Fraunhofer Institute for Digital Media Technology IDMTAs part of the HybridDigital project (subproject: Machine Learning Algorithms and Ontologies), Fraunhofer IDMT demonstrated that acoustic signals can provide valuable information about the quality and mechanical properties of hybrid materials. Using AI-supported analysis methods, the team could classify components by batch and predict the tensile-shear strength of CFRP-steel hybrid joints based on audio data. The institute also developed an AI-based long-term recording system for industrial testing and tools for standardising and evaluating large acoustic datasets. These results are an important foundation for future non-destructive quality assessment in industrial manufacturing.
Current situation and relevance to the industry
Lightweight and hybrid construction materials are becoming increasingly important in industries such as automotive manufacturing, aerospace and mechanical engineering. In particular, combining fibre-reinforced composites with metals enables high strength with reduced weight. At the same time, quality assurance requirements for these complex materials are increasing. Conventional testing methods are often time-consuming and costly, or can only be used on a random sample basis. The HybridDigital project therefore investigated new digital approaches for assessing the condition and quality of hybrid materials.
Challenge
The greatest challenge was identifying correlations between acoustic signals and the mechanical properties of the hybrid materials. While loading and ageing processes cause a wide range of acoustic changes, these often vary from sample to sample. Predicting imminent component failure in particular proved to be complex. While the project results demonstrate that universally applicable failure prediction is currently only possible to a limited extent, the audio data does contain a wealth of information relevant to quality.
Acoustic AI for long-term monitoring
Acoustic AI enables processes to be monitored non-contact, continuously and non-destructively. In this project, Fraunhofer IDMT developed technologies for long-term recording and processing of acoustic measurement data, as well as AI-based analysis. Using modern methods of feature extraction, data visualisation and machine learning, quality and strength characteristics could be reliably derived from the audio signals. Methods were also developed to efficiently store and standardise large amounts of data for future AI applications.
Benefits for industry and research
The results demonstrate the potential of acoustic AI for providing cost-effective, scalable quality assurance. Companies can automate inspection processes, detect quality deviations early on and gain valuable insights into material properties, all without destroying components. Additionally, a data and analytics infrastructure has been established to serve as the foundation for further industrial applications of acoustic condition monitoring. Work on transfer learning, domain adaptation and large standardised audio datasets, in particular, opens up new possibilities for AI-based quality monitoring in production, material development and predictive maintenance.