Rough neural network based data-driven model-free adaptive fault-tolerant control for discrete-time nonlinear systems

This paper presents a data-driven Fault-Tolerant Control (FTC) system based on Model-Free Adaptive Control (MFAC) for a class of nonlinear discrete-time systems. First, the original system is converted in to a Compact Form Dynamic Linearization (CFDL) data model using the Pseudo-Partial-Derivative (PPD) technique. Second, a Rough Neural Network (RNN) is employed as an observer for Fault Detection (FD) by generating residual. Moreover, the obtained residual is incorporated into the CFDL and the optimality criterion to reconstruct the FTC strategy. The key contributions of this research include: 1) Considering and compensating the sensor and actuator faults simultaneously, which improves the overall system's robustness.; 2) the use of RNN as a powerful predictor for noisy and uncertain industrial data for FD, thereby enhancing the accuracy of FD; 3) the direct embedding of the generated residual from RNN into the controller without the need to estimate the fault function, simplifying the control process and 4) the research relies on input and output data from the controlled system for FD and FTC processes, which reduces the computational burden and increases efficiency. Finally, the simulation results show the effectiveness of the proposed data-driven FTC approach in comparision with the existing data-driven FTC approaches in fault control of some sample systems including high-speed train and CSTR system.