Plants that make fuels, chemicals, cement and metals are expected to use less energy and emit less while they keep producing. New process technology and capital projects take years to arrive. In the meantime, the energy a plant uses depends on how its existing units run from one minute to the next: how hard a furnace or kiln is fired, how much steam goes to a reboiler, how much power a mill draws. Running those units closer to their best operating point lowers the carbon footprint now.
Imubit's Deep Learning Process Control® (DLPC®) learns a dynamic model of each unit from its own operating history and uses it to control the unit's key parameters directly. It runs in closed loop on top of the plant's existing APC and DCS, writing setpoints minute by minute, inside limits your engineers set. The models can be trained to meet product demand and quality targets while they raise high-value yield, cut energy use, respect environmental limits and run renewable processes efficiently.
Yield and waste
When a unit runs below its best yield, the plant has to process more feed to meet the same product demand. The extra feed brings more low-value and waste streams, more energy per tonne of product, and more greenhouse gas emissions.
The models learn the actual relationships in the unit and control its key parameters continuously. The unit then makes more high-value product and less waste from each tonne of feed, and meets demand at a lower carbon footprint. Across Imubit deployments, that shows up as a 1 to 3% average yield improvement.
The models also pick up small changes in feed composition. That matters most in feed-flexible and co-processing units, where it lets the plant capture more of the economic and environmental value of each unit of throughput.
Oxbow, a producer of calcined petroleum coke, runs its rotary kiln operations in closed loop with Imubit. Higher yield means more calcined coke from the same kiln, and the kiln uses less natural gas while it meets product specifications. Oxbow's SmartKiln® decarbonizes through that yield improvement. Read the Oxbow story.
Energy efficiency
The models can be trained to minimize energy consumption and greenhouse gas emissions while they meet quality standards. They control the critical parameters to find the most efficient operating point for the plant, and they keep adapting as feed composition, fouling, catalyst activity and weather change. Where energy is the optimization target, Imubit deployments have delivered a 15 to 30% reduction in energy consumption. Fuel burned on site counts toward Scope 1 emissions, so firing less fuel for the same product lowers those emissions directly.
The main energy levers differ by industry.
- Chemicals and polymers. Running more efficiently, with less downtime and higher yields, takes less energy per pound of finished product. The models also learn how operating strategies drive energy consumption, which helps build carbon reduction into the case for future debottlenecking projects.
- Cement. A more stable kiln reduces the need to overburn clinker as a safety margin, and consistent clinker quality lowers energy use in calcination and grinding.
- Mining and minerals. Grinding is the most energy-intensive step in mineral processing. Its losses come from variable ore and the lack of a real-time particle size measurement in the mill. Holding particle size distribution on target lets the circuit push to the mill's power-draw limit, raise throughput and use less reagent in flotation. In mining, minerals and metals, Imubit's published results include 5 to 10% grinding-energy savings.
In refining, Preem, Sweden's largest fuel company, partnered with Imubit in 2025 to apply closed-loop optimization to the fluid catalytic cracking (FCC) unit at its Lysekil refinery. The aim is to reduce emissions and fuel use by improving the unit's yield profile and stabilizing the refinery's fuel gas balance.
Working with Imubit allows us to leverage cutting-edge AI to optimize our renewable processes while minimizing our carbon footprint.
Respecting environmental limits
Limits on emissions such as SOx, NOx and CO protect the people who live near a plant. Managing them over time is often a manual task that is not done continuously. Control is hard because the plant is interdependent and its process relationships are nonlinear. Limits are hardest to hold during transients, which are frequent in refineries and petrochemical plants. Plants then either violate environmental constraints or run conservatively to avoid violations, and the conservative margin costs efficiency elsewhere.
The models are trained to respect critical environmental limits in real time while they maximize margin. A limit can be adjusted, or calculated over a longer duration, so the plant runs efficiently while environmental deviations stay small. Because the models are dynamic, they predict key emissions variables more accurately through transients, which lowers environmental impact and makes control safer.
Renewable fuels and ethanol
Renewable processes add their own complexity: feeds that change in composition from one source to the next, co-processing with conventional feeds, and sensitivity to ambient conditions. The models account for changes in composition and ambient conditions and adjust the key handles of the plant to keep it efficient and profitable. They apply to continuous renewable processes such as ethanol production and renewable fuels.
The diagram shows two examples. In ethanol drying, two feed trains pass through dryers in series, with syrup added, and the controller moves each dryer's manipulated variables to hold product moisture. In renewable fuels processing, several feed sources pass through two reactors with hydrogen recycle and then a fractionator, and the controller moves the reactor handles to optimize the naphtha, kero and jet, and diesel yields.
For a closer look at one renewable process, read Renewable diesel optimization.
Start with where your plant stands
We start by mapping your highest-impact plan-to-plant gaps, including energy and emissions, and the operating decisions to model and control to close them.
