Rockwool
Predicting production quality issues
Machine Learning & AIManufacturing & Industrials
Key resultQuality-issue causes caught in the first production phase
Challenge
Understanding the causes of burn-mark quality issues meant tracing them as far back in the production line as possible — applying ML classifiers to time-series data with uncertain alignment due to volatility and product differences.
Solution
A set of ML classifiers implemented as a containerised solution that predicts quality issues and identifies their causes within the existing cloud framework.
- Automation. Runs as a micro-service in Docker on the client's Azure cloud, fed by an automated ETL pipeline.
- Early detection. Causes identified early enough to take remedial action — with clear cost and quality benefits.
- Product-aware. The solution adjusts for varying product characteristics (thickness, composition) and their impact on the line.
Results
- Significant ability to predict quality issues across product categories
- Causal factors identified as early as the first production phase
Involving us throughout the development phase with an active role providing feedback during the prototyping through test and refinement was needed in order to qualify use cases.
Topics
- Machine Learning
- Causal Analysis
- Azure