IMRPE和AO-SVM在往复压缩机故障识别中的应用

IMRPE and AO-SVM in reciprocating compressor fault identification

针对常规故障诊断方法不适用于提取往复压缩机声音信号的故障特征,导致往复压缩机的故障识别精度不高的问题,提出了基于改进多尺度反向排列熵(IMRPE)、t-分布邻域嵌入(t-SNE)和天鹰优化器(AO)优化支持向量机(SVM)的往复压缩机故障诊断方法。首先,采用具有优异特征表达性能的IMRPE方法来提取往复压缩机声音信号的故障信息,构建了反映样本故障特征属性的故障特征向量;然后,利用t-SNE方法对故障特征进行了特征降维处理,以降低故障特征维数和去除冗余特征,从而获得了低维的敏感特征;最后,利用AO方法对SVM的惩罚系数和核参数进行了自适应搜索,从而建立了结构参数最优的分类器,并将低维的敏感故障特征输入至AO-SVM分类器中,进行了训练和分类,依据测试样本的输出标签完成了样本的故障识别;以往复压缩机声音信号故障数据为对象开展了研究,并评估了IMRPE-t-SNE-AO-SVM方法的有效性和稳定性。研究结果表明:IMRPE-t-SNE-AO-SVM方法的故障识别精度达到了97%,不仅能够用于准确且稳定地识别往复压缩机的故障类型,提高故障识别的精度,而且在准确率和稳定性方面优于其它对比方法。

Aiming at the problem that the conventional fault diagnosis method is not suitable for extracting the fault characteristics of the sound signal of the reciprocating compressor,and the fault identification accuracy of the reciprocating compressor is not high.A reciprocating compressor fault diagnosis method based on improved multiscale reverse permutation entropy(IMRPE),t-distribution stochastic neighborhood embedding(t-SNE)and aquila optimizer(AO)optimized support vector machine(SVM)was proposed.Firstly,IMRPE method with excellent feature representation performance was used to extract fault information of reciprocating compressor sound signal,and fault feature vector reflecting sample fault feature attributes was constructed.Then,t-SNE method was used to reduce the dimension of fault features dimension to reduce the dimension of fault features and remove redundant features,so as to obtain low-dimensional sensitive features.Finally,AO method was used to adaptively search the penalty coefficients and kernel parameters of SVM classifier,so as to establish a classifier with optimal structural parameters,and input the low dimensional sensitive fault features into the AO-SVM classifier for training and classification,and the fault identification of the test samples was completed according to the output label of the test samples.Taking the acoustic signal fault data of reciprocating compressor as the object of study,and the effectiveness of the IMRPE-t-SNE-AO-SVM method were evaluated.The research results show that the proposed fault diagnosis method can not only accurately and stably identify the fault types of reciprocating compressors,improve the accuracy of fault identification,but also outperform the comparison methods in terms of accuracy and stability.