A Cloud-Edge Collaborative Intelligent Fault Diagnosis Method Based on LSTM-VAE Hybrid Model

EasyChair Preprint 5466, version 2

Abstract

Fault diagnosis is of great significance for timely detection of safety hazards of machinery and ensure normal operation of production. To address the problems of low accuracy and poor robustness of mechanical fault diagnosis methods in general, the paper proposes a cloud-edge collaborative intelligent fault diagnosis method based on the LSTM-VAE hybrid model. The method trains the LSTM-VAE hybrid model in the cloud by using the vibration signals of mechanical components at the early stage of operation, and then reconstructs the real-time vibration signals in the edge by using the trained LSTM-VAE, calculates the difference degree between the original signal and the reconstructed signal, compares them with the adaptive threshold, and combines the "3/5" strategy to achieve fault warning. The experimental results show that, compared with other fault diagnosis methods, the proposed method can accurately diagnose the fault of rolling bearings with different degradation modes, and significantly improve the fault warning time in slow degradation modes, with high timeliness and strong adaptability.

Keyphrases: Cloud-Edge Collaborative, Long Short-Term Memory, deep learning, intelligent fault diagnosis, variational auto-encoder

@booklet{EasyChair:5466,
  author    = {Shaofei Lu and Xiaolin Tang and Yajun Zhu and Jingke She},
  title     = {A Cloud-Edge Collaborative Intelligent Fault Diagnosis Method Based on LSTM-VAE Hybrid Model},
  howpublished = {EasyChair Preprint 5466},
  year      = {EasyChair, 2021}}