关键词: 容错控制, 控制分配, 多旋翼无人机, 强化学习, 电机故障

Abstract: Progressive motor degradation faults pose a serious threat to the flight safety of multi-rotor UAVs. To address this issue, this paper proposes a fault-tolerant control strategy based on Channel-wise Teacher-Policy Constrained Reinforcement Learning (CTCRL). First, a progressive motor degradation and temperature rise model is established to simulate the dynamic response of faults. Then, a channel-wise control allocation algorithm is designed to achieve effective isolation of faulty motors and mission reconfiguration. On this basis, a teacher policy constraint mechanism is introduced to further enhance the control reconfiguration capability under faults. Experimental results show that under the prescribed mild fault conditions, the CTCRL algorithm achieves a peak tracking error of 0.254 m, outperforming the compared non-crash algorithms. Under the prescribed moderate fault conditions, CTCRL reduces the average temperature of faulty motors by up to 5.95℃ compared to the contrast algorithm without temperature constraints. Under the prescribed severe fault conditions, CTCRL lands with the lowest impact kinetic energy of 199.76 J, which is only 14.6% of the free-fall kinetic energy, and exhibits the best trajectory controllability. The prescribed Hardware-in-the-loop experiments further verify that the proposed algorithm can effectively ensure the reliability and robustness of faulty UAV under progressive motor degradation, thereby providing underlying technical support for the mission sustainability of swarm formations.

Key words: Fault-Tolerant Control, Control allocation, Multi-rotor UAV, Reinforcement learning, Motor faults