University Events

Description: TOPIC: ANALYSIS OF FAULT TOLERANT CLASSIFIERS

LOCATION: Lester W. Cory Conference Room, Science & Engineering Building (Group II), Room 213A

ABSTRACT:
Prognostics and Health Management (PHM) has emerged as a promising approach to predict the deterioration of critical components prior to failure so that they can be maintained on an as-needed basis. PHM provides human maintenance personnel with advanced warning of potential failures and the possible fault locations. This information prepares maintenance personnel to inspect a system upon return from a mission, thereby accelerating fault isolation and diagnosis, reducing system down time.

Much of the existing research has focused on improving the accuracy of individual PHM methods. However, few studies have considered the potential benefit of utilizing multiple diverse classification techniques in a fault-tolerant architecture to increase the probability of fault detection. Moreover, these previous studies do not examine the negative impact of correlated failures, where failure in a majority of the individual PHM units can lead to system failure and possibly safety violations.

This project studied the application of fault tolerance to machine learning with an eye toward enhancing PHM and fault classification. Specifically, we apply ensemble classifiers in a majority voting scheme to failure data from the reliability research literature. Results indicate that redundancy can mask single misclassifications but correlated misclassification in two or more algorithms would lead to overall misclassification.

To mitigate the risk of correlated misclassification, we apply diversity measures, which quantify the difference between the outputs of these classification algorithms. These measures can identify better combinations of techniques from which to build an ensemble classifier, thus increasing the overall diversity and lowering the frequency of correlated failures. Six pair wise and non-pair wise diversity measures are studied in the context of an ensemble classifier for prognostics.

NOTE: All ECE Graduate Students are ENCOURAGED to attend.
All interested parties are invited to attend. Open to the public.

Committee Members: Dr. Honggang Wang, Department of Electrical & Computer Engineering and Dr. Iren Valova, Department of Computer and Information Science

*For further information, please contact Dr. Lance Fiondella at 508.999.8596, or via email at lfiondella@umassd.edu.