Automatic time series segmentation and clustering for process monitoring in series production

Due to high expenses for data analytics and implementation of individual process monitoring applications, potentials for data-driven process optimization often remain unused. We present a transferable method for automatic preprocessing for characteristic current and acceleration sensor signals of production plants. The method includes semi-automated segmentation, feature extraction and clustering of high sampling sensor signals. The clustered segments enable interpretation by process experts for further applications. This procedure enables low-effort preprocessing of data and allows the extraction of relevant process information from raw signals for monitoring, trend analysis and anomaly detection. Evaluation is performed on a production process for coil springs.