Parallelsession: AI and Systems Integration - Best Practice and Discussion

Moderation: Georg Alber, Business Upper Austria

Talk 1: Predict and maximize the dressing interval of the grinding wheel with AI, Bojan Miličić, MEDIUS

We collaborated with Volvo engineers to predict and extend the sharpening and maintenance intervals for tools used in the production of drive shafts.

By analyzing input data, regarding materials, hardening processes, and coolants—alongside all metric data from the production machinery, such as rotation speeds and pressures—and correlating this with product quality data, we are able to calculate in real-time how long a tool can remain in operation before requiring maintenance,  using Digital Twins and custom AI models.

Extending these intervals saves €300,000 per tool, per machine, as premature maintenance is extremely costly. Most interestingly, this model is scalable and can be transferred to all other machines.

Talk 2: Multilevel Industrial AI - from digital to Physical and from device to Big Data, Matthias Kornexl, KEBA Digital

At KEBA we apply machine learning across a range of real production settings, both inside our own operations and in the products and services we deliver to customers. In each case the motivation is concrete rather than exploratory. The tasks are familiar ones: catching defects, spotting anomalies in sensor data before they become failures, or making sense of dense diagnostic information faster than a human could. The value is equally concrete: less manual rework, earlier warning, and expert time saved. Drawing on selected examples, the talk focuses less on the algorithms themselves - which range from classical machine learning to deep learning to physics-informed rules - than on what it takes to run them robustly and unattended in production. That is trustworthy data foundations, calibrated behaviour under uncertainty, and being explicit about what a model doesn't know.