Modeling of thermal conductivity of ethylene glycol nanofluids containing carbon nanotubes by Multilayer Perceptron neural network

In our previous work [1], pristine and functionalized carbon nanotubes with 1, 2 and 4 hours refluxing times and concentrations of 0.1, 0.25 and 0.5 Vol% were used to prepare nanofluids and their thermal conductivity was measured at 20, 30, 40 and 50 °C. Lots of empirical works cannot be done because they are time consuming and costly. One of the best methods for investigation of low cost and wide range of empirical works is using the modeling methods. The artificial neural network model is a method which replicates the initial sensory processes of the brain. It is possible to design a virtual laboratory using artificial neural networks to predict the results for the same conditions even not measured experimentally. In this work, a multilayer perceptron (MLP) neural network was used to design a virtual lab and modeling the experimental data including the thermal conductivity of ethylene glycol nanofluids containing CNTs. In order to achieve a minimum error, neural networks with different hidden layers (1, 2 and 3 layers) and different number of neurons in each layer (2, 3, 4, 5, 6, 10 and 15 neurons) were studied. The minimum error of 6.5% was obtained for the neural network with two hidden layers by 3 neurons in the first layer and 2 neurons in the second one. This network was used to predict the results in the conditions which were closed to experimental conditions and it was observed that the predicted results were in good agreement with the experimental results.