A Novel Fault Diagnosis Method Based on Improved Negative Selection Algorithm

Abstract

In this article, an improved negative selection algorithm (NSA) is proposed and then applied to fault detection and fault isolation (also called as fault diagnosis). Coverage rate and overlap rate are the two important performance indexes on evaluating NSA. A higher coverage rate means a lower fault missing rate, and a lower overlap rate means a lower computation cost. However, the existing NSAs cannot keep a good balance between the coverage rate and the overlap rate. In our work, a new formula for overlap rate is first proposed, which is more accurate than the existing works. Then, the screening rule is introduced to decrease the overlap rate. More importantly, the detector radius can change automatically according to the overlap rate, and simultaneously, the detector movement is adopted to increase the coverage rate. Hence, the proposed improved NSA can not only increase the coverage rate but also decrease the overlap rate, which is then applied to fault detection and isolation. Finally, simulations on the data from the three-tank system are conducted to verify the effectiveness of the proposed scheme.