Machine-Learning-Driven Synthesis of the Next Generation Carbon Dots with Tunable Fluorescence/Band-Gap

无机量子点表现出有趣的荧光和半导体特性，目前正在探索生物成像剂或太阳能电池板/光催化过程以及电子显示器的生物应用。然而，它们是基于有毒金属（例如Cd， Pb），合成过程复杂，并且在正常大气条件下储存时表现出低稳定性。生物质衍生的碳点（CDs）已成为无机量子点的有前途和可持续的候选材料。然而，尽管在新型和可持续碳点的开发方面取得了很大进展，但在其带隙工程以适应可见光吸附的不同要求，以及生产具有可调荧光特性和高量子产率的碳点方面，该领域仍然存在巨大的挑战。为了解决这些问题并生产出具有可调荧光和带隙的碳点，我们建议在碳点的制备过程中关联反应参数，以探索结构-性能关系和潜在的应用前景。关于cd的合成和性能有大量的实验数据（已经存在于Titirici的实验室和更广泛的文献中），因此我们将应用机器学习（ML）筛选高性能材料，首先选择合成cd的不同条件，然后权衡合成变量的重要性并优化所需的性能。在这个项目中，我们将展示基于机器学习的技术如何为成功预测、优化和加速cd的合成过程和特性提供见解，从而导致新兴的应用。建立了基于不同生物质前体的水热合成CDs的回归ML模型，揭示了各种合成参数与实验结果之间的关系，并增强了荧光量子产率（QY）和可调谐带隙等工艺相关特性。合成将首先分批进行，然后最终转移到包含先进过程分析技术（PAT）的流动反应器中，该反应器将用于原位表征碳点，并在ML算法的监督下使用计算机控制调整其性质。

Inorganic quantum dots exhibit interesting fluorescent and semiconductor properties and are currently explored in biological applications as bioimaging agents or in solar panels/photocatalytic processes, as well as in electronics displays. However, they are based on toxic metals (e.g. Cd, Pb), complicated synthetic processes and exhibit low stability when stored under normal atmospheric conditions. Biomass-derived carbon dots (CDs) have emerged as promising and sustainable candidates to inorganic quantum dots. Yet there remains grand challenges in the field, despite great progress made in the development of novel and sustainable carbon dots, in the engineering of their band gap to fit different requirements for visible light adsorption, as well as producing them with tunable fluorescent properties and high quantum yields. To address these challenges and produce carbon dots with tunable fluorescence and band gap, we propose to correlate reaction parameters in the preparation process of carbon dots to explore structure-properties relationship and potential applications. There is a plethora of experimental data (already existing in Titirici's lab and wider literature) on the synthesis and properties of CDs, so we will apply machine learning (ML) for screening of high-performance materials, first to select diverse conditions under which to synthesise the CDs, and then to weight the importance of synthesis variables and to optimise the desired properties. In this project we will demonstrate how ML-based techniques can offer insights into the successful prediction, optimisation, and acceleration of CDs' synthesis processes and properties leading to emerging applications. A regression ML model on hydrothermally-synthesised CDs from various biomass precursors will be established to reveal the relationship between various synthesis parameters and experimental outcomes as well as enhancing the process-related properties such as the fluorescent quantum yield (QY) and tunable bandgap. The synthesis will first be done in batch, and then ultimately transferred to a flow reactor incorporating advanced process analytical technology (PAT), which will be used to characterise the carbon dots in-situ and tune their properties using computer control, supervised by the ML algorithm.