Neural network-based robust actuator fault diagnosis for a non-linear multi-tank system

Area of Science:

• Control Systems Engineering
• Non-linear Dynamics
• Fault Diagnosis

Background:

• Dynamic non-linear systems are susceptible to actuator faults, impacting operational reliability.
• Existing fault diagnosis methods often struggle with robustness and handling complex non-linear dynamics.

Purpose of the Study:

• To develop a robust actuator fault diagnosis method for dynamic non-linear systems.
• To design an unknown input observer capable of estimating actuator faults under uncertainty.

Main Methods:

• Modeling the non-linear system using recurrent neural networks (RNNs).
• Transforming the RNN model into a linear parameter varying (LPV) form.
• Designing a robust unknown input observer within the H∞ framework.

Main Results:

• The proposed observer achieves a prescribed disturbance attenuation level for actuator fault estimation error.
• Guaranteed convergence of the observer dynamics is demonstrated.
• The method enables accurate actuator fault estimation.

Conclusions:

• The developed robust unknown input observer effectively diagnoses actuator faults in non-linear systems.
• The approach provides a foundation for implementing fault-tolerant control strategies.
• The RNN-to-LPV transformation facilitates robust observer design for complex systems.