基于支持向量机的离心泵初生空化监测

为利用机器学习的方法对离心泵运行状态进行监测,于离心泵发生空化故障前对离心泵初生空化状态做出判断,从而为离心泵运行状态在线监测提供一定的技术参考.针对基于支持向量机(SVM)的离心泵初生空化监测进行研究,采集离心泵运行振动信号,分析并选取均值、标准偏差、偏度、峭度等特征为特征向量训练模型,同时采用网格寻参与K-CV交叉验证的方式寻找最优组合参数.研究结果表明:网格寻优与交叉验证结合的方式能较好地寻找到最优参数;选取单一特征训练模型情况下,标准偏差的平均识别率最高,识别准确率为94.58%,以标准偏差、偏度、峭度两两组合的特征训练模型的平均识别率达到90%以上;该方法对离心泵初生空化识别具有较高准确率,具有一定鲁棒性,有较好的实用价值.

Monitoring of primary cavitation of centrifugal pump based on support vector machine

The purpose of this paper is to use machine learning methods to monitor the running status of centrifugal pump, and to make judgments on the initial cavitation status before cavitation failures occur in the centrifugal pumps, so as to provide a research foundation for online monitoring technology of centrifugal pumps. The primary cavitation of centrifugal pump was studied based on support vector machine monitoring. The vibration signals of centrifugal pump were collected and the features selection of the mean, standard deviation, skewness, kurtosis were selected as the eigenvector training model. At the same time, the grid search was used to participate in K-CV cross-validation way to find the optimal combination of parameters. The results show that the grid optimization combined with cross validation method can find the optimal parameters. In the case of single feature training model, the average recognition rate of the standard deviation is the highest, and the recognition accuracy rate is 94.58%. The average recognition rate of feature training model with combination of the model with standard deviation, skeuteness and kurtosis is more than 90%. This method has high accuracy, robustness and good application value for the identification of primary cavitation of centrifugal pump.