一种改进BP算法及其在滚动轴承故障诊断中的应用  

Diagnostic Technology of Rolling Bearing Based on Dynamical Model of Neural Networks

作　　者：王太勇[1] 商同[1] 任成祖[1] 刘兴荣[2] 

机构地区：[1]天津大学 [2]河北工业大学

出　　处：《中国机械工程》2001年第10期1179-1181,共3页China Mechanical Engineering

基　　金：国家自然科学基金资助项目 ( 5 9875 0 6 7;天津市重点基金资助项目 ( 99380 2 411)

摘　　要：分析了前向型神经网络动力系统模型 ,根据该模型的特点提出了能够克服传统 BP算法学习速度慢、容易陷入局部极小的新算法。改进后的算法用于滚动轴承故障诊断 ,试验结果表明 ,该算法可以有效缩短网络在训练过程中滞留于局部极小区域的时间 。BP neural networks have such disadvantages as too many learning times and easy getting into local minimum. In order to overcome these shortcomings, in this paper,a dynamical model of feed-forward neural networks is analyzed,and is a new algorithm put forward,which can overcome the shortcoming of traditional BP algorithm such as the occurrence of temporary minimum and total training time is too long. The improved algorithm has been used in the diagnosis for rolling bearings. Results show that the new algorithm can minimize the time of network trapping in a temporary minimum and improve the learning speed greatly.

关 键 词：前向型神经网络 动力系统 Jacobian矩阵 局部极小 特征值 滚动轴承 故障诊断 

分 类 号：TH133.33[机械工程—机械制造及自动化] TP183[自动化与计算机技术—控制理论与控制工程]