Koopman Fault-Tolerant Model Predictive Control for Multirotor with Actuator Fault

This article proposes a Koopman operator-based fault-tolerant model predictive control (K-FTC) for a multi-rotor with loss of effectiveness (LoE) fault. Using the Koopman operator, nonlinear system is transformed to an equivalent lifted linear system. The lifted linear system allows linear model predictive control (MPC) to be applied to the nonlinear system by reducing the required computational burden. When a fault occurs, a model should be updated to guarantee the performance of MPC. In this paper, weight window extended dynamic mode decomposition (WW-EDMD) is implemented to recursively update the input matrix in the lifted linear model, giving a higher weight to the latest data. The simulation results show that the proposed method can compensate for the LoE fault situation. In addition, an average computation time is calculated to confirm the possibility of implementation to a multirotor.