Machine Learning for NCC's Concrete Pile Production
In this thesis, the usefulness of machine learning (ML) is evaluated for the processes of NCC's subsidiary company Hercules. It is evaluated with regard to ML's ability to assist reduction of CO2 footprint and costs. The work comprises analysis of Hercules' processes and analysis of data from these processes as well as a search for an appropriate ML model for predicting compressive strength of concrete. Results show that gradient boosted trees through the CatBoost library is a suitable ML model. However, additional data is needed to develop any such an ML model that is t for use. A general example of how the CatBoost library can be used to predict strength of concrete is given. This example can be used as a starting point for future work on predicting compressive strength of concrete and for other ML problems at NCC. It was also found that Hercules' logistical system would bene t from further investigation. Short order times stresses the organisation and there may be a case for ML to improve the logistical system.