Metamodeling the deposition process in oil preprocessing to optimize heat exchanger cleaning: a deep neural network approach

Resumo

The accumulation of deposits in heat exchangers during oil preprocessing, known as fouling, compromises thermal efficiency and increases maintenance and energy costs. This study aimed to develop effective strategies for managing these deposits to optimize heat exchanger operations. Metamodeling combined with Artificial Intelligence was employed, using machine learning models to process sequential data, which is crucial for understanding the evolution of deposits over time. Two machine learning models, Deep Neural Networks (DNN) and Long Short-Term Memory (LSTM), were trained and evaluated. The DNN achieved a Mean Squared Error (MSE) of 0.05345, outperforming the LSTM, which had an MSE of 0.8763. The results suggest that the DNN was more effective in predicting fouling, although both models showed signs of overfitting. Advanced monitoring technologies, such as IoT and remote sensing, as well as predictive maintenance, were found to be essential for improving operational efficiency and reducing costs, highlighting the importance of digitalization and automation in oil preprocessing.