Yutime

AI can be utilised to fundamentally change the way plants in the downstream oil and gas and petrochemical businesses are built and operated, achieving true digital transformation. Until this year, AI projects in petrochemicals and refining have been limited in scale and scope to incremental change with the goal of assisting human operations rather than creating an AI operated plant. Now however, a true Smart Plant will be built in China driven by AI matched with the capability of a private 5G network to deploy sensors everywhere. The AI system uses both traditional forms of control data and new unstructured data; video, thermal imaging, 3D models. Overall plant start-up times will be reduced by two years, profit margins increased by 7–12 percent and emissions and waste reduced by 15–20 percent.

Historically, AI utilisation in these process industries has been limited. Owners are very risk averse; the capital cost of plants is three to twelve billion dollars (US) and few want to be the first to use (and risk failing with) new technology. Operating and maintaining a plant is more complex than for manufacturing plants: everything is interconnected, changing pressure or temperature in one processing unit effects other units in ways that are not obvious and not easy to model. Imagine trying to land twenty passenger jets simultaneously whilst all are joined together with steel cables. This is an incredibly hard problem to solve.

Holistic Design, Start-up, Operation and Maintenance

AI cannot be an afterthought to building the plant and AI cannot be perceived as an IT project. New ways of designing, building, operating and maintaining the plant have to be developed by a cross functional team working before and during the plant being built. The CIO’s team and engineering and operations have to work as an integrated team on the plant design and operations before it is built. The CIO’s team needs a detailed understanding of petrochemical plants functioning at theoretical and practical levels: deep domain knowledge is critical. The CIO’s team has to be involved beyond the data aspects and be staffed with experts in Distributed Control Systems, Plant Operations, Business and AI: IT needs to speak the language of all these groups. In turn, the team needs to teach Engineering and Operations the limits and benefits of AI.

The entire design process has to be digital for physical design, process design and automation design. Complex tools enabling this are required, and processes developed to understand and control change impact across tightly coupled equipment, chemical processes, controls and AI. This is critical, especially as vendors have no experience in building an AI run Plant; requirements have to be very precise.

Individual chemical processes require different AI approaches. Machine Learning is not a silver bullet for the process industry but rather one of many approaches. A nested approach is required with optimisation at control, process unit, and plant levels.

Project progress is mapped against the digital models. AI data requirements and limitations are fed into the physical design at the early planning stage, in turn AI and 5G allows the physical design to be simplified (e.g. fewer elevated walkways may be needed as AI performs work previously performed by operations staff walking around the plant). Detailed Digital Twins simulating the whole plant allowsfor plant optimisation before construction and running millions of scenarios on the simulations allows data to synthesised for AI applications before real data exists.

Typical plants have individual controls managed by a distributed control system that balances a single process; state-of-the-art plants have some processes enhanced with an Advanced Control System (ACS) that optimises and stabilise one process, often using historical data from controls systems and implemented years after the plant is first operational. A true Smart Plant builds on these previous technologies with ACS on all process units to first stabilise the plant; however, the Smart Plant then uses a complex integration of a mixture of AI approaches, hybrid models, physics and chemistry-based models and simulations to delivera full AI driven plant control. The AI balances across all the process units and performs trade-off benefits for each individual process against the overall desired condition. Instead of operators changing individual temperatures and pressures of hundreds of thousands of controls, the plant is run by inputting high-level business parameters such as the cost of raw materials and energy and customer demand.

The AI systems create a plant that is optimised automatically based upon total cost of ownership, as example, when demand and process are high the plant can be tuned for higher output if the increased operational and maintenance costs are lower than the additional profit. Through multiple levels of controls, these high-level drivers are translated in real-time into actions for over one hundred thousand individual sensors and controls.

AI for the Smart Plant is only the first phase, once implemented, further links to the inbound and outbound supply chain create an AI driven ecosystem. Ethane carrying ships will be optimised for fuel consumption and storage will be optimised based upon the plant’s raw material demand. Similarly, product pricing will be managed based upon the plant status to help balance plant supply with customer demand. Plant maintenance will be planned across the supply chain and expensive operational costs such as catalysts and energy can be optimised.