Comparison of the analytical model with neural network model on the case of heat exchanger behavior under fouling

The article presents mathematical modeling of the impact of fouling on heat transfer in shell-and-tube exchangers using analytical methods and an artificial neural network (ANN). The article shows the limitations of using ANN diagnosis of fouling on the behavior of industrial shell and tube heat exchangers. The analytical heat exchanger model is a reference model used for the comparative assessment of the ANN model. The analytical model is treated by the authors as a model providing accurate numerical calculations, aiming at diagnosis of fouling on the heat exchanger behavior during the operation of an industrial installation. The proposed model of the exchanger is characterized by sufficient accuracy from the point of view of the exchanger diagnosis under fouling, and the calculation error does not exceed 1%. The article compares the numerical results of the heat exchanger duty calculated by the use of the proposed analytical method with the heat exchanger duty obtained by ANN. Comparing the analytical model and the ANN model, the coefficient of determination R² is about 7%, which indicates a significant discrepancy in results between the analytical model and the ANN model. The presented article states that the ANN fouling model operates as an extrapolator. Thus, the application of ANN is limited to the case of small variations in ​​measured inlet process parameters. Meanwhile, the analytical model is more universal because it is less sensitive to the variations in ​​measured process parameters.