Hochschule Mittweida - University of Applied Sciences
The Hochschule Mittweida - University of Applied Sciences is located in the midst of the state Saxony in the east of Germany. It comprises faculties for engineering, sciences, computer science, business administration and social work. Most research happens within the scope of the faculties but it also has mostly research-oriented institutes.
At the moment the Hochschule Mittweida (HSMW) is celebrating its 150th anniversary. During the last 1 1/2 centuries it has maintained a focus on science and engineering. Recently, new topics like interactive media, game programming and Germany's first study program on traditional and digital forensics have earned the University a reputation well beyond its local neighborhood.
In addition to its schools, the HSMW has some institutes which are mainly dedicated to research. Among those, one of the flagships it its Laser-Institute which mostly does third-party-funded research in areas like e.g., laser-cutting or laser-sintering.
We want to introduce computer-vision approaches in wafer-production in such areas where they are not yet employed.
Within WP 4.4 we mostly tackle computer-vision problems in wafer production. While being an independent partner within the consortium we closely working with Infineon Dresden where the wafer fab is located. Another key-partner is the Infineon branch in Villach, Austria which is also working on image-processing and CV applications to wafer production. In order to better understand how machine-learning could complement CV, another close partner is MCL (Austria) which has specialized in the application of machine learning to industrial applications.
We want to introduce computer-vision approaches in wafer-production in such areas where they are not yet employed. This involves both the inspection of product surfaces and of production tools alike. From a more scientific perspective we want to gear computer-vision into an ad-hoc direction. Traditional applications of CV in the industry require the careful adjustment of the camera alignment, lighting or they requires to shield the background to prevent reflections in glossy surfaces. However, humans can often mask out these effects and still detect flaws during wafer-production. With the advances in computational capabilities and in the field of machine-learning we believe, that algorithms also need to catch up more to the human perception. In the future, personnel will want to attach cameras on a case-by-case basis when and where needed. This is particularly important for small production lots where setup-cost needs to be low. Algorithms will be expected to cope with given light conditions, perspective distortions, noise or vibrations.
Another more practical challenge will be to merge visual features into an existing Process Control System which today processes scalar data only.