Prediction of the minimum film boiling temperature using artificial neural network

This work studies the application of artificial neural network (ANN) to predict the minimum film boiling temperature (T-min) for various substrate rods quenched in high- and low-pressure distilled water pools. The ANN was trained using 379 experimental data collected from literature. Liquid subcooling temperature (T-sub), fluid to the substrate material thermophysical properties (beta(f)/beta(w)), system saturated pressure (P-sat) and length to diameter ratio (L/D) are utilized as inputs, whereas T min was viewed as the output. The number of neurons and hidden layers were determined based on the precision of results. The trained ANN predicted the experimental data with a mean absolute error (MAE) of approximately 5%, and a determination coefficient (R-2) greater than 0.95 for all data, utilizing an arrangement of 12 neurons inside 2 hidden layers. The estimated results were less than +/- 10% for 90% of all data, and +/- 20% for 99% of all data. Among the available correlations for T-min in literature, this work shows a success in the first attempt in implementing ANN to predict T-min accurately. (C) 2020 Elsevier Ltd. All rights reserved.