Machine Learning Tools for Biofuel Creation and Purification using Ionic Fluids

迈阿密大学的Orlando Acevedo获得了化学系化学理论、模型和计算方法项目的奖励，以开发有助于设计生物质衍生燃料的人工智能工具。生物质作为一种可再生能源具有巨大的潜力。例如，据估计，到2030年，20%的运输燃料和25%的化学品将由生物质生产。尽管年产量约为1700亿吨，但用于各种用途的生物质不超过5%。从植物材料生产的增值化学品和高质量燃料产品的潜在市场是巨大的。然而，目前，钻探和提炼矿物柴油比种植、提取和纯化生物柴油更具成本效益。被称为离子液体（ILs）和深共晶溶剂（DESs）的环境友好型溶剂有潜力成为创造经济上可行的生物燃料解决方案的重要组成部分。Acevedo博士将开发和应用基于人工智能的软件（即机器学习）来设计新的ILs和DESs，有效地将生物质衍生的构建块转化为液体燃料，并从柴油和生物柴油燃料中提取杂质。该项目将通过强调使用具有成本效益和可回收的离子流体，帮助加速从化石燃料向生物质能源的过渡。Acevedo博士开发的计算工具将免费提供给科学界。此外，多个社区外展活动将通过位于劳德代尔堡和佛罗里达州迈阿密的科学博物馆的演示，研究网络社交聚会和K-12年级的插图科学诗歌比赛来突出人工智能在化学中的应用。在第一个主要研究重点中，Acevedo博士将研究使用n -杂环碳烯（NHC）催化将高度通用的生物质衍生的5-羟甲基糠醛（HMF）化合物转化为(a) 5,5 ' -二（羟甲基）呋喃（DHMF），以及(b) 2,5-二（羟甲基）呋喃（BHMF）通过Cannizzaro反应。这两种方法都是高效、温和、环保的。为了预测溶剂衍生的速率和选择性在这些HMF转换上的增强，将开发一种新的混合机器学习，量子力学和分子力学（ML/QM/MM）方法，该方法将提供从头算QM计算的准确性，并显着提高计算速度。在第二个主要研究重点中，Acevedo博士将探索使用ILs和DESs从柴油中提取(a)二苯并噻吩和其他芳香硫化合物，以及(b)从生物柴油中提取甘油，生物柴油是植物甘油三酯油酯交换过程中产生的废物。il和DESs都具有非常适合从运输燃料中提取杂质的特性：不溶于油，无毒，对环境无害。为了研究这些燃料基杂质与ILs/DESs之间有利的分子间相互作用的起源，我们的替代方法将使用具有代表性的从头算分子动力学数据训练人工神经网络（ANN），并将新优化的ANN电位与蒙特卡罗模拟相结合。开发的基于机器学习的方法将广泛适用于化学研究的其他领域，作为该项目一部分生成的所有软件/产品将随时可供科学界使用。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Orlando Acevedo of the University of Miami is supported by an award from the Chemical Theory, Models and Computational Methods program in the Division of Chemistry to develop artificial intelligence tools that will aid in the design of biomass-derived fuels. Biomass has significant potential as a renewable source of energy. For example, it has been estimated that by 2030, 20% of transportation fuel and 25% of chemicals will be produced from biomass. Despite an annual production of around 170 billion metric tons, no more than 5% of this produced biomass is used for diverse purposes. The potential market for value-added chemicals and high-quality fuel products produced from plant materials is immense. However, at present it is more cost-effective to drill and refine mineral diesel oil than grow, extract, and purify biodiesel. Environmentally friendly solvents known as ionic liquids (ILs) and deep eutectic solvents (DESs) have the potential to serve as a significant part of the solution in the creation of economically viable biofuels. Dr. Acevedo will develop and apply artificial intelligence-based software (i.e., machine learning) to design new ILs and DESs that efficiently convert biomass-derived building blocks into liquid fuel and extract impurities from diesel and biodiesel fuels. This project will help accelerate the transition from fossil fuels to biomass-derived energy by emphasizing the use of cost-effective and recyclable ionic fluids. The computational tools developed by Dr. Acevedo will be freely available to the scientific community. In addition, multiple community outreach events will highlight the use of artificial intelligence in chemistry through demonstrations at science museums located in Ft. Lauderdale and Miami, Florida, research networking social gatherings, and an illustrated science poem contest for K-12 grades.In the first major research thrust, Dr. Acevedo will study the conversion of the highly versatile biomass-derived 5-hydroxymethylfurfural (HMF) compound into (a) 5,5’-di(hydroxymethyl)furoin (DHMF) using N-heterocyclic carbene (NHC) catalysis, and (b) 2,5-bis(hydroxymethyl)furan (BHMF) via a Cannizzaro reaction. Both methods are efficient, mild, and eco-friendly. To predict solvent-derived rate and selectivity enhancements upon these HMF transformations, a new mixed machine learning, quantum mechanics, and molecular mechanics (ML/QM/MM) methodology will be developed here that provides the accuracy of ab initio QM calculations with dramatically increased calculation speeds. In the second major research thrust, Dr. Acevedo will explore the use of ILs and DESs to extract (a) dibenzothiophene and other aromatic sulfur-compounds from diesel, and (b) glycerol from biodiesel, a waste product developed during the transesterification of plant triglyceride oil to biofuel. Both ILs and DESs have properties ideally suited for impurity extraction from transportation fuel: insoluble in oil, non-toxic, and environmentally benign. To study the origin of the favorable intermolecular interactions between these fuel-based impurities and ILs/DESs, our alternative method will train artificial neural networks (ANNs) using representative ab initio molecular dynamics data and couple the newly optimized ANN potentials to Monte Carlo simulations. Machine learning based methodology developed will be broadly applicable to other areas of chemical research and all software/products generated as part of this project will be readily available to the scientific community.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.