基于LS-SVM多分类器融合决策的混合故障诊断算法

Hybrid fault diagnosis algorithm based on fusion decision of multiple LS-SVM classifiers

故障诊断的关键是特征向量提取和分类器的选择,提出一种综合运用多特征提取和多分类器组融合决策的故障诊断算法。多特征提取选择小波包变换、总体平均经验模式分解方法(Empirical Mode Decomposition,EEMD)和改进小波能熵方法,得到三组不同的故障特征信息;将这三组特征信息输入由3个最小二乘支持向量机(Least Square Support Vector Machine,LS-SVM)组成的分类器组进行初步诊断;采用自整定权值的决策模板法(Self-adjusting weighted Decision Templates,SWDT)进行多分类器诊断结果的融合决策。实验证明,该方法能实现轴承不同故障类型,尤其是复合故障的可靠识别,验证了该算法提取轴承故障特征信息的完备性,以及分类器组融合决策的可靠性。

The key of fault diagnosis is feature extraction and classifier selection. Here, a hybrid fault diagnosis algorithm was presented, using features extraction and fusion decision of multiple classifiers. The feature extraction methods included the wavelet packet transformation, EEMD, and the improved wavelet energy entropy, with them three different groups of fault feature information were obtained. The feature information was input into a classifier group consisting of three LS-SVM classifiers to make an initial diagnosis. SWDT was chosen to make a fusion decision of the initial diagnosis results. The tests indicated that the proposed method can realize a reliable identification of different bearing faults, even the compound faults; the completeness of the feature information extracted with this method and the reliability of the fusion decision of multiple LS-SVM classifiers are verified.