摘要：The computational fluid dynamics (CFD) modeling is an applicable tool for the prediction of fluid flow characteristics (velocity, temperature, pressure, etc.). However, CFD requires a lot of time, computational efforts and as a result, much more expenses for complicated cases (e.g. turbulent flow, 3-dimensional calculations, etc.). The present work tries to conduct an investigation on the potential of the artificial intelligence algorithms in overcoming such barriers of CFD modeling. Turbulent forced convection of Cu/water nanofluid in a tube under constant heat flux is considered as a case study for the model development. The density and viscosity of the based fluid are enhanced by the suspension of the nanoparticles. This makes more pressure drop and as a result, imposes more pumping power. So, this paper is focused on a way to facilitate the prediction of the pressure of the nanofluid convective flow. The results of the CFD modeling are learned by the adaptive network-based fuzzy inference system (ANFIS), as the artificial intelligence method. The CFD modeling is done for several Cu nanoparticle volume fractions (i.e. 0.3, 0.5, 0.8, 1, and 2). Several types of variables as inputs (i.e. x, y, z, and nanoparticle volume fraction) and different kinds of membership functions (i.e. Generalized bell-shaped membership function (gbellmf), Gaussian membership function (gaussmf), Gaussian combination membership function (gauss2mf), Difference between two sigmoidal membership functions (dsigmf), Product of two sigmoidal membership functions (psigmf)) are examined until the intelligence requirements of the ANFIS are satisfied. Once the best intelligence of ANFIS has been achieved, there is no need for complicated CFD modeling. The ANFIS predictions show the highest compatibility with the CFD results. The maximum pressure drop was predicted around 1500 Pa. The results also revealed that for checking the intelligence condition the coefficient of determination (R 2 ) is not solely sufficient and the error values must be considered as well. Considering the gauss2mf as the membership function and the nanoparticle volume fraction as the fourth input, the ANFIS could distinguish intelligently the pattern of data (changing the pressure with coordinates and particle fraction). For the best intelligence, the mean standard error was around zero, while the coefficient of determination was close to one. At this condition, the pressure of the nanofluid could be determined as a function of the nanoparticle volume fraction and anywhere inside the tube without using the CFD modeling.
