Rapid, autonomous high-throughput characterization of hydrogel rheological properties via automated sensing and physics-guided machine learning
Rapid, autonomous high-throughput characterization of hydrogel rheological properties via automated sensing and physics-guided machine learning
复制标题
DOI：
10.1016/j.apmt.2022.101720
复制
发表时间：
2023-02
影响因子：
8.3
通讯作者：
Junru Zhang;Yang Liu;Durga Chandra Sekhar.P;Manjot Singh;Yuxin Tong;Ezgi Kucukdeger;H. Yoon;Alexander P. Haring;M. Roman;Zhenyu Kong;Blake N. Johnson
中科院分区：
文献类型：
--
作者：
Junru Zhang;Yang Liu;Durga Chandra Sekhar.P;Manjot Singh;Yuxin Tong;Ezgi Kucukdeger;H. Yoon;Alexander P. Haring;M. Roman;Zhenyu Kong;Blake N. Johnson
成分-工艺-结构-性能关系的高通量表征(HTC)对于加速分子和材料的发现和制造范例至关重要。在这里，我们提出了一种通过自动传感和物理指导的有监督的机器学习来快速、自主地测量井板格式的水凝胶流变性的HTC方法。新的高温超导方法有助于快速、自主地表征96孔板格式中与凝胶和网络互穿相关的水凝胶流变性和渗流过程，速度为24 S/样品(比最先进的速度快70倍)。用该方法获得的粘弹性能和相行为与传统的流变学研究进行了对比。Pluronic F127、胶原和海藻酸盐-PNIPAM水凝胶在96孔板中的凝胶点分别为0.31wt%(Pluronic F127)、0.031 mg/ml(胶原)和0.069 wt%(NIPAM)的高分辨率表征证明了该方法的快速和实用性。从传感器多变量时间序列数据、校准数据和流体-结构相互作用模型生成的实验组成-性质关系数据使用有监督的机器学习实现了对样品相的准确分类。使用传感器物理的特征增强，这里是流体-结构相互作用模型，相对于在没有基于物理的特征增强的情况下获得的材料(即样品)相分类精度。最终，创造快速、自主的HTC方法，与常见的高通量实验模式(如井板)协同工作，可以加快跨多个学科的研究步伐，并为新兴行业的质量保证和控制产生新的工具。
High-throughput characterization (HTC) of composition-process-structure-property relations is essential for accelerating molecular and material discovery and manufacturing paradigms. Here, we present a rapid, autonomous method for HTC of hydrogel rheological properties in well plate formats via automated sensing and physics-guided supervised machine learning. The novel HTC method facilitates rapid, autonomous characterization of hydrogel rheological properties and percolation processes associated with gelation and network interpenetration in 96-well plate formats at a rate of 24 s/sample (70 times faster than the state-of-the-art). Viscoelastic properties and phase behavior obtained by the method were benchmarked against traditional rheology studies. The speed and utility of the method were demonstrated by high-resolution characterization of the gel point of Pluronic F127, collagen, and alginate-PNIPAM hydrogels in 96-well plate formats at resolutions of 0.31 wt% (Pluronic F127), 0.031 mg/ml (collagen), and 0.069 wt% (NIPAM), respectively. Experimental composition-property relation data generated from sensor multivariate time-series data, calibration data, and fluid-structure interaction models enabled accurate classification of sample phase using supervised machine learning. Feature augmentation using sensor physics, here, a fluid-structure interaction model, improved material (i.e., sample) phase classification accuracy relative to that obtained in the absence of physics-based feature augmentation. Ultimately, creating rapid, autonomous HTC methods that synergize with common high-throughput experimentation formats, such as well plates, can accelerate the pace of research across several disciplines as well as generate new tools for quality assurance and control across emerging industries.