An Intelligent Compound Fault Diagnosis Method Using One-Dimensional Deep Convolutional Neural Network With Multi-Label Classifier

Abstract

Intelligent compound fault diagnosis technology is significantly important to ensure that rotating machinery works in high-efficiency, security and reliability, and it remains a great challenge in this field. A lot of traditional intelligent fault diagnosis techniques have been developed with certain achievements in recent years, however, these methods inherently suffer from the obvious shortcoming that the traditional classifier only outputs one label for a testing sample of compound fault, rather than multiple labels. Consequently, it cannot classify a compound fault as two or more single faults. To solve this problem, a novel method named 1D DCNN-MLC, One-Dimensional Deep Convolutional Neural Network (1D DCNN) with a Multi-Label Classifier (MLC), is proposed for intelligent compound fault identification. 1D DCNN is employed to learn the representations from the vibration raw signals effectively. MLC is then designed to discriminate and predict the single or compound fault by outputting single or multiple labels. The proposed method is validated by a gearbox dataset with bearing fault, gear fault and compound fault. The experimental results demonstrate that the proposed method can effectively detect and identify compound fault. To the best knowledge of the authors, this work is the first effort to identify compound fault for rotating machinery via outputs multiple labels.