Iranian Journal of Chemical Engineering (IJChE)

Abstract This study presents a neuro-fuzzy inference system for predicting the pressure loss in coiled flow inverter (CFI) tubes. Computational fluid dynamics (CFD) simulations were conducted to obtain the amounts of the pressure loss across nine distinct configurations of CFI. The neuro-fuzzy model utilized three key input parameters of the Reynolds number (Re), number of 90° bends (N), and tube-to-coil diameter ratio (L/D). Following CFD validation, the dataset was partitioned into training (two-thirds) and testing (one-third) subsets. The model achieved an outstanding mean relative error (MRE) of 0.549%, demonstrating its high predictive accuracy and reliability for the estimation of the pressure loss in coiled flow inverter systems. These results highlight the neuro-fuzzy approach as a suitable tool for optimizing CFI designs in industrial applications. This study ultimately demonstrates how the strategic combination of numerical simulation and machine learning can accelerate development cycles while maintaining rigorous accuracy standards, providing engineers with a powerful tool for system design and optimization. 

Abstract Prediction of the dynamic crossflow ultrafiltration rate of a protein solution such as milk poses a complex non-linear problem as the filtration rate has a strong dependence on both the solution physicochemical conditions and the operating conditions. As a result, the development of general physics-based models has proved extremely challenging. In this study an alternative dynamic neuro-fuzzy model for milk ultrafiltration that describes the variation in dynamic permeate flux decline with temperature, transmembrane pressure (TMP), fat percentage, pH and molecular weight cut off (MWCO) has been developed with the experimental data of the pilot spiral wound membrane test rig. By increasing the temperature, TMP, and pH the permeate flux is increased, and by increasing fat concentration the permeate flux is decreased. The MWCO variation indicates a paradoxical permeate flux. Additionally, a hybrid physical model for dynamic prediction of total resistance in the milk ultrafiltration by combination of two neuro-fuzzy (ANFIS) models and a physical model (BLA model) is developed. By increasing the TMP and fat concentration, the total resistance is increased. But by increasing the pH and temperature, the total resistance is decreased. Also, MWCO variation indicates a paradoxical total resistance value.