Applications of deep learning for fault detection in industrial cold forging
Andrew Glaeser, Vignesh Selvaraj, Soo Young Lee, Yunseob Hwang, Kangsan Lee, Namjeong Lee, Seung‐Chul Lee, Sangkee Min
Abstract
The feasibility of using deep learning techniques in industrial cold forging for fault detection was investigated. In this work, vibration data were collected from an industrial setting to detect machine conditions resulting in defective products (faults). After collecting data from several commonly encountered faults, a Convolutional Neural Network classifier detected fault conditions with 99.02% accuracy and further classified each fault with 92.66% accuracy. A decision tree (DT) model was also used in an attempt to detect and classify faults using time domain features. The model was able to detect faults with 92.5% accuracy but was unable to classify them. In addition, DT feature importance analysis was performed to understand how various faults impacted the machine signal for future refinement of the proposed system. The results suggest that the proposed deep learning method has the potential to detect faults in cold forging, but future work is required to validate the method.