Date of Award

2025

Document Type

Open Access Dissertation

Degree Name

Doctor of Philosophy in Mechanical Engineering-Engineering Mechanics (PhD)

Administrative Home Department

Department of Mechanical and Aerospace Engineering

Advisor 1

Shangyan Zou

Committee Member 1

Zequn Wang

Committee Member 2

Trisha Sain

Committee Member 3

Chunpei Cai

Committee Member 4

Xinyu Lei

Abstract

This dissertation develops a knowledge-informed deep learning framework for robust fault diagnosis and prognosis across various engineered systems, including bearings, lithium-ion batteries, and mooring systems in wave energy converters (WECs). A deep transfer learning (DTL) method is first developed to improve bearing fault diagnosis by fusing data from multiple sources and extracting key features using convolutional neural networks (CNNs). Building upon this, a knowledge-informed deep network (KIDN) framework is proposed, integrating physics-based features with deep learning to enhance diagnostic accuracy and generalizability. A constrained Gaussian process (CGP) model is further employed to guide adaptive network design and reduce model tuning efforts.

For lithium-ion battery prognostics, a knowledge-constrained machine learning (KcML) framework is developed to predict remaining useful life (RUL). Prior knowledge of battery degradation is used to constrain the learning process, improving accuracy under diverse operational conditions.

The dissertation also extends the knowledge-informed strategy to fault diagnosis of WEC mooring systems. By combining autoregressive (AR) modeling and CNNs, the framework detects faults such as corrosion and biofouling and classifies fault severity based on dynamic responses like surge, heave, and pitch motions.

Overall, this research advances fault diagnosis and prognostics through the integration of domain knowledge and data-driven learning, leading to more accurate, generalizable, and efficient models for critical mechanical and energy systems.

Download

Share

COinS