Fault detection and diagnosis based on particle filters combined with interactive multiple-model estimation in...

Area of Science:

• Process Systems Engineering
• Control Systems Theory
• Data-driven Fault Diagnosis

Background:

• Complex nonlinear stochastic dynamic processes often require advanced fault detection and diagnosis (FDD) methods.
• Single models struggle to accurately represent systems with abrupt operational changes.
• Existing FDD techniques may lack sensitivity to faults or robustness against noise.

Purpose of the Study:

• To introduce a new integrated framework, PFs-IMM, for fault detection and diagnosis in nonlinear stochastic dynamic process systems.
• To enhance the accuracy and reliability of state estimation and measurement reconciliation.
• To improve sensitivity to faults while maintaining robustness against disturbances and noise.

Main Methods:

• Combining particle filters (PFs) with interactive multiple model (IMM) estimation for a multiple-model approach.
• Utilizing IMM to handle abrupt changes in process behavior and adaptively select the most appropriate model.
• Employing parallel PFs for state and measurement estimation, incorporating a probabilistic weighting scheme.
• Developing a new fault index function for improved FDD sensitivity and robustness.

Main Results:

• The PFs-IMM approach effectively estimates the system's operational or faulty mode.
• It provides accurate overall state estimation and measurement reconciliation, even during mode changes.
• Simulation results demonstrate the method's effectiveness in highly nonlinear dynamic processes.

Conclusions:

• The proposed PFs-IMM method offers an integrated and effective solution for FDD in complex nonlinear systems.
• It successfully addresses limitations of single-model approaches and enhances robustness to process variations.
• This framework provides a valuable tool for improving the safety and reliability of industrial processes.