CHALLENGE
MAINTAINING HIGH QUALITY AND YIELD IN PRODUCTION
A glass product manufacturer had been looking for ways to increase manufacturing quality and the amount of non-defective products throughout the manufacturing process. In order to achieve this goal, the quality level should be visible to all machine operators and other stakeholders.
Unikie understood the quality level of a manufactured glass product varies depending on the human operator in charge of operating the production line, as well as on the glass batch.
PROCESS
Managing manufacturing yield and consistency
At the start of the project, Unikie and the client made an assessment of the current situation containing aspects such as:
- Existing process
- Data connectivity to equipment in use
- Available manufacturing data sets
- Sensors in use
- Product and batch information
Based on the information from the assessment and understanding of the key quality parameters, Unikie proposed how to minimize the impact of human operators and differences between glass batches.
No installation of new sensors, nor changes to the manufacturing process were needed during the process. Existing equipment parameters and product attributes (e.g. glass type, thickness, quality measurement) were processed with Unikie’s framework for real-time sensor fusion powered by AI. Machine learning was also used to understand existing parameter correlation to output quality.
SOLUTION
Optimized process control, fewer non-defective products
The solution is based on a data and complex ML-based digital twin of the production line that is used for predicting quality and generating recommendations on optimal machine settings. The project had two major outcomes for the client:
Realtime performance dashboard
AI-based operator assistant
While the realtime performance dashboard provides the machine operators an overall view of the process quality and yield, the AI-based Operator assistant provides them real-time suggestions of possible process parameters before a manufacturing of a new batch is started. Operators can then adjust the process parameters accordingly to produce optimal quality and yield regardless of their experience level.
The solution was taken into use in a single factory where the amount of defective and failing glasses decreased even over 20% compared to the level before implementing the solution, increasing the overall yield in the same proportion. Based on the results, the solution is being rolled out to several other manufacturing sites.
Key results of
the project
- Real-time monitoring of the holistic production performance
- Predicting production bottlenecks and quality issues
- Identifying trends and forecasting manufacturing yield
- Providing recommendations of optimal process parameters
