Artificial Intelligence and Machine Learning to Crack Crystal Growth

**Crystal**_Grower_Ltd是一家科学软件公司，从曼彻斯特大学剥离出来，生产模拟晶体生长的软件。每个人都知道自然界中的水晶、钻石、盐、雪花，但也知道软体动物的贝壳，甚至鲨鱼眼睛中的聚焦成分。人们可能不太意识到的是，晶体是我们日常生活中使用的几乎每一种功能材料的活性成分：药物，如阿司匹林和扑热息痛；电脑中的电子元件；用于储存氢以实现氢经济的材料，或用于从大气中消除二氧化碳以改善环境的材料。所有这些晶体都是通过在实验室中种植来生产的，代表着一个数十亿英镑的全球和英国制造市场。然而，为了精确地控制这些晶体的功能，关键是首先在纳米/分子尺度上控制它们的生长。这确保了所生产的晶体具有所需的质量和功能特性。**Crystal**_Grower_Ltd生产的软件通过模拟晶体如何生长，以及这种生长如何随着实验室条件的变化而变化，从而产生更好的生产结果，帮助公司准确地实现这一点。这种模拟的大部分依赖于模拟晶体与实验生产的晶体的比较，目前，这必须在大量人工干预下完成。目前这个项目的目的是开发基于图像的分类器，使我们的客户能够以更智能、更自动化的方式处理大量数据。这将大大减少处理模拟的时间，减少人为错误，特别是在规模上，并最终实现更快的洞察。这将允许我们的客户花更多的时间专注于开发他们的产品，而不是数据分析。图像分类技术还可以使高级模拟转向技术的未来发展成为可能，减少解决问题的时间，同时消耗更少的计算资源。

**Crystal**_Grower_ Ltd is a scientific software company, spun out from The University of Manchester, that produces software to model how crystals grow. Everyone is aware of crystals in nature, diamonds, salt, snowflakes but also the shells of molluscs and even the focussing components in shark eyes. What people may be less aware of is that crystals are the active components of almost every functional material that we use in our everyday life: pharmaceuticals, such as aspirin and paracetamol; the electronic elements inside your computer; the materials used for storing hydrogen for a hydrogen economy or removing carbon dioxide from the atmosphere for an improved environment. All these crystals are produced by growing them in the laboratory and represent a multi-billion pound global and UK manufacturing market. However, in order to control precisely the functionality of these crystals it is critical first to control their growth at the nano/molecular scale. This ensures that the crystals are produced with the required quality and functional properties. The software produced by **Crystal**_Grower_ Ltd helps companies do exactly that by simulating how crystals grow and how this growth changes upon altering laboratory conditions thereby leading to better production outcomes.Much of this simulation relies on comparison of simulated crystals with those produced experimentally which, at present, must be done with significant human intervention. The purpose of this current project is to develop image-based classifiers that would enable our clients to process substantial volumes of data in a more intelligent, automated way. It would take significantly less time to process simulations, reduce human error, particularly at scale, and ultimately enable more rapid insights. This would allow our clients to spend more time focused on developing their products, instead of data analytics. Image classification technology can also enable the future development of advanced simulation steering techniques, reducing the time to solution, whilst consuming fewer computational resources.