SEMINAR: Dr Maximilian Christ Event as iCalendar

23 November 2016


Venue: G10, 70 Symonds Street

Machine Learning for continuous steel casting

Blue Yonder GmbH, Karlsruhe, Germany

The computerization and automation of manufacturing in the so called industry 4.0 trend is promising to improve existing value chains and yield new business models in the industrial sector. The research project iPRODICT explores the partly automated adaptation and improvement of business processes during the continuous casting of steel billets at the German steel manufacturer Saarstahl.

Both iPRODICT and the industry 4.0 trend rely on the aid of Machine Learning techniques to make industrial operations smart. Often in those fields the statistical models need to incorporate changing states of processes and objects, recorded as time series. There are two ways on how to deal with temporal structured input for classification, regression, clustering and related tasks. Either the machine learning algorithm incorporates such time series directly or maps it to another, possibly lower dimensional, representation.

In this talk I will discuss an approach of the second type, the so called filtered feature extraction from time series. This approach and the respective Python package "tsfresh" allow for automatic extraction of a huge number of relevant characteristics from time series which can be used as input for the subsequent model. Such a features-based approach enables us to for example make predictions about the quality of steel billets in the iPRODICT application.