Automatic determination of optimal fault detection filter

Abstract

Optimal detection filters can greatly enhance fault detection performance, but designing these filters requires fault data which is difficult to obtain in practice. This paper proposes a scheme that automatically determines the optimal detection filter from a filter bank online without using fault data. The method can improve fault detection rate and accelerate detection speed. In order to reduce the false alarm rate, a method of threshold setting is introduced based on kernel density estimation. Implementation issues concerning filter bank design and online decision rule are also discussed. The method is validated in a numerical example and Tennessee Eastman process, and its performance is compared to those of other state-of-the-art methods.