Detection of fatigue defect in sustainable thermoplastic composite using infrared non-destructive testing and convolution neural network

Detection of the fatigue damage caused by cyclic loading, which is the most common cause of damage to parts in industry, can be predicted without damage occurring, using various instruments. In this study, the temperature rise during the fatigue test of the novel Elium thermoplastic composite material reinforced with glass fiber has been monitored using an infrared camera, and a set of videos has been recorded. Fatigue tests have been completed for three different fiber orientations (0/90, 0/90/45, ±45) at three different stress levels with low-cycle fatigue life. The frames in these captured videos have been trained using a 13-layer convolutional neural network (CNN) model. In this training, the point at which the sample reached its highest temperature and whether damage occurred there have been determined. At the same time, the temperature increase trend has been established, and the temperature at which the level occurs without damage has been determined. The validation accuracy rate for ±45 fiber orientation is over 93%, while the same rate for 0/90/45 is around 90%. These results were also confirmed by feature maps. When it comes to evaluation based on the confusion matrix F 1 score, the material with a 0/90/45 fiber orientation reached 91%, while this value remained at 86% for the ±45 orientation.