神经网络在高速超精轴系故障诊断中的应用

讨论了ＢＰ神经网络神经元开关函数饱和参数的选择问题，建立了ＢＰ神经网络的容错能力、全局性能与收敛速率之间的关系，并提出了一个如何应用这些关系来合理选择神经元开关函数饱和参数的方法模拟试验研究和分析结果表明，本文提出的合理选择神经元开关函数的方法可以有效地帮助ＢＰ神经网络技术解决诊断中复杂的模式分类问题。

BP neural networks for fault diagnosison high-speed super-precision bearing

How to select the saturation coefficient of a neuron I/O function in BP neural net-works is discussed and the relations among the fault toleration ability, the generation capabilityand the contraction speed of BP neural networks are studied. A method to reasonably selectthe saturation coefficient is suggested. The result of analysis and simulation study shows thatthis method is helpful to solve the complicated pattern recognition problems while using BPneural networks in machinery fault diagnosis.