Extraction of Interpretable Features from Temporal Measurements using Approximate Prototypes

Abstract:
Measured data in the manufacturing industry becomes increasingly diverse. To capture every detail in the production process, sensors provide locally and temporally resolved measurements, thus producing time series like data. In order to make this data usable, a feature extraction is mandatory. In this work, we firstly discuss time series clustering and prototype generation in the context of real world data from a tube production process. Secondly, we present a method to convert the distances between time series and the previously calculated prototypes to meaningful similarities. This enables us to infer process knowledge of domain experts to unseen data.