TY - JOUR

T1 - Back propagation modeling of shear stress and viscosity of aqueous Ionic-MXene nanofluids

AU - Afzal, A.

AU - Yashawantha, K.M.

AU - Aslfattahi, N.

AU - Saidur, R.

AU - Abdul Razak, R.K.

AU - Subbiah, R.

N1 - The final publication is available at Springer via http://dx.doi.org/10.1007/s10973-021-10743-0

PY - 2021/8/31

Y1 - 2021/8/31

N2 - Back-propagation modeling of viscosity and shear stress of Ionic-MXene nanofluid is carried out in this work. The data for Ionic-MXene nanofluid of 0.05, 0.1, and 0.2 mass concentration (mass%) are collected from the experimental analysis. Shear stress and viscosity as a function of shear rate and mass% of MXene nanoparticles is used as input. Additionally, viscosity as a function of temperature and % of MXene nanoparticles is collected separately. Based on the possible combinations, five back-propagation algorithms are developed. In each algorithm, five models depending upon the number of neurons in the hidden layer are used. The training and testing of all the models in each algorithm are performed. Statistical analysis of the network output is done to evaluate the accuracy of models by finding the losses in terms of mean squared error (MAE), root-mean-squared error, mean absolute error, (MAE), and error deviation. Model 1 is found to have lower accuracy than the remaining models as the number of neurons in its hidden layer is only one. The performance evaluation metrices of the back-propagation model show that the error involved is acceptable. The training and testing of the algorithms are satisfactory as the network output is found to be in comfortably good agreement with the desired experimental output.

AB - Back-propagation modeling of viscosity and shear stress of Ionic-MXene nanofluid is carried out in this work. The data for Ionic-MXene nanofluid of 0.05, 0.1, and 0.2 mass concentration (mass%) are collected from the experimental analysis. Shear stress and viscosity as a function of shear rate and mass% of MXene nanoparticles is used as input. Additionally, viscosity as a function of temperature and % of MXene nanoparticles is collected separately. Based on the possible combinations, five back-propagation algorithms are developed. In each algorithm, five models depending upon the number of neurons in the hidden layer are used. The training and testing of all the models in each algorithm are performed. Statistical analysis of the network output is done to evaluate the accuracy of models by finding the losses in terms of mean squared error (MAE), root-mean-squared error, mean absolute error, (MAE), and error deviation. Model 1 is found to have lower accuracy than the remaining models as the number of neurons in its hidden layer is only one. The performance evaluation metrices of the back-propagation model show that the error involved is acceptable. The training and testing of the algorithms are satisfactory as the network output is found to be in comfortably good agreement with the desired experimental output.

KW - Algorithms

KW - MXene

KW - Nanofluids

KW - Neural networks

KW - Shear stress

KW - Viscosity

KW - Backpropagation

KW - Errors

KW - Function evaluation

KW - Mean square error

KW - Nanoparticles

KW - Experimental analysis

KW - Hidden layers

KW - Mass concentration

KW - Mean absolute error

KW - Mean squared error

KW - Root mean squared errors

KW - Training and testing

KW - Nanofluidics

U2 - 10.1007/s10973-021-10743-0

DO - 10.1007/s10973-021-10743-0

M3 - Journal article

VL - 145

SP - 2129

EP - 2149

JO - Journal of Thermal Analysis and Calorimetry

JF - Journal of Thermal Analysis and Calorimetry

SN - 1388-6150

ER -