A Game Theoretic Approach to the Design of Multi-Agent Process Planning Systems

本研究计划的目的是建议如何设计制造过程计划系统。在自动化制造中，工艺规划是将设计转换为制造指令有序列表的中间步骤。例如，在机械加工中，一块均质金属（坯料或初始工件）必须通过适当地去除材料来变形。对于每一步，必须指定工件的位置，如何保持到位（夹具），要使用的特定刀具，刀具进入方向和精确的刀具路径。计算机辅助工艺规划系统以标准格式定义设计的计算机文件——计算机辅助设计系统的输出——并为计算机辅助制造设备找到计算机指令，以制造所需的工件。越来越多地，这样的过程规划系统包含多个人工智能代理（或分布式决策者）。过程计划者（以及一般的多代理系统）定义了特定的游戏。对这类博弈的研究有助于人们理解过程计划者和多代理系统是如何工作的，为什么它们会失败，以及如何改进它们。博弈论机制设计分析的是如何设计博弈，使预期结果作为博弈中的均衡出现。本研究将首先利用博弈论和机制设计的现有成果来研究过程规划者和多智能体系统。然后在网络和分布式博弈中得到新的结果，并分析智能体的潜在分解。目标是建立严格的理论原则，基于战略优化行为，指导多智能体过程规划系统和一般多智能体系统的设计。如果成功，该研究项目将为改进工艺规划系统的设计提供原则。流程规划既耗时又昂贵；当它失败时，由于重新设计，它可能导致进一步的成本和延迟。此外，在制造业（机械加工）中，工艺规划是至关重要的一步，而制造业通常占发达经济体国内生产总值的5%左右。因此，很好地设计过程计划系统是很重要的，这样他们就能成功地找到可行的——并且，在理想情况下，是好的，这里的“好”意味着相对便宜的制造公差规格——当设计是可制造的时，过程计划平均很快（如果设计是不可制造的，不要浪费过多的时间和资源）。更广泛的影响包括学生（包括那些代表性不足的群体）参与这项研究，并在明尼苏达大学开设了一门针对工程师的博弈论新课程。此外，该项目本质上是跨学科的，涉及博弈论在制造业中的新应用。

The objective of this research project is to suggest how manufacturing process planning systems should be designed. In automated manufacturing, process planning is the intermediate step that transforms a design into an ordered list of instructions for fabrication. For instance, in machining, a block of homogeneous metal (the stock or initial workpiece) must be transformed by removing material appropriately. For each step, the position of the workpiece, how it is held in place (fixturing), the specific tool to be used, the tool access direction, and the precise tool path must be specified. A computer aided process planning system takes a computer file defining the design in a standard format---the output of a computer aided design system---and finds computerized instructions for computer assisted manufacturing equipment to make the desired artifact. Increasingly, such process planning systems contain multiple artificially intelligent agents (or distributed decision makers). Process planners (and multi-agent systems in general) define particular games. The examination of such games helps one to understand how process planners and multi-agent systems work, why they can fail, and how they can be improved. Game-theoretic mechanism design analyzes how games can be designed so that the desired outcomes arise as equilibria in the games. The proposed research will first use existing results in game theory and mechanism design to study process planners and multi-agent systems. Then new results will be derived for network and distributed games and the potential disaggregation of agents will be analyzed. The goal is to establish rigorous theoretical principles, based on strategic optimization behavior, to guide the design of multi-agent process planning systems and multi-agent systems in general.If successful, this research project will provide principles to improve the design of process planning systems. Process planning can be time-consuming and expensive; when it fails, it can lead to further costs and delays due to redesign. Moreover, process planning is a crucial step in a manufacturing industry (machining) that typically represents about five percent of gross domestic product in developed economies. Hence, it is important to design process planning systems well, so that they successfully find feasible---and, ideally, good, where "good'' means relatively inexpensive to manufacture to tolerance specifications---process plans quickly on average when the design is manufacturable (and don't waste excessive amounts of time and resources if the design is not manufacturable). Broader impacts include involvement of students (including those from underrepresented groups) in this research and the initiation of a new course at the University of Minnesota on game theory for engineers. Moreover, the project is inherently interdisciplinary and involves novel applications of game theory in manufacturing.