Data-Driven Distributed Kalman Filter-Based Sensor Fault Isolation and Estimation for Large-Scale Interconnected...

Area of Science:

• Control Systems Engineering
• Networked Systems Analysis
• Data-Driven Fault Diagnosis

Background:

• Large-scale interconnected systems face challenges in sensor fault management due to complexity and unknown interactions.
• Existing methods often require centralized processing, limiting scalability and robustness.

Purpose of the Study:

• To propose a novel data-driven distributed Kalman filter (DKF) scheme for sensor fault isolation and estimation.
• To enable effective fault diagnosis in heterogeneous subsystems coupled via a directed graph.
• To achieve fully distributed sensor fault isolation and estimation without requiring global system knowledge.

Main Methods:

• Development of a data-driven DKF-based residual generator within local diagnosis units (LDUs) using local and neighboring data.
• Implementation of distributed sensor fault isolation at subsystem and element levels, including simultaneous faults.
• Design of a DKF-based estimator for multi-subsystem sensor fault estimation with distributed Kalman gain computation.
• Local stability analysis performed without reliance on overall system information.

Main Results:

• Successful decoupling of unknown interaction components through localized data utilization.
• Realization of fully distributed sensor fault isolation, enabling global isolation with key LDUs.
• Accurate estimation of sensor faults across multiple subsystems using the proposed DKF estimator.
• Validation of the scheme's effectiveness and performance on a power network system.

Conclusions:

• The proposed data-driven DKF scheme provides a robust and scalable solution for sensor fault diagnosis in large interconnected systems.
• Distributed fault isolation and estimation enhance system reliability and maintainability.
• The method's local processing approach ensures stability and reduces communication overhead.