Approaches to robust process identification: A review and tutorial of probabilistic methods

Abstract

Industrial data sets are often contaminated with outliers due to sensor malfunctions, signal interference, and other disturbances as well as interplay of various other factors. The effect of data abnormalities due to the outliers has to be systematically accounted while developing models that are resistant towards unforeseen effects of the outliers. The spectrum of methods that account for irregularities in process data while modeling are collectively known as robust identification methods. Even though, there are various non-probabilistic methods to tackle robust identification, few of them have considered the effect of outliers explicitly. In contrast to that, probabilistic identification methods ensure that these effects are given due attention. Despite these advantages, the probabilistic robust identification strategies have hardly been explored by practitioners. This review paper provides a general introduction to the probabilistic methods for robust identification, illustrates the main steps involved in the development of models, and reviews the related literature. Further, the paper contains two tutorial examples, including an industrial case study, to highlight various steps involved in the robust identification process.