A New Predictive Maintenance Approach: novel integration of GMM-LSTM for prediction of latent state and failure location of Rotating Machinery

Authors:

Abstract
This paper introduces a novel integration of Gaussian Mixture Models (GMMs) with Long Short-Term Memory (LSTM) networks to enhance predictive maintenance strategies for industrial machinery. The contribution lies in using GMM for the initial classification of bearing states and LSTM to predict the precise location of potential bearing failures. This hybrid model leverages Fast Fourier Transform (FFT) and Wavelet Packet Decomposition (WPD) for robust feature extraction, creating a nuanced input representation for the LSTM. Our approach delineates a state-of-the-art method for real-time health monitoring and failure prognosis in rotating machinery, offering substantial advancements in predictive maintenance. The model is rigorously validated on distinct datasets, confirming its ability to generalize with high precision and recall rates while effectively circumventing overfitting. It successfully determines not only the condition of bearings but also identifies the exact bearing that may fail, providing a critical tool for preemptive maintenance actions. This integrated methodology of GMMs for state determination combined with the temporal prediction capabilities of LSTM sets a new precedent in the predictive maintenance domain, showcasing the potential for significant reductions in unplanned downtime and maintenance costs. The findings presented herein are instrumental for the field, proposing a practical and effective solution to a long-standing challenge in the maintenance of industrial rotating equipment.

This publication has 11 references indexed in Scilit: