Giner, J. (2019). Machine learning in manufacturing : an outline of machine learning fundamentals and concepts in manufacturing, validated by the implementation of a real-life use case in the production of a midsize company [Diploma Thesis, Technische Universität Wien]. reposiTUm. https://doi.org/10.34726/hss.2019.69703
E307 - Institut für Konstruktionswissenschaften und Produktentwicklung
-
Date (published):
2019
-
Number of Pages:
87
-
Keywords:
Maschinelles Lernen
de
Machine Learning
en
Abstract:
Machine learning has been applied successfully in recent years in applications in manufacturing such as monitoring of real-time data, detecting and interpreting patterns in data or as part of decision-making support systems. Industrial manufacturing is currently experiencing an unprece-dented trend towards digitalization, commonly referred to as Industry 4.0 and machine learning, as an aspect of artificial intelligence, is believed to play a decisive role in it. The work is divided into a theoretical and a practical part. In the theoretical part an outline of machine learning with a special focus on machine learning in manufacturing is given. Different learning types are explained and different classes of algorithms are described and summed up in an illustrative overview. Challenges and opportunities of machine learning in manufacturing are identified as well as concepts to facilitate the execution of machine learning projects. Finally, the current state-of-the-art of machine learning in manufacturing is investigated. In the practical part results and concepts found in the theoretical part are tested and proofed on their applicability in a real-life environment by implementing an unsupervised outlier detection for welding data. The machine learning project is carried out according to the cross-industrial standard process for data mining (CRISP-DM). In a first step welding data is analysed, pre-processed and fit to a standard-ized format. Subsequently, two different approaches for outlier detection models are implemented and compared, namely local outlier factor (LOF) and one-class SVM. To conclude, both parts of the work are summed up and main findings are pointed out. Furthermore, an outlook on possible further actions is given.