Deep wavelet sequence-based gated recurrent units for the prognosis of rotating machinery

Abstract

<jats:p> Prognostics and health management (PHM) is an emerging technique which aims to improve the reliability and safety of machinery systems. Remaining useful life (RUL) prediction is the key part of PHM which provides operators how long the machine keeps working without breakdowns. In this study, a novel prognostic model is proposed for RUL prediction using deep wavelet sequence-based gated recurrent units (GRU). This proposed wavelet sequence-based gated recurrent unit (WSGRU) specifically adopts a wavelet layer and generates wavelet sequences at different scales. Since vibration signals exhibit non-stationary characteristics, wavelet analysis is thereby needed to capture both the time and frequency domain information to fully identify the degradation of the rotating components. In the proposed WSGRU, the vibration signals are decomposed into different frequency sub-bands via wavelet transformation, and then a deep GRU architecture is designed to predict the RUL taking advantage of the temporal dependencies that naturally exist in the waveforms. Experimental studies have been performed for RUL prediction of bearings with collection of vibration signals during the run-to-failure tests. The prediction results show that deep WSGRU outperforms traditional models due to the multi-level feature extraction on the transformed multiscale wavelet sequences. </jats:p>

High-Rate Structural Health Monitoring and Prognostics: An Overview

“…To address the second and third challenges of lack of physics and inconsistency of the data observations, approaches may include developing physics-enhanced machine learning (PEML) models, which integrates the learning model with any known physics. Machine learning using data-based methods alone currently prove very powerful tools for predicting remainable useful life for rotational machinery [18,19] and prediction of high-rate dynamics [20]. Additional current trends include PINN (physical-informed neural networks), physics-informed machine learning (PIML) [21], and digital twin interpretability [22].…”

A new fault diagnosis method based on adaptive spectrum mode extraction

“…1–3 Once a failure occurs, it may cause huge economic losses and personal safety problems. 4–6 In recent years, with the rapid development of rotating machinery toward complexity and intelligence, fault diagnosis of rolling bearings has become a huge challenge. 7 In particular, when the fault signal is disturbed by noise, how to effectively extract the fault features in the fault signal is the key to diagnose the fault.…”

Scalability, Explainability and Performance of Data-Driven Algorithms in Predicting the Remaining Useful Life: A Comprehensive Review

“…Ma et al presented Wavelet Sequence-based GRU (WSGRU), where a wavelet layer is added prior to feeding the data into a GRU which performs continuous wavelet transform to project the signal into different scales. This single layer addition handles nonstationary signals by extracting the time-frequency features, allowing the network to better learn the fault information[109].…”