Deep Convolutional Auto-encoders Based Fault Diagnosis for Diesel Generator Set

Abstract

As a kind of power supply equipment, diesel generator set has the characteristics of good mobility, fast start, stable power supply, convenient operation and maintenance. Diesel generator set is very important for power supply applications. The research on automatic fault diagnosis of diesel generator set is of great significance for monitoring the operation status of diesel generator and timely maintenance. A deep convolutional auto-encoders based fault diagnosis for diesel generator set is developed. Convolutional auto-encoders replace the full-connected layers with convolutional layers, and provide a lower dimension latent representation to extract the invariable features for fault classification. The sensor data collected from diesel generator set are processed to form a dataset for training and test, and an architecture for deep convolutional auto-encoders is designed. The experimental results show that the deep convolutional auto-encoders based method has the best fault diagnosis performance in recall, precision, accuracy and F1-score than other learning based methods.