Machine learning

The main objective of the research in this work package was to investigate how to apply design data-driven analysis using Machine Learning (ML) methods and algorithms to predict future machine performance and to support human intelligence and learning to further digital transformation processes.

Upon looking into possible cases together with the partner companies it was decided that the relevant candidate for a first machine learning Proof of concept (PoC) implementation consisted of building a ML solution for a preemptive quality control of Computer Numerical Control (CNC) manufactured parts. The CNC-equipment was already fitted with thousands of interesting measuring points (sensors) and Industrial Internet of Things (IoT) connectivity capabilities.

The machines had already been connected to the first version of an IoT data platform and pipeline. Furthermore, this was one of the highest volume components that the partner company manufactured, which is important since ML methods require large amounts of training data. The data feeds from the machines into the IoT-platform consisted of manufacturing logs and sensor readings. As expected, the data needed cleaning, it was also a sparce set where some values were missing and had to be patched and complemented. The big-data research resulted in an increased understanding of the collected data and methodology as well as lessons learned and recommendations for enhancements.

An extensive systematic literature review was conducted overlapping all work packages of the project to towards a research agenda for industrial work-integrated learning. Findings show that predictive maintenance and predictive quality was, combined, the largest area found in the literature. However, research covering the human-centric aspects of it was, comparatively, non-existent.