Data-Driven Soft Sensor for Crude Oil Fouling Monitoring in Heat Exchanger Networks

Crude oil fouling within heat exchanger networks (HENs) is a long-standing challenge that detrimentally affects heat transfer efficiency, increases operating and maintenance costs, and burdens regular cleaning operations. Previous approaches to mitigate the oil fouling involved regular mechanical or chemical cleaning of the HENs. While researchers have developed some fouling models to calculate the crude oil fouling in HENs, these models primarily focused on the chemical reaction fouling and often neglected the influence of flow velocity and temperatures. Consequently, they may not offer precise theoretical guidance for optimizing cleaning operations. To address this gap, this study presents a data-driven model for monitoring crude oil fouling in HENs using an online dataset obtained from real-world industrial operations and then predicts the fouling development in the future to optimize the cleaning operation schedule. By analyzing all available measured data from industrial HENs, we explore the correlations among various operational parameters. The obtained soft sensor models, which include the coupling of the main influencing parameters, were obtained by training the measured data. Subsequently, we validate the proposed soft sensor models by comparing their predictions of fouling in different periods. The results exhibit strong agreement between the predicted and measured fouling levels, demonstrating the effectiveness of the proposed soft sensor model in real-time monitoring of crude oil fouling, with an
${R}^{{2}}$
value of around 0.90. These findings substantiate the model’s capacity to calculate future fouling development, thereby providing valuable insights for the optimization of cleaning schedules of oil refineries.