一种基于PCA与SVM的往复压缩机典型故障诊断方法

摘要

As the vibration signals of reciprocating compressor are nonlinear and non-stationary, a fault diagnosis method based on principal components analysis (PCA) and support vector machine (SVM) was proposed. Aiming at the typical faults of reciprocating compressor, such as scuffing of cylinder bore, piston rod fastening nut loose and cylinder collision faults, extracting the fault charac-teristics in time and frequency domain at first. Then use principal components analysis (PCA) method to reduce the dimension of characteristic matrix, and extract the sensitive characteristics as the new feature vector. Finally, the extracted sensitive characteristics will be used as the input of support vector machine(SVM). As a result, one structure model will be obtained after training the support vector machine. Using the model to predict the test feature vector, and therefore the type which the data belongs could be deter-mined. The real failure data are used to verify the effectiveness of the method for reciprocating compressor fault diagnosis.%往复压缩机的振动信号具有非线性、非平稳性的特点，对此提出一种基于PCA与SVM的典型故障诊断方法。针对往复压缩机典型的拉缸、活塞杆紧固螺母松动、撞缸等故障，首先提取出振动信号的时频域特征参数，再利用主成分分析（PCA）方法缩减特征参数的维度，提取出故障敏感特征，作为新的特征向量，最后再将提取出的敏感特征输入支持向量机（SVM）中，对其进行训练获得SVM结构模型，并将这个模型用于处理待测试的特征向量，由此判决数据所属的类别；利用真实的故障案例数据验证了此方法对往复压缩机故障诊断的有效性。