基于极限学习机的柱塞泵空化状态智能诊断研究  

Intelligent Diagnosis Reserch on Cavitation State of Piston Pump Based on Extreme Learning Machine

作　　者：李延民[1] 李明[2] LI Yanmin;LI Ming(Henan Shenma Nylon Chemical Co.,Ltd.,Pingdingshan 467013,China;School of Mechanical Engineering,Henan University of Science and Technology,Zhengzhou 471023,China)

机构地区：[1]河南神马尼龙化工有限责任公司,河南平顶山467013 [2]河南科技大学机械工程学院,河南郑州471023

出　　处：《机械制造与自动化》2023年第3期245-248,共4页Machine Building & Automation

摘　　要：针对传统依靠单一指标判断空化状态方法中存在的效果差和滞后性缺点,在多类特征融合的基础上,对不同工况下产生的空化外特性信号进行探讨,开发出一种以极限学习机(ELM)方法评价液压柱塞泵空化程度的新技术,能够非常准确地测定液压柱塞泵运行阶段出现的空化情况。研究结果表明:通过EMD完成空化状态的分解,再对IMF分量样本熵实施归一化,完成空化处理后形成了不同的样本熵。提高样本熵值后,形成了更复杂变化特征的样本序列。通过与BP神经网络、随机森林方法进行对比,表明采用ELM分类器处理时只需设置简单的结构和系统参数就可以消除人为因素造成的误差波动。用户也可以自主设置隐层节点的个数,极大地增加了系统的适应性。With regard to the poor effect and hysteresis of the traditional method of judging cavitation state by a single index,the cavitation external characteristic signals generated under different working conditions were discussed based on the fusion of multiple kinds of features,and a new technology was developed to evaluate the cavitation degree of hydraulic piston pump by extreme learning machine(ELM)method,which can accurately measure the cavitation of hydraulic piston pump during operation.Reserch results show that the decomposition of cavitation state is completed by EMD,and the sample entropy of IMF component is normalized.After the completion of cavitation processing,different sample entropy is formed,and the sample sequence with more complex changing features is formed with the increase of sample entropy.Compared with BP neural network and random forest,it is shown that ELM classifier can eliminate error fluctuation caused by human factors only by setting simple structure and system parameters.Users can also set the number of hidden layer nodes independently,which greatly enhances the adaptability of the system.

关 键 词：极限学习机 柱塞泵 空化 振动信号 特征提取 检测 

分 类 号：TH322[机械工程—机械制造及自动化]