Argyri Smaragda-Maria, Evenäs Lars, Bordes Romain
Department of Chemistry and Chemical Engineering, Chalmers University of Technology, Gothenburg 41296, Sweden.
Department of Chemistry and Chemical Engineering, Chalmers University of Technology, Gothenburg 41296, Sweden.
J Colloid Interface Sci. 2023 Jun 15;640:637-646. doi: 10.1016/j.jcis.2023.02.077. Epub 2023 Feb 23.
Acoustic levitation provides the possibility to deform levitated droplets in a controllable, and quantifiable manner, thus offering a means to measure the surface tension of a liquid droplet based on its deviation from sphericity. However, for new generation of multi-source and highly stable acoustic levitators, no model relates the acoustic pressure field to the deformation and surface tension. Utilizing a machine learning algorithm is expected to identify correlations between the experimental data without any set preconditions.
A series of aqueous surfactant solutions with a large range of surface tensions were prepared, and evaporated under levitation, while the acoustic pressure was varied. A dataset of over 50,000 images was used for the training and evaluation of the machine learning algorithm. Prior to that, the machine learning approach was validated on in silico data that also included artificial noise.
We achieved high accuracy in predicting the surface tension of single standing droplets (±0.88 mN/m), and we surpassed certain physical conditions related to the size, and shape of the suspended samples that simpler theoretical models are subject to.
声悬浮提供了以可控且可量化的方式使悬浮液滴变形的可能性，从而提供了一种基于液滴偏离球形的程度来测量其表面张力的方法。然而，对于新一代多源且高度稳定的声悬浮器，尚无模型将声压场与变形及表面张力联系起来。利用机器学习算法有望在没有任何预设条件的情况下识别实验数据之间的相关性。
制备了一系列表面张力范围广泛的水性表面活性剂溶液，并在悬浮状态下使其蒸发，同时改变声压。一个包含超过50000张图像的数据集用于机器学习算法的训练和评估。在此之前，该机器学习方法在包含人工噪声的计算机模拟数据上得到了验证。
我们在预测单个静置液滴的表面张力方面达到了高精度（±0.88毫牛/米），并且超越了一些与悬浮样品的尺寸和形状相关的物理条件，而更简单的理论模型会受到这些条件的限制。
