Motivation

In recent years, enormous progress has been made in the field of artificial intelligence (AI).
New methods support people in their daily and professional tasks and make it possible to automate many processes that previously required strenuous manual work.
A particularly important area of application for AI is anomaly detection, i.e. identifying data that deviates from the norm.
Examples include mutations in virology, cases of fraud in finance or faulty components in industrial production.
However, current anomaly detectors are limited to applications where enormous amounts of training data are available.
This is precisely where the Physics-informed anomaly detection (PIAD) research project comes in.

Aims and approach

The aim is to develop a new type of physically-informed AI for areas that do not have a large available database,
such as the laser beam melting application pursued in the project.
Physics-informed means that physical knowledge and rules are incorporated into the AI.
This is based on state-of-the-art neural recognition algorithms, which have recently been able to drastically reduce error rates.
Model architectures are being developed that specifically incorporate information from the data domain in order to further increase data efficiency.
Methods are being developed to expand and improve the existing training data.
The prototype application will be evaluated as an example in additive manufacturing, particularly in efficient process monitoring.