中国机械工程
中國機械工程
중국궤계공정
CHINA MECHANICAl ENGINEERING
2015年
11期
1532-1537
,共6页
齿轮箱%主成分分析%支持向量机%BP 神经网络%特征融合
齒輪箱%主成分分析%支持嚮量機%BP 神經網絡%特徵融閤
치륜상%주성분분석%지지향량궤%BP 신경망락%특정융합
gearbox%principal component analysis%support vector machine%BP neural network%feature fusion
为有效降低齿轮箱故障特征的维数并提高诊断准确率，提出了基于主成分分析法的齿轮箱故障特征融合方法，并结合支持向量机和 BP 神经网络对诊断的准确率进行了分析。以齿轮箱中不同裂纹齿轮为对象，选取能够表征齿轮箱故障状态的时域、频域和基于希尔伯特变换的36个特征，提取累积贡献率达到95%以上的主成分并输入支持向量机分类器中进行分类识别，用 BP 神经网络分类器进行结果的比较分析。结果表明，采用主成分分析法与支持向量机相结合的方法，既能降低特征维数，降低计算的复杂性，又能有效地表征齿轮箱的运行状态，识别不同裂纹水平的齿轮，比单独使用支持向量机分类器的方法诊断准确率更高，训练时间更短。
為有效降低齒輪箱故障特徵的維數併提高診斷準確率，提齣瞭基于主成分分析法的齒輪箱故障特徵融閤方法，併結閤支持嚮量機和 BP 神經網絡對診斷的準確率進行瞭分析。以齒輪箱中不同裂紋齒輪為對象，選取能夠錶徵齒輪箱故障狀態的時域、頻域和基于希爾伯特變換的36箇特徵，提取纍積貢獻率達到95%以上的主成分併輸入支持嚮量機分類器中進行分類識彆，用 BP 神經網絡分類器進行結果的比較分析。結果錶明，採用主成分分析法與支持嚮量機相結閤的方法，既能降低特徵維數，降低計算的複雜性，又能有效地錶徵齒輪箱的運行狀態，識彆不同裂紋水平的齒輪，比單獨使用支持嚮量機分類器的方法診斷準確率更高，訓練時間更短。
위유효강저치륜상고장특정적유수병제고진단준학솔，제출료기우주성분분석법적치륜상고장특정융합방법，병결합지지향량궤화 BP 신경망락대진단적준학솔진행료분석。이치륜상중불동렬문치륜위대상，선취능구표정치륜상고장상태적시역、빈역화기우희이백특변환적36개특정，제취루적공헌솔체도95%이상적주성분병수입지지향량궤분류기중진행분류식별，용 BP 신경망락분류기진행결과적비교분석。결과표명，채용주성분분석법여지지향량궤상결합적방법，기능강저특정유수，강저계산적복잡성，우능유효지표정치륜상적운행상태，식별불동렬문수평적치륜，비단독사용지지향량궤분류기적방법진단준학솔경고，훈련시간경단。
To effectively reduce the dimension of gearbox fault feature and improve the accuracy of diagnosis,a fault signal feature fusion method of gearbox was proposed based on principal component analysis,and the support vector machine and BP neural network were used to analyze the diagnosis accuracy.The 36 features with different crack gears in gearbox were selected based on time-domain, frequency-domain and Hilbert transform,which could be used to characterize the fault states of gear-box.The principal components which had more than 95% cumulative contribution rate were extracted and input into support vector machine classifier for identification.BP neural network classifier was used for comparative analysis of the results.Results show that a combination of principal component analysis and support vector machine method can reduce the feature dimension and computational com-plexity,characterize the gearbox running status effectively,and identify the different levels of gear crack.The diagnosis accuracy is higher and the training time is shorter than that of individual support vector machine classifiers.
