The Fraunhofer IKTS OpenZfp-KI-Portal (Open AI portal for non-destructive testing (NDT)) now makes it possible to examine audio data from machines and systems for anomalies and have them analyzed using artificial intelligence (AI). Anomalies are deviations from the normal state that provide indications of wear, loose or defective components. These deviations do not necessarily have an impact on the process, but could lead to reduced quality, disruptions in the process or even failure.
Until now, such data has only been checked on a random basis, usually manually, and in many cases is based on the experience of the inspectors. The OpenZfp-KI-Portal is now intended to automate the inspection and make it traceable. The probability of error detection depends on the respective application and is up to 100 percent.
Audio recordings of the process must be uploaded to the OpenZfp-KI-Portal for analysis. A specially trained algorithm then carries out the anomaly detection and outputs the results as a spectrogram. Markings in the graph allow even inexperienced users to recognize the area in which deviations occur and thus narrow down the search for errors.
The portal offers an automatic mode that is preset with standard parameters and is particularly suitable for users who are just starting out with AI. Experienced AI users, on the other hand, can use the professional mode and adjust individual parameters themselves. The experimental nature of the data analysis is intended to reduce the fear of contact with AI.
Registration is required to use the portal. The data will only be used for your own data classification; it will not be passed on or used by third parties. The data is automatically deleted after 30 days. Until then, users have the option of having their own data reclassified and, if necessary, evaluated using other parameters. It is also possible to “donate“ your own data anonymously and thus help the portal to obtain a broader database in order to further train the AI algorithms.