Prediction of specific heat of hybrid nanofluids with ANN Optimization of hyperparameters with Bayesian optimization Development of a specific heat estimator alternative to correlations Figure A. Schematic representation of the network structure used in the present study.
Purpose:The aim of this study is to develop an Artificial Neural Networks (ANN) based estimator that can be used to predict the specific heat of deionized water-based CuO + MWCNT, MgO + MWCNT, and SnO2 + MWCNT hybrid nanofluids and to investigate the usability of the ANN-based estimators instead of the commonly used correlations in the literature.
Theory and Methods:Experimentally obtained data found in the literature on the specific heat of deionized water-based CuO + MWCNT, MgO + MWCNT and SnO2 + MWCNT hybrid nanofluids measured for various temperature T (25 -50℃ ), volume ratio φ (0.25% -1.50%), and particle diameter dp (20 -50 nm) were used in the present study. The training algorithm's and the network's hyperparameters such as neuron number, hidden layer number, transfer function, epoch number, and learning rate, and the best training algorithm for the problem addressed among various training algorithms were determined by employing the Bayes optimization. K-fold cross-validation was applied as a precaution against overfitting.
Results:It has been determined as a result of the optimization that the LM training algorithm gives the best result with the hyperparameter combination where the hidden layer number is 1, the number of neurons is 24, the epoch number is 498, the learning rate is 0.15542 and the transfer function is logsig. The determination coefficient (R 2 ) of the optimized network structure for training, validation, test data, and the whole data was found to be 0.998415, 0.995497, 0.995023, and 0.997504, respectively.
Conclusion:It was concluded that the ANN-based estimator obtained in the present study can be used to predict the specific heat of deionized water-based CuO + MWCNT, MgO + MWCNT, and SnO2 + MWCNT hybrid nanofluids with high accuracy. It has been determined that the ANN-based estimator reveals better performance to predict the specific heat of the hybrid nanofluids than the classical correlation. Therefore, it has been concluded that more precise and realistic calculations can be made by using the ANN-based estimator in determining the specific heat of nanofluids required in experimental and numerical studies.