China Mechanical Engineering ›› 2022, Vol. 33 ›› Issue (07): 825-833.DOI: 10.3969/j.issn.1004-132X.2022.07.009

Previous Articles     Next Articles

An Assembly Quality Prediction Method for Automotive Instrument Clusters Using CNN-SVR

HE Yan1;XIAO Zhen1;LI Yufeng1;WU Pengcheng1;LIU Degao2;DU Jiang2   

  1. 1.State Key Laboratory of Mechanical Transmissions,Chongqing University,Chongqing,400030
    2.Chongqing Yazaki Meter Co.,Ltd.,Chongqing,401123
  • Online:2022-04-10 Published:2022-05-04

使用CNN-SVR的汽车组合仪表组装质量预测方法

何彦1;肖圳1;李育锋1;吴鹏程1;刘德高2;杜江2   

  1. 1. 重庆大学机械传动国家重点实验室,重庆,400030
    2.重庆矢崎仪表有限公司,重庆,401123
  • 作者简介:何彦,女,1981 年生,教授、博士研究生导师。研究方向为数字化制造与装备智能化、绿色设计与制造。E-mail:heyan@cqu.edu.cn。
  • 基金资助:
    重庆市技术创新与应用示范产业类重点研发项目(cstc2018jszx-cyzdX0147)

Abstract: Due to the long quality inspection time during assembly and lower production efficiency of automotive instrument clusters, an assembly quality prediction method for automotive instrument clusters using CNN-SVR was proposed. Combined with the assembly processes of instrument products, the production data features were extracted through CNN, which were used as the inputs of SVR to predict the pointer deflection angle that characterizesd the quality of instruments. The original production data of the instruments were obtained through the quality inspection systems of the assembly workshops, and the pointer deflection angles under different quality inspection conditions were predicted. The results indicate that proposed method has smaller prediction errors and strong generalization ability, which may accurately and effectively predict the assembly quality of automobile instrument clusters.

Key words: quality prediction, automotive instrument cluster, convolutional neural network(CNN), support vector regression(SVR)

摘要: 汽车组合仪表组装过程质检时间长、效率低,因此提出卷积神经网络与支持向量回归相结合的汽车组合仪表组装质量预测方法。结合仪表组装工艺,将卷积神经网络提取的生产数据特征作为支持向量回归的输入,对表征仪表质量的指针偏转角度做出预测。通过车间质检系统获取了仪表原始生产数据,对不同质检情况下的指针偏转角度进行了预测;结果表明所提方法预测误差较小,且具备较强的泛化能力,能够准确有效地预测汽车组合仪表的组装质量。

关键词: 质量预测, 汽车组合仪表, 卷积神经网络, 支持向量回归

CLC Number: