Intelligent fault diagnosis of rolling bearings based on LSTM with large margin nearest neighbor algorithm

In industrial machinery, rolling bearings are often rated as the most likely to fail in mechanical systems due to excessive working stress. Therefore, effective methods to diagnose the faults in rolling bearings are becoming necessary and required to ensure economic efficiency and manufacturing reliability. Recently, several studies tried to develop deep learning models for intelligent fault diagnosis based on traditional methods. However, creating an effective method for fault recognition is still a major obstacle due to varying operating conditions, large amount of collected data and redundant noise in measured vibration signals. Advanced signal processing techniques and traditional strategies often consist of shallow constructs that suffer from learning a huge amount of data, losing valuable fault information, yielding in low accuracy and labor time losses. In this paper, a combination method of long short-term memory with large margin nearest neighbor (LSTM-LMNN) is designed to address the above issues and effectively recognize multi-faults in mechanical rotating machines. Different from traditional LSTMs, the proposed LSTM-LMNN utilizes a powerful orthogonal weight initialization technique to memorize the critical information of faults during parameters updating and strongly organizing the samples of each condition in pattern classification process. Two experimental studies of bearing fault diagnosis demonstrate that the proposed LSTM-LMNN model outperformed other existing methods in terms of diagnostic efficiency, stability, and reliability.