Prognostics and Maintenance Decision-making for Mechanical Systems based on Condition Monitoring Data

• Author / Creator

  He, Rui

• Condition-based maintenance (CBM) is a maintenance approach that uses condition monitoring data to make maintenance decisions. The goal of CBM is to avoid machine shutdowns, reduce maintenance costs, and improve system safety. In modern mechanical systems, a wide range of sensors are used to collect data on the condition of different components. This has led to the proposals of various models, ranging from physics-based to data-driven methods, for predicting the remaining useful life (RUL) of components. Artificial intelligence techniques, specifically deep learning methods, have become increasingly popular in prognostics and health management due to their ability to process large amounts of data. The computational capabilities of machines have also advanced, allowing for numerical simulations that closely resemble real-world maintenance actions. These advancements provide more opportunities to utilize prognostic techniques and real-time data to support system maintenance.
  However, in order to efficiently schedule maintenance using prognostic information, it is crucial to approach issues from both the component and system perspectives. Specifically, at the component level, it is necessary to develop prognostic techniques that can handle situations where there is a scarcity of data. In practice, engineering assets are often replaced before failures, resulting in a small number of failure histories being available for analysis. This lack of data poses challenges in building prognostic models. On the other hand, it is difficult to determine a maintenance strategy at a system level, considering limited and inaccurate prognostic information. Integrating multiple data sources, such as predicted RUL, economic dependencies, and time-to-failure knowledge, into system-level maintenance optimization remains a challenge. These limitations form the main focus of this thesis.
  To bridge these gaps, the overarching objectives of this thesis are twofold, focusing on prognostics at the component level and making maintenance decisions at the system level. In the first three topics, deep learning methods are developed for prognostics on machines under different data availability scenarios. Specifically, when there is limited failure data but a larger amount of suspension data, a semi-supervised learning method is proposed to predict RUL by utilizing both types of data. Additionally, a novel method based on transfer learning is proposed for cross-domain prognostics, allowing for RUL predictions of machines operating in new conditions with only a small amount of suspension history data collected. Finally, a RUL prediction method based on state-space modeling and reinforcement learning is presented for situations where no data can be collected from machines operating in new environments. This method enables us to predict the RUL for machines operating beyond historical records.
  The last two topics explore strategies for maintaining mechanical systems using prognostic information. The main focus is on addressing the challenge of having limited monitoring capability for the system and inaccurate predictions of RUL during maintenance schedules. A maintenance optimization approach is proposed that considers not all components can be monitored. This method helps in identifying predictive thresholds and optimizing planned inspection together for the entire system. In addition, a novel maintenance basis is introduced to determine preventive maintenance actions when prognostics are not sufficiently accurate. By incorporating error modeling into the optimization model, insights into the impact of prognostic errors on maintenance costs are obtained.
  The research conducted in the thesis provides novel techniques for prognostics and maintenance decision-making using condition monitoring data. The developed methods will contribute to reducing the expenses associated with the operation and maintenance of mechanical systems.

• Subjects / Keywords

  • Prognostics
  • Condition-based maintenance
  • Wind farm
  • Deep learning
  • RUL prediction

• Graduation date

  Spring 2024

• Type of Item

  Thesis

• Degree

  Doctor of Philosophy

• DOI

  https://doi.org/10.7939/r3-7cve-bj41

• License

  This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.