A CNN-Based Predictive Maintenance for DC-DC Converter

This paper presents a predictive maintenance of DC-DC converter by using a Convolutional Neural Network (CNN). The goal of this research is the implementation of an embedded system of predictive maintenance process which is incorporated on a DC-DC converter with a micro-controller. Therefore, the algorithm must be compact and fast. Considering this requirement, a very small CNN is suggested and, moreover, for its training data augmentation, temperature-variant PWM (Pulse Width Modulation) Duty signals are generated based on a theoretical expression. In the experiments, the prediction accuracy is evaluated against PWM Duty signals with high and low temperatures and it attained 100% in all cases. The suggested predictive maintenance process makes it possible to implement on actual DC-DC converter as an embedded system in the next step.