基于GA优化SVM的滚动轴承故障诊断方法研究

Fault Diagnosis of Rolling Bearing Based on Genetic Algorithm and Support Vector Machine

在滚动轴承故障诊断研究中,常采用时域、频域或者时频域分析方法对振动监测数据进行故障诊断。时域中的无量纲指标因对故障敏感,而被广泛运用于机械故障诊断中,但目前无量纲指标在诊断过程中存在严重重叠问题,造成诊断准确率低。为了解决这个问题,研究了基于互无量纲指标和支持向量机(SVM)结合的滚动轴承故障诊断方法。针对SVM对参数依赖性强,且在参数选择上没有系统理论而导致欠学习或过学习的问题,提出了一种基于遗传算法优化支持向量机(GA-SVM)的滚动轴承故障诊断方法。利用遗传算法进化搜索原理,以预测的准确率作为适应值,对SVM参数进行寻优,从而得到较优的支持向量机分类模型。实验表明,基于互无量纲指标和GA-SVM算法的故障诊断方法能够准确地识别旋转机械滚动轴承的状态。

In the research of rolling bearing fault diagnosis,the time domain,frequency domain or time-frequency domain analysis method is usually used to diagnose the vibration monitoring data.Dimensionless indices widely used in mechanical fault diagnosis for the reason that the time domain signal is the basic and original signal,and they are sensitive to faults.However,there is a serious overlapped phenomenon in the diagnosis process of dimensionless indicators,causing lower diagnosis accuracy in practice.To address the above problem,a fault diagnosis method for rolling bearings based on mutual dimensionless index and support vector machine(SVM)is proposed in this paper.To solve the problem that the performance of SVM is largely depends on the selection of relevant parameters and there is not systematic theory of parameter selection which will lead to the problem of under-learning or over-learning,SVM based on genetic algorithms(GA)is proposed,which uses the evolutionary search principle of genetic algorithm to optimize the relevant parameters with the prediction accuracy as the fitness of genetic algorithm,and then obtain superior support vector machine model.The experimental results show that the method combining mutual dimensionless and GA-SVM can realize the fault diagnosis of rotating machinery bearings effectively.