.png){kind=icon}
.png){kind=icon}
Return
Particle Filter-Based Fault Diagnosis With Hidden Markov Models for Nonlinear and Uncertain Dynamical Systems
.png){kind=icon}
PRE
AI
V
I
DOI：10.1109/TCST.2026.3682345.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This article presents a fault diagnosis (FD) scheme for nonlinear discrete-time stochastic systems that combines particle filtering with a hidden Markov model. Fault modes are represented as binary states embedded in the system dynamics, and the posterior expectation of these states provides a principled decision rule for diagnosis. The formulation eliminates the need for parallel estimator banks, reducing computational complexity while preserving accuracy. The hybrid filter is derived using sequential Bayesian inference and realized via the bootstrap particle filter (PF), chosen for its deployment simplicity and design flexibility for nonlinear, non-Gaussian systems. A central feature is the treatment of fault transitions: rather than requiring explicit transition probabilities, the method introduces a single mode retention parameter that regulates exploratory transitions of binary particles. The likelihood function drives their evolution, allowing fault-indicative particles to proliferate when faults occur. Validation through benchmark simulations and a real-world process control experiment demonstrates the effectiveness of the approach.
Keywords:
Fault diagnosis (FD)
hybrid state estimation
particle filtering
real-time monitoring
sequential Bayesian inference
Journal
IF:
3.9
Papers:
4.8K
Citations:
1.7W
Organization
Cited Papers