The distributed configuration is an effective approach for implementing Hotelling's T2 test to detect faults in large-scale systems. Multiple T2 tests are employed in the distributed configuration in which each test uses measurements from a different set of sensors. A decomposition (i.e. an allocation of the sensors among different sets called subsystems) must first be selected to implement the multiple T2 tests. The detection performance of the multiple T2 tests and their ability to satisfy certain operational constraints depend on the decomposition. We propose two methods that find a feasible decomposition of a large-scale system for which the detection performance of the multiple T2 tests is near optimal. The methods first use greedy algorithms to generate many feasible candidate subsystems and decompositions. The methods then quickly evaluate the detection performance of these feasible candidates using easy to compute analytic expressions and finally select a feasible decomposition with the best performance.
Bibliographical noteFunding Information:
Partial financial support from NSF-CBET is gratefully acknowledged.
© 2020 Elsevier Ltd
- Distributed fault detection
- Hotelling's T test
- Large-scale systems
- System decomposition