Hopfield/ART-1 neural network-based fault detection and isolation

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Linear dynamic systems are prone to faults that can compromise performance.
• Existing fault detection methods may lack accuracy in parameter estimation and fault isolation.
• Real-time monitoring and adaptation are crucial for maintaining system integrity.

Purpose of the Study:

• To develop an advanced fault detection and isolation (FDI) scheme for linear dynamic systems.
• To leverage neural networks for robust system parameter estimation and fault classification.
• To enhance the reliability and accuracy of fault diagnosis in dynamic systems.

Main Methods:

• System parameters are estimated using a Hopfield-type neural network during normal operation.
• Faults are detected by observing parameter transitions during changes in system dynamics.
• Statistical tests on residuals in a moving window determine the transition zone exit.
• An ART-1 based neural network classifies estimated parameters for fault isolation.

Main Results:

• The proposed method effectively detects changes in system dynamics.
• Parameter estimation transitions reliably indicate the occurrence of faults.
• Statistical tests accurately identify when the system exits the transition zone.
• ART-1 network successfully classifies parameters for fault isolation in a position servo system.

Conclusions:

• The integrated approach of Hopfield and ART-1 neural networks provides a robust FDI solution.
• The method demonstrates practical applicability in complex systems like position servo systems.
• This technique enhances system safety and operational efficiency through accurate fault diagnosis.