Neural network based correlation for critical heat flux in steam-water flows in pipes

This paper presents a new approach using artificial neural networks (ANN) to predict the critical heat flux (CHF) for a steam–water mixture through pipes. A large number of experimental measurements are used for training and testing the developed network. The Levenberg–Marquardt algorithm was used to train the developed feed forward ANN. The training and validation are performed with good accuracy. The correlation coefficient obtained with unknown data applied to the network is 0.998 which is satisfactory and verifying the fidelity of the developed network. The present methodology proved to be much better than the traditional table and best-fit methods. Using the weights and biases obtained from the trained network, a new formulation is proposed for determination of the CHF Qcr. Experimental results of Qcr are compared with both the results obtained by the developed ANN based correlation and the results obtained by a best-fit correlation. Deviations between the results are found to be less than 5.5% and 30%, respectively. The developed ANN based correlation may make use of the dedicated ANN software unnecessary to use for each calculation time. As seen from the results obtained, the calculated Qcr is obviously within acceptable uncertainties. This ANN based correlation can be employed with any programming language or spreadsheet program for estimating the CHF Qcr. If the validity range changes, the ANN based correlation can be updated in terms of new sets of weights and biases using the same network architecture (same no. of hidden layers and no. of neurons).