The Fraunhofer Alliance for Big Data and Artificial Intelligence annually supports practical AI solutions that address specific industrial challenges. A jury of business unit leaders has now selected five winning projects from 20 ideas—Fraunhofer IMS is represented with two projects.

In the food and pharmaceutical industries, rotating tank cleaning nozzles are indispensable for the hygienic cleaning of process and storage tanks. The problem: these systems are not visible during operation. Blockages, bearing damage, or reduced rotational speeds often go undetected for long periods, with potentially serious consequences for product quality and production safety—in the worst case, leading to product recalls.

Together with Fraunhofer IVV, the IMS is developing an AI sensor that is mounted on the outside of the tank and analyzes the characteristic structure-borne sound signatures of the spray jets. The AI detects deviations from normal operation and identifies potential causes of failure in real time, without any intervention in the ongoing process. Using domain adaptation and few-shot learning methods, the system adapts to different tank geometries and operating conditions with just a few data points. The AI evaluation runs directly on the sensor node, based on the open-source framework AIfES developed at IMS, thus enabling easy retrofitting of existing systems.

The demonstrator brings the concept to life: The sensor is being tested under real-world conditions on a 10,000-liter industrial tank at Fraunhofer IVV. Deliberately induced faults such as clogged nozzles, jams, or bearing damage provide the training data for the AI models. The detection results are displayed live on a dashboard at the tank. Since the system is designed as a self-contained retrofit solution, it can later be deployed directly at customer sites or on a portable demonstration tank.

Vibrations are an important indicator of machine condition, but analyzing them has typically required expensive specialized cameras or complex sensor systems. The VIVID demonstrator, a joint project with Fraunhofer IEM, takes a different approach.

The project is developing AI-based algorithms that make high-frequency vibrations visible and measurable using cost-effective standard cameras. The solution captures minute movements from video data, precisely determines frequencies, and detects vibration-induced changes directly on edge systems for use in industrial real-time applications.

The approach is clearly demonstrated using a guitar: the vibrations of the strings are visually captured and analyzed—a compelling example of the technology’s potential in predictive maintenance.

Both projects start in April 2026 and run for eight months. They reinforce Fraunhofer IMS’s strategic focus in the field of industrial sensor-based AI and expand the portfolio to include new application areas in the food industry and condition monitoring.

Funded by the Fraunhofer Cluster of Excellence Cognitive Internet Technologies