(28c) A Hybrid Approach for Fouling Monitoring and Forecasting with Application to an Industrial Heat Exchanger

Heat exchanger fouling in ethylene oxide (EO) plants is a safety hazard that can lead to unplanned outages and temperature excursions or decomposition events, therefore constituting a major risk for safe operations (Hess & Tilton, 1950; Cavitt, 1983). Fouling of the heat exchangers reduces energy recovery and increases pressure drop. The increase in pressure drop eventually leads to added expenditures in utilities to maintain the process flows at a target design level. Furthermore, fouling deposits originate localized heat transfer dead zones where hotspots can form leading to strong temperature excursions and decomposition events (Crocco et al., 1959). All these effects are detrimental to the plant operation and should be managed as closely as possible to secure the stability, economy, and safety of the process. However, in most circumstances, fouling is not directly observable or measured, and the only alternative available is to infer it from the analysis of collected data or field measurements, usually in a rather ad hoc fashion, depending on the operators' experience (Sundar et al., 2020). Therefore, the development of a framework for advanced monitoring and forecasting of heat exchanger fouling is both opportune and important to improve the reliability safety of the operation (Trafczynski et al., 2021).

We propose a hybrid approach, where knowledge-based feature generation is integrated with data-driven methods, to forecast a key performance indicator (KPI) that acts as a fouling surrogate (Diaz-Bejarano et al., 2020). Knowledge-based feature generation allowed to monitor the evolution of fouling over time and enabled the use of off-the-shelf data-driven forecasting methods. Among the KPIs tested, two were selected to act as fouling surrogates. We advise monitoring them in tandem with the Reynolds number and the ratio ΔT/ΔT_ml as KPIs because the first one reflects the flow conditions in the heat exchanger while the ratio provides information on the heat transfer phenomena. The developed forecasting model, a time series multiple regression model, was capable to predict the KPI one-month ahead with a testing accuracy of R²=0.7. Furthermore, we showed that long-term forecasting is also possible with this model, always with the caveat that the furthest we look into the future, the more uncertainty the forecast carries. Nevertheless, the model can still be applied for process optimization and maintenance scheduling.

In the future, we believe it will be opportune to further develop the model as well as the optimization framework by including uncertainty quantification on the inputs and in the model. In this way, more robust decisions can be made that consider process safety, reliability, and profitability.

References

Cavitt SB. Ethylene oxide production. US Patent 4,374,260 (Feb. 15, 1983).

Crocco L, Glassman I, Smith I. Kinetics and mechanism of ethylene oxide decomposition at high temperatures. The Journal of Chemical Physics. 1959;31.

Diaz-Bejarano E, Coletti F, Macchietto S. A model-based method for visualization, monitoring, and diagnosis of fouling in heat exchangers. Industrial & Engineering Chemisty Research. 2020;59.

Hess L, Tilton V. Ethylene oxide-hazards and methods of handling. Industrial & Engineering Chemistry. 1950;42.

Montgomery DC, Peck EA, Vining GG. Introduction to Linear Regression Analysis. 5th Edition, John Wiley & Sons, Inc., Hoboken, New Jersey, 2012.

Sundar S, Rajagopal MC, Zhao H, Kuntumalla G, Meng Y, Chang HC, Shao C, Ferreira P,
Miljkovic N, Sinha S, Salapaka S. Fouling modeling and prediction approach for heat exchangers using deep learning. International Journal of Heat and Mass Transfer. 2020;159.

Trafczynski M, Markowski M, Urbaniec K, Trzcinski P, Alabrudzinski S, Suchecki W. Estimation of thermal effects of fouling growth for application in the scheduling of heat exchangers cleaning. Applied Thermal Engineering. 2021;182.