Comparison of advanced set-based fault detection methods with classical data-driven and observer-based methods for uncertain nonlinear processes

Abstract

Automated fault detection (FD) methods are essential for safe and profitable operation of complex engineered systems. Both data-driven and model-based methods have been extensively studied, and some are widely used in practice. However, distinguishing faults from acceptable process variations remains a critical challenge, making both false alarms and missed faults commonplace. In principle, set-based FD methods can rigorously address this challenge. However, existing methods are often much too conservative, particularly for nonlinear systems. Moreover, few if any published comparisons clearly demonstrate the supposed advantages of set-based methods relative to conventional methods. This paper first presents a new set-based FD method based on discrete-time differential inequalities and demonstrates increased fault sensitivity through several case studies. Next, a detailed comparison of set-based methods with representative data-driven and model-based approaches is presented. The results verify some key advantages of the set-based approaches, but also highlight key challenges for future work.