基于制造大数据融合建模的多材料车身装配质量预测控制

The multi-material vehicle body design is an inevitable trend for the light weight new-energy vehicles. The theoretical assembly variation model and statistical process control methods are hard to solve the quality control problems for the dynamic multi-state, nonlinear, and uncertain assembly process for multi-material bodies. Therefore, based on the big data in the manufacturing process, this project proposes a systematic analysis flow including associated modeling, fusion of multi-source information and predicted quality control in the assembly process, and it will provide theoretical foundation and reference for the assembly process optimization and quality control. The studies mainly include: 1) with the multi-layer data collected in the multi-station, the pretreatment and clustering methods are studied for quality evaluation and data preparation. 2) the input and output features are chosen based on an associated modeling method. Then, a combined method based on pattern matching method for multi-state recognition and partial least squares regression is proposed to model the multi-state, irregularly sampling and nonlinear assembly process. 3) the multi-source information including assembly variation theoretical models, process simulation models etc., and also the data-based model are ensembling for a more accurate quality prediction. 4) in the uncertain condition, the decision tree model for assembly process optimization decision is constructed and solved to provide a data-driven automation optimization decision, such as combinations of incoming parts quality improvement, predicted maintenance schedules of fixtures or the other variation sources in the process. The proposed multi-information fusion modeling method and quality control methods will provide references for the quality control strategies in the production process in the mode of intelligent manufacturing and is probable to be used in the real assembly shops.

多材料混合车身是新能源汽车轻量化发展的必然趋势，复杂异材装配过程多模态、非线性和不确定性特点使得金属件装配偏差机理模型与统计过程控制不再适用。本项目从制造大数据角度探索装配偏差关联、融合建模与质量预测控制新方法，为多材料车身工艺优化、质量控制提供理论基础和参考。研究内容：1）研究多工位、多层面源数据的预处理与聚类方法，挖掘质量缺陷分布规律；2）基于频繁项集法提取质量输入输出要素，研究基于主元相似度评价和偏最小二乘回归结合的装配过程潜变量建模方法，解决不规则采样下多模态、非线性和动态制造过程建模难题；3）研究多源数据、装配机理、工艺仿真等多源信息融合的集成预测方法，实现车身质量的可靠预测与监控；4）研究不确定性预测下装配工艺优化决策建模与求解方法，实现装配质量的预测控制。本项目提出的融合建模与质量控制理论将为智能制造模式下装配质量保证提供参考和依据。研究成果有望实际应用于多材料车身装配过程。

本项目针对异材车身装配质量控制的工程难题，以车身薄板件装配为对象，从数据驱动角度提出基于多源信息融合的装配偏差数据建模与精度控制方法。提出多目标变量下基于最大相关最小冗余方法的检测特征提取方法，实现车身装配过程多层次不匹配数据条件下检测特征自适应布局与关键变量提取。针对工艺不确定性的装配偏差控制问题，提出基于潜结构建模与随机Kriging代理模型的在线装配工艺的优化控制方法，解决不规则采样下动态制造过程的尺寸数据建模难题。在上述模型的基础上，本项目提出一种基于多源信息融合建模的装配合格率预测和基于主动学习的装配质量控制方法，解决了现有质量控制方法难以满足高质量与生产节拍的要求，同时考虑高维点云建模存在的样本不足及点云无序性等问题。进一步，提出了基于装配工艺知识驱动的装配偏差源诊断方法，为多材料车身质量预测、工艺优化与维护决策提供理论依据和指导。本项目研究成果将推动制造大数据的应用理论发展，对于大数据建模理论在装配精度控制领域的技术落地有着重要意义。经过四年的研究已经完成项目的研究目标，共计发表标注本项目资助的论文20篇，其中在IEEE/ASME TM, JIM, JMS等期刊发表SCI/EI论文12篇，获机械工程领域知名学术会议ASME IMECE最佳论文奖（2019）、第十五届设计与制造前沿国际会议（2022）及上海汽车工程学会优秀论文奖（2021）。授权发明专利7项，项目负责人入选上海市“浦江人才计划”并依托本项目培养研究生毕业10名，3人获上海理工大学优秀学位论文奖且获市级校级优秀毕业生等荣誉称号。研究成果落地应用于上汽通用等汽车制造企业，获上海市质量协会优秀成果奖，项目组获上汽通用汽车“智造先锋团队”荣誉称号。