本文介绍采用自回归模型参数作故障特征,用模式识别中的 Perceptron 算法求解的线性判定函数对各故障状态进行分类的方法,介绍 Perceptron 算法的具体过程,讨论结束学习过程的条件.用这种方法对简单泵源系统故障诊断模拟试验研究表明,这种诊断方法原理正确、效果良好、具有工程应用价值. This paper introduces the method of using autoregressive model parameters as fault features and using the linear decision function of Perceptron algorithm in pattern recognition to classify each fault state, introduces the specific process of Perceptron algorithm, and discusses the conditions for ending the learning process. The simulation research on the fault diagnosis of simple pump system shows that this method has the correct principle and good effect, and has the value of engineering application.