Date of Award

8-2022

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Automotive Engineering

Committee Chair/Advisor

Rahul Rai

Committee Member

Venkat Krovi

Committee Member

Bing Li

Committee Member

Laine Mears

Abstract

System health monitoring aids in the longevity of fielded systems or products. Providing a fault diagnosis or a prognosis can evaluate a system's current health. A diagnosis is the type of issue that could lead to a system's end-of-life (EOL); a prognosis is the remaining useful life (RUL) between the current state and the EOL. Fault diagnosis and RUL prediction can be acquired through (1) physics-based methods (PbM), (2) data-driven methods (DDM), or (3) hybrid modeling methods. DDM accurately provide a fault diagnosis, but the amount of data required is significant. This study reduces the amount of required data by upgrading the human-machine interaction between humans and DDM. The method is segmented into three distinct steps: (1) data collection, (2) a deep learning model, and (3) deployment/usability. The data collection develops a dataset with the highest difference between classes, controlling location and data type. After collection, the data is provided to a few-shot learning (FSL) model. The FSL model is trained to predict the difference between classes. After training, the requirements for prediction are deployed on a mobile device through an augmented reality (AR) application that illustrates what the user must do to create the optimal data sequence on the real-world system.

For prognostics, hybrid modeling is gaining traction. Generative Adversarial Networks (GANs) are well-known tools for data generation and semi-supervised classification. GANs, with less labeled data, outperform Deep Neural Networks (DNNs) and Convolutional Neural Networks (CNNs) in classification tasks. The success of GANs in classification tasks motivates the development of GAN-based techniques for semi-supervised regression tasks. However, developing GANs for regression introduces two major challenges: (1) inherent instability in the GAN formulation and (2) performing regression and achieving stability simultaneously. So this study explores improving the GAN architecture for regression tasks and deploying the improved GAN to perform RUL predictions. Lastly, the fault diagnosis and RUL prediction pipelines are combined to show improved RUL predictions with a provided fault condition and develop a complete profile of the system's health.

Available for download on Thursday, August 31, 2023

Share

COinS