Answers are grounded in this paper and cite their sources.
In fineblanking, a mass production process for often safety-critical components, maintaining high-quality sheared surfaces without tears across a tool maintenance cycle is essential. Current modeling techniques assume static process conditions and fail to capture the stochastic variations in real-world sensor data, termed process noise. This work introduces a framework for systematically attributing process noise, establishing an analysis chain from phenomena, through interpretable correlations, to latent driving factors. It combines quality evolution analysis, explainable predictive modeling (XGBoost with SHAP), and latent variable modeling (VAE variant). For 40,450 strokes across eight coils and 1,624 inspected parts at nine geometry features, regression on stroke count and a punch wear proxy shows features ranging from predominantly wear-associated ( R² ≈ 0.74 ) to almost entirely unexplained by these trend variables ( R² ≈ 0.02 ). For an aggregated tear indicator (relative tear frequency, mean 0.37), progressive coil-wise cross-validation on inspected samples yields a mean MAE of 0.108; errors peak at the first coil with exponential tear accumulation, a regime unseen in training. Restricting features to the shearing phase improves generalization and confirms physical plausibility. Two identified latent factors capture coil-related force offsets and progressive force curve changes, hypothesized to reflect material batch variation and wear dynamics. The framework offers a blueprint for physically attributing process variability in sheet-metal forming.
No takes yet. Share an insight, caveat, or question.
Unterberg et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: