Objective

In this article Optimize Global Solutions presents a thorough review of how to quantitatively model key refinery processes for the crude distillation unit with an objective to balance the business needs and technology tools via the use of data science, machine learning, deep learning methods to provide an insight into operational agility to ensure successful and full-scale transition towards industrial AI.

Introduction & Overview

Refinery modeling guides the optimization plan to maximize margins and reduce environmental impacts. Process adjustments can be tuned in the model, which means there is no need to risk production exploring different operating settings. Planners explore a large number of different scenarios before bringing the changes to the real world, allowing for much wider experimentation. If the refinery is operating as a SMART REFINERY (AI-enabled process), the gap between the business needs and technological requirements will be largely gapped.

Refinery modeling is a type of process simulation. Producing a refinery model requires specialized process simulation software In this software, refiners create a dynamic digital copy of the refinery. Process Simulation Software holds a numerical representation of the refinery; different settings may be adjusted within the software, then the simulation produces realistic reactions and results. In short, a properly constructed refinery model will behave the same as the refinery being modeled.

One important caveat is that the model must be created using accurate data. Programming the refinery model with parameters based on the refinery’s original engineering designs creates a gap between the model and the real world. To create a truly connected and accurate refinery model, this form of process simulation requires live sensor data. By incorporating the actual operating conditions, technicians and operators will have greater assurance that the model they have created is accurate.

ML-based Modeling

By implementing Machine Learning-enabled technology in a thoughtful way that targets specific business needs, companies will gain the ability to optimize each critical asset. Embedded AI applications allow users to efficiently and successfully perform their domain-specific operations with increased accuracy, quality, reliability and sustainability of model throughout the asset lifecycle.

Prior to the development of refinery modeling, engineers and operators were limited in the amount of changes they could realistically deploy. Refineries require significant capital expenditures – and experimentation can put those investments at risk. Therefore, any changes to the refining process had real-world financial consequences, leading to a conservative approach.

With accurate digital models, the difficulty comes from the sheer volume of choices. Any one human can never have enough time to explore every combination of adjustments to a refinery model, but an artificial intelligence program can try hundreds of combinations at once. The AI can be programmed with different goals to achieve as it tweaks the model. Refinery optimization goals might be to increase margins, increase reliability by reducing equipment failures, reduce energy usage or find a balance of all of these demands. Specialized ML in oil & gas algorithms can provide recommendations and demonstrate them in the model for human operators to appraise.

Next generation, AI-powered applications augment the value of existing software solutions and help companies transcend functional silos within the enterprise to drive greater productivity, efficiency and reliability across their operations. There are some examples of how companies can leverage industerial AI solutions.

• A refinery could apply industerial AI technology to simultenously evaluate thousands of different scenarios to identify the optimum crude oil slate for processing. Coupled with the cognitive capabilities to improve decision-making and ease-of-use, the technology would free up planning personnel to focus on more strategic tasks.
• A process plant could deploy an advanced class of AI, namely Deep Learning model that combines machine learning and first principles to deliver more comprehensive, more accurate and more performant models

Conclusion

In light of tectonic workforce shifts and unprecedentend market volatility happening now, industerial organizations will need capabilities to drive business outcomes from industerial AI applications to remain relevant in the future. They will also have to implement semi-autonomous or autonomous systems to act on those outcomes, along with advanced decision-support capabilities to enable greater agility. More importantly, Machine Learningenabled solutions as operational excellence will facilitate the successful transition to the new business model that will be necessary to remain competitive. Process Industry companies need to adopt to a world where oil is increasingly used to produce chemicals and the need to recycle plastic waste is ever more pressing. Accelerated digitalization will be required to address these two disruptors, and industerial AI can enable that acceleration.