提出了一种基于均匀试验设计和BP神经网络模型构建随机参数与结构响应之间的近似非线性关系的概率有限元方法，该方法较直接MonteCarlo法需要的样本数据大幅减少，更适合大规模计算的结构概率有限元分析。采用该方法以某型柴油机气缸盖结构疲劳失效分析为算例，进行了可靠度期望寿命计算，并通过对影响气缸盖应力的燃烧压力、各螺栓预紧力和喷油器压紧力等22个载荷参数的均值和标准差的敏感性分析，初步确定了造成气缸盖本体结构疲劳开裂的主要影响因素。

A probability finite element method used to construct
the approximate nonlinear relation between random parameters and structure
responds was supposed based on uniform design and BP neural network. Compared
with direct MonteCarlo method, this method reduced
sampled data greatly and was better to apply to the largescale engineering structure probability finite element calculation.
Fatigue failure case of diesel engine cylinder head was carried out by this
method. The fatigue reliability life was calculated and the sensitivity
analysis of 22 loading parameters such as burning pressure, bolts preloading and fuel injector preloading was
performed too. The main influencing parameters which caused cylinder head
structure fatigue failure were determined preliminarily in the end.