基于gcForest算法的液压泵多传感信息融合健康状态诊断  

Health Status Diagnosis of Hydraulic Pump Based on Multi-information Fusion Based on Cascade Forest Model

作　　者：赵亚丽 路泽永 沙洲[3] ZHAO Yali;LU Zeyong;SHA Zhou(Hebei Petroleum Technical University,Chengde Hebei 067000,China;Hebei Institute of Instrument and Meter Industry Technology,Chengde Hebei 067000,China;State Key Laboratory of Precision Testing Technology and Instruments,Tianjin University,Tianjin 300072,China)

机构地区：[1]河北石油职业技术大学,河北承德067000 [2]河北省仪器仪表产业技术研究院,河北承德067000 [3]天津大学精密测试技术及仪器国家重点实验室,天津300072

出　　处：《机械设计与研究》2024年第4期226-229,234,共5页Machine Design And Research

基　　金：河北省重点研发计划资助项目(21375502D)。

摘　　要：由于液压泵运行的复杂特性,单一的传感器信号源检测存在故障识别低的问题。为了提高对液压泵复杂条件下的故障诊断能力,设计了一种基于多粒度级联森林(gcForest)算法的液压泵多传感信息融合健康状态诊断方法。在深度神经网络的多粒度分级方法中引入到森林分类器中实施运算,层叠森林各层次中都包含了传统结构与完整的随机森林分类器。开展实验平台测试分析,研究结果表明:利用多源数据的多粒径串级森林模型实现液压泵真实工况的精确诊断,使液压泵故障诊断的精度达到了99.6%,尤其适合于高维的重要特征提取。选择压力+流量特征作为指标不能达到理想诊断结果;以温度与流量参数组合获得预测精确率和召回率较高;选取压力+流量+温度组合达到几乎接近100%的诊断准确率。Due to the complex characteristics of hydraulic pump operation,using a single sensor for detection has the problem of low fault identification rate.In order to improve the fault diagnosis ability of hydraulic pumps under complex conditions,a multi-sensor information fusion health state diagnosis method based on multi-particle cascade forest(gcForest)algorithm is designed.In this paper,the multi-granularity classification method of deep neural network is introduced into the forest classifier,and the traditional structure and complete random forest classifier are included in each level of cascade forest.The experimental platform is tested and analyzed.The research results show that the accuracy of fault diagnosis is up to 99.6%,which is especially suitable for high-dimensional important feature extraction.The choice of pressure and flow characteristics as indicators cannot reach the ideal diagnostic results.The prediction accuracy and recall rates are higher by combining the temperature and flow parameters.The selection of pressure,flow and temperature combination achieves almost 100%diagnostic accuracy.

关 键 词：液压泵 信息融合 多粒度级联森林算法 健康状态诊断 

分 类 号：TH113[机械工程—机械设计及理论]