Modeling and optimization the effective parameters of nanofluid heat transfer performance using artificial neural network and genetic algorithm method

The heat transfer coefficient and, as a result, the Nusselt (Nu) number for nanofluids are affected by parameters such as thermal conductivity, thermal capacity of the fluid and nanoparticles, flow pattern, nanofluid viscosity, volume fraction of suspended particles, particle shape, and size. Since the parameters affecting the heat transfer coefficient of nanofluids are more than two or three parameters, and optimization requires calculating the value of the function at different points, on the other hand, there are no suitable equations for the relationship between these parameters, so for this purpose, existing experimental data was modeled using neural networks. After validating the created networks, using these networks, the optimal points in the existing information range have been determined using the genetic algorithm optimization method. All leading networks, with an intermediate layer, have the Levenberg–Marquardt (LM) training method, and the method of least sum of squares error has been used for validation. In addition, other networks were trained using data for all three nanofluids (Al2O3/water, CuO/water, and Cu/water nanofluids) by adding a particle size parameter. The obtained networks were also optimized in the genetic algorithm. The results showed that increasing Reynolds (Re) from 729 to 1995 at a concentration of 1 wt%, T = 96 °C, and Prandtl (Pr) = 2.97 increased the Nu by nearly 82% for Al2O3/water nanofluids. For CuO/water nanofluids, with increasing Reynolds from 617 to 2053 at a concentration of 3 wt%, T = 95.9 °C, and Pr = 4.59, the variations of the ratio of convection heat transfer coefficient of nanofluid to water increased by almost 38%. For Cu/water nanofluids, increasing the Reynolds number from 626 to 1895 at a concentration of 3 wt%, T = 65 °C, and Pr = 2.95, increased the Nu by nearly 77%. •Available experimental data was modeled using neural networks.•The optimal points in the existing data range specified by the genetic algorithm.•The Levenberg–Marquardt training method was used for validation.•The Nu for fluids increased by raising Re and Pr number at various concentrations.