Machinery Fault Diagnosis Using Recurrent Neural Network: A Review

Fault diagnosis is a kind of process to discover whether the continuous running mechanical system and equipment have faults using various inspection and test methods. It is widely applied in industrial engineering and has drew much attention recently since people need higher reliability and safety of machines to complete complex tasks, and the cost of repairment and maintenance is also high. While traditional fault diagnosis methods consume a lot of time and energy, those methods used deep learning which has excellent performance in feature extraction and pattern recognition become extremely important, thus a quantity of research work has been concluded in different networks of deep learning in this paper, a comprehensive and systemic review of machinery fault diagnosis using various forms of Recurrent Neural Network (RNN) and their combinations with other networks is introduced. First, some theoretical backgrounds of RNN are presented following its development, from RNN to Long Short-Term Memory (LSTM), bidirectional forms and Gate Recurrent Unit (GRU). Then, detailed literature survey of these networks and some combinations of RNN with Convolutional Neural Network (CNN) and Support Vector Machine (SVM) in practical applications are given. Finally, some limitations and discussions of these networks are discussed.