A new fault diagnosis model for rotary machines based on MWPE and ELM

Automated fault diagnosis techniques have been a popular demand in many industrial applications over the past decades. However, a practical diagnosis support system to be used in the field generally requires a high dregree of both diagnostic accuracy and efficiency, of which efficiency, ie fast learning and identification speed, is of greater importance for online and real-time fault detection and diagnosis. Therefore, this paper develops a new intelligent fault diagnosis model based on the use of multiscale weighted permutation entropy (MWPE) and an extreme learning machine (ELM), named MWPE-ELM, which is expected to achieve fast and accurate diagnostic performance. First, MWPE is utilised to analyse vibration data and extract the hidden fault-related features over multiple temporal scales; then, an ELM is employed to identify various health conditions. Diagnostic experimental results for both bearings and gears demonstrate that the proposed MWPE-ELM model not only achieved higher recognition accuracy but also a faster computation speed than the use of back-propagation neural networks (BPNNs) and support vector machines (SVMs) combined with different multiscale feature extraction approaches.