Development of a novel model to estimate the separation of organic compounds via porous membranes through artificial intelligence technique

Abstract

We have carried out modeling and computation of mass transfer in a membrane contactor for removal of organic compounds from aqueous solutions. Both computational fluid dynamics (CFD) and Artificial Intelligence (AI) methods were utilized for modeling separation process. For the AI, we explored the application of three distinct regression models, namely Kernel Ridge Regression, Gaussian Process Regression, and Poisson Regression to predict the concentration of a component, C, based on r and z. To enhance the performance of these models, the hyper-parameter tuning process employs Glowworm Swarm Optimization (GSO). The findings illustrate the effectiveness of the utilized models. Gaussian Process Regression achieves a noteworthy R2 score of 0.99791, with a RMSE of 3.9666×101(mol/m3) and an AARD% of 4.52000×10-1. Kernel Ridge Regression, while slightly less accurate, achieves a commendable R2 value of 0.97865, with an RMSE of 1.2446×102(mol/m3) and an AARD% of 2.63808. Poisson Regression offers a respectable performance, yielding an R2 score of 0.95509, along with an RMSE of 1.8011×102(mol/m3) and an AARD% of 4.28969. Moreover, the separation efficiency was estimated to be greater than 70 % using the membrane process.

Keywords

• Membrane contactor
• Machine learning
• Separation
• CFD
• Modeling