A Flare System Digital Twin for CO2e Emissions Inference

Summary

Flare systems are among the most important devices to achieve safe gas release in the hydrocarbon and petrochemical industries. A flare system, alternatively known as a flare stack, is a gas combustion device used in industrial plants such as petroleum refineries, chemical plants, and oil and gas production sites, which during a plant upset or emergency, deals with a flow of waste gas that must be burned completely.

As part of the carbon footprint reduction within zero routine flaring by 2030, an accurate and online measurement of exhaust emissions is necessary to identify the sources of hydrocarbons burned to take actions to reduce their combustion to as low a value as possible. Considering the extensive use of flare systems and their emission composition measurement requirement, a novel digital twin using artificial intelligence (AI) and engineering models merged in a hybrid modeling framework was developed to inference emissions in a flare system (Figure 1).

Methods for emissions estimation

A digital twin is a virtual representation of a system, a process, or an asset, which is developed using engineering models or real-time data. The hybrid modeling framework combines both the engineering models and real-time data from sensors using artificial intelligence (AI) /machine learning (ML) models to create a robust, consistent, and up-to-date view of a process. Since the physical phenomena of flare systems is complex, the hybrid modeling framework is a straightforward alternative to develop an accurate digital twin for emissions inference in a flare system.

Direct measurements, inference models, mass balances, and emission factors are some of the most used methods to estimate emissions of CO₂, methane, and other common greenhouse gases produced by a flare system. Direct measurements can be done using special infrared (IR) cameras or space imaging that can accurately determine the composition of gases produced by burning. This method requires a large investment in equipment, appropriate personnel to carry out the task, and is not usually done with enough frequency.

Inference methods using computational fluid dynamics (CFD) allow the online monitoring of the burned gases but require expensive hardware resources for computational power, requiring expert personnel to develop these kinds of solutions.

Oil and gas facilities monitor emissions either by using emission factors or mass balances. Even though these methods are simple to implement and widely recommended by different international environmental agencies, these alternatives have different assumptions such as certain operating conditions or high values of combustion efficiency. These may not always represent all the possible cases and can cause errors in the emission inference.

Digital twin input information

The following information is required to tie the virtual and the real flare systems (Figure 2) for modeling:

• Online measurement of flow, temperature, and pressure of gas to the flare.
• Gas composition to flare.
• Low heating value (LHV) of the gas to the flare.
• Wind speed.

Gas production plant emission inference

Results

The digital twin was allowed to identify the state of the emissions in the gas production plant. Figure 6 presents the hourly emissions of CO₂e from 1 August 2023 to 4 September 2023. The gas was emitting an average of 1 metric ton of CO₂e per hour, represented by the red line in the graph, with a minimum of 0.018 tons of CO₂e and a maximum of 4.92 tons of CO₂e. The average emissions were heavily affected by the spikes. Even though at the beginning of the analysis, it was supposed that the spikes represented the activation of safety valves, PSVs, etc., by further inspection it was found that the last security valve opening was three years ago, and that the emissions generated from the plant come from another source.

Final thoughts

The flare’s digital twin allows constant monitoring of the emissions in the plant. Being a tool for identifying opportunities to decrease emissions and to make the appropriate decisions, it serves as a soft sensor and is a solution that is much cheaper than the implementation of a physical sensor or a continuous measurement using cameras or even more complex inferences using CFD.

In the pilot project presented, this online monitoring helped to identify how the emissions are behaving, and through the analysis of the data and the interviews with the operators, a problematic circumstance was found that is increasing the emissions of the plant. The use of the digital twin as a soft sensor allowed the identification of this situation but also gave enough information for taking the appropriate corrective action.

Finally, an advanced process control (APC) optimization algorithm is in the implementation phase in the gas plant. When commissioned, the next step is using the flare system’s digital twin to obtain insights on how the optimization affected the gas production plant emissions if a reduction is obtained, or what additional measures must be taken.

_{References
1. Bahadori, A. (2014). Blow-down and flare systems. In Elsevier eBooks (pp. 275–312). https://doi.org/10.1016/b978-0-08-099971-5.00006-4
2. Wakil, W., “Minimize Flaring in Oil and Gas Facilities”, Chemical Engineering Progress (CEP), p. 22 (Jan. 2023).
3. What is a digital twin? | IBM. (2023). https://www.ibm.com/topics/what-is-a-digital-twin
4. Abbas, A., Guevara, V., & Romagnoli, J. A. (2002). Hybrid modeling: architecture for the solution of complex process systems. In Elsevier eBooks (pp. 409–414). https://doi.org/10.1016/s1570-7946(02)80096-7
5. Emam, E.A. (2015). GAS flaring in industry: An overview. 57. 532-555.
6. New and Revised Emissions Factors for Flares and New Emissions Factors for Certain Refinery Process Units and Determination for No Changes to VOC Emissions Factors for Tanks and Wastewater Treatment Systems | US EPA. (2022, December 19). US EPA. https://www.epa.gov/air-emissions-factors-and-quantification/new-and-revised-emissions-factors-flares-and-new-emissions}