Comparative Analysis of Different Predictive Models for Fouling Factor Prediction in HE

This project presents a comparative analysis of different predictive models for fouling factor prediction in heat exchanger (HE). HE have use cases in the mechanical industry where they are used as a preheating device in the hydrotreating unit of oil refineries. If fouling is not detected in time, it leads to the HE shutdown and plant explosion. Frequent cleaning causes the loss of time, energy, and money. Therefore, in time prediction is necessary. Shell and tube heat exchanger (STHE) is used in this paper for prediction. Prediction models such as Multiple Linear Regression (MLR), Recurrent Neural Network (RNN), Random Forest, and Artificial Neural Network (ANN) algorithms are used. Through the notification, users will be notified to stop the operation and start the cleaning process. Power BI is used for visual analysis by integrating with the Python code. Comparative analysis is done for all four above-mentioned algorithms based on accuracy, time complexity, the efficiency of the algorithm, ease of training the model, space complexity, etc., to predict the fouling resistance. All types of scenarios are considered while predicting the fouling resistance such as temperature, U-Transfer rate, logarithmic mean temperature difference (LMTD), area, fluid flow rate increase or decrease in heat exchange. The accuracy of MLR and Random Forest are 98.9 % and 99.52 % and the R2 score of RNN and ANN are −34.81 & −0.395 respectively. R2 and accuracy are the performance measures of the model. The comparison of the results shows that the MLR-based approach can be successfully used for predicting the rate of fouling factor in the HEs for theupcoming 6 months.