Machine Learning-Based Heat Transfer Prediction in Spiral Tube Heat Exchangers Using Bayesian-Regularized Neural Networks

In this proposed research, an ANN with Bayesian regularization for forecasting heat transfer rates in Spiral Tube Heat Exchangers (STHEs) is presented. Nonetheless, the objective is to assure high prediction quality for various operating conditions without overfitting and enhance the predictive model’s generality. The ANN is shown to be very effective having achieved correlation coefficients (R-values) of 0.99769 for the training set and 0.99882 for the testing set thus showing a good linearity between the predicted and target variables. The overall regression plots are very close to the Y = T line and the best line equations are also not much deviated from 1 (y = ax+b), which implies the correct scaling of the output and almost negligible systematic error. Analyzing the overall data set gives the model an R-value of 0.99535 meaning the model is stable and reliable. That way, these results corroborate the applicability of the ANN model to forecast STHE heat transfer, and how it can be useful in precise regression analysis applications. The addition of the Bayesian regularization part helps generalize the model and places the developed tool in the array of tools and methods to be used in thermal prediction of intricate heat exchanger systems.