基于状态-计划-故障信息的视情维修和生产计划的集成优化

Production process has a variety of deterioration states due to different characteristics such as operating costs, yield rates and failure possibilities. As a result, losses such as production breakdown and product quality fault are more likely to occur. Process state monitoring, production plan, together with failure history record can provide useful information from multiple perspectives. This brings possible benefits of integrated optimization based on deterioration state identification, classification and control. In face of a generalized production process, this project studies the following topics, (1) the state identification and classification based on fusion of multi-source data, to address the difficulty of maintenance timing and lack of useful information; (2) the control limits strategy and integrated optimization of condition-based maintenance (CBM), to address the difficulty of product varieties’ constraint on CBM; and (3) distributional robust optimization of CBM and production planning, to address the difficulty of multi-dimensional uncertainties’ requirement of robustness. The research team has achieved some premier results of relatively high level. The expected output of this project is to propose new integrated optimization method for CBM and production planning based on multi-source data. The enterprises are also boosted to make use of accumulated big data to optimize CBM, make the digitalized shift, and meet the strategic requirement of intelligent manufacturing.

由于运营成本、良品率和故障发生率等特征的不同，生产过程呈献多种劣化状态，容易导致生产中断、产品缺陷等损失。状态监测、生产计划和故障历史记录可以多角度提供有效信息，对生产过程劣化状态的识别、分类和管控提供集成优化的可能便利。项目面向一般生产过程展开，研究（1）多源数据信息融合下的状态识别分类，解决维修时效性、有效信息缺失性的难点；（2）视情维修(condition-based maintenance)的控制阈值策略、与生产计划的集成优化，解决产品类型多样性制约视情维修的难点；（3）视情维修和生产计划的分布式鲁棒优化，解决多维不确定性对方案鲁棒性要求的难点。项目组已取得较高质量的前期成果，研究产出在理论上提出基于多源信息的视情维修和生产计划的集成优化新方法；在实践上有利于企业累积大数据信息进行视情维修优化，进而开展数字化转型，符合智能制造的战略需求。

项目组按照申请书拟定的技术路线方案，从石化、印刷电路板等行业中的企业实际需求出发，提炼共性科学问题展开具体工作。首先，从日常运营层面，根据石化企业关键设备的电流、振幅和温度等运行状态数据，结合生产计划和历史故障信息，研究了生产过程的视情维修策略。针对表面贴装技术生产线的维护和调度集成优化问题，将设备故障、产出良品率等因素的影响概括为订单处理时间的不确定性，构建鲁棒优化模型，并设计分支-定价精确型算法。之后，从战略决策层面，考虑竞争、服务成本和设备可靠性，研究了生产企业的战略性维修服务选择。在基本完成预设研究目标之后，项目组根据上述研究取得的智能调度优化和数据分析等研究方法层面的成果，将基于机器学习的智能调度方法用于求解赛汝(Seru)生产系统中的维护和调度问题、“东数西算”国家工程的边缘算力链调度优化和O2O社区超市的订单在线拣选与配送研究；并将基于数据分析的预测方法应用于原油期货价格预测；借鉴生产企业选择维修服务外包的分析框架，项目组还在跨国公司的转移定价和零售渠道运营等方面取得研究成果。最后，通过对中国智能制造研究现状的可视化分析，发现本项目完成的基于数据分析的视情维修研究符合智能制造的战略发展方向。