In a recently completed Vinnova- and PiiA-funded project, SCA together with Calejo Industrial Intelligence has managed to gain a better understanding and more accurate forecast of the use of water vapor. The project has the potential to lead to a more even energy consumption over time and a significantly reduced consumption of fossil fuel in SCA Obbola site outside Umeå. The result shows the great potential of using AI in industry.
In SCA's sulphate pulp mill in Obbola, water vapor is both a residual product and something used in the production process by, among other things, operating a turbine generator for electricity production. Usually there is a balance between need and supply of water vapor, but in case of disruptions in parts of production, a deficit of water vapor may occur. The imbalance that then arises can be regulated by either a reduction in electricity production or by creating water vapor through the combustion of fossil fuels. The latter is not very desirable.
Against this background, this Vinnova project produced an accurate forecast of the process's need for water vapor, as well as a digital twin over the process. The goal was to be able to understand how water vapor can be created in good time before the need arises - through a better understanding of the process - without reducing electricity production or burning fossil fuels.
Impressive results
The project shows that it is theoretically possible to reduce oil consumption in the bark boilers by means of self-learning AI technology and more accurate forecasts through a changed control of the soda boiler, the cookery and the bark boiler.
"This technology is applicable in several areas, but the just completed project is one of the few where the AI technology has been successfully tested in the global process industry," says Johannes Holmberg, CEO of Calejo Industrial Intelligence.
The model proves to be able to predict about 70 percent of the occasions when oil is used in the bark boilers. With an improved analysis and understanding of the process, a large part of the remaining 30 per cent can be eliminated.
“An optimization of the entire time period confirms the theory that it is possible to use energy in the process, so that the need for oil is greatly reduced. The 70 percent oil demand in the bark boilers, as the model can predict, will disappear completely after optimization. In addition, the evaluation clearly shows that the remaining 30 per cent of the oil consumption has occurred in addition to normal operation. This proves our original theory that steam can be produced without the addition of fossil oil in normal operation”, says Johannes Holmberg.
Several new projects are planned
The digital twin used in this project has been built entirely from a black box model, which means that it consists solely of neural networks that are trained on historical process data.
In the future, the model can also advantageously be converted to a so-called gray box model, that is, a model consisting of both trained neural networks and mathematical calculations. This will create an even better model understanding between direct and indirect process events, which will further increase the accuracy without the need for more data. A next step that is now being discussed is to test to using the model in order to provide operators with a better process support.
“A model that can predict and give indications on how the process should be managed is, from a productivity perspective, clearly interesting to us as plant owners. We are very pleased with the results and will continue with these experiences of AI in upcoming development projects”, says Magnus Viström, Innovation Manager at SCA.