基于RF-PCA-改进SVM模型的齿轮故障诊断方法

为提升齿轮故障诊断的正确率,提出了基于随机森林(Random Forest,RF)和主成分分析法(Principal Components Analysis,PCA)对齿轮振动信号进行特征降维处理,并采用粒子群算法(Particle Swarm Optimization,PSO)求解支持向量机(Support Vector Machine,SVM)特征参数最佳取值的齿轮故障诊断模型(RF-PCA-改进SVM模型)。对齿轮箱实例中正常、断齿、齿根开裂、剥落、削尖等9种不同状态进行了验证,证明了RF-PCA-改进SVM模型对齿轮故障诊断的有效性。试验结果表明:通过对比不同诊断模型识别率,证明RF-PCA-改进SVM模型具有更优的齿轮故障识别率,平均达到了99.66%,且计算效率较高;样本数量改变虽然会影响模型正确识别率,但不同的改变方式对识别率影响的程度不同,对于RF-PCA-改进SVM模型,当齿轮状态数据大于40个时即可达到88%以上的正确识别率。

Gear Fault Diagnosis Method Based on RF-PCA and Improved SVM Model

In order to improve the accuracy recognition rate of gear fault diagnosis,a gear fault diagnosis model (RF-PCA-improved SVM model) is proposed,which is based on random forest (RF) and principal components analysis (PCA) to reduce the dimension of gear vibration signals,and uses particle swarm optimization (PSO) algorithm to solve for the optimal value of support vector machine (SVM) feature parameters.Nine different states,including normal gear box,broken gear,cracked gear root,peeling and sharpening,are verified in the gearbox example,which proves the effectiveness of the RF-PCA-improved SVM model for gear fault diagnosis.The experimental results show that by comparing the recognition rates of different diagnostic models,it is proven that the RF-PCA-improved SVM model has a better gear fault recognition rate,with an average of 99.66%,and high computational efficiency.Although the number of samples may affect the correct recognition rate of the model,the degree of impact varies depending on the method of change.For the RF-PCA-improved SVM model,when the gear status data is greater than 40 groups,the correct recognition rate can reach over 88%.