Deep Learning Based Fault Tolerant Thrust Vector Control

Designing a fault tolerant control system under several actuator failures and external disturbances is a challenging problem for rockets. Previous methods struggle with providing immediate responses and recovering the vehicle in the case of a failure. In this study, we propose a deep learning based fault tolerant thrust vectoring control system using nonlinear dynamic inversion as the underlying control methodology for the loss of effectiveness and float type of failures. LSTM, as the deep neural network, is used to capture long time dependencies and understand the underlying pattern of the state information. For training the network, data set which is gathered from numerous simulations is created by considering different failure modes at different time steps during burn phase of the rocket. Superiority of the proposed method over the NDI based fault tolerant controller is demonstrated with example fault scenarios using high fidelity 6-DOF generic rocket model.