Advanced Composites Manufacturing and Repair Using Integrated Distributed Actuation and Dynamic Network Control

由于其高刚度和重量轻，复合材料如碳纤维增强聚合物在汽车、航空航天、海洋运输部门和可再生能源系统（如大型风力涡轮机）中的应用越来越多。制造许多复合材料的一个重要步骤是高温固化，这涉及到在一个被称为高压灭菌器的密封压力容器中通过精确的温度序列循环零件。这个项目的重点是将这些成品的单个零件连接起来，形成一个完整的产品。虽然机械紧固件或低温粘合剂的性能不如高温复合粘合剂，但组装系统太大，无法作为一个单元放置在高压灭菌器中。该项目考虑使用分布在整个关节的嵌入式加热器来执行必要的热循环。为了不损害结构的完整性，这些加热器与复合结构的其余部分使用相同的碳纤维材料制成。在初步测试中，这种方法产生的接头在质量上与蒸压零件相当；该项目面临的挑战是在大型复杂结构（如飞机机翼）上实现这样的结果。这将通过重新利用最初为动态系统网络（如移动机器人）开发的先进控制技术来实现。同样的方法可以用于修复损坏或疲劳的复合系统，经济和可持续地延长昂贵的基础设施的使用寿命。即使是最先进的技术，如果没有训练有素的劳动力来使用，也是毫无用处的。该项目将为本科生和研究生提供先进制造和控制方面的培训和教育，其成果将用于向高中生推广和招聘。一个由多个分布式加热器组成的网络，可以解释热损失的变化，例如，由于不均匀的亚结构，并在粘合线上实现均匀的温度，以实现粘合剂的一致固化。由于材料特性（例如，复合材料的厚度）和边界条件（例如，由于子结构的存在）的差异，网络中每个分布式加热器系统的热动力学（输入电压到局部温度）的差异引起了控制挑战。目前网络多智能体系统的迭代方法大多适用于同质智能体动态。当不同智能体的动力学是一般线性系统时，证明异构、网络化多智能体系统的迭代学习收敛性是具有挑战性的。设计了一种新的基于逆的迭代控制方法，该方法可以校正每个智能体的动态特性，从而达到网络指定的期望输出。该研究的智力价值在于开发了量化可接受的建模不确定性的条件，以确保在存在异质性的情况下迭代方法的收敛。因此，该研究将推动网络多智能体系统迭代学习理论的发展，并将其应用于制造业等重要技术问题。

Because of their high stiffness and light weight, composites such as carbon-fiber reinforced polymers are finding increased use in the automotive, aerospace, and marine transportation sectors and in renewable energy systems such as large wind turbines. An important step in the manufacture of many composites is high-temperature curing, which involves cycling the part through a precise temperature sequence in a sealed pressure vessel called an autoclave. This project focuses on the joining of these finished individual parts to form a complete product. While mechanical fasteners or low-temperature adhesives have inferior performance to high-temperature composite bonds, the assembled system is far too large to place in an autoclave as a unit. This project considers the use of embedded heaters distributed throughout the joint to execute the necessary thermal cycle. So as not to compromise the integrity of the structure, these heaters are made using the same carbon-fiber material as the rest of the composite structure. In preliminary tests this approach has produced joints that equal autoclaved parts in quality; the challenge addressed by this project is achieving such a result over a large, complex structure, such as an airplane wing. This will be done by repurposing advanced control techniques originally developed for networks of dynamic systems, such as mobile robots. The same approach may be used to repair damaged or fatigued composite systems, economically and sustainably extending the useful life of expensive infrastructure. The most advanced technology is useless without a trained workforce capable of employing it. This project will provide training and education to undergraduate and graduate students in advanced manufacturing and controls, and the results will be used in outreach and recruitment to high-school students.A network of multiple distributed heaters that can account for variations in heat loss, e.g., due to uneven sub-structures, and enable uniform temperatures at the bondline for consistent curing of the adhesive are planned. The control challenge arises from differences in the thermal dynamics of each of the distributed heater system (input voltage to local temperature) in the network due to differences in material properties (e.g., thickness of composites) and boundary conditions (e.g., due to the presence of substructures). Current iterative approaches for networked multi-agent systems are mostly applicable to homogenous agent dynamics. It is challenging to prove convergence of iterative learning for heterogeneous, networked multi-agent systems when the dynamics of the different agents are general linear systems. A novel inversion-based iterative control is planned that can correct for the dynamics of each agent and, thereby, can achieve the desired output specified through the network. The intellectual merit of the research is to develop conditions that quantify the acceptable modeling uncertainty for ensuring convergence of iterative approaches in the presence of heterogeneity. Thus, the research will advance the state-of-the-art in iterative-learning theory for networked multi-agent systems and its applicability to important technological problems such as manufacturing.