Daring More Intelligence – Design Assistants in Mechanics and Dynamics (SPP 2353) - Coordination Proposal

为了尊重生态和社会的责任和挑战，以及考虑到更严格和更复杂的法规，未来的系统设计必须变得越来越多学科。今天在力学和动力学中使用的基于计算机的支持，主要局限于系统分析，已经不够了。即使在高级设计工作流程中，通常也只能手动进行大规模的仿真驱动参数研究和检查，以根据经验和专家知识迭代地更改候选设计。这个过程不仅非常耗时，而且通常基于主观而不是形式化的数学目标。既定优先计划中的研究应旨在开发设计辅助系统，结合优化，人工智能和动力学/力学的方法，以协助和部分自动化工程系统的跨学科设计。这可能不仅会导致设计在形式化标准方面实际上是最优的，而且这种设计助手可能会为设计工程师提供人工直觉，以补充他们自己的专业知识。这样，现在只在后期设计阶段考虑的标准可以在早期考虑，以更基本的方式改进结果系统，而不是今天遵循既定设计范例的增量改进。实现对动力学和力学有实际影响的设计辅助系统的关键是通过集成人工智能和机器学习的方法，在系统分析、优化和设计方面超越目前的技术水平。例如，机器学习方法可以作为有价值的工具来推断代理模型和响应面，这些模型和响应面可以用于管理大规模分析的计算工作，作为依赖多标准优化的自动化设计过程的一部分。来自人工智能的方法甚至可以直接做出某些创造性的设计决策。然而，由于机器学习和人工智能最近主要在远离动态系统设计的领域蓬勃发展，迄今为止，尚不清楚哪种方法最适合，特别是如何将它们与系统分析和优化相结合以实现更好的设计。理想情况下，设计辅助组件应该高度灵活，具有易于访问的界面，以便它们可以模块化地组合起来，以建立越来越全面的辅助设计过程，并作为第二个资助期继续研究的基础。

To respect ecological and societal responsibilities and challenges as well as to account for stricter and more complex regulations, future systems design has to become increasingly multidisciplinary. Computer-based support as employed today in mechanics and dynamics, mostly limited to system analysis only, is not sufficient anymore. Even in advanced design workflows, usually large-scale, simulation-driven parameter studies are conducted and inspected only manually to iteratively alter a candidate design based on experience and expert knowledge. This process is not only very time consuming but also typically based on subjective rather than on formalised mathematical objectives.The research in the established Priority Programme shall aim at the development of design assistance systems combining methods from optimisation, artificial intelligence, and dynamics/mechanics to assist in and partially automate the interdisciplinary design of engineering systems. This may not only result in designs that are actually optimal with respect to formalized criteria, but such design assistants may equip design engineers with an artificial intuition supplementing their own specialized expertise. This way, criteria nowadays only considered in later design stages may be taken into account early on, improving resulting systems in a much more fundamental manner than today’s incremental improvements following established design paradigms.The key to realizing design assistant systems of practical impact in dynamics and mechanics is to go beyond the state of the art in system analysis, optimisation, and design by integrating methods from artificial intelligence and machine learning. For instance, machine learning methods can be valuable tools to infer surrogate models and response surfaces that can be used to make manageable the calculation effort for large-scale analysis as part of automated design procedures relying on multicriteria optimisation. Methods from artificial intelligence may even directly make certain creative design decisions. However, since machine learning and artificial intelligence have recently thrived mostly in fields far from the design of dynamic systems, it is, as yet, rather unclear which methods will be best-suited and, in particular, how they can be combined with system analysis and optimisation to achieve better designs. Ideally, the design assistant components should be highly flexible with easily accessible interfaces so that they can be combined modularly to build up increasingly holistic, assisted design procedures, and to serve as a foundation for continued research in the second funding period.