Stochastic Event-Triggered Fault Detection and Isolation Based on Kalman Filter

Area of Science:

• Control Systems Engineering
• Signal Processing
• Electrical Engineering

Background:

• Discrete linear systems with unknown inputs and colored measurement noise present challenges for robust fault detection and isolation (FDI).
• Traditional Kalman filters may require approximations when dealing with non-zero mean, colored noise, and unknown inputs, impacting performance.
• Event-triggered control strategies offer potential for reducing communication load in FDI systems.

Purpose of the Study:

• To develop a robust FDI Kalman filter enhanced with stochastic event-triggered schedulers for discrete linear systems.
• To address challenges posed by deterministic/stochastic unknown inputs with nonzero mean and colored measurement noise.
• To optimize the trade-off between communication rate and FDI performance.

Main Methods:

• A subspace projection method is employed to attenuate disturbance effects.
• A fusion method is utilized to manage colored measurement noise in Kalman filter design.
• Stochastic event-triggered schedulers, modeled as Gaussian functions, are integrated to preserve innovation sequence properties.
• Convex optimization is used to determine design parameters for minimizing communication and maximizing FDI performance.

Main Results:

• The proposed FDI method effectively detects and isolates faults, specifically stator interturn short circuits and broken rotor bars in three-phase induction motors.
• The use of Gaussian-based stochastic event-triggered schedulers preserves the innovation sequence's Gaussian property, avoiding approximate recursive equations.
• Convex optimization successfully achieved a low communication rate between sensor nodes and the FDI filter while ensuring high FDI performance.

Conclusions:

• The developed robust FDI Kalman filter with stochastic event-triggered schedulers provides an effective solution for fault diagnosis in discrete linear systems.
• The method demonstrates superior performance in handling unknown inputs and colored measurement noise, validated through induction motor fault analysis.
• The approach offers a promising strategy for efficient and reliable condition monitoring in industrial applications.