Designing bioinspired green nanosilicas using statistical and machine learning approaches
Designing bioinspired green nanosilicas using statistical and machine learning approaches
复制标题
DOI：
10.1039/d0me00167h
复制
发表时间：
2021-04-01
影响因子：
3.6
通讯作者：
Patwardhan, Siddharth, V
中科院分区：
文献类型：
--
作者：
Dewulf, Luc;Chiacchia, Mauro;Patwardhan, Siddharth, V
二氧化硅的体外生物启发合成受到体内生物硅化的启发，是高价值多孔二氧化硅传统生产的可持续替代方案。反应时间短，室温温和的反应条件及其使用的良性前体使其成为一种具有巨大工业潜力的生态友好，经济和可扩展的路线。然而，缺乏对生物启发二氧化硅的关键工艺参数和材料属性的系统优化。具体而言，使用机器学习的统计方法，如实验设计（DoE）和全局敏感性分析（GSA）可能非常有效，但尚未应用于这种“绿色”纳米材料。在此，第一次，一个连续的DoE策略开发与预筛选实验，以勾勒出可行的设计空间。使用2（3）全析因设计的连续筛选确定，从最初研究的三个因素（反应物浓度的比率、pH和前体浓度）中，仅前两个对于二氧化硅产率和表面积具有统计学显著性。随后使用中心复合设计进行的级联优化确定了最大产率为90 mol%，最大表面积为300-400 m2 g-1。由于成功的商业化，高产率和大比表面积是可取的，他们的同时优化也实现了高预测性回归模型。对于互补，基于方差的GSA首次成功应用于生物启发二氧化硅。该方法快速识别了控制理化性质的关键参数和相互作用，并在广泛的参数空间中提供了见解，并通过广泛的DoE活动进行了验证。这项工作是在大型实验空间内对多维因素-响应关系进行整体建模的起点，以补充生物启发二氧化硅及其他产品的资源高效型产品和工艺开发和优化的努力。
The in vitro bioinspired synthesis of silica, inspired from in vivo biosilicification, is a sustainable alternative to the conventional production of high value porous silicas. The short reaction time, mild reaction conditions of room temperature and its use of benign precursors make this an eco-friendly, economical and scalable route with great industrial potential. However, a systematic optimisation of critical process parameters and material attributes of bioinspired silica is lacking. Specifically, statistical approaches such as design of experiments (DoE) and global sensitivity analysis (GSA) using machine learning could be highly effective but have not been applied to this "green" nanomaterial yet. Herein, for the first time, a sequential DoE strategy was developed with pre-screening experiments to outline the feasible design space. A successive screening using 2(3) full factorial design determined that from the initially investigated three factors (the ratio of the reactant concentrations, pH, and precursor concentration), only the first two were statistically significant for silica yield and surface area. The subsequent concatenated optimisation using central composite design located a maximum yield of 90 mol% and a maximum surface area of 300-400 m(2) g(-1). Since for successful commercialisation, high yields and large specific surface areas are desirable, their simultaneous optimisation was also achieved with high predictability regression models. For complementation, a variance-based GSA was successfully applied to bioinspired silica for the first time. This method rapidly identified key parameters and interactions that control the physicochemical properties and provided insights in the wide parameter space, which was validated by the extensive DoE campaign. This work is the starting point in holistically modelling the multidimensional factor-response relationship over a large experimental space in order to complement efforts for resource-efficient product and process development and optimisation of bioinspired silica and beyond.