»
»
Fouling
Table 1. Pearson correlation scores.
These models were assessed using metrics such as Mean Absolute Error( MAE), Mean Squared Error( MSE), or Root Mean Squared Error( RMSE). Hyperparameter tuning was performed to optimize the model’ s performance by exploring different combinations of settings in Azure. Once the model predicts the UA of the heat exchanger, this value will be graphed to determine cleaning period. Results will be validated with the equipment inspection report to confirm the accuracy.
Digital twin dashboard integration Power BI Dashboard is developed to showcase heat exchanger operating parameters and fouling prediction monitoring. The deployment strategy in future will focuses on sustainable and scalable manner.
Results and discussion HTRI automation system HTRI automation successfully processes one year of simulated PI data to perform comprehensive thermal calculations for CW heat exchangers. HTRI output results are systematically organized for each variable such as tube skin temperature, fluid flowrate, temperature and velocity profiles, duty, etc. This ensures smooth data delivered to Azure for fouling prediction analysis. This in-house solution has delivered significant direct cost savings without subscribing to external third-party technology, while achieving similar results and performance. Manual data entry duration and human error is reduced to the minimum. This demonstrates comparable performance can be achieved while operating at limited time and resources, showcasing the efficiency of the automated approach.
Fouling prediction – Variables selection by Pearson correlation test This analysis helps identify which variables are dominantly correlated with heat transfer coefficient( UA) or fouling in the case, providing insights into which factors are likely significant predictors in the regression model. Top five variables with the highest absolute correlation coefficients will be focused for accurate fouling prediction. 1. According to the table above, the top five influential variables are: 2. Tube max skin temperature( K) 3. Heat transfer area( m2) 4. Temperature inlet cold fluid(° C) 5. Cooling water flowrate( m3 / hr) 6. Tube maximum velocity( m / s)
Fouling prediction- Regression model selection Multiple regression models showcase in Table 2, were trained and tested on the same data. This analysis aims to achieve the second objective of building and evaluating several regression models to identify the best performer. The results indicated that the linear regression model had the highest R2 and the lowest MAPE. The R2 value shows how well the model’ s predictions align with the actual data, with a higher R2 indicating a better fit. MAPE measures prediction accuracy as a percentage, with a lower MAPE suggesting fewer errors. Therefore, the linear regression model demonstrated the highest accuracy and reliability, a preferred choice for predicting UA. The strong performance of the linear regression model can be attributed to its simplicity and the nature of the data. The relationships between the variables in the dataset are primarily linear, which is well-suited
Table 2. Regression models comparison.
48 Heat Exchanger World July 2025 www. heat-exchanger-world. com