Prediction of heat exchanger fouling for predictive maintenance using artificial neural networks

The petroleum refining business consumes approximately 0.2 MMBTU/BBL of energy annually. This consumption is mitigated using heat integration techniques. However, a significant challenge in this process is fouling in the preheat train network of heat exchangers. Fouling necessitates regular cleaning, leading to substantial operational inefficiencies and costs, with annual losses estimated at nearly $16.5 billion. To address this issue, implementing a predictive maintenance model is crucial for performing maintenance at optimal periods, thereby reducing these losses. The study proposes an artificial neural network (ANN) developed using MATLAB’s nntool, trained on industrial heat exchanger samples that were preprocessed in Microsoft Excel. This ANN model is designed to forecast fouling patterns in shell and tube heat exchangers. The model’s accuracy and effectiveness were validated using R² (coefficient of determination) and RMSE (root mean square error) measures. The results indicated that the EA-307 Feed-Forward Back-Propagation Neural Network (FFBPNN) model delivered satisfactory performance, with an R² value of 0.9961. This high level of accuracy underscores the significant impact of the number of neurons on the model’s predictive output. Furthermore, the model’s testing on a new dataset yielded impressive results, achieving an R² value of 0.966. This demonstrates the model’s robustness and reliability in predicting fouling patterns, facilitating improved maintenance schedules, and minimizing the financial losses associated with fouling. The study highlights the potential of advanced neural network models to significantly enhance the operational efficiency of petroleum refineries by enabling more precise and timely maintenance interventions.

Graphical abstract