Fraunhofer Institute for Digital Media Technology IDMTNeue Messe Stuttgart / May 06, 2025 - May 09, 2025
Control 2025 – International Trade Fair for Quality Assurance
Meet us in hall 7, at the Fraunhofer Vision booth 7301 and learn all about “Acoustic Monitoring and Federated Learning in Production”
Acoustic AI for cross-location error detection in production
At the trade fair, we will be demonstrating how airborne sound analysis and artificial intelligence can be used to optimize production across multiple locations.
Higher efficiency through intelligent machine monitoring
Predictive maintenance plays an important role in modern industrial plants. Machine and plant manufacturers are increasingly relying on the intelligent evaluation of machine and process data for condition monitoring. Unstable process parameters lead to quality variations in production, which is why operators of welding and milling machines rely on the knowledge of their machine operators, who recognize deviations from the normal condition of their machine by the noise it produces. Whether engines, gears or spindles - wherever something moves, sound is generated that provides information about the condition. The use of an acoustic system for the automatic detection of machine faults is a valuable advantage in times of a shortage of skilled personnel.
Video Player is loading.
This is a modal window.
The media could not be loaded, either because the server or network failed or because the format is not supported.
Interactive demonstrator for acoustic condition monitoring across locations
The exhibit shows how an AI-based acoustic monitoring system analyzes machine conditions, detects faults and relies on cross-location learning. Identical machines are operating at each of the system's locations and their operating sounds are analyzed by an acoustic AI. The pre-trained models classify three different states.
As errors rarely occur, the amount of data for AI training at a single location is limited. This is where distributed learning (federated learning) comes in: Instead of sharing confidential audio data directly, the AI models exclusively share learned knowledge in terms of model parameters. This improves error detection across all locations without incurring data security risks.
The demonstrator shows an innovative combination of intelligent acoustic condition monitoring and distributed learning - in this example specifically for classifying engine sounds.
Condition monitoring using airborne sound analysis and AI is conceivable for a wide range of applications in industrial production - whether for continuous monitoring of engines and gearboxes or for monitoring individual production steps, such as the welding of battery boxes. Thanks to the optimal selection of acoustic sensors and pre-trained AI models, anomalies and errors can be reliably detected even in noisy industrial environments.
This technology sets new standards for efficient and safe AI-based quality assurance in production - for all locations.