Optimising ANN architecture for shell and tube heat exchanger modelling

Abstract

Heat exchangers have a special place in chemical process industries. The shell and tube heat exchanger is commonly used for heating or cooling of process fluids. The various parameters to be taken into account for developing a model are inlet and outlet temperatures of shell and tube side fluids and their flow rates. Artificial Neural Networks (ANN) are effective in modeling of non-linear multi variable relationships and also referred to as the black box models. In the present work, various ANN models have been developed with single, two and three hidden layers for estimation of exit temperature of both the fluids as a function of inlet temperature conditions and also flow rates. The data used for training of ANN is generated on a small shell and tube heat exchanger, fabricated for this purpose. The ANN models thus developed are validated for test data that was not used for training of these models. The comparisons between models have been carried out. It is observed that ANN model with three hidden layers (15-15-15 neurons) has good level of accuracy (95-98%) for predicted values of training and test data set.