Machine learning approaches for prognostics and design of energy systems

Efficient health monitoring for identifying and quantifying damages can substantially improve the performance and structural integrity of engineered systems. Many energy system applications, such as Lithium-ion batteries and permanent magnet synchronous motors (PMSM), require accurate and efficient online state of health (SoH) estimations. Recent studies in the field of prognostics and health management (PHM) have shown that hybrid methodologies perform quite well, specifically methods that combine physics-based and data driven models. These approaches have been applied to many engineering concentrations; however, some of these techniques require substantial amounts of data, which is not always available in certain complex applications. In this dissertation, we focus on physics-informed machine learning (PIML) methodologies that address the challenges of fault detection accuracy, limited data, and overfitting. Chapter 2 centers on SoH estimation of Lithium-ion batteries, specifically real-time and independent estimations. The developed methodologies provides a bridge for combining PIML and filtering techniques, while addressing the experimental data requirement for machine learning functionality. Though these proposed prediction techniques can be generalized to many applications, the methods require continuous reliable data for accurate real-time estimations. Therefore, Chapter 3 focuses on optimal sensor placement for data quality and fault diagnosis. Specifically, a reliability based design optimization (RBDO) approach is employed for sensor placement and fault detection within a PMSM. This framework simultaneously determines the minimum number of sensors while training a classifier for fault detection, and can be implemented on various designs that require sensor networks. The proposed formulation was also extended to be “N-1” resilient, which means it satisfies a detectability level for when one of the sensors fails. While Chapter 3 provides a framework that can successfully detect faults and their severity level, determining the location of the fault can be critical for the maintenance and safety of the system. Thus, Chapter 4 demonstrates PIML methods that can not only detect faults, but identify the area where the fault occurred. Specifically, ensemble learning methods are implemented to provide a probabilistic estimate to the locale of the fault. Overall, this research aims to link the gap between machine learning and engineering applications with approaches varying from signals processing, reliability engineering, and design optimization. Moreover, the performances of the proposed PIML methods are demonstrated through energy applications, namely Lithium-ion batteries and permanent synchronous motors, and can be generalized to numerous high-impact engineered systems.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Industrial Engineering

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2024

Author dc:creator

• Kohtz, Sara

Contributors dc:contributor

• Wang, Pingfeng
• Miljkovic, Nenad
• Allison, James
• Chronopoulou, Alexandra

• Physics Informed Machine Learning
• Energy Systems
• Prognostics And Health Management
• Reliability

Statement dc:rights

• 2024 Sara Kohtz

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/124542

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F124542