Neuro-dynamic programming-based event-triggered fault tolerant control for nonlinear systems with multiple faults.

Area of Science:

• Control Systems Engineering
• Artificial Intelligence in Control
• Nonlinear System Dynamics

Background:

• Existing neuro-dynamic programming (NDP)-based fault-tolerant control (FTC) methods primarily address actuator faults, overlooking sensor faults.
• Online implementation of NDP-FTC is hindered by strict persistence of excitation (PE) conditions and initial admissible control requirements.

Purpose of the Study:

• To develop an online FTC scheme for uncertain nonlinear systems capable of handling multiple actuator and sensor faults.
• To integrate neural networks for simultaneous state reconstruction and fault estimation.
• To reduce computational resource consumption through an event-triggered control (ETC) strategy.

Main Methods:

• A neuro-observer with two neural networks (NNs) is employed for state and fault estimation.
• A critic NN derives an event-triggered control (ETC) policy indirectly.
• The NDP-based FTC strategy combines NDP-based ETC with an actuator fault compensator.
• An additional stabilizing term and experience replay technique relax PE and initial control conditions.

Main Results:

• Simultaneous accurate reconstruction of system states and estimation of actuator/sensor faults.
• Significant reduction in computational resource consumption via the event-triggered FTC strategy.
• Demonstrated uniform ultimate boundedness of observer errors, fault estimation errors, and the closed-loop system using Lyapunov's direct method.

Conclusions:

• The proposed online FTC scheme effectively handles multiple faults in uncertain nonlinear systems.
• The integration of NNs and event-triggered control offers a computationally efficient solution.
• The method relaxes traditional constraints, enabling practical online implementation.