Intelligent condition monitoring of bearings for early material damage detection using acoustic emission signals
Doctoral thesis
Permanent lenke
https://hdl.handle.net/11250/3099096Utgivelsesdato
2023Metadata
Vis full innførselSamlinger
Sammendrag
Rotating machinery is a vital component in maritime vessels and wind turbines, and bearing is one of the most important parts in rotating machinery providing support, reducing friction, and facilitating smooth operation. However, bearings are susceptible to wear, degradation, and various other faults as they are subjected to continuous operation, escalating into catastrophic failures over time. Hence, early detection of incipient damages is an effective way to avoid downtime and loss of revenue as well as to protect both assets and employees.
Today, manual inspections and condition monitoring (CM) are still frequently used, however, it is not easily feasible in autonomous and remote-controlled vessels and subsea installations. In such scenarios, remote CM becomes the preferred option, allowing for continuous monitoring over long periods of time. This requires sensor data processing and analysis techniques to identify patterns and anomalies that indicate the presence of damages. In recent year, significant advancements in Artificial Intelligence (AI) techniques have emerged, offering a promising solution to address this challenge. This has sparked considerable interest and discussion in the field.
Some research gaps are to be mentioned: (i) The majority of existing studies focus on vibration analysis; (ii) The unsupervised early damage detection is relatively understudied compared to the supervised paradigm; (iii) The real-time monitoring of bearing requires further investigation. It has been reported that the current vibration-based CM systems for rotating machinery have limited sensitivity and capability for detecting pre-failure damages. As a result, by the time damage is detected, catastrophic failure becomes imminent, leaving little to no time for adjusting operational parameters to prevent further damage. This often necessitates shutting down operations until repairs or component replacements can be carried out. As an alternative non-destructive monitoring technique, Acoustic Emission (AE) has been found superior to vibration monitoring, as it can pick up signals from early damage before it propagates to the surface and become detectable by vibration sensors. Besides, AE detection also shows superiority in slowly rotating machinery where the hit energy is far too low to be detectable with vibration methods.
We aim to integrate the AE technology and power of AI algorithms, providing real-time insights into the condition monitoring of bearings in this Ph.D. work. The research process encompasses six research papers that contribute to the development of novel CM frameworks using AE signals and intelligent analytics. These frameworks are designed for early damage detection in machinery. By the introduction of sensitive and intelligent CM systems with real-time analytics capabilities, this research aims to advance the digitalization of the maritime sector. The goal is to reduce operational costs associated with maintenance.
The thesis is presented as a collection of publications that build towards the goal of intelligent early material damage detection of bearings with AE signals. The structure of the thesis is consisted of five components: The Chapter 1 provides a comprehensive overview of the background and research questions of this Ph.D. work. Chapter 2 investigates the existing literatures of related topics and outlines the research gaps and challenges of the present study. Chapter 3 summarizes the main contributions of each research paper. Chapter 4 elaborates the connections between the conducted research papers and the derived three research questions. A brief summary of the entire Ph.D. research and the prospect of future research are given in Chapter 5.
Består av
Paper A: Wang, Yu; Hestmo, Rune Harald; Vinogradov, Alexei. Early sub-surface fault detection in rolling element bearing using acoustic emission signal based on a hybrid parameter of energy entropy and deep autoencoder. Measurement Science and Technology 2023 ;Volum 34 https://doi.org/10.1088/1361-6501/acc1f8 Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence (CC BY 4.0)Paper B: Wang, Yu; Vinogradov, Alexei. Simple is good: Investigation of history-state ensemble deep neural networks and their validation on rotating machinery fault diagnosis. Neurocomputing 2023 ;Volum 548.https://doi.org/10.1016/j.neucom.2023.126353 This is an open access article under the CC BY license
Paper C: Wang, Yu; Vinogradov, Alexei. Improving the Performance of Convolutional GAN Using History-State Ensemble for Unsupervised Early Fault Detection with Acoustic Emission Signals. Applied Sciences 2023 ;Volum 13.(5) https://doi.org/10.3390/app13053136 This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license
Paper D: Wang, Yu; Bernat,Szymon; Vinogradov, Alexei. BC-GAN: a threshold-free framework for unsupervised early fault detection in rotating machinery - preprint Available at SSRN: https://ssrn.com/abstract=4276565 or http://dx.doi.org/10.2139/ssrn.4276565
Paper E: Wang, Yu; Wang, Qingbo; Vinogradov, Alexei. Ensembled multi-classification generative adversarial network for condition monitoring in streaming data with emerging new classes - OES 2023
Paper F: Wang,Yu; Wang, Qingbo; Bernat,Szymon; Vinogradov, Alexei. Ensembled multi-task generative adversarial network (EMT-GAN): a deep architecture for classification in streaming data with emerging new classes and its application to condition monitoring of rotating machinery