Theses and Dissertations

ORCID

https://orcid.org/0000-0002-3875-4137

Advisor

Bethel, Cindy

Committee Member

Moss, Jarrod

Committee Member

Rahimi, Shahram

Committee Member

Jones, Adam

Committee Member

Vrantsidis, Thalia

Date of Degree

5-15-2026

Original embargo terms

Visible MSU Only 6 months

Document Type

Dissertation - Campus Access Only

Major

Computer Science

Degree Name

Doctor of Philosophy (Ph.D.)

College

James Worth Bagley College of Engineering

Department

Department of Computer Science and Engineering

Abstract

Predictive maintenance (PdM) is the idea of utilizing historical data of a mechanical system with machine learning (ML) to predict when a maintenance event will need to occur. This is done to maximize the up-time of the system and to minimize the costs that can be associated with maintenance related to accidental damages. These ML systems have to report to a human user to make a decision about whether to perform maintenance or store the vehicle for later maintenance. This interaction with humans would require the use of explainable artificial intelligence (XAI) as a way to explain the behaviors of the ML system in a way that humans can understand. Explainable predictive maintenance (XPM) is the field that applies the XAI systems to PdM; however, the field has not taken into account the human being on the receiving end of the explanations when studying the vast number of approaches. The goal of this research is to utilize a driving simulation and XPM techniques to study the impact of XAI on the participant’s accuracy at which they identify vehicle faults, confidence of their ability to identify the fault, speed at which they can complete a mission’s drive while maintaining the vehicle’s health, and satisfaction using the XPM system. The simulation had each participant drive a vehicle through an environment. At some point in the simulation, the vehicle began to break down due to a predetermined fault. The participant utilized a combination of the random forest-based PdM system and one of four conditions: (1) no XAI, (2) graphical XAI, (3) textual XAI, or (4) a combination XAI, to identify the fault and served as a proxy to fix the vehicle. With each fault fixed, the participant would report their confidence and satisfaction with the XPM system. The results of the experiment are reported in terms of statistical analyses. In addition, the discussion goes into how the results relate to the larger field of XPM and what considerations need to be in place when creating an XPM system. Lastly, future work is presented that details the different directions to take this particular simulation and what to study in the future in the field of XPM.

Download

Share

COinS