PTQ Q3 2025 Issue

Plantwide AI optimisation and beyond

Case study shows how Big West Oil’s digital transformation journey is spearheading AI-enabled optimisation for margin improvement while training the next generation

Travis Legrande Big West Oil LLC Mitchell McCloud Imubit

A cross the US refining landscape, digital transforma - tion is well underway. Companies face mounting pressure to invest in new ways of unlocking value from the enormous amount of data generated by highly instrumented industrial processes. Continually decreasing data storage and compute costs are exacerbating these pressures. The digital strategies developed in response to this wealth of data have uncovered pathways to rethink unit operations, empowering operators and engineers with robust tools to make better, quicker decisions. These rich data streams can also be leveraged to automate process optimisation in new ways. Moreover, machine learning (ML) and artificial intelli - gence (AI) present opportunities to exploit the inherent nonlinearity of refining processes, far beyond the capability of traditional step-testing or linear model-based methods. Yet, many early adopters have found that scaling from pilot programmes to full-plant implementations can be difficult without clear strategies for governance, data quality, and organisational change management. Big West Oil LLC (Big West) operates a high-conversion 36 MBPD refinery located in North Salt Lake City, Utah. Its team identified the need for a modern process optimisation solution to enhance operational efficiency and help it stay competitive with larger oil and gas companies, benefiting from scale economies and vertical integration. Big West partnered with AI optimisation (AIO) provider Imubit for this targeted digital transformation journey, beginning with the development and deployment of AI applications targeting key refinery constraints and oper - ational pain points. What started as a means of doing closed-loop process optimisation quickly demonstrated potential in other areas. Beyond the direct economic ben - efits, the refinery operator has also capitalised on the AI models developed to build internal competencies, thereby training the next generation of operators and engineers. Challenges Initially, Big West put its primary focus on optimising the fluidised catalytic cracker (FCC) unit, which had historically been constrained by debutaniser throughput. It believed that alleviating the debutaniser bottleneck would allow operations to maximise conversion, capture higher-value product yields, and thereby improve the overall unit margin. Beyond the FCC, a broader solution to optimise its diesel pool was desired. This included reducing ultra-low sulphur

diesel (ULSD) flash giveaway, hydroprocessing reactor optimisation, molecular management improvements in the crude distillation unit (CDU), as well as correctly accounting for trade-offs between distillate hydrotreating feedstocks as seasonal specifications shifted. Beyond the economic objectives of closed-loop optimi - sation, the refiner also recognised the role that technology plays in attracting and retaining the new generation of operators and engineers. Chemical engineers and process technicians experience the best-in-class technology in uni - versity programmes, and then frequently find themselves in industry roles operating on technology paradigms of past decades. Introducing AI to all levels of refinery employees in the context of process optimisation demonstrates a com - mitment to keeping employees and facilities up to date on the latest technological advancements. Rapid deployment of this modern AIO technology over - came various organisational and technological hurdles along the way. These challenges fell broadly into the buck - ets of data governance and infrastructure, change manage - ment, and scale.  Data quality and governance High-fidelity historical data is crucial for training deep learn - ing models. Data compression, missing data, and infrequent setpoint changes can limit the model’s ability to capture the full range of useful process variability. Companies consid - ering this type of solution should consider de-compressing data where resources allow. Since training deep learning controllers requires learning the process relationships between signals, it is best to have many years of histori - cal data capturing different modes of operation in different economic cycles or feedstock incentives. This helps ensure that there is setpoint movement over time, which is key to creation of a robust data-based controller. If setpoints are stagnant over the full historised life of a unit, the AI model will fail to learn the full extent of potential process relation - ships. As Big West began exploring AIO solutions, it found that its historical data was compressed in a way that would hinder AI models from learning relationships. The team first had to overcome the hurdle of changing how data was stored before the project could move forward. v Resistance to change Despite the promise and proven success of AI solutions, there is inherent resistance to change when deploying any

PTQ Q3 2025