Answers are grounded in this paper and cite their sources.
This paper presents a data-driven cycle-to-cycle quality control strategy for injection molding that automatically adjusts machine settings to maintain consistent product quality despite process variations. Unlike conventional approaches that rely solely on part weight, the proposed framework integrates inline visual quality inspection, specifically filling level and burr count, to capture defects that weight alone cannot detect. A neural network with a physics-informed monotonicity penalty models the nonlinear relationship between machine settings and multi-objective quality features. Two cycle-to-cycle machine-setting update strategies of increasing sophistication are compared: a gradient-based iterative learning controller (GB-ILC), and a norm-optimal iterative learning controller (NO-ILC). Experimental validation on an injection molding machine demonstrates that both controllers recover complete filling from suboptimal initial conditions within roughly 10–24 cycles, i.e. , within at most five controller updates. In a material-change experiment, the NO-ILC compensated the resulting filling loss through the quality feedback alone. Sustained complete filling, however, required a manual adjustment of the weight target, since the original target was inconsistent with complete filling for the new material. The experiments further reveal that identical part weights can correspond to different visual quality, indicating that inline visual feedback is an important step towards automated, zero-defect manufacturing.
No takes yet. Share an insight, caveat, or question.
Jung et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: