Data-driven glass viscosity soft sensor development and validation in a glass container manufacturing line

Abstract

Viscosity plays a key role in glass container manufacturing, directly impacting product quality and consistency. To date, online measuring of this property during the glass manufacturing process has been both difficult and costly. This study proposes and validates a data-driven approach to develop a soft sensor for measuring glass viscosity.

This method employs data on the height of the rotating tube at the forehearth outlet, along with the corresponding glass temperatures. To validate the approach, viscosity estimates are applied to predict glass gob length. Analysis of over 70 production days across various operations demonstrates high predictive accuracy on a per-job basis, with $R^2$ and MSE values consistently above 0.80 and below 1 millimeters, respectively, and reaching $R^2$ over 0.95 for certain jobs. Here, job refers to the continuous production of a single type of container on the production line. Additionally, an aggregate model across all data achieves a predictive accuracy of MSE = 3.80 millimeters.

The proposed methodology offers a reliable means to monitor and control glass viscosity, enhancing production efficiency and product quality in the glass container industry.