基于内积延拓LMD及SVM的轴承故障诊断方法研究

针对特征提取中局域均值分解(Local Mean Decomposition,LMD)存在端点效应缺陷及模式识别中人工神经网络(Artificial Neural Network,ANN)存在收敛速度慢、过学习等不足,提出基于内积延拓LMD及支持向量机(Support Vector Machine,SVM)的轴承故障诊断方法。利用内积延拓LMD方法对信号延拓分解抑制LMD端点效应;利用分解的可描述信号特性主分量PF(Product Function)构建初始特征向量矩阵;用SVD(Singular Value Decomposition)方法对初始特征向量矩阵进行奇异值分解,获得信号特征参数作为SVM的输入进行训练;对训练的SVM进行测试及模式分类。通过实际轴承故障信号分析及故障类型分类表明,该方法不仅能抑制LMD端点效应缺陷,且在故障模式识别中能有效避免ANN网络结构难确定、收敛速度慢等不足,能较好实现轴承故障准确分类,可用于轴承故障诊断。

Research on bearing fault diagnosis method based on integral waveform extension LMD and SVM

Aimed at Local Mean Decomposition(LMD) end effect in feature extraction and Artificial Neural Network(ANN) having disadvantages of convergence slow and over learning in pattern recognition, the paper proposes a bearing fault diagnosis method based on integral waveform extension LMD and Support Vector Machine(SVM). Firstly, it extends the analyzed signal and decomposes it by the method based on integral waveform extension LMD to inhibit end effect; And use the main component , describing the signal characteristics, to establish the initial eigenvector matrix; Then decompose the initial eigenvector matrix by Singular Value Decomposition (SVD) method to achieve characteristic parameters and train the SVM. Finally, use the trained SVM to test and pattern classify. Through the experiments of analyzing actual bearing fault signals and fault types classification, the method not only inhibits the LMD end effects better, but also it avoids the ANN disadvantages, convergence slow and over learning in pattern recognition, and realizes the fault type classify accurately. The method can be used to bearing fault diagnose.