The petroleum refining industry's crude oil fouling of heat exchangers has been a challenge, resulting in inefficient heating of crudes in the upstream process. It has been observed that fouling due to various reasons, such as crude composition, process parameter effects, working conditions, etc., has been the main cause of degradation of these heat exchangers' performances. We have therefore proposed Machine Learning (ML)-based models and algorithms development that can take essential steps to prevent fouling and thus apprehend proper maintenance schedules of these unit operations in refineries.

This report describes in detail the various forms of fouling that typically affect a heat exchanger in the refinery and proposes an ML-based algorithm that can predict the effectiveness of each heat exchanger in the network through energy and mass balance error estimations. The present algorithm takes into account a network factor on energy and mass balance calculations, which optimizes the flow rates and temperatures of various flow streams to enhance the overall effectiveness of the heat exchanger network.

This study also investigates the application of machine learning techniques to estimate fouling resistance in water filtration systems. Two methods are analyzed in this work to study the performance of ML models on process industrial problems. In the first methodology, a linear regression model using various parameters to predict fouling resistance has been proposed, while in the second study, the performance of different regression models has been examined regarding an industrial problem. The results show that random forest regression outperforms linear regression.

The results of this study suggest that machine learning models can effectively predict fouling resistance in water filtration systems, enabling optimization of the systems to reduce fouling, improve efficiency, and reduce costs.