Celsian Glass & Solar B.V. is a leading glass operations consultancy company. They spin out of TNO Glass Group, built on multiple years of high precision glass moulding at Royal Dutch Philips.
Glass manufacturers use the Celsian modelling software to optimize their glass melting processes. When a furnace is designed and deployed, most settings go precisely according to the process program. However, subtle changes in the glass production process lead to quality issues such as blisters in the eventual product which for example in case of bottles can cause them to burst later in the filling process. To assure a high quality product, there are multiple quality checks in the final stage of the production. However, the challenge is to prevent the occurrence of faulty production from the beginning of the process.
After learning about the challenge, Machine2Learn took the advantage of the availability of large, though incomplete database of historic sensor readings from a glass furnace. An anomaly detection model was developed that links the occurrence of blisters to sensor readings 24 hours earlier in the process.
Machine2Learn algorithm proposes the corrections in the furnace variables that results in 20% improvement in the quality of the final production.