Sensor fault diagnosis in nonlinear stochastic systems: a hybrid system formulation

This paper introduces a novel Fault Diagnosis (FD) approach for nonlinear stochastic dynamic systems susceptible to multiple, potentially concurrent sensor faults. We formulate the FD problem as a hybrid state estimation problem. To achieve this, a vector of binary state variables is introduced to augment the continuous-valued state variables. These binary variables act as an on-off switching mechanism, providing a framework for independent reasoning about the presence or absence of each sensor fault. Leveraging this hybrid dynamic system formulation, encompassing both binary and continuous-valued states within a single Particle Filter (PF) framework, enables our approach to efficiently co-estimate fault modes alongside dynamic system states. This innovation ensures efficient diagnosis of multiple, potentially concurrent sensor faults while circumventing the need for the cumbersome bank of estimators characteristic of traditional Multiple Model (MM) approaches used for FD. The performance of the proposed algorithm in diagnosing multiple sensor faults is evaluated using a benchmark nonlinear system from the literature. The results demonstrably validate the algorithm's efficacy, solidifying its potential for practical sensor FD applications under conditions of single or multiple faults.