British technology company MatAlytics has received £619,000 in public funding to test artificial intelligence in steel production. The system is designed to calculate rapidly how steel heats up and changes inside an industrial furnace. The aim is to reduce gas consumption without compromising product quality.
The funding comes from the UK government’s innovation agency, Innovate UK. It will be used to develop and test the company’s CITRUS software under conditions intended to reflect real steel production.
MatAlytics previously received £100,000 from Innovate UK to develop an initial version of the technology. The latest award is intended to move the project from a technical proof of concept towards industrial trials and validation.
The company grew out of research at the University of Nottingham. Its technology is based on what is known as physics-based AI. This means that the system is trained using established physical principles and results from advanced computer simulations.
It is therefore not a conventional chatbot. The software does not produce text or answer general questions. Its purpose is to calculate what happens to steel when the material is exposed to high temperatures and strong mechanical forces.
The furnace is crucial to steel quality
Before steel can be rolled, large semi-finished pieces known as slabs or billets are heated in a furnace. A slab can be described simply as a thick, heavy block of steel that will later be pressed and rolled into products such as steel plate.
The steel must reach the correct temperature before processing begins. Heating the surface is not enough. The required temperature must penetrate the material evenly and in a controlled manner.
If a slab remains in a furnace longer than necessary, the plant consumes additional fuel. This increases both operating costs and emissions. Excessively long heating times can also create a bottleneck that limits the amount of steel the mill can produce.
Removing the steel too early creates a different problem. Parts of the material may still be too cold or may not have developed the properties needed for the next production stage. This can cause difficulties during rolling and, in the worst case, affect the quality of the finished product.
Steelmakers therefore need to find a careful balance. Heating must continue long enough to achieve the correct temperature and material properties, but not longer than necessary.
Engineers currently use advanced computer models to calculate how heat, pressure and movement affect steel. One widely used technique is known as the finite element method.
In simplified terms, the computer divides a physical object into a very large number of small sections. It then calculates what happens in each section when the steel is heated, cooled, bent or placed under pressure.
These simulations can provide detailed information about temperature, internal stresses, deformation and the risk of damage. The disadvantage is that highly detailed calculations can take several hours or even days and may require substantial computing power.
This is acceptable for planned engineering studies. It is less useful when operators need to adapt production quickly to a different steel grade, slab size or furnace temperature.
AI could provide results within seconds
MatAlytics’ proposed solution is to train a neural network using results from a large number of conventional simulations. A neural network is a computing system trained to identify complex patterns in large volumes of information.
Once trained, CITRUS should be able to make a new prediction without repeating the entire time-consuming simulation from the beginning.
MatAlytics says calculations that normally require hours or days could instead be completed within seconds. This could make the technology useful when a steelworks is planning production or adjusting a process already underway.
The system could, for example, calculate how long a particular steel slab needs to remain in the furnace and what temperature is required. If heating time can be shortened without reducing quality, the steelworks could use less natural gas and increase its production rate.
The AI model is also intended to predict changes in the steel’s microstructure. Microstructure describes the arrangement of the material’s extremely small crystals and other internal components.
This structure cannot be seen with the naked eye, but it has a major influence on the steel’s properties. It affects characteristics such as strength, hardness, toughness and resistance to wear.
The microstructure changes as steel is heated, cooled and mechanically processed. Two pieces of steel with the same chemical composition can therefore develop different properties depending on how they are treated during production.
If CITRUS can predict these changes accurately, its value could extend beyond energy savings. The technology might also support the development of new steel grades, production troubleshooting and quality control.
Industrial performance has yet to be demonstrated
The promise of results within seconds remains a performance claim from MatAlytics. The new project must demonstrate whether the system can also produce sufficiently accurate and dependable predictions in a working steel-production environment.
An AI model depends heavily on the information used during training. If it encounters a steel grade, furnace or manufacturing process that differs substantially from its training cases, its calculations may become less reliable.
Plant data must also be accurate. Incorrect temperature readings or incomplete information about the size and composition of the steel could result in unsuitable recommendations.
Speed alone is therefore not enough. Engineers must be able to determine when a result is reliable and when it needs to be checked using traditional calculations or physical testing.
Clear safety and responsibility procedures would also be required if the software were connected to furnace controls. The plant would need to establish who reviews the recommendations and who makes the final production decision.
Integration represents a further challenge. Sensor readings must be collected, organised and supplied to the model in the correct context. Its output must then become part of existing operating procedures rather than remaining on a separate screen that employees rarely use.
The possible savings will also depend on the flexibility of each plant. A faster calculation cannot reduce energy consumption if the furnace, production schedule or downstream rolling mill cannot respond to the recommended changes.
MatAlytics also identifies possible applications in power generation, nuclear and fusion facilities, hydrogen infrastructure, aerospace, automotive production and defence. These sectors face similar requirements to understand how materials and structures respond to heat, pressure and mechanical stress.
The project must now establish whether physics-based AI can deliver more than fast and technically impressive calculations. Its industrial value will ultimately depend on whether it can produce measurable energy savings while maintaining consistent steel quality and safe plant operation.
Sources: MatAlytics, Innovate UK, the Institute of Materials, Minerals and Mining, and Industrial News.
