USKI
USKI
Project duration
From: 01.11.2025 To: 31.10.2027Description
Ultrasonic welding (US) is a commonly used joining process for plastics components with high quantities, often with welding times of less than one second. A fundamental problem with US and other welding processes is that, according to the current state of the art, the quality of a manufactured joint can only be fully assessed by destructive testing. In industrial production this often leads to a conflict between comprehensive quality assurance and low scrap rate due to destructive testing. Accordingly, destructive tests are usually only performed on a random sample of the manufactured components. This is particularly critical in the case of US, as the quality of the joint can vary significantly compared to other welding processes.
In order to enable inspection of all produced, non-destructive testing is usually employed. This can identify component- or application-specific fault patterns with a certain degree of probability. Non-destructive testing is carried out downstream of the welding process in the same way as destructive testing and sometimes requires considerable resources, which must be minimized for economic reasons.
In this context, an alternative approach to quality assurance with low resource requirements is process monitoring of the welding machine itself, i.e., sensor data from force, displacement, or power sensors. This data is recorded and stored individually for each weld by modern ultrasonic welding systems with a temporal resolution of approx. 1 ms and, if necessary, evaluated according to simple criteria specified by the user, such as minimum or maximum values. However, since process monitoring variables in industrial applications often exhibit complex time dependencies, the current, rather simple evaluation is in many cases insufficient for robust quality assurance. Employees with a high level of process understanding and experience, on the other hand, can often identify a whole range of error patterns from the process monitoring data of US systems. With respect to the cycle time of the US process and the availability of welding specialists, this approach is only possible downstream and only in the case of very critical errors. For this reason, to date the potential of the current US process monitoring for quality assurance has only been exploited to a very limited extent.
Against this backdrop, the USKI project aims to develop an AI-supported, data-based, fully automated, inline quality assurance method for the ultrasonic welding processes. The AI models to be developed will use standard process monitoring data from US systems to predict weld seam strengths immediately after completion of the welding process. In addition, it will be examined whether the predictive capability of such AI models can be improved by additional data extracted from thermograms recorded immediately after the welding process. The data generated in the course of the project, the AI models trained with it, and, in particular, the methodologies developed for training the AI models are intended to enable companies to implement their own quality controls for 100% of their ultrasonically welded components. The benefits for companies lie in the reduction of damage claims, complaints, and recourse claims, as well as an increase in resource efficiency.
Industrial implementation of the obtained results is to be tested in the research project for various component geometries and materials. Furthermore, all data collected in the project, programs developed, and evaluation algorithms are to be published for free use in order to enable the training of the AI models created for other products.
Contact person:
