Process industrial machine learning

Important results from the project

The purpose of this feasibility study was to study process industrial applications of machine learning and Big Data analytics. The project studied data analysis of large amounts of experimental data and operating data, to evaluate the choice of tools and development of methodology. The project worked with examples, one from the petrochemical process industry and Haldor Topsøe, and one from the food industry and Skånemejerier. The purpose was to apply methods for machine learning and this has been done, especially with the packages Scikit-learn and Keras, with good results.

Expected long term effects

A large number of studies have been done with several different tools and methods for machine learning. The main result is the models based on WaveNet that provide good prediction, but the most important effects are the built-up competence and insight into how the technology can be used for the different types of applications that have been studied. Another unexpected effect of the project was that one of the case studies provided deep insights into the nature of the problem, which resulted in the problem being solved by redoing the process control system.

Approach and implementation

The feasibility study was based on two case studies with completely different problems and companies. Case study-based projects have the advantage that the companies can relatively easily generate in-kind in the project and that the results are directly applicable. But it turned out to be too far away in both corporate culture and in problem solving, so that synergies could arise between the different case studies in this project.