ML-Pipeline for the Quality Assessment of Screwdriving Processes

Current quality assessment methods of screwdriving processes evaluate the final torque-angle combination only. Consequently, anomalies and errors during the screwdriving process may remain unnoticed. Furthermore, the cause of an error cannot always be determined. We present an ML-pipeline for the assessment of screwing operations based on time series data of the entire process. The analysis is performed using the publicly available AURSAD dataset. The transfer onto screwdriving processes of the commercial vehicle production is likewise discussed. We find that the ML-pipeline performance is highly dependent on the combined consideration of preprocessing methods and model training.