Fault-tolerant control for nonlinear systems with a dead zone: Reinforcement learning approach

Area of Science:

• Control Systems Engineering
• Nonlinear Dynamics
• Artificial Intelligence

Background:

• Nonlinear systems with nonstrict-feedback structures present significant control challenges.
• Actuator faults and unknown dead zones further complicate control system design and stability.
• Traditional control methods often struggle with high computational complexity and local optima.

Purpose of the Study:

• To develop an adaptive reinforcement learning-based optimal control strategy for nonlinear nonstrict-feedback systems.
• To address actuator faults and unknown dead zones within the control framework.
• To mitigate computational complexity and avoid local optimal solutions inherent in traditional methods.

Main Methods:

• A novel neural network weight updated algorithm is proposed, replacing gradient descent for improved efficiency.
• Backstepping techniques are employed to design an actor-critic-based reinforcement learning controller.
• Two auxiliary parameters are introduced to effectively manage actuator faults and input dead zones.

Main Results:

• The proposed control strategy ensures semi-globally uniformly ultimately boundedness for all system signals.
• Simulation results demonstrate the effectiveness and remarkable performance of the developed approach.
• The novel algorithm successfully reduces computational load and avoids local optima.

Conclusions:

• The adaptive reinforcement learning control offers a robust solution for complex nonlinear systems with uncertainties.
• The proposed method provides a significant advancement in handling actuator faults and dead zones.
• The study validates the efficacy of the novel neural network update algorithm and backstepping technique.