Prediction of pore-scale clogging using artificial intelligence algorithms

• DOI: 10.1007/s00477-023-02551-9
• Publication Date: 2023-09-05T07:01:47Z
• Authors: Chao Lei; Mandana Samari-Ke...; Hamed Aslannejad; Alraune Zech
• Keywords: Clogging; Lattice Boltzmann methods

Abstract We use five established, but conceptually different artificial intelligence algorithms for analysing clogging and quantifying colloid transport at pore scale: neural networks, decision tree, random forest, linear regression, support vector regression. test how these algorithm can predict by interpolating physics based simulation data. Our training data set is on results from Lattice Boltzmann simulations reproducing the of through a typical throat present in glass beads or medium sized sand. perform hyperparameter optimization cross validation all algorithms. The tree methods have highest Nash–Sutcliffe efficiencies among tested with values mostly above 0.9 independent event be predicted even 100% accuracy. indicate non-linear, rather categorial nature (simulation) This contrast to network while are often applied observational partly link this small size our dataset. application porous media research shows that time-consuming easily supplemented extended computational costs predictability quantitative effects process specific parameters colloidal given high reliability.