ProjectDeep process learning

Deep process learning

Swedish process industries produce enormous amounts of data and have a high automation degree, but many remaining challenges are hard to address using traditional analytics used in optimization and automation. Hence, the data multiplies, but the tools to use the information are lacking. Big data analytics have the potential to bring the next leap in productivity, quality, and automation to Swedish process industry. More specifically, Deep Learning introduces a great new opportunity to handle the challenges of process industry.

This project will aim to show how Deep Learning can be used to introduce a big leap for automation in process industry. This in turn will inspire the Swedish industry how big data analytics can enable a new phase in process optimization. The proposed approach will take advantage of the data already gathered in the process control system and use that to suggest the needed action to improve the desired KPI. The results can be used in the first step as a decision support for operators, and later completely automate the control of the process. The project will demonstrate the approach via a specific case in the pulp and paper industry, namely, predicting the running speed of a paper machine and paper qualities adjusted to particular fibre characteristics.

The goal of the project can only be reached by tightly bringing together competences from several different domains; process industry expertise from BillerudKorsnäs, frontier deep learning technology from Peltarion, the latest measurement technology from PulpEye, big data, cloud, and digital services user experience from SICS Västerås, and cross industrial know-how and networks from FindIT. The project will be executed in close collaboration with the strategic innovation program PiiA (processindustrial IT & automation) that has been an active partner in the pre-study phase and project preparation.