1. Motivation
Following the casting of aluminum die-cast parts, several tens of thousands of thermographic images of the mold, along with associated quality data (OK/NOK + defect type), are available.
Currently, the quality assessment only takes place after the complete process. An automatic, image-based early warning system could reduce scrap, optimize process parameters, and lower costs.
2. Objectives of the Work
- Proof of a statistically robust correlation between thermograms and quality labels.
- Development, training, and evaluation of a Deep Learning model for classification (good/bad) and, if applicable, multi-label defect detection.
- Determination of relevant image areas/features using Explainable AI (XAI) methods.
- Documentation of best-practice recommendations for the industrial partner.