上海交通大学学报

上海交通大学学报 ›› 2019, Vol. 53 ›› Issue (7): 838-843.doi: 10.16183/j.cnki.jsjtu.2019.07.010

• 学报（中文） • 上一篇    下一篇

基于稀疏自编码神经网络的产品再设计模块识别方法

马斌彬1，马红占1，褚学宁1，李玉鹏2   

  1. 1. 上海交通大学 机械与动力工程学院， 上海 200240； 2. 中国矿业大学 矿业工程学院， 江苏 徐州 221116
  • 出版日期:2019-07-28 发布日期:2019-08-02
  • 通讯作者: 褚学宁，男，教授，博士生导师，E-mail: xnchu@sjtu.edu.cn.
  • 作者简介:马斌彬（1993-），男，浙江省宁波市人，硕士生，研究方向为产品再设计，E-mail: mabinbin@sjtu.edu.cn.
  • 基金资助:
    国家自然科学基金项目（51875345，51475290，51075261，51505480）

Identification of Product Redesign Modules Based on Neural Network of Sparse Auto-Encoder

MA Binbin 1,MA Hongzhan 1,CHU Xuening 1,LI Yupeng 2   

  1. 1. School of Mechanical Engineering, Shanghai Jiao Tong University, Shanghai 200240, China; 2. School of Mines, China University of Mining and Technology, Xuzhou 221116, Jiangsu, China
  • Online:2019-07-28 Published:2019-08-02

PDF (PC)

212

摘要： 提出了基于性能时变数据分析的再设计模块识别方法.利用产品在健康状态下的性能时变数据构建无监督学习的稀疏自编码神经网络(SAENN)模型，以用于健康状态下产品性能数据的特征提取以及产品功能退化程度的评估；将产品在健康状态下的性能数据用于训练SAENN模型，使用运行期间的性能时变数据更新产品的状态特征，以反映功能的退化过程；通过对比功能间的退化差异来识别需要再设计模块；同时，以某制造企业水平定向钻产品再设计功能模块的识别为例验证了所提方法的可行性.结果表明，所提出的再设计模块识别方法具有较好的准确性，能够识别需改进的功能模块，识别结果可作为产品再设计的依据.

关键词: 产品再设计； 模块识别； 性能时变数据； 稀疏自编码神经网络； 功能退化

Abstract: This paper presents a redesign-module identification method based on time-varying product usage data. The method is conducted in three steps：Building a model based on sparse auto-encoder (SAE) neural network using performance feature data from the product health state; Assessing functional performance degradation by using the data during the actual operation; Identifying the redesign module by comparing the difference in functional degradation. Then, a case study of horizontal directional drilling redesign module identification is presented to illustrate feasibility of the proposed method. The result shows the effectiveness of proposed method so that it can identify the weak function modules, while the identified result can provide support for redesign decision-making.

Key words: This paper presents a redesign-module identification method based on time-varying product usage data. The method is conducted in three steps：Building a model based on sparse auto-encoder (SAE) neural network using performance feature data from the product health state; Assessing functional performance degradation by using the data during the actual operation; Identifying the redesign module by comparing the difference in functional degradation. Then, a case study of horizontal directional drilling redesign module identification is presented to illustrate feasibility of the proposed method. The result shows the effectiveness of proposed method so that it can identify the weak function modules, while the identified result can provide support for redesign decision-making.

中图分类号: