One of the biggest points of emphasis for the energy production industry is how to effectively decarbonize and reduce the footprint of generation activities while still maintaining sufficient capacity to fulfill the energy needs of the world at large. Unlike many other sectors that can pivot to alternate forms of zero-emission fuel, the very nature of fossil fuel extraction, production, and transmission makes this transition particularly difficult and expensive. Small, immediate gains in efficiency with minimal investment can play a significant role in both smoothing the energy transition as well as extending the window where climate effects and increases in global temperatures above 3.5°F can still be mitigated or eliminated.
Previous work and field trials presented by the authors have demonstrated that efficiency losses associated with fouling of heat transfer surfaces are a significant contributor to carbon emissions; in steam generation plants, fouling of the main condenser can lead to increased backpressure and reductions in power output. Hard deposit buildup on the pre-heat train (PHT) of a refinery can result in dramatically increased fuel use to raise the temperature of production fluid so that is ready for separation and distillation. New materials capable of imparting low-surface energy properties and greatly reduced surface roughness have been demonstrated to significantly decrease fouling in many of these cases, opening untapped operational capacity. However, without careful monitoring of the exchanger itself, this capacity may go entirely unrealized and un-utilized.
This paper presents a new strategy in developing a monitoring and prescriptive maintenance solution that can specifically work as a complement to determine improved heat transfer performance after refurbishment by an anti-fouling surface treatment. The thermal sensor intelligence module (TSIM) was designed to be a lightweight and self-contained system, with the ability to be easily deployed on heat transfer equipment.
To make accurate and precise predictions for the absence or presence of fouling on a treated system, where both historical and real-time data may be limited, an ensemble learning method was utilized in conjunction with a subscale condenser system whereby the TSIM could be rapidly trained on a variety of simulated fouling conditions, and the presence or absence of treatment. The learning method demonstrated in this work allowed for the TSIM to improve its fouling predictions through a model that allows it to impute the values of different parameters if the deployed exchanger or condenser does not have the necessary instrumentation. This imputation and prediction of the missing exchanger parameters allows for the accuracy be improved by nearly 20%, and the precision and F1 scores to be comparable to the model with a full set of input features. Finally, results gathered from this test condenser system, and the calculation of heat transfer efficiency showed good correlation with previously reported field data gathered under similar conditions, with a roughly 3-7% improvement after the addition of the anti-fouling treatment.