Hybrid AI-Simulation Mesoscopic Material Design

混合人工智能仿真细观材料设计

• 批准号: 2327699
• 金额: --
• 依托单位: University of Manchester
• 依托单位国家: 英国
• 项目类别: Studentship
• 财政年份: 2019
• 资助国家: 英国
• 起止时间: 2019 至 无数据
• 项目状态: 已结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-2327699
• 关键词: Hybrid; AI; Simulation; Mesoscopic; Material

BackgroundThe project investigates colloidal suspensions of complex particles, which have the ability to cluster together and form long-range ordered structures. Such colloidal nanomaterials are ubiquitous in industrial and technological applications ranging from foods, pesticides, paints, oil recovery, lubricants, medicines, pastes, detergents, cosmetics and creams. They are especially popular as thickening agents used for tuning the viscoelastic properties of various formulations. ObjectivesThe PhD student will focus on the development and application of techniques that can automatically set up, simulate and accurately measure industrial mesoscopic colloidal structure property relationships, as well as coupling them to AI regression optimisation algorithms. The project adapts a dual hybrid AI-Simulation approach, whereby the colloidal simulations provide direct insight into the intricate relationship between the macroscopic of mesoscopic materials, and their microscopic formulation ingredients. At the same time, AI machine learning regression algorithms will automatically define and execute a series of these simulations to efficiently calculate the optimal formulation recipe for producing a colloidal mesoscopic material with the prescribed macroscopic properties.Main Research questions1. What is the phase behaviour of colloidal trimers interacting via the Mie potential?2. What are the crystal phases that are observed at equilibrium?3. What is their rheological response when an external field is applied?4. What are the parameters that most influence the formation of these structures and their rheology?5. How can the kinetics of self-assembly be optimized by AI regression optimisation algorithms?6. How can we use these algorithms to minimise structural defects in nanomaterials? ApproachThe student will employ molecular simulation and modelling, as well AI regression optimisation algorithms to study equilibrium and out-of-equilibrium properties of colloidal suspensions. Spe cifically:1. Molecular Dynamics simulations to calculate the phase diagram of Mie-like colloidal trimers (on-going - liquid/vapour region already identified)2. Non-equilibrium MD simulations to study the effect of an external field (e.g. shear) on the stability of the structures3. AI regression optimisation algorithmsNovel contentThe project aims at bridging colloidal science with computer science by providing a computational apparatus able to identify the most suitable kinetic paths for the synthesis of nanomaterials. Once a material property of interest is identified, our vision is to find the most suitable set of particles, inter-particle interactions and physico-chemical properties, that would provide a nanomaterial with specific characteristics, performance and response. To this end, we need algorithms able to manage substantial amount of data and to identify the most suitable parameters leading to the target property.

背景该项目研究复杂颗粒的胶体悬浮液，这些颗粒具有聚集在一起并形成远程有序结构的能力。这种胶体纳米材料在工业和技术应用中无处不在，从食品，农药，油漆，石油回收，润滑剂，药品，糊剂，洗涤剂，化妆品和面霜。它们作为增稠剂特别受欢迎，用于调节各种制剂的粘弹性能。博士生将专注于开发和应用技术，可以自动设置，模拟和准确测量工业介观胶体结构属性关系，以及将它们与AI回归优化算法相结合。该项目采用双混合AI模拟方法，胶体模拟可直接洞察宏观介观材料及其微观配方成分之间的复杂关系。与此同时，AI机器学习回归算法将自动定义和执行一系列模拟，以有效地计算出生产具有规定宏观特性的胶体介观材料的最佳配方。主要研究问题1.胶体三聚体通过米氏势相互作用的相行为是什么？2.在平衡状态下观察到的晶相是什么？3.当外加磁场时，它们的流变学响应是什么？4.哪些参数对这些结构的形成及其流变性影响最大？5.如何通过AI回归优化算法优化自组装的动力学？6.我们如何使用这些算法来最大限度地减少纳米材料中的结构缺陷？学生将采用分子模拟和建模，以及AI回归优化算法来研究胶体悬浮液的平衡和平衡特性。特殊类别：1.分子动力学模拟计算米氏样胶体三聚体的相图（正在进行的液体/蒸汽区域已经确定）2。非平衡MD模拟研究外场（例如剪切）对结构稳定性的影响3. AI回归优化算法Novel contentThe project aims at bridging colloidal science with computer science by providing a computational apparatus capable to identify the most suitable kinetic paths for the synthesis of nanomaterials.一旦确定了感兴趣的材料特性，我们的愿景是找到最合适的颗粒集、颗粒间相互作用和物理化学特性，这将提供具有特定特征、性能和响应的纳米材料。为此，我们需要能够管理大量数据并确定导致目标属性的最合适参数的算法。