SBIR Phase I: Implementing AL-enhanced Machine-Learning for Advanced Electrochemical Manufacturing

这个小企业创新研究（SBIR）第一阶段项目的更广泛的影响/商业潜力是证明机器学习（ML）指导的实验活动的可行性，这些活动可以预测，评估和优化具有小实验数据集的电有机转化。化工行业的公司已经将电化学确定为实施更可持续和节能的制造工艺的一个有前途的途径。然而，新工艺开发所需的巨大成本和努力阻碍了电化学技术的实施。ML预测算法可以成为加速开发和优化更可持续的化学过程的强大工具，但需要大量的实验数据来训练模型。这些大型数据集通常不可用且获取成本高昂，这极大地限制了ML在化学工业中的使用。该项目将通过减少50%的实验来开发新的、更可持续的化学生产路线，从而推动未来的制造业，最终解锁社会应用中新分子、药物和材料的制造。此外，通过减少所需的实验次数，该技术将显著降低行业的排放和资源消耗。拟议的项目引入了一个能够指导实验活动和数据收集的ML平台，以便能够使用尽可能小的数据集准确预测反应行为。该方法依赖于化学工程和机器学习知识的结合，以克服每个领域中存在的优化限制。它将使用对甲氧基甲苯的电氧化作为模型反应进行验证，并将阐明ML预测模型的基本局限性和优势，这些模型通过小数据集捕获物理系统的复杂性。该奖项反映了NSF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。

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase I project is to demonstrate feasibility of machine-learning (ML)-guided experimental campaigns that predict, assess, and optimize electroorganic transformations with small experimental datasets. Companies across the chemical industry have pinpointed electrochemistry as a promising avenue for the implementation of more sustainable and energy-efficient manufacturing processes. However, the large cost and effort required in new process development hinders the implementation of electrochemical technologies. ML predictive algorithms can be a powerful tool to accelerate the development and optimization of more sustainable chemical processes, but repeatedly require large amounts of experimental data to train the models. These large datasets are often unavailable and expensive to obtain, which significantly limits the use of ML in the chemical industry. The project will advance future manufacturing by enabling the development of new and more sustainable chemical production routes using 50% less experiments, ultimately unlocking the manufacture of new molecules, medicines, and materials in societal applications. Moreover, by reducing the number of experiments required, the technology will significantly lower emissions and resource consumption in the industry.The proposed project introduces a ML platform capable of guiding experimental campaigns and data collection to enable accurate predictions of reaction behavior with the smallest possible datasets. The approach relies on the combination of chemical engineering and ML knowledge to overcome the optimization limitations found within each field. It will be validated using the electrooxidation of p-methoxytoluene as a model reaction and will elucidate the fundamental limitations and strengths of ML predictive models capturing the complexity of physical systems with small datasets.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.