中文摘要：

针对无人水下机器人（UUV）传感器常见故障，采用一种基于有限脉冲响应（FIR）滤波器模型的在线故障诊断方法。根据该模型的故障检测结果，提出一种基于BP神经网络模型的容错控制策略，实现水下机器人首向角的估计以及传感器信号的在线重构。将重构的信号替代故障传感器信号，实现水下机器人在线容错控制。在OUTLAND 1000水下机器人定向控制系统中，首向角传感器（罗经）发生故障情形下，给出机器人的故障检测以及容错控制结果。

英文摘要：

A finite-impulse-response （FIR） filter was applied for on-line adaptive modeling and fault detection. A strategy of tolerant control based on BP neural network was proposed to estimate the actual position of an UUV and reconfigure the sensor signal. The reconfigured signal replaced the faulty sensor signal to implement the tolerant control of the UUV online. The technique of sensor fault tolerant control was implemented to the course keeping subsystem of OUTLAND 1000 on the occurrence of compass faults. Pool test results have demonstrated the effectiveness of the strategy.