关键词: 往复式压缩机, 故障诊断, 虚拟样本生成, 故障机理, 卷积神经网络, 角域分析

Abstract: Aimed at the problem that existing fault diagnosis methods for reciprocating compressors were faced with difficulties in constructing a high precision diagnosis model when fault samples were lacked,a fault diagnosis method for reciprocating compressors was proposed. This method combined fault mechanism driven Virtual Sample Generation (VSG) and Convolutional Neural Networks (CNN). First,a fault common parameter distribution model was established based on the fault mechanism of reciprocating compressors and the source domain data set. Next,fault features were calculated through common parameters and the normal angular domain data of the target domain. By combining the fault features with the normal angular domain data of the target domain,personalized fault virtual samples were generated. Finally,a fault diagnosis model that did not require real fault samples of the target domain reciprocating compressors was constructed using fault virtual samples and target domain normal samples. In the comparative analysis,the data from the reciprocating compressor fault simulation test bench and the production site data of a petrochemical enterprise were used to verify the proposed method. The results indicated that the average accuracy of the proposed method reached 88.92 %. This was 49.19 %,22.09 %,18.99 %,14.59 % higher than the comparative methods TCA-BPNN,LMD-SDAE,WKCL and RSTRN respectively,proving that the proposed method had a significant improvement in performance.

Key words: reciprocating compressor, fault diagnosis, virtual sample generation, fault mechanism, convolutional neural network, angular domain analysis