Fault Diagnosis of Nonlinear Systems Using a Hybrid-Degree Dual Cubature-based Estimation Scheme

TL;DR

This paper introduces a hybrid dual estimation scheme using cubature rules for fault diagnosis in nonlinear systems, demonstrating improved robustness and accuracy over existing methods like UKF and Particle Filters.

Contribution

The paper presents a novel hybrid-degree dual cubature-based estimation approach tailored for nonlinear system fault diagnosis, incorporating case-dependent rules and robustness enhancements.

Findings

Successfully applied to a nonlinear gas turbine engine.

Outperforms UKF and Particle Filters in robustness and accuracy.

Effective in diagnosing component faults like fouling, erosion, and abrupt failures.

Abstract

In this paper, a novel hybrid-degree dual estimation approach based on cubature rules and cubature-based nonlinear filters is proposed for fault diagnosis of nonlinear systems through simultaneous state and time-varying parameter estimation. Our proposed dual nonlinear filtering scheme is developed based on case-dependent cubature rules that are motivated by the following observations and facts, namely (i) dynamic characteristics of nonlinear system states and parameters generally are distinct and posses different degrees of complexities, and (ii) performance of cubature rules depend on the system dynamics and vary due to handling of high-dimensional integrations approximations. For improving the robustness capability of our proposed methodologies modified cubature point propagation method is incorporated. The performance of our proposed dual estimation strategy is demonstrated and evaluated by application to a nonlinear gas turbine engine for addressing the component fault diagnosis problem within an integrated fault detection, isolation and identification framework. Robustness analysis is implemented to verify the capability of our proposed approaches to deal with parametric uncertainties and unmodeled dynamics. Extensive simulation case studies and discussions with respect to component fouling, erosion or abrupt faults are provided to substantiate and justify the superiority of our proposed fault diagnosis methodology when compared to other well-known alternative diagnostic techniques such as the Unscented Kalman Filters (UKF) and Particle Filters (PF) that are commonly available in the literature.