Abstract

Proton exchange membrane fuel cells are a promising technology for the automotive industry. However, it is necessary to develop effective diagnostic tools to improve system reliability and operational life to be competitive in the automotive market. Early detection and diagnosis of fuel cell faults may lead to increased system reliability and performance. An efficient on-line diagnosis system may prevent irreparable damage due to poor control and system fatigue. Current attempts to monitor fuel cell stack health are limited to specialized tests that require numerous parameters. An increased effort exists to minimize parameter input and maximize diagnostic robustness. Most methods use complex models or black-box methods to determine a singular fault mode. Limited research exists with pre-processing or statistical methods. This research examines the effectiveness of a Na¨ıve Bayes classifier on determining multiple states of health; such as healthy, dry, degraded catalyst, and inert gas build-up. Independent component analysis and principal component analysis are investigated for pre-processing. An automotive style fuel cell model is developed to generate data for these purposes. Since automotive applications have limited computational power, a system that minimizes the number of inputs and computational complexity is preferred.

Library of Congress Subject Headings

Proton exchange membrane fuel cells--Testing--Data processing; Bayesian statistical decision theory

Publication Date

8-2013

Document Type

Thesis

Student Type

Graduate

Degree Name

Mechanical Engineering (MS)

Department, Program, or Center

Mechanical Engineering (KGCOE)

Advisor

Jason Kolodziej

Advisor/Committee Member

Margaret Bailey

Advisor/Committee Member

Satish Kandlikar

Comments

Physical copy available from RIT's Wallace Library at TK2933.P76 R84 2013

Campus

RIT – Main Campus

Plan Codes

MECE-MS

Share

COinS