Data analysis of diesel engine faults

Benjamin A Warman, Purdue University

Abstract

In recent years, On-Board-Diagnostics regulations have put more emphasis on reducing nitrogen oxide emissions and detecting and diagnosing faults in diesel engines. It is beneficial to investigate techniques to solve this problem. An algorithm called Sparse Linear Discriminant Analysis (SLDA) can be used to select relevant sensor signals in the detection and classification of faults. The work in this thesis applied those methods to data collected from three similar but separate engines used in the field that incurred identical faults. Healthy and faulty data samples were collected from each truck, and the data was processed to reduce selection bias between the two samples. The SLDA algorithm was then used on the processed data to select the most relevant sensor signals that defined the separation between healthy and faulty data. A classification algorithm was then used to determine the accuracy of the classifier. The combined use of the algorithms classified healthy data from faulty data with an accuracy of greater than 90%. The success of the methods used in this work shows the potential for real time implementation on engines in the field.

Degree

M.S.M.E.

Advisors

King, Purdue University.

Subject Area

Mechanical engineering

COinS