합성곱 오토인코더를 이용한 체인 전동 장치의 고장 결함 감지 및 진단

초록

This paper presents a method to detect the mechanical faults of a chain drive power transmission system (CDPTS) using a convolutional auto-encoder (CAE). In previous research, it was known that the methods to detect faults of the CDPTS based on an artificial neural network (ANN) and convolutional neural network (CNN) were useful. In this paper, an advanced application of CNN, the CAE function of CNN is employed to detect faults. This method uses the characteristics of reconstruction of CAE. Difference of input images of the CNN and reconstructed images extracted by CAE were used as the guideline of fault detection. In the fault condition of the system, the difference was larger than the predetermined threshold of error. The encoder of CAE can be fine-tuned to classify the fault types of CDPTS. Finally, this method was well applied to diagnose the fault types of the test CDPTS installed in the laboratory.

키워드