Adaptive thresholding-a robust fault detection approach

Abstract

A new non-statistical adaptive thresholding technique is proposed to address the problem of detection of "abrupt" fault in the presence of system uncertainties due to variabilities such as usage, life cycle, environment, installation, build-to-build, product configuration, and product line. Computationally efficient algorithms are presented using set-membership identification and multi-step ahead uncertainty prediction to find a 100% confident bound that specifies region of nominal system behavior. This bound produces the adaptive threshold for abrupt fault detection scheme. Examples of abrupt fault and outlier detections using the field data are given to demonstrate the proposed approach.