剖析了基于 BP神经网络和径向基函数网络的故障诊断模型的诊断性能和应用中的局限性 ,针对这些诊断模型的局限性 ,提出了基于椭球单元 (Ellipsoid Unit)高阶网络的诊断模型 ,并对网络训练算法进行了研究 ,提出了基于模糊聚类算法的网络权重初始化方法和网络动态训练策略 ,有效地改善了网络的学习性能和诊断性能 ;最后对该网络在旋转机械故障诊断中的应用进行了研究。结果表明 :比之经典前馈网络 ,椭球单元网络在故障分类方面因其能形成封闭有界的决策区域而具有明显的聚类的优越性和分类的合理性 ,很适合故障诊断领域的分类问题 The diagnostic performance and application limitations of fault diagnosis models based on BP neural networks and radial basis function networks are analyzed. Based on the limitations of these diagnostic models, a diagnostic model based on Ellipsoid Unit higher order networks , And studied the network training algorithm, and proposed a network weight initialization method and a network dynamic training strategy based on the fuzzy clustering algorithm, which effectively improved the learning performance and diagnostic performance of the network. Finally, the network in the rotating machinery fault diagnosis The application has been studied. The results show that compared with the classical feedforward networks, the ellipsoid unit network has obvious advantages of clustering and classification rationality in terms of fault classification because of its ability to form a closed bounded decision region, and is very suitable for classification in the field of fault diagnosis problem