|
Activity Number:
|
321
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 14, 2002 : 12:00 PM to 1:50 PM
|
|
Sponsor:
|
Section on Quality & Productivity*
|
| Abstract - #301458 |
|
Title:
|
Statistical Substantiation of an Approach to the Selection of Fault Detection Algorithms
|
|
Author(s):
|
Roman Shapovalov*+
|
|
Affiliation(s):
|
Oklahoma State University
|
|
Address:
|
423 Engineering North, Stillwater, Oklahoma, 74078, USA
|
|
Keywords:
|
Bayes Risk ; Fault Detection ; Algorithm Selection ; Multiple Comparisons
|
|
Abstract:
|
Detection of process faults is a special case of pattern recognition. There are many very different by nature factors that affect the suitability of different pattern recognition algorithms for different process monitoring problems. In addition, the number of available pattern recognition algorithms is astronomical; hence, there is a need for a consistent approach to the selection of proper algorithms for process monitoring problems. We propose such an approach that reduces all the algorithms applicable for fault monitoring to a hierarchical set of simple components. The applicability of a certain algorithm for solving a particular fault detection problem is verified using a set of rules expressing the effects of different problem factors on each algorithm component and algorithm structure. This work presents a consistent approach to the generation and statistical validation of the algorithm applicability rules using statistical multiple comparison techniques. A special attention is given to the rule regarding a compromise between the required robustness of fault detector and time to detection that can be achieved using our proposed modification of the Bayes Risk.
|
