracy compared with the experimental measurements. Additionally, the Grad-CAM technique was used to visualize the regions influencing the CNN prediction of compressive YS, providing qualitative insight into the structural features contributing to the predictions. This study demonstrates the potential of predicting mechanical properties from micro-CT images without performing destructive compression tests.