提出和建立了一种用于液体火箭发动机（ＬＲＥ）故障检测的神经网络系统，这种系统包括两层：第一层由ＷＴＡ（Ｗｉｎｎｅｒ－Ｔａｋｅ－Ａｌｌ）神经网络组成，ＷＴＡ网络用于检测发动机故障输出模式；第二层由ＢＰ（Ｂａｃｋ－ｐｒｏｐａｇａｔｉｏｎ）神经网络组成，ＢＰ网络利用第一层次的输出结果作为输入显示故障大小。文中对ＬＲＥ故障检测进行了数值仿真，仿真结果验证了神经网络故障检测系统的优越性能。 A neural network system for fault detection of Liquid Rocket Engine (LRE) is proposed and established. The system consists of two layers: the first layer is composed of WTA (Winner-Take-All) neural network, which is used to detect the engine Fault output mode; the second layer by the BP (Back-propagation) neural network, BP network using the first level of the output as an input to display the size of the fault. In this paper, the numerical simulation of LRE fault detection is carried out. The simulation results verify the superior performance of the neural network fault detection system.