中文摘要：

由于海洋深处工作环境的复杂性、不可预测性，水下机器人可靠性控制技术一直备受关注。在常规无人水下机器人推进器故障诊断中，均假设推进器处于几种固定故障模式，离线设计控制律，在线调度控制律实现容错控制，这与实际推进器故障情况有较大差别，因此采用 ICMAC（Improved Credit Assignment Cerebellar Model Articulation Controllers） ）神经网络在线辨识推进器的实际故障大小，再利用伪逆重构控制策略，产生容错控制矩阵，实现无人水下机器人在线自适应容错控制。并针对推进器突发性故障，给出了故障诊断与容错控制仿真结果。

英文摘要：

The reliability control technology for unmanned underwater vehicles is still a research hot topic because of complexity and unforeseen feature of underwater environment. Normal state and several different fault patterns were considered in conventional thrusters fault diagnosis of unmanned underwater vehicles, and the control law was designed off-line, and FTC （Fault-Tolerant Control） for unmanned underwater vehicles was realized by allocating control law on-line. But it was not different from actual thrusters fault situation. ICMAC （Improved Credit Assignment Cerebellar Model Articulation Controllers） neural network was used to realize the on-line fault identification for thrusters. The fault accommodation unit used information of the fault provided by the neural network to accommodate fault and perform an appropriate control reallocation. The system stability and performance were analyzed under failure scenarios. The numerical simulation was proposed for abrupt thruster fault.