Design and optimisation of metasurface materials using AI/machine learning algorithms

现代技术的进步增加了对多功能组件的需求。在射频（RF）世界中，无线通信需要高效、可调谐、廉价和更小封装的天线，其可以被放入越来越小的设备中。此外，随着我们进入下一代无线通信技术，这种天线的设计变得更加复杂。由于复杂性的增加，需要使用人工智能和机器学习工具等自动化技术来处理这些即将到来的新技术所涉及的计算量和复杂性。由于复杂性和某些应用，天线的设计者必须在尺寸、功能和制造成本之间进行折衷。偏铁氧体材料已成为解决大多数这些问题的有吸引力的选择，并且已成为从天线到吸收器以及复杂空间或频域器件构造的各种应用的非常有用的工具[1]。超材料用于天线设计的主要好处是，它们可能实现宽角度扫描和出色的波束性能、电子控制指向和极化、低功耗的设计，并且可以是扁平的、重量轻的和尺寸小的[2]，回答了新技术设备需求所面临的许多主要问题。这些超材料是工程材料，也称为左手（LH）材料或反向波（BW）介质或负折射率材料（NIM）或双负（DNG）介质，[2]它们表现出在自然发生的材料中看不到的有趣特性。由于超材料设计的复杂性增加，这些材料的设计者使用机器学习技术并不罕见。超材料由夹在两个金属层之间的衬底组成，其中一个通常用单元阵列图案化。这就是机器学习技术主要集中在设计和优化上的地方。这项研究的意义在于，它将开发一种机器学习辅助方法，帮助快速制造用于测试的样品。另外，设想在对材料性质的几何优化中改进先前超材料设计的性能。所提出的方法是非常新颖的，因为它试图全面开发一个平台的工具与最终用户的应用程序。研究目的和目标：研究机器学习技术在超材料天线设计中的应用。研究使用黏模算法（SMA）进行超材料设计和优化，因为这是一种相对较新的技术，据研究人员所知，目前对其超材料优化的使用研究很少。为了改进由Yihan Ma小组的前博士建立的神经网络框架，通过实施混合的元启发式算法来将超材料设计升级到更大的规模。目的是推动知识的生产的超材料设计的自动化整体工具，考虑到多个目标参数。

Modern advances in technology have increased the demand for multifunctional components across the spectrum. In the Radio frequency (RF) world, wireless communications require efficient, tunable, inexpensive and smaller package antenna that can be put into smaller and smaller devices. Additionally, as we move into the next generation of wireless communication technology the designs of such antenna are becoming more complex. Due to this increased complexity automated techniques such as, AI and machine learning tools need to be used in order to deal with the amount of computation and complexity involved in these new upcoming technologies. Due to the complexity and certain applications designers of antennas have to compromise either size, functionality and fabrication cost. Metaferrite materials have become an attractive option for addressing most of these issues and have become a very useful tool for various applications from antennas to absorbers and the construction of complex spatial or frequency domain devices [1]. The main benefits of a metamaterial for antenna design is that they potentially enable the design of wide angle scanning and excellent beam performance, electronically controlled pointing and polarisation, low power consumption and can be flat, lightweight and small in size[2], answering many of the main issues that are facing the demand for new technological devices. These metamaterials are engineered materials also called left-handed (LH) materials or backward wave (BW) media or negative index materials (NIM) or double negative (DNG) media, [2] that exhibit interesting properties not otherwise seen in naturally occurring materials. Due to the increase in complexity of metamaterial designs it is not uncommon for machine learning techniques to be used by designers of these materials. Metamaterials consists of a substrate sandwiched between two metal layers where one is usually patterned with a unit cell array. This is where the machine learning techniques are mainly focused on designing and optimising. The significance of the research is that it will develop a machine learning assisted method that will assist with the rapid fabrication of samples for testing. Additionally, improving upon the performance of previous metamaterial designs in the geometric optimisation to material properties is envisaged. The proposed approach is highly novel as it seeks to holistically develop a platform of tools with end user applications. Research Aims and Objectives:To investigate the use of machine learning techniques in their functionality for metamaterial antenna design. To investigate the use of Slime mould algorithm (SMA) for metamaterial design and optimisation as this is a relatively new technique and has very little research currently into its use of metamaterial optimisation, to the researchers knowledge. To improve upon previous Neural Network frameworks built by a previous PhD in the group Yihan Ma, by implementing a mix of metaheuristic algorithms to upscale metamaterial designs to a much larger scale. Aim to push knowledge forward for the production of an automated holistic tool for metamaterial design, taking into account multiple objective parameters.