Deep Learning-Based Control Optimization for Glass Bottle Forming

TL;DR

This work addresses the challenge of autonomously controlling the glass gob forming process by proposing a fully data-driven framework that combines a forward regression model with a model-inversion algorithm to optimize Cam deadpoints in real time. Trained on historical plant data, the regression model maps the current machine state and per-section adjustments to gob properties, while the inversion step searches for deadpoint configurations that achieve target weight $W$ and length $L$, enforcing cycle-wide continuity. The approach demonstrates practical feasibility on multiple production lines, achieving reasonable regression accuracy and stable, interpretable inverse solutions with notable reductions in setup time and waste, albeit with limitations in extrapolation beyond the training domain and in regions of sparse data. Overall, the method shows strong potential to augment operator expertise, improve process stability, and move toward smarter, data-driven glass manufacturing processes, while suggesting avenues for physics-informed extensions and richer feature integration.

Abstract

In glass bottle manufacturing, precise control of forming machines is critical for ensuring quality and minimizing defects. This study presents a deep learning-based control algorithm designed to optimize the forming process in real production environments. Using real operational data from active manufacturing plants, our neural network predicts the effects of parameter changes based on the current production setup. Through a specifically designed inversion mechanism, the algorithm identifies the optimal machine settings required to achieve the desired glass gob characteristics. Experimental results on historical datasets from multiple production lines show that the proposed method yields promising outcomes, suggesting potential for enhanced process stability, reduced waste, and improved product consistency. These results highlight the potential of deep learning to process control in glass manufacturing.

Figures (15)

• Figure 1: Schema of the hollow glass forming machine (left) and focus on the four-step gob feeder process (right) glass_book.
• Figure 2: Example of a plunger movement profile within a cycle. Each Cam is constrained by its neighbors to comply with the mechanical constraints of the apparatus.
• Figure 3: Distributions of the Weight-Length working points of the available lines.
• Figure 4: Distributions of the differences, in term of Weight and Length, between the sections within a same production. Hereafter, such variations will be used as targets by the regression model to learn general gobs transformations.
• Figure 5: Simulated inversion on a sample cycle. The image reports the convergence progress observed on one of the eight sections for both the targets and the inputs. The transformation requested is random in the range [-40g, +40g] for the weight and [-20mm, +20mm] for the length.