Convolutional Autoencoder-Based Sensor Fault Classification

Automation machines perform not only simple operations but also operations requiring high accuracy. Sensors are essential to carry out the delicate operations. Therefore if there is a fault in sensors the machine can malfunction and the process-line will be damaged. To prevent this sensors should be monitored and diagnosed in real time. In the paper we propose a convolutional autoencoder-based sensor fault classification scheme in which time-domain statistical features and convolutional autoencoder features of sensor data are both utilized to classify types of sensor faults. Through simulation it is shown that the proposed scheme can improve classification performance of the sensor faults.