• CN:11-2187/TH
  • ISSN:0577-6686

机械工程学报 ›› 2024, Vol. 60 ›› Issue (24): 317-329.doi: 10.3901/JME.2024.24.317

• 交叉与前沿 • 上一篇    下一篇

扫码分享

基于预训练模型与相似度算法的复杂产品装配技术问题风险预警方法

张晋1,2, 刘检华1, 赵文浩3, 庄存波1,2, 张旭1   

  1. 1. 北京理工大学机械与车辆学院 北京 100081;
    2. 北京理工大学重庆创新中心 重庆 401151;
    3. 上海卫星装备研究所 上海 200240
  • 收稿日期:2024-01-20 修回日期:2024-07-30 出版日期:2024-12-20 发布日期:2025-02-01
  • 作者简介:张晋,男,1996年出生。主要研究方向为装配MES、装配质量管控技术。E-mail:3120210404@bit.edu.cn;刘检华,男,1977年出生,博士,教授,博士研究生导师。主要研究方向为数字化制造技术。E-mail:jeffliu@bit.edu.cn;赵文浩,男,1995年出生。主要研究方向为数字化制造。E-mail:1789559473@qq.com;庄存波(通信作者),男,1991年出生,博士,副研究员,硕士研究生导师。主要研究方向为装配MES、数字孪生技术。E-mail:zhuangdavid@bit.edu.cn;张旭,男,1969年出生,博士,副教授,硕士研究生导师。主要研究方向为工程设计与设计自动化、产品设计与工艺集成技术、产品生命周期管理与企业集成技术。E-mail:zhangxu@bit.edu.cn
  • 基金资助:
    国家重点研发计划(2020YFB1901401)、佛山-清华产学研合作协同创新专项(2020THFS0107)、水圈科学与水利工程全国重点实验室开放基金(sklhse-2023-E-03)、国家自然科学基金(51809148)和北京市自然科学基金(3192015)资助项目。

Risk Warning Method for Complex Product Assembly Technology Problems Based on Pre-training Model and Similarity Algorithm

ZHANG Jin1,2, LIU Jianhua1, ZHAO Wenhao3, ZHUANG Cunbo1,2, ZHANG Xu1   

  1. 1. School of Mechanical Engineering, Beijing Institute of Technology, Beijing 100081;
    2. Beijing Institute of Technology Chongqing Innovation Center, Chongqing 401151;
    3. Shanghai Institute of Spacecraft Equipment, Shanghai 200240
  • Received:2024-01-20 Revised:2024-07-30 Online:2024-12-20 Published:2025-02-01

摘要: 在复杂产品研制过程中,常因装配工艺不完善而出现各种装配技术问题。传统的管控方案是在装配技术问题出现后,对其进行多轮反馈处理,严重影响装配效率。为了减少装配技术问题的出现,提出一种基于预训练模型与相似度算法的复杂产品装配技术问题风险预警方法,实现在装配工艺设计阶段对潜在技术问题的风险预警。首先,对复杂产品装配流程进行分析。其次,针对短文本,使用Chinese-roberta-wwm-ext-large预训练模型;针对长文本,使用paraphrase-multilingual-mpnet- base-v2预训练模型,对历史技术问题反馈信息进行语义提取。在此基础上,分别使用改进词转移距离算法和余弦相似度算法计算历史问题工艺文本与新工艺文本的语义相似度,从而检索出新工艺中可能存在的风险点位并进行预警。最后,开发复杂产品装配技术问题管控系统并应用于某航天企业,验证了所提方法的有效性和准确性。

关键词: 复杂产品, 装配, 风险预警, 自然语言处理, 预训练模型, 相似度算法

Abstract: In the process of complex product development, various assembly technical problems often arise due to imperfect assembly process design. The traditional control scheme is to deal with them through multiple rounds of feedback after they occur, which seriously affects the assembly efficiency. In order to reduce the occurrence of assembly technical problems, this paper proposes a complex product assembly technical problem risk early warning method based on pre-trained models and similarity algorithms, which can realize the risk early warning of potential technical problems in the assembly process design stage. Firstly, the complex product assembly process is analyzed. Secondly, for short texts, the Chinese-roberta-wwm-ext-large pre-trained model is used; for long texts, the paraphrase-multilingual-mpnet-base-v2 pre-trained model is used to extract the semantics of historical technical problem feedback information. On this basis, the improved word mover’s distance algorithm and the cosine similarity algorithm are used respectively to calculate the semantic similarity between the historical problem process text and the new process text, and to retrieve the possible risk points in the new process and give early warning. Finally, a complex product assembly technical problem control system is developed and applied to a space enterprise, which verifies the effectiveness and accuracy of the proposed method.

Key words: complex products, assembly, risk warning, natural language processing, pre-training models, similarity algorithms

中图分类号: