Back

Question

Submit an Answer

What role does artificial intelligence (AI) play in revamping downstream facilities that are scaling back on conventional fuels production while upgrading to capture value from new products?

Apr-2024

Answers

Ujjal Mukherjee, Lummus Technology,

AI, an integral part of a digital strategy, can be used effectively in retrofitting downstream units. We have a joint venture with TCG Digital, called Lummus Digital, which leverages AI-based platforms such as tcgmcube. We couple this with rigorous first principle-based process technology tools to take available data, quickly create digital twins, and develop unique hybrid solutions. These AI-driven digital solutions can be adjusted within equipment, product specification, and utility constraints to minimise energy and CO₂ emissions while maximising production of the most valuable products. These techniques are applicable to optimising existing operations, designing new plants, or evaluating potential retrofit options.

Once the products have been maximised within existing constraints, new solutions are developed with retrofits. This can include the change-out of catalyst systems, the addition of new equipment, and sometimes entire upstream or downstream process technology. All of these impact the overall product slate while remaining within the turndown constraints of the base complex.

Examples of such strategies include:
• The introduction of biofeedstocks and waste plastic- derived pyrolysis oils to existing refineries or petrochemical complexes.
• The elimination of gasoline production while maximising jet and diesel production, high-sulphur fuel oil production, and high-sulphur coke production.
• Increasing chemicals production from 10-30% and sometimes 50% production of needle coke and anode coke from an existing coker, which is a dramatic shift in crude slates and more.

The AI digital twin assesses plant performance under various configurations and integrates with economic and capital cost models to evaluate the full impact of the revamp considered.

Apr-2024
Philippe Mege, Axens, Philippe.MEGE@axens.net

Proposing a client upgrade a refinery plant from a conventional market to new products is generally based on consumer preferences, market trend analyses, and regulatory changes. AI can use machine learning algorithms to analyse historical data and make predictions about future market trends. It can also help by running different market scenarios to assess potential risks and associated impacts while developing a new product. NLP can also be used to extract valuable information from unstructured data sources to get a better understanding of the market and, therefore, stay ahead of the trends.

Apr-2024
Bradley Ford, KBC (A Yokogawa Company),

The growing scarcity of skilled labour is impacting the performance of facilities worldwide. In fact, a study by Deloitte and The Manufacturing Institute reports that the manufacturing skills gap in the US alone could result in 2.1 million unfilled jobs by 2030, resulting in a projected cost totalling $1 trillion. As the industry pushes for optimisation, the infrastructure’s increasing complexity poses a challenge. KBC is observing the emergence of various types of AI technologies that are starting to address these challenges.

For example, process simulation technologies are prevalent at nearly all global assets, operating as process digital twins or online real-time optimisers. However, simulation models that reflect reality still require calibration from engineers. KBC now sees AI handling this critical task to:
- Monitor the asset and models
- Identify when calibration is lost
- Automatically recalibrate it.

The critical impact is allowing the available finite human resources to focus on higher-value tasks.

Looking into the next steps, generative AI’s capabilities are potentially game-changing in capturing organisational knowledge that is dispersed across silos, contextualising that knowledge, and allowing junior staff to use it for idea generation. Careful oversight is needed to prevent generative AI systems from ‘hallucinating’ or producing theoretical outputs that conflict with the data on which the algorithm has been trained. Hence, training programmes are required to educate staff on how to leverage generative AI to create ideas for improvement, which still requires peer reviews before implementation.

Apr-2024