Contextualization of design tasks for automation of the embodiment design and dimensioning phase with Artificial Intelligence

由于高工资国家的高竞争压力，越来越需要将产品开发过程中的日常活动自动化。与此同时，快速的技术进步使得人工智能（AI）研究领域的最先进工具在工业上的使用越来越多。通常，这些算法作为黑盒集成到产品开发过程中，而产品开发人员对算法没有详细的了解。因此，人工智能解决方案的行为和结果只能很好地解释，导致错误的预测。因此，产品开发中的人工智能方法应该与产品开发人员的思维方式更紧密地结合起来，以便可以直接测试生成的结果和实现它们的过程。在申请人的初步工作中，可以证明，在产品开发中使用AI方法强化学习的好处为给定的任务提供了巨大的潜力。在强化学习中，代理执行产品开发人员的部分设计任务。因此，该项目的目标是通过强化学习来提高自动改进产品的可理解性。特别重点将放在产品开发中的虚拟几何生成，而不会对组件设计造成不利影响。将回答三个研究问题，涵盖不同的方面。首先，重点是状态描述的扩展，应使其更接近产品开发的需要。之后，应该制定一个新的开发策略，该策略以有经验的产品开发人员为导向。在最后一个问题的帮助下，分析了在发展过程中通过内在奖励映射隐含关系的可能性。纤维增强塑料（FRP）组件作为演示组件，因为它们的设计是基于复杂的相互关系。然而，在玻璃钢构件的设计中，申请人可以利用广泛的专业知识，这构成了研究项目的基础。通过自动化工具来改进自动化几何生成并提高所执行操作的可理解性，改进产品开发过程的潜力巨大。一方面，在使用AI方法方面经验较少的产品开发人员可以使用研究的工具，更好地了解产品形状的变化。另一方面，标准或常规活动可以自动化，从而节省产品设计的时间和成本。

Due to the high competitive pressure in high-wage countries, there is an increasing need to automate routine activities from the product development process. At the same time, rapid technological advancements are enabling the increasing industrial use of state-of-the-art tools from the research field of artificial intelligence (AI). Often, these are integrated into the product development process as black boxes without the product developers having a detailed understanding of the algorithms. Consequently, the behavior and results of the AI solution can only be poorly interpreted, resulting in erroneous predictions. Therefore, the AI methods in product development should be more closely aligned with the mindset of the product developers, so that the generated results and the journey to them can be directly tested. In preliminary work of the applicant it could be shown that the use and benefit of the AI method Reinforcement Learning in product development offers great potential for the given task. In Reinforcement Learning, an agent performs part of the design tasks of product developers. Therefore, the goal of the project is to increase the understandability of automatically improved products by means of reinforcing learning. Special focus will be put on the virtual geometry generation in product development without disadvantages for the component design. Three research questions will be answered, which cover different aspects. First, the focus is on the extension of the state description, which should be brought closer to the needs of the product development. Afterwards, a new exploration strategy should be developed, which is oriented towards the exploration of experienced product developers. With the help of the last question the extent to which it is possible to map implicit relationships in the development process via intrinsic rewards is analyzed. Fiber-reinforced plastic (FRP) components serve as demonstrator components because their design is based on complex interrelationships. However, in the design of FRP components, extensive expertise can be drawn on at the applicant, which forms the basis for the research project. With an automation tool for improved automated geometry generation and increased comprehensibility of the performed actions, there is enormous potential to improve the product development process. On the one hand, product developers with less experience in using AI methods can use the researched tool and better understand the changes in the product shape. On the other hand, standard or routine activities can be automated and thus lead to time and cost savings in the design of products.