Áron Somogyi on Advancing Refinery Performance Through Digital Twins and Advanced Modelling

Áron Somogyi, Business Support Expert in Quality Control & Advanced Modeling at MOL Group, explores how advanced modelling is becoming central to modern refinery decision-making. He highlights the growing role of digital twins, real-time optimisation, and data-driven models in improving process accuracy, energy efficiency, and operational visibility, while emphasising the importance of user adoption and data quality in delivering tangible results.

Looking ahead, Somogyi sees advanced modelling as a key enabler for handling evolving feedstocks and increasingly complex refinery configurations, supporting more informed, agile operations through practical, low-CAPEX digital solutions that can be scaled across the refinery.

From your perspective, how is the role of advanced modelling evolving within refinery operations today?

In my experience, process simulation models, which have been used for a longer period, are also gaining more acceptance since more and more experts are working in refining who are familiar with process simulation tools and are aware of their strengths and limitations. The increasing amount of available data has also paved the way for the application of data-driven models in refinery systems, thereby enabling the modeling of processes that we could previously approach with much less accuracy. The development and increasing use of RTO, digital twin and other model-based decision support systems is also due to the rise of advanced modeling.

Many refiners are investing in digital twins and advanced analytics. In your experience, what differentiates projects that deliver real value from those that remain pilots?

Based on my experience so far, the most difficult part of any development in the field of digital twin solutions is identifying the right user base within the refinery and building their trust in these tools. Obviously, digital twin solutions initially raise doubts among users, but positive examples from their application that provide a sense of achievement can help build trust and faith that these tools are an advantage and that their daily use provides positive results. I believe that this can be the key to a successful digital twin implementation.

How can advanced modelling and AI contribute to improving energy efficiency across refinery operations? Where do you see the most immediate and practical opportunities for impact?

Using an accurate model can enable proper monitoring of product qualities and the current state of equipment, thereby helping to reduce quality giveaway and energy consumption. An excellent area to demonstrate this is the monitoring of heat exchanger fouling and thereby the detection of excess energy used – in most cases, monitoring heat exchangers does not require particularly complex models, the cost of excess energy due to fouling can be quickly detected, thereby optimizing the time and method of heat exchanger cleaning.

Predictive maintenance is often highlighted as a key application of AI. What are the real limitations refiners should be aware of?

The most critical point in predictive maintenance, as with any system using AI solutions, is the appropriate amount and quality of data. Many older refineries have a demand for predictive maintenance, but the lack of data due to poor instrumentation does not allow for proper monitoring of the equipment and the construction of an accurate predictive model.

Data quality and integration remain a challenge for many operators. How are leading refineries overcoming these barriers?

Leading refineries are overcoming data barriers by transitioning from siloed data architectures to a Unified Namespace to ensure a “single source of truth”. To improve data quality, they utilize advanced data reconciliation, which uses physical laws like mass and energy balance to mathematically correct sensor inaccuracies. Furthermore, digital twins and soft sensors can be deployed to simulate process conditions in environments where physical instrumentation is difficult or impossible to maintain.

How do you balance model complexity with usability for operations teams? What ensures that insights are actually used on the ground rather than remaining theoretical? 

During the development of digital twin applications, it is important that the system is designed according to the needs of the users, since these systems are made for them, so we definitely strive to make the system understandable, transparent and easy to use for them. To this end, during each digital twin development, we regularly ask for feedback from them on whether any additions are necessary, what parameters they are interested in, whether the built models are sufficiently accurate, or whether the parameters and objective function used during optimization have been selected appropriately. If the usage of such a system is clear to users, it can be integrated into their everyday activities with greater success.

Looking ahead, what role do you see advanced modelling playing in supporting new feedstocks and evolving refinery configurations? 

In recent times, especially in Central Europe, great emphasis has been placed on comparing crude oils purchased from different markets, and refinery models are also of great help in the selection process. With the help of properly constructed models, it is possible to examine what product yields and energy consumption can be achieved for different crude oil types, which can significantly facilitate their subsequent selection. Advanced modeling methods, such as hybrid models built for refinery reactors, can describe with greater accuracy than before how the parameters of the products obtained during conversion technologies change in the case of new type of feedstocks.

If you had to prioritise one digital or modelling initiative for a refinery operating under tight budgets, what would it be and why?

Any digitalization development used for optimization purposes that has a low CAPEX and can achieve quick results. For example: There is a distillation column, where the quantity of impurities in the overhead product is lower than the maximum allowed value. With an accurate model, it is possible to examine how much the amount of energy used can be reduced of the product yield can be maximized if we reduce the quality giveaway. Such a decision support solution, if the plant is properly instrumented, does not require high CAPEX cost, the profit it achieves can be quickly demonstrated, and a larger benefit can be reached by combining several smaller such cases, which makes it especially effective for refineries working on a tight budget.

What are you most looking forward to discussing with your peers at ERTC: Ask the Experts this year?

I’m really curious about how other refineries planning digital twin development, what successes they achieved, what difficulties they experienced and how they solved them.