Industrial Deep Learning
AI-based defect detection in part inspection using our annotation tool Catalyst
AI methods that work in real production systems - not just on benchmarks.
We develop deep learning methods for industry: robust under shifting process conditions, data-efficient when labels are scarce, and explainable for the people who act on them. We publish at the top venues of AI research and validate our methods directly in our partners' production environments.
Who we are
Industrial Deep Learning is a research group at the TMDT with 15 doctoral researchers and three postdocs. Our work rests on two mutually reinforcing pillars: scientific substance (contributions at leading international conferences) and industrial depth (long-term collaborations in which our methods move from prototype to productive use). The first pillar builds the methodological credibility that convinces partners, the second provides the real problems and data that make substantial research possible.
Our focus areas
Visual Perception & Understanding
In industrial environments, seeing is not enough
Automated damage detection on freight cars
Vision models that interpret ambiguous scenes, cope with scarce data, and explain their decisions. From industrial inspection through automotive perception to media forensics.
→ Learn more
Temporal Representation & System Understanding
The behavior of industrial systems is encoded in their sensor signals.
Real-time explainable quality prediction in the welding process
Temporal models that monitor systems, assess their state, and detect anomalies. From welding processes through glass manufacturing to flood forecasting.
→ Learn more
Sequential Decision Making & Planning
Real-world planning problems are dynamic and full of constraints.
Energy-cost-optimized machine utilization through reinforcement learning
Learning systems that find solutions in large search spaces and react quickly as conditions change. From production planning in aerospace to human-robot collaboration.
→ Learn more
Vom Benchmark in die Anlage
Our ambition does not end with the academic dataset. We validate our methods on real industrial systems together with our partners, under real conditions.
- Windshield manufacturing: Online quality prediction at Saint-Gobain Sekurit, running in live production for years.
- Welding technology: Inline quality prediction for arc welding, validated on real process data from FEF Aachen.
- Automotive perception: Radar- and LiDAR-based object detection in a multi-year collaboration with Tier-1 supplier Aptiv.
Research with visibility
We publish at leading international venues in AI and applied research. Selected contributions:
- Efficient 3D object detection for edge hardware; Best Paper Award, ICCV 2025 Workshop
- Quality prediction in arc welding with discrete representations; CIKM 2024 & 2025
- Deep reinforcement learning for machine scheduling: a survey of 143 works; Journal of Manufacturing Systems
- Predictive quality in windshield manufacturing; KDD 2023
Working with us
You're from industry? From the first problem sketch to a productive system, in funded collaborative projects as well as direct R&D contracts.
→ To the Industry page
You want to do a PhD or join us?
→ Open positions and → theses topics
