lu.se

Matematikcentrum

Lunds universitet

Denna sida på svenska This page in English

Kalendarium

Examensarbete "Tillämpad maskininlärning i stålindustri" av Oscar Fredriksson

Seminarium

Tid: 2020-02-25 13:15
Plats: MH:309A
Kontakt: magnuso [at] maths [dot] lth [dot] se
Spara händelsen till din kalender


Oscar Fredriksson presenterar sitt examensarbete "Tillämpad maskininlärning i stålindustri - Undersökning av möjliga tillämpningar av maskininlärning för SSAB Oxelösund"

Abstract:

 In modern industry there are high demands for increased resource efficiency that in part are motivated by the stronger competition on the global market, as well as by the societal necessity for a more sustainable industry. This master thesis project examines applied machine learning as a potential solution for an improved resoruce efficiency in the steel industry. The thesis was performed at SSAB Oxelösund and involved an evaluation of the possibility of applying ''Predictive Maintenance'' and ''Quality Predicition'' with the help of available historical logged data. The method used consisted of collecting and connecting process data and target variables from different sources of data in the company and performing an initial analysis to determine whether a dataset for training of a machine learning model could be created. Then features were extracted from the raw data in the created dataset and used to train 30 different Random Forest Classifiers with different methods for feature selection and imbalance handling. The performance of the models were then evaluated with evaluation methods suitable for imbalanced datasets, for example with AUC. This thesis showed that only one of the examined applications of machine learning, ''Quality Prediction'', was implementable with available data at SSAB Oxelösund. The performance of the created models could be improved with the use of feature selection and imbalance handling. Mainly, this thesis showed that the quality of the data is crucial in applied machine learning. An improvement in the quality of the data and the introduction of new and relevant signals in the logging of data should be of high priority when machine learning is to be applied in practice.