Operational AI for process industry

Purpose and goal

The project has developed a process industrial pilot installation with AI-based decision support that can provide continuous guidance to operators in real time. The decision support and its AI algorithms are implemented in a cloud solution and give users in the plant an idea of how the process should be controlled and the expected outcome during the next several hours. To understand the system consequences of the AI-based optimizations and for further energy optimization, models have been developed for prediction of average indoor temperature and energy consumption in buildings.

Expected results and effects

With machine learning, the project has succeeded in producing predictions of detailed energy use, distribution delays, and temperature losses, in the district heating network. Regarding process optimization and ML, there is a risk that the algorithms capture other patterns from training data than what you want and that there may be few data points on the parts where you want the process to be after optimization. Here we see needs in industrial applications of partly other AI methods.

Planned approach and implementation

A stated goal was to handle how AI solutions would be made more transparent. The supply temperature AI algorithm suggests which temperature to produce, but it does not answer why this is so. Through feedback from operators, a separate AI algorithm was developed with the task of making an explicit prediction of the process delay. This is not needed for process control, but increases the understanding of why the proposed temperature is the right choice at this time. An even greater participation from operators would have brought additional benefits.