[4F4-GS-10o] AI応用：マテリアル情報

[4F4-GS-10o-02] Prediction of low cycle fatigue damage on thin metal film by using deep learning

〇Michiaki Kamiyama¹, Kazuteru Shimizu², Yoshiaki Akiniwa¹ (1. Yokohama National University, 2. NIPPON STEEL TECHNOLOGY Co., Ltd.)

Keywords:Metal Thin Film, Fatigue Fracture, Deep Learning

Metal thin films are used in a wide range of components of devices. For instance, there are usage such as wiring, electrodes and conductive films. Since the films are applied significant components in a product reliability, a prediction of its fatigue damage is important from the viewpoint of product reliability. The damage of the films on FPCs (Flexible Printed Circuits) are, however, difficult to predict it despite the severe fatigue conditions because of high complexity of the fatigue process. A method to predict the damage was considered by using CNN (Convolutional Neural Network). There is a possibility of prediction of the damage with high accuracy because the CNN directly evaluate fatigue cracks observed on the surface.The low cycle fatigue damage on the microscopic images of the copper thin film surface was predicted. A sum of an MSE and a VGG loss were used as a loss function in the training of the CNN. As a result, crack propagation caused by the fatigue damage was able to be predicted approximately. In addition, electric resistance ration which increase with progress of the fatigue damage were able to be predicted within 6% error.