Abstract Facing the increasing amount of available data, supervision of processes is experiencing a vast upheaval. Especially time series recorded during high-speed manufacturing processes like shear-cutting challenge the interpretation of… Click to show full abstract
Abstract Facing the increasing amount of available data, supervision of processes is experiencing a vast upheaval. Especially time series recorded during high-speed manufacturing processes like shear-cutting challenge the interpretation of the data. This work shows how to extract features from shear cutting force data that help to explain process variations. The ability to predict the product quality based on these features, however, plays a decisive role. Here the classic approach of feature engineering, in which features are selected using domain-specific knowledge of the engineer, is compared to statistical feature extraction which only bases on the actual process data. The use of these features aims at identifying the process state and product properties using predictive models. Both feature extraction methods are applied on force data and evaluated empirically in three different shear cutting processes. It turns out that both methods perform similar but differ in the presence of measurement uncertainty. Although simple prediction models have been used in this study, the features provide an excellent basis for predicting process or product properties.
               
Click one of the above tabs to view related content.