Quality assurance Check using AI from the first workpiece

Related Vendors

FAULHABER Drive Systems

Wuxi InfiMotion Technology Co., Ltd.

Optical inspection systems require images for actual-to-target comparison. These can be pre-generated through simulation – allowing the inspection system to control from the first workpiece.

BMW uses AI-based systems at its German Dingolfing plant for automated quality control. Since the summer of 2023, the automaker has been inspecting car bodies using camera-based methods. Specifically, it checks whether there are three rubber plugs present and correctly inserted in the area of the two taillights.

How does it work technically? Based on the automaker's CAD files for the product and production environment, the system provider, Vision Tools, creates a simulation. Additionally, images of the actual product can be incorporated into the simulation: for example, from pre-production or product development.

Realistically simulate testing influences

Within the simulation, so-called randomization parameters are available. This means, for example, position and material tolerances, color and lighting differences, effects of extraneous light, image blurs, and image distortions. The simulation thus realistically covers the physical testing process, including occurring influences in the environment and on the product. After the start of production, the virtually generated images can be supplemented with real images. The advantage is a higher detection rate.

To prepare the artificial intelligence for as many deviations as possible, a random generator creates a specified number of virtual images within given boundaries. According to the software manufacturer, the images contain label data or annotations and are suitable for training an AI algorithm.

CAD data crucial for recognition

The reliability of the inspection process mainly depends on the quality of the CAD data. For example, missing product details negatively affect the recognition rate. Two cameras operate in the inspection station. Employees receive a notification on a display if there is an error.

According to the manufacturer, the upfront effort usually pays off: Constructing all conceivable errors as workpieces and keeping real images for training and validation purposes is significantly more labor-intensive.

Use datasets in other inspection stations

According to the manufacturer, users can adopt the virtually created environment and its corresponding dataset as a digital twin in other inspection stations with similar requirements. If different inspection scopes are required there, the digital image data basis is adaptable.

Vision Tools developed the foundations for the virtual commissioning of AI-based testing and inspection processes in collaboration with the Karlsruhe Institute of Technology, the Karlsruhe University of Applied Sciences, and other industrial partners.

Details about the system at the Smart Factory Day

More information about the system will be available at the Smart Factory Day event on May 14 and 15 in Karlsruhe/Germany.