基于改进高效偏最小二乘的质量相关故障诊断
Quality-related fault diagnosis based on improved efficient partial least squares
摘要点击 446  全文点击 52  投稿时间:2020-04-16  修订日期:2020-08-12
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  10.7641/CTA.2020.00197
  2020,37(12):2645-2653
中文关键词  过程监控  质量相关  故障诊断  改进高效偏最小二乘  新息贡献率  TE 过程
英文关键词  process monitoring  quality-related  fault diagnosis  improved efficient partial least squares  new information contribution rate  TE process
基金项目  国家自然科学基金项目(61673387, 61833016, 61903375), 陕西省自然科学基金项目(2020JM–356)资助.
作者单位E-mail
孔祥玉 火箭军工程大学,导弹工程学院 xiangyukong01@163.com 
解建 火箭军工程大学,导弹工程学院 hdxiejian@163.com 
罗家宇 火箭军工程大学,导弹工程学院  
李强 火箭军工程大学,导弹工程学院  
中文摘要
      高效偏最小二乘(EPLS)作为偏最小二乘(PLS)的扩展算法之一, 在质量相关故障检测中取得了良好的应用 效果. 然而, 研究发现当系统中存在一些与产品质量无关的信息时会导致EPLS的检测率降低, 影响工业生产安全及 效益. 同时, 传统的基于贡献图的故障诊断方法在无故障时输入变量会对故障检测指标的贡献值不均等, 从而影响 故障诊断效果. 针对上述问题, 本文提出了一种改进高效偏最小二乘(IEPLS)的质量相关故障诊断方法. 所提方法首 先用正常数据建立IEPLS算法模型, 利用获得的模型参数对过程变量进行空间分解. 然后在分解后的空间中定义局 部信息增量均值和局部动态阈值, 结合故障判据进行故障检测. 当故障发生后, 利用每个变量的新息矩阵计算对故 障总体的新息贡献率, 根据各个变量新息贡献率大小实现对故障变量的定位. 最后, 使用田纳西伊士曼过程(TEP)对 算法性能进行了验证.
英文摘要
      Efficient partial least squares (EPLS), as one of the extended algorithms of partial least squares (PLS), has achieved good application results in the quality-related fault detection. However, it is found that when there is information unrelated to the product quality in the system, the detection rate of EPLS is reduced, which can affect the safety and efficiency of the industrial production. Meanwhile, the traditional contribution graph-based fault diagnosis method has unequal contribution values to the fault detection index when there is no fault, thereby affecting the fault diagnosis effect. As such, an improved EPLS (IEPLS) quality-related fault diagnosis method is proposed in this paper. Firstly, the normal data are used to establish the IEPLS-based model, and the obtained model parameters are employed to spatially decompose the process variables. Secondly, the local mean value of the incremental information and the local dynamic threshold are defined to detect the fault in the decomposed space. When there is a fault, the new information matrix of each variable is applied to calculate the new information contribution rate to the total failure, and the fault variable is located based on the new information contribution rate of each variable. Finally, the performance of the proposed algorithm is verified by using the Tennessee Eastman process (TEP).