中国机械工程 ›› 2021, Vol. 32 ›› Issue (19): 2374-2382.DOI: 10.3969/j.issn.1004-132X.2021.19.013

• 工程前沿 • 上一篇    下一篇

基于多传感器信息融合和多粒度级联森林模型的液压泵健康状态评估

单增海1;李志远2;张旭2;黄亦翔2;李彦明2;刘成良2;张轩2   

  1. 1.徐州重型机械有限公司,徐州,221004
    2.上海交通大学机械系统与振动国家重点实验室,上海,200240
  • 出版日期:2021-10-10 发布日期:2021-11-05
  • 通讯作者: 黄亦翔(通信作者),男,1980年生,副研究员。研究方向为设备智能维护与故障诊断。E-mail:huang.yixiang@sjtu.edu.cn。
  • 作者简介:单增海,男,1972年生,高级工程师。研究方向为整车模型模拟、PID控制、性能预测。E-mail:zx_sjtu@126.com。
  • 基金资助:
    国家重点研发计划(2017YFB1302004);
    国家自然科学基金(51975356)

Health Status Assessment of Hydraulic Pumps Based on Multi-sensor Information Fusion and Multi-grained Cascade Forest Model

SHAN Zenghai1;LI Zhiyuan2;ZHANG Xu2;HUANG Yixiang2;LI Yanming2;LIU Chengliang2;ZHANG Xuan2   

  1. 1.Xuzhou Heavy Machinery Limited Company,Xuzhou,Jiangsu,221004
    2. State Key Laboratory of Mechanical System and Vibration,Shanghai Jiao Tong University,Shanghai,200240
  • Online:2021-10-10 Published:2021-11-05

摘要: 液压泵健康状态评估对工程设备的运行状态监测有极其重要的意义,现有基于振动信号分析的方法数据来源单一、可靠性低,为此,提出了一种基于多传感器信息融合和多粒度级联森林模型的液压泵健康状态评估方法。通过试验系统采集了不同工作时间下液压泵的压力、温度、流量等信号,分别提取信号的时域特征组成初步特征。使用多个分类器获取初步特征的类别概率向量,将其与随机森林模型选出来的重要特征进行拼接形成最终特征,并使用多粒度级联森林模型进行健康状态评估。试验结果表明,所提方法在仅有5%训练比例的情况下分类精确率仍可达99.5%,可以有效提高液压泵健康状态评估的准确度。

关键词: 液压泵, 多传感器融合, 多粒度级联森林模型, 健康评估

Abstract:  The health status assessment of hydraulic pumps was of great significance to the operating state monitoring of engineering equipment. The existing methods based on vibration signal analysis had single data source and low reliability, therefore a method was proposed for evaluating the health status of hydraulic pumps based on multi-sensor information fusion and multi-grained cascade forest. The pressure, temperature, flow and other signals of the hydraulic pumps were collected under different working hours through the test system, and the time-domain features of the signals were extracted to form preliminary features. Multiple classifiers were used to obtain the category probability vectors of the preliminary features, and the results were stitched with the important features selected by the random forest model to form the final features, and then the multi-grained cascade forest was used to evaluate the health status.The testing results show that the classification precision of the proposed method may still reach 99.5% when the training ratio is only 5%, which may effectively improve the accuracy of hydraulic pump health assessment.

Key words:  , hydraulic pump, multi-sensor fusion, multi-grained cascade forest model, health assessment

中图分类号: