2024 AIChE Annual Meeting

(375v) Machine Learning, Modelling, and Simulation for Problem-Solving in the Hydrogen Production Process

Authors

Morales Medina, G. Sr. - Presenter, Universidad Industrial de Santander
Colombian economy relies on fuel sector income. A fundamental process in the clean-fuels chain corresponds to the steam methane reforming (SMR), which generates the hydrogen needed for production of low-sulphur fuels. The identification of opportunities to increase hydrogen production involves the analysis of variables that affects. This document presents the main results of an analysis of heat supply in an industrial SMR furnace based on both, data analysis and simulation with Aspen HYSYS. To such end, eight-year-process-operation data were collected and analysed with kmeans multivariate and random forest algorithm. The simulation was used to calculate the Duty´s in the reformer and Preheater train and was validated with design data and compared to process data. Next, the Random Forest Algorithm explore the main variable in H2 production and Data Analysis show the deviations process in timeline. According to the combined results of the AI techniques and process simulations applied, it was possible to identify the root causes of bottlenecks in H2 production, and based on this, to mitigate them. Finally, the results obtained in the H2 production and economic analysis are shown in
  1. INTRODUCTION

The operating efficiencies of industrial SMR units are limited in EOR operation, mainly due to catalyst and preheater train equipment fouling (Janbarari & Najafabadi, 2023). This fouling causes the flue gases to remove the heat available for the reforming reactions and for it to be discharged into the atmosphere at an elevated temperature, thus decreasing the efficiency of the process. Unfortunately, the maintenance of the preheating train and the change of the catalyst in the reactor require a total shutdown of the unit. The dependence on fuels makes the reforming operation critical for the economic and social sustainability of the country, restricting the unit shutdowns to the minimum possible. Therefore, the extended operation time, and the increase in H2 generation in EOR condition are required, while the respective supply plans are being executed.

One possibility for the operation of SMR units in EOR conditions - maintaining methane, steam and fuel flows and considering heat transfer restrictions - is the identification of conditions that lead to an increase in H2 production. Such identification of conditions can be based on the analysis of operating data and on the results of a simulation for the furnace-train preheater system.

    DATA ANALYSIS

    Process data for a reforming furnace and its preheater train at an SMR industrial unit of a national refinery were collected for an eight-year operating window. Samples with outliers were eliminated by interquartile range (Abebe at al., 2001). Also, the database was analyzed with the kmeans clustering method for the identification of schemes or operating modes representative of the unit. The different performances in H2 generation obtained with the identified operating modes were validated by means of ANOVA and F statistical tests, taking as H0 the equality between means and the ratio between variances equal to unity, respectively (Abebe at al., 2001; Box et al., 2008). The statistical procedures were applied according to the codes of the open-source R program (R Core Team, 2021) and its Rcommander package (Fox, 2017).

    SIMULATION OF THE REFORMING FURNACE AND PREHEATING TRAIN

    The reforming furnace and the preheating train of the SMR unit were simulated in the Aspen HYSYS v.10 program. The geometrical aspects of the equipment were taken from the respective specification sheets. Likewise, the operational conditions for the simulation were established according to the design condition. The thermodynamic package named Peng-Robinson was chosen due to different literature supports (Vlădan et al., 2011; Ehteshami & Chan, 2014; Challiwala et al., 2017). The reforming kinetics was established according to the Xu & Froment model for chemical equations (1-3), predominant in the furnace conditions (Barelli et al., 2008; Rostrup-Nielsen & Christiansen, 2011). The steam/natural gas to reformed ratio was set to 3.33, while the air/fuel ratio corresponded to 3.7. The natural gas and steam streams were mixed generating a 54000 lb/h stream at 510 °F and 425 psia to feed the reforming reactor, defined as a PFR reactor. The PFR was detailed with a length and total volume of 40 ft and 400 ft3, respectively. The PFR object was thermally connected to a conversion reactor. This reactor develops the burning of a 34000 lb/hr fuel flow at 1900 °F and 20 psia, supplying ca. 80 MMBTU/hr to the PFR reactor. Hence, the PFR and conversion reactor objects are thermally coupled, representing the operation of the reforming furnace. This coupling has been used in various literature works (see Posada & Manousiouthakis, 2005; Fan et al., 2016; Shalliwala et al., 2017; Amran et al., 2017).

    VALIDATION OF THE SIMULATION

    The simulation developed was validated using samples of the process data for each operational scheme explained in the statistical analysis (Section 4.1). The selection of the data was carried out according to the following formula for the calculation of the size of a finite sample (Valdivieso et al., 2011)