摘要
提出了一种利用光栅尺和编码器采集丝杠内部伺服信息的方法;针对滚珠丝杠副故障信号的非线性、非平稳性特征,引入经验模态分解法(EMD),对丝杠4种故障状态下的内部伺服信息进行时频域分析,并将峰度、频率、方差等时、频域特征组成原始特征集,以该特征集为输入,建立概率神经网络(PNN)模型,对滚珠丝杠副的故障状态进行模式识别。通过分析比较EMD-PNN与EMD-BP两种网络模型的性能和诊断结果,验证了EMD-PNN网络模型对滚珠丝杠副故障诊断的优越性及可行性。
A new method of using grating ruler and encoder acquisition the internal servo information of screw is proposed; according to the nonlinear and non stationarity characteristics of ball screw pair of fault signal,introduces empirical mode decomposition( EMD) of the 4 kinds of fault state of ball screw pair to time domain and frequency analysis,and combined with the frequency variance,kurtosis,time domain and the frequency characteristics of the composition of the original feature set. With the feature set as input,the PNN neural network model is established,the pattern recognition of fault state of ball screw is taken. Through the performance analysis of EMD- PNN network and EMD- BP network in the two diagnosis models,the superiority and feasibility of the EMD- PNN network model for fault diagnosis are validated.
出处
《制造技术与机床》
北大核心
2015年第3期62-67,共6页
Manufacturing Technology & Machine Tool
基金
国家自然科学基金项目(51075220)
山东省高等学校科技计划项目(J13LB11)
高等学校博士学科点专项科研基金(20123721110001)
青岛市科技计划基础研究项目(12-1-4-4-(3)-JCH)