A Study on Motor Poor Maintenance Detection Based on DT-CNN

Signals of the motor failure, such as a bearing or gear fault, are proportional to its mechanical characteristics. These characteristics generate similar signal patterns for the same fault. However, signals in poor installation maintenance depend on how to install them. This means that a different signal pattern occurs in the same fault state. In this paper, DT-CNN (Decision Tree Convolutional Neural Network) algorithm is proposed to solve the above problems in applying deep learning to motor fault diagnosis. The supervised learning cannot verify outlier data. The proposed algorithm complements this disadvantage by using over fitted CNN in the selection process of the decision tree. In order to verify the performance of this algorithm, normal and gear fault data were collected. The data were collected by varying the motor installation state. Using these data, DT-CNN algorithm was implemented and it succeeded in detecting the maintenance failure signal almost similar to the normal state and the validity was confirmed.