Sensor faults have always been an essential and challenging problem in data-driven fault-tolerant control (DDFTC). Nevertheless, few studies in DDFTC have simultaneously considered the FTC problem under disturbance. To address the issue with both measurement noise and sensor fault, an iterative learning-based fault-tolerant control (ILFTC) strategy is presented when dynamics are completely unknown. First, an improved data-based unscented Kalman filter (UKF) mechanism is designed for the FTC scheme, which enhances the feasibility of the data-driven approaches under measurement disturbance. Then, the adaptive estimation algorithm with a switching mechanism is designed to improve the flexibility of the controller. Furthermore, a data-driven fault detection method is constructed under the framework of a broad learning system (BLS), which improves the detection accuracy while avoiding the need for precise model parameters in designing the fault detection observer. Finally, simulations and an application to the Franka-Panda robot demonstrate the superiority of the presented algorithm. Compared with the existing DDFTC schemes, smaller tracking errors and superior data signal-to-noise ratio (SNR) can be obtained.
