Glas Trösch: predicting glass quality while the pane is still being coated

The coating process had always been controlled the way experienced operators control a process they know well, by reading the line and adjusting. The limitation was not skill. It was timing. Quality measurement happens at the end of the line, so a problem is identified after the panes carrying it have been coated. The correction is then applied to the next batch, and the material already affected is waste.

The question Glas Trösch brought to us was whether artificial intelligence could be applied to an established process at all, given that their research team worked with material scientists rather than data scientists.

ThirdEye Data built an open loop system that runs alongside the production line. It takes live data from the process, predicts the end-of-line quality values the current run is heading towards, and puts that prediction, together with suggested adjustments to the process parameters, in front of the operator through an interface on the shop floor, while the panes are still being coated.

Open loop was a deliberate choice rather than a limitation. The system advises and the operator decides. In a process where a century of production knowledge sits with the people running the line, a system that overrides them would have been harder to trust, harder to certify, and worse at its job.

Operators act on a predicted quality problem while the batch is still in process rather than discovering it at the end of the line, which moves quality control from inspection to prevention and reduces waste.