A framework for decomposing process noise in fineblanking

In fineblanking, a mass production process for often safety-critical components, the objective is to achieve and maintain high-quality sheared surfaces without defects such as tears across as many processing cycles as possible. Fineblanking relies on modeling techniques that assume static process conditions. Consequently, these models struggle to capture the stochastic variations observed in real-world measurements from in-situ sensor systems, which typically involve substantial uncertain components, subsumed as process noise. This work introduces a systematic framework to decompose process noise into explainable and potentially unexplainable components, utilizing a comprehensive dataset, consisting of fineblanking process and quality data. Three complementary analytical approaches are employed: quality evolution analysis to track tear development over time, explainable predictive modeling to link force signal variations to tear formation, and deep latent variable modeling to uncover underlying drivers of process variations. The findings provide novel insights into the shearing dynamics under industrial conditions. For instance, the analysis reveals that tool wear drives distinct tear patterns across part geometries, linking systematic process variations to quality outcomes. This work offers a blueprint for decomposing process noise in sheet-metal forming processes.