The latest news and updates from Imubit.

Complexities of Applying AI in Refining

Dec 1, 2020

Imubit blog header 12.1.2020

As industry increasingly recognizes the benefits of artificial intelligence (AI), companies are seeking how to best fit AI into their business strategy. AI, however, is relatively young and the extensive benefits it provides demand creative solutions to unique challenges. Successfully implementing AI requires a deliberate and dedicated approach that sometimes differs from implementing more standard technologies. The most expedient, cost-effective, and secure option to control refineries and chemical plants with artificial intelligence is to utilize a team of experts that are accountable for the end-to-end project success and its maintenance.

Developing skills in data science and machine learning is time consuming and typically requires training that has only become mainstream over the last two decades. A recent study by MIT Technology Review found that 45% of senior business and technology decision makers believed their organizations lacked the skilled personnel needed to implement AI. New applications frequently require specialized experts to transition the technology from theoretical to practical.

Another distinct challenge to introducing AI is that it requires a completely different data architecture than traditional business strategy analytics. Even if a company has access to the immense amounts of data required for AI, significant “data cleaning” efforts might be required to make the data useful. The same study by MIT Technology Review found that 42% of senior business and technology decision-makers believed a lack of quality unbiased data is the greatest digital barrier to AI adoption, while 38% believed that the greatest barrier is a lack of the required infrastructure.

When it comes to the oil refining industry, adopting AI has specific data-related challenges that must be solved:

• Refiners commonly adopt a practice of “data compression” which essentially deletes data deemed inconsequential and only stores data that represents more important events. The discrepancies caused by “data compression” can be detrimental to AI algorithms and need to be carefully evaluated by data scientists and process engineers.

• Standard data historians – such as PI (Process Information) – are not built to effectively export years of minute-by-minute data in a format required for AI.

• AI models can optimize broader (and more profitable) problems and sometimes require data from multiple sources. Merging and organizing data that is typically stored in different formats and with different tag IDs, requires specialized knowledge and tools.

• AI models are inherently challenging to understand and require robust and specific tools to visualize and comprehend.

The difficulties associated with AI are magnified when the technology is applied to a complex process. This is especially the case in refining and petrochemicals, where years of technical industry experience are required to efficiently run a plant. In this case, a team of people with domain expertise integrated with a team of data scientists are required to properly incorporate process knowledge with AI models. For more commentary on this subject, see our previous blog post, The Right AI Solution: Vertical Domain-Driven or Generic Industry 4.0?

AI applications often unlock opportunities that are significantly more valuable than previously thought possible. In these cases, the costs associated with the required consulting and maintenance is insignificant compared to the opportunity. Changes in plant configuration or operating strategy are addressed quickly if one team of experts is accountable for the end-to-end value, including on-going maintenance. There are also indirect savings that can be achieved by using a dedicated team instead of in-house controls engineers, who are stretched in multiple directions and typically have limited capacity.

Capturing the enormous potential of AI requires solving a series of several complex challenges, and the current state of the technology justifies teams of dedicated data scientists collaborating with refinery industry experts. Over years of iterative learning, Imubit has developed robust and sometimes specialized solutions to these challenges. As new challenges arise in the refining operations, our team of refinery experts, data scientists and software engineers are not burdened by the competing priorities associated with running the plant, and additionally understand the complexities and implications. They are therefore able to work immediately to find a robust solution to any changes required. Imubit utilizes a full-service subscription model that makes us accountable for ensuring the AI application is maintained and is continuously optimizing your plant to maximize profit on a 24/7 basis.

For more information, visit

Unlock your plant's Untapped Value.

It all starts with Imubit.