改进BP算法及其在滚动轴承故障诊断中的应用

对每次权值和阈值的调整均采用固定不变的学习率，是导致传统ＢＰ算法收敛速度慢的一个主要原因。本文从提高收敛速度及精度出发，对改进ＢＰ算法进行了深入研究，在ＢＰ算法中引入统计思想，给出相关系数定义。基于相关系数，采用变学习率策略，提出两种学习率自适应调整算法，并将其具体应用于滚动轴承的故障诊断中。试验证明，此改进算法的收敛速度比传统ＢＰ算法显著提高。

Improved BP Algorithm and its Application to Fault Diagnosis of RollingComponent Bearing

The constant learning rate is adopted for weights and thresholds adjustment each time is the main reason that results in low learning convergence speed of traditional BP algorithm. In this paper in order to improve convergence speed and accuracy of BP algorithm, the idea of statistics is introduced into BP algorithm and the definition of relevancy coefficient is given out. Based on relevancy coefficient and changeable learning strategy, two kinds of adaptive learning rate algorithms are put forward and applied to fault diagnosis of rollingcomponent bearings. The experiments show that this improved method improves more greatly than traditional BP algorithm in convergence speed.