Deep neural network-based adaptive supervisory control for strict-feedback nonlinear systems with sensor and actuator faults

This paper investigates the adaptive supervisory control problem for strict-feedback nonlinear systems with sensor and actuator faults, where some healthy actuators serve as backups. A deep neural network whose weights are updated in real-time is introduced to approximate the unknown nonlinearities. Based on this deep neural network, an adaptive supervisory control scheme without overparameterization is developed to ensure the prescribed performance of the resulting closed-loop systems by switching from the current faulty actuator to the subsequent healthy one. It is shown that the proposed deep neural network-based adaptive supervisory control scheme can achieve superior tracking performance to the traditional two-layer neural network-based adaptive supervisory control scheme. Finally, a numerical example is provided to validate the effectiveness of the presented control scheme.