In industrial production, the testing of machines and products with acoustic signals still plays a niche role. At this year’s Hannover Messe, the Fraunhofer Institute for Digital Media Technology is exhibiting a cognitive system that is intended to detect erroneous sounds more objectively than the human ear. This technological approach combines intelligent acoustic measurement technology with signal analysis, machine learning, as well as secure, flexible data storage. The researchers have been able to detect up to 99 percent of the defects in initial pilot projects with industry.
The scientists identify possible sources of noises and analyse their causes, create a noise model of the environment, and focus their microphones there. The system removes interfering sounds, such as voices or noise, from the overall signal. The signal is then repeatedly compared with previously determined reference sounds obtained in a laboratory environment. With the help of artificial neural networks, the scientists are gradually developing algorithms that are able to detect errors from the noise they generate. The technology is so sensitive that it also indicates nuances in error intensity and manages complex tasks. For example, it can be used to perform end-of-line inspection of car parts and check motors for car seats.
The Fraunhofer researchers are able to ensure the data security of the collected acoustic signals by authorising users and managing rights and identities. For example, real and virtual identities are decoupled in order not to infringe user rights when different people evaluate the data. Machines and test systems are usually installed on a production line. The researchers therefore store their acoustic data records in a secure cloud.