Robust intelligent paradigms for estimating fouling in phosphoric acid / steam heat exchanger

Heat exchanger fouling has several adverse effects on the performance of chemical process industry, including reducing operating cycles, deterioration of apparatus and increase energy losses. Precise prediction of fouling resistance is crucial for managing the heat transfer effectiveness of heat exchangers. The present work aimed at developing an accurate model which predict fouling resistance from several feature variables. Artificial Neural Network (ANN), Adaptive Neuro- Fuzzy Inference System (ANFIS), Support Vector Machine (SVM) and K-Nearest Neighbors (KNN) approaches are highly accurate methodologies for comprehending and analyzing the most intricate systems. The Multi-Layer Perpectron (MLP), Cascade Feed-Forward (CFF), Radial Basis Function (RBF), Recurrent (RNN), and Elman back propagation (ENN) neural networks, Adaptive Neuro- Fuzzy Inference System (ANFIS), Support Vector Machine (SVM), and K-Nearest Neighbors (KNN) are used in this study as intelligent paradigms to predict the fouling resistance in the cross flow heat exchanger of phosphoric acid concentration unit. 70 % of the experimental data collected are used to design and train the intelligent paradigms (253 data points), 15 % for validation (54 data points) and 15 % for the test (54 data points). The highest relationship between considered features and the first order of fouling resistance was found by using Pearson's approach. According to the results obtained, the best intelligent paradigm model was a cascade feed-forward neural network, composed of nine hidden neurons, with excellent overall AARD = 3.44 %, MSE = 2.779 10−12, RMSE = 1.667 10−6 and r2 = 0.999. The significance of this study lies in its provision of a practical and precise tool for predicting fouling resistance to both the research community and the industry. This model can be regarded as a trustworthy substitute for empirical analyses, which are frequently costly and time-consuming.