A new comprehensive model for relative viscosity of various nanofluids using feed-forward back-propagation MLP neural networks

In this study, a comprehensive model was proposed to predict the nanofluids relative viscosity on the basis of feedforward back-propagation network by utilizing various training algorithm including Scaled Conjugate Gradient (SCG), Levenberg-Marquardt (LM), Bayesian Regulation back propagation (BR), and Resilient back Propagation (RP). Moreover, different transfer functions such as Radial basis (radbas), Tan-sigmoid (tansig), Log-sigmoid (logsig), Hard-limit (hardlim), Triangular basis (tribas) and Soft max transfer function (softmax) were investigated in hidden layer and their effects on network precision were estimated. A total of 1620 experimental data points that was chosen from a broad range of nanoparticles suspension with various basefluids was gathered from reliable literature to train, test and validate the proposed network. Temperature, shear rate values, nanoparticle size, particle nature (nanoparticle density) and particle concentration were chosen as input variables of the developed network. The network with 1 hidden layer and 23 neurons with tan-sigmoid and purelin transfer functions in the hidden and output layers was determined to have the optimum performance. Furthermore, Levenberg-Marquardt (LM) was the best train algorithm while in this case the model was more precise than others. The results revealed that the proposed network has the ability to correlate and predict the relative viscosity accurately with an overall Mean Square Error (MSE) value of 0.00901 and correlation coefficient (R2) of 0.9954.