Artificial intelligence for performance prediction of organic solvent nanofiltration membranes
Artificial intelligence for performance prediction of organic solvent nanofiltration membranes
复制标题
DOI：
10.1016/j.memsci.2020.118513
复制
发表时间：
2021-02-01
影响因子：
9.5
通讯作者：
Szekely, Gyorgy
中科院分区：
文献类型：
--
作者：
Hu, Jiahui;Kim, Changsu;Szekely, Gyorgy
迫切需要开发预测方法，以快速跟踪有机溶剂纳滤（OSN）的工业实施。然而，OSN膜的性能预测一直是一个艰巨的和具有挑战性的任务，由于大量的可能的溶剂和溶剂-膜，溶质-溶剂和溶质-膜相互作用之间的复杂关系。因此，我们没有开发基本的数学方程，而是通过编译大型数据集并基于收集的数据集（包含38，430个数据点，超过18个维度（参数））构建基于人工智能（AI）的预测模型来打破常规。为了阐明影响膜性能的重要参数，我们进行了彻底的主成分分析（PCA），结果表明，影响渗透率和截留率的因素是惊人的相似。我们已经训练了三种不同的AI模型（人工神经网络，支持向量机，随机森林），以前所未有的准确度预测膜性能，高达98%（渗透率）和91%（截留率）。我们的研究结果为适当的数据标准化铺平了道路，不仅用于性能预测，还用于更好的膜设计和开发。
There is an urgent need to develop predictive methodologies that will fast-track the industrial implementation of organic solvent nanofiltration (OSN). However, the performance prediction of OSN membranes has been a daunting and challenging task, due to the high number of possible solvents and the complex relationship between solvent-membrane, solute-solvent, and solute-membrane interactions. Therefore, instead of developing fundamental mathematical equations, we have broken away from conventions by compiling a large dataset and building artificial intelligence (AI) based predictive models for both rejection and permeance, based on a collected dataset containing 38,430 datapoints with more than 18 dimensions (parameters). To elucidate the important parameters that affect membrane performance, we have carried out a thorough principal component analysis (PCA), which revealed that the factors affecting both permeance and rejection are surprisingly similar. We have trained three different AI models (artificial neural network, support vector machine, random forest) that predicted the membrane performance with unprecedented accuracy, as high as 98% (permeance) and 91% (rejection). Our findings pave the way towards appropriate data standardization, not only for performance prediction, but also for better membrane design and development.