Efficiency increment of CFD modeling by using ANFIS artificial intelligence for thermal-based separation modeling

This study presents a comprehensive analysis of predicting the temperature in a vacuum membrane distillation (VMD) process. The simulation was carried out via computational fluid dynamics (CFD) as well as machine learning. CFD was performed for obtaining temperature distribution in the feed solution, and the calculated temperature was used for training several machine learning models. The dataset comprises over 13,000 observations, which provide a rich source for modeling complex relationships. Three sophisticated regression models were employed: Adaptive Neuro-Fuzzy Inference System (ANFIS), Kernel Ridge Regression (KRR), and Multi-Layer Perceptron (MLP). ANFIS was chosen for its hybrid nature, combining neural networks and fuzzy logic, effectively capturing intricate non-linear relationships in data. ANFIS performance in fitting the data was compared with the other models. Hyper-parameter optimization for these models was conducted using the Tabu Search algorithm to ensure optimal performance. The ANFIS model demonstrated superior performance with an R2 score of 0.9964513 on the training set and 0.9964507 on the test set, alongside a MSE of 0.037655 and a MAE of 0.168272. The robustness of ANFIS was further confirmed by a 3-fold cross-validation mean R2 score of 0.9964579 and a standard deviation of 3.3619616e−05.