计算机集成制造系统 ›› 2024, Vol. 30 ›› Issue (1): 217-226.DOI: 10.13196/j.cims.2021.0434

• • 上一篇    下一篇

不均衡小样本下的设备状态与寿命预测

陈扬,刘勤明+,郑伊寒   

  1. 上海理工大学管理学院
  • 出版日期:2024-01-31 发布日期:2024-02-04
  • 基金资助:
    国家自然科学基金资助项目(71632008,71840003);上海市自然科学基金资助项目(19ZR1435600);教育部人文社会科学研究规划基金资助项目(20YJAZH068);上海理工大学科技发展资助项目(2020KJFZ038);2021年上海理工大学大学生创新创业训练计划资助项目(XJ2021196)。

Equipment status and life prediction under unbalanced small samples

CHEN Yang,LIU Qinming+,ZHENG Yihan   

  1. School of Business,Shanghai University of Technology
  • Online:2024-01-31 Published:2024-02-04
  • Supported by:
    Project supported by the National Natural Science Foundation,China(No.71632008,71840003),the Shanghai Municipal Natural Science Foundation,China(No.19ZR1435600),the Humanities and Social Sciences Research Planning Fund of the Ministry of Education,China(No.20YJAZH068),the Shanghai University of Technology Science and Technology Development Fund,China(No.2020KJFZ038),and the Shanghai University of Technology College Student Innovation and Entrepreneurship Training Program in 2021,China(No.XJ2021196).

摘要: 针对面向小样本不均衡设备健康监测数据时AdaBoost处理效果差的问题,提出了基于裁剪过采样新增AdaBoost算法的设备健康状态分析以及寿命预测模型。首先,基于AdaBoost计算出样本权值分布和容量,根据样本最大权值与样本个数生成改进裁剪系数,选择性地对权值大于裁剪系数的样本进行处理从而提高计算效率。其次,通过类k近邻法则过滤出错分类样本权值,随后引入合成少数类过采样技术提升该种类样本权值个数,有效规避迭代过程中不均衡数据集可能引起的过拟合问题。最后,通过对设备运行状态进行准确分类并拟合出与时间相关的设备寿命曲线预测设备寿命。算例结果表明,所提模型能够有效分析出不均衡数据下的设备健康状况,同时也可以对剩余寿命进行有效预测。

关键词: 小样本, 不均衡数据, AdaBoost算法, 合成少数类过采样技术, 剩余寿命预测

Abstract: In view of the poor processing effect of AdaBoost for small sample unbalanced equipment health monitoring data,an equipment health state analysis and life prediction model were proposed based on clipping oversampling add AdaBoost algorithm.The sample weight distribution and capacity were calculated based on AdaBoost,the improved clipping coefficient was generated according to the maximum weight and the number of samples,and the samples with weight greater than the clipping coefficient were selectively processed,so as to improve the calculation efficiency.The error classification sample weights were filtered by the class k nearest neighbor rule,and then the synthetic minority oversampling technology was introduced to improve the number of sample weights,so as to effectively avoid the over fitting problem caused by unbalanced data sets in the iterative process.The equipment life was predicted by accurately classifying the equipment operation state and fitting the time-related equipment life curve.The example results showed that the proposed model could effectively analyze the equipment health status under unbalanced data,and could also effectively predict the remaining life.

Key words: small sample, unbalanced data, Adaboost, synthetic minority oversampling technology, remaining life prediction

中图分类号: