Abstract: As for the uncertainty and complex association in the process of fault diagnosis of Autonomous Underwater Vehicle（AUV） software system，an improved Failure Mode and Effects Analysis（FMEA） method is applied in the intelligent planning and decision-making control system.On the basis of analyzing main software failure mode，a three-layered Bayesian diagnosis model is proposed.By the inference mechanism of Bayesian Networks，experiments of single fault and multi-fault are given separately，and the results prove that the above methods can effectively improve the reliability of AUV system，as well as the ability of fault diagnosis.

Key words: Autonomous Underwater Vehicle（AUV）, software reliability, Failure Mode and Effects Analysis（FMEA）, fault diagnosis

摘要： 针对于智能水下机器人在软件系统故障诊断过程中广泛存在的不确定性和复杂关联性，采用改进的软件FMEA方法，对AUV智能规划决策控制系统进行了可靠性分析和研究，在总结了AUV主要软件故障模式的基础上，提出了一种基于FMEA的三层贝叶斯网络诊断模型。通过贝叶斯网络的推理机制，分别对单一故障和复合故障进行了推理实验。实验结果表明，采用上述方法能有效地提高水下机器人系统软件可靠性以及故障诊断能力。

关键词: 智能水下机器人, 软件可靠性, 失效模式影响分析, 故障诊断