Physics-Informed (and -informative) Reinforcement Learning and Bio-Inspired Design of a Smart Morphing Flapping Wing for Dual Aerial/Aquatic Propulsion and Maneuvering

双空中/水上（DA 2）车辆，允许快速空中旅行穿插与水下勘探被设想为许多海洋任务的最佳候选人，如水质采样，搜索和救援，以及海洋领土渗透。开发这样一个系统可以大大提高加拿大在未来海洋或内陆水勘探和开发方面的全球竞争力。关键的障碍之一是设计一种能够在空气和水中都能最佳工作的推进系统。虽然使用传统的推进器来实现这一使命是具有挑战性的，但大自然已经提供了自己的解决方案，即在海鸟中看到的变形扑翼/箔。为了为可行的DA 2飞行器设计创建和控制变形扑翼致动器，在技术上，除了1）了解扑翼周围的涡流外，还需要2）在不确定环境中快速解决大量变量问题的学习策略，3）多功能和鲁棒智能变形结构的设计和制造工具包。因此，我们提出两项关键任务，以满足上述要求。第一个任务是为扑翼翼开发一个物理信息（和信息）强化学习（Phi 2 RL）框架，该框架能够使用稀疏分布的压力传感器来感测近体尾流，并在湍流/阵风环境中快速执行轨迹规划。Phi 2 RL包括1）一个系统动力学模型，其中包含物理嵌入式结构降阶模型以及数据辅助组件的学习灵活性，以补偿未建模的动态，以及2）实用的强化学习和迁移学习算法，以探索和利用实时的最佳力（升力和推力）分布生成。第二个任务是使用离散的细胞超材料和炭黑-聚二甲基硅氧烷（CB-PDMS）设计一个智能变形扑翼致动器与软压力传感器阵列的皮肤，能够自适应地交替翼的形状，面积和扑翼运动。拟议的研究将为未来DA 2车辆的可行推进解决方案提供巨大的见解。此外，Phi 2 RL将是一个强大的人工智能（AI）增强的流体实验解决方案，可以推广到更广泛和更大规模的各种流体问题，例如流线和海崖体的减阻。此外，包括2名博士在内的HQP，3个硕士和1个本科生将在这个多学科项目上合作。他们将学习非定常空气动力学/流体动力学，降阶建模，实验测试，人工智能算法，稀疏传感和数字制造方面的知识，并将获得强大的沟通和团队合作技能，使他们成为成功的科学家和工程师，为加拿大学术界和工业界做出贡献。

Dual aerial/aquatic (DA2) vehicles that allow fast aerial travel interspersed with underwater exploration are envisioned as the best candidate for many oceanic missions, such as water quality sampling, search and rescue, and ocean territory infiltration. Developing such a system can significantly advance Canadian global competitiveness in the future ocean or inland water exploration and exploitation. One of the key obstacles is to design a propulsion system that can work optimally in both air and water. While it is challenging to use traditional propulsor to achieve this mission, nature has provided its own solution of a morphing flapping wing/foil, seen in seabirds. To create and control a morphing flapping actuator for viable DA2 vehicle designs, technologically, apart from 1) an understanding of the vortical flow around flapping foils, it also requires 2) a learning strategy to quickly solve problems with a large number of variables in an uncertain environment, 3) a design and fabrication toolkit for multi-functional and robust smart morphing structures. Therefore, we propose two key tasks to address the aforementioned requirements. The first task is to develop a physics-informed (and -informative) reinforcement learning (Phi2RL) framework for the flapping foil capable of using sparsely distributed pressure sensors to sense the near-body wake and swiftly performing trajectory planning in a turbulent/gusty environment. The Phi2RL includes 1) a system dynamics model that contains both the physics-embedded-as-structure reduced-order model as well as the learning flexibility of the data-assisted component to compensate the unmodeled dynamics, and 2) practical reinforcement learning and transfer learning algorithms to explore and exploit the optimal force (lift and thrust) profile generation in real-time. The second task is to use discrete cellular metamaterial and carbon-black-polydimethylsiloxane (CB-PDMS) to design a smart morphing flapping actuator with a skin of soft pressure sensor arrays that is capable of adaptively alternating wing shapes, areas, and flapping kinematics. The proposed research will provide tremendous insights into a viable propulsion solution for a DA2 vehicle in the future. Additionally, the Phi2RL will be a powerful artificial intelligence (AI)-enhanced fluid experiment solution that can be generalized to address a variety of fluid problems at a broader scope and greater scale, such as drag reduction of streamline and bluff bodies. Furthermore, HQPs, including 2 Ph.D., 3 MSc, and 1 undergraduate, will work collaboratively on this multi-disciplinary project. They will learn knowledge on unsteady aerodynamics/hydrodynamics, reduced-order modeling, experimental testing, AI algorithms, sparse sensing, and digital fabrications and will acquire strong communication and teamwork skills, which transfers them to be successful scientists and engineers contributing to the Canadian academia and industry.