Virtual commissioning of AI-based camera systems

AI image processing system without time-consuming and costly collection of image data

From CAD model to virtual training image in just a few steps: Synthetically generated camera images from CAD data for AI training for inspection from workpiece one in the real image processing.

 

Planning of the inspection system, taking interference contours into account using CAD data. Optimisation of the camera and lighting arrangement using the generated virtual image data enables precise assembly and commissioning.

Colour, position or other material properties are generated at random: synthetically generated camera images with simulated workpiece variants for robust AI training.

Assembly check through AI-supported error evaluation of the test features on the real component from the very first vehicle.

VisionTools Bildanalyse Systeme GmbH has developed a unique, industry-compatible complete solution comprising AI-based analysis and virtual commissioning for optical quality control, identification and robot guidance. Based on the CAD data provided, this solution generates any number of artificial reference images before the start of series production. At car manufacturer BMW, the camera-based inspection process already starts with the first vehicle.

In one of around 70 inspection cells at the BMW plant in Dingolfing, two cameras use virtual image data to check whether three rubber plugs are present or correctly inserted in the area of the two rear lights on the almost fully assembled 4 Series Convertibles.

The final commissioning of camera-based image processing systems previously required a sufficient number of reference images of the object to be inspected. However, testable components are only available to a limited extent before the start of series production, so effective training could only begin with the start of production. The smaller the series, the greater the time and cost required per workpiece.

The innovative AI image processing solution developed by Waghäusel-based VisionTools Bildanalyse Systeme GmbH and integrated into the production line at BMW heralds a paradigm shift in inspection processes: Based on the available component data, the innovative software generates any number of virtual, realistic image variants of the inspection area before the start of series production. The simulation incorporates all criteria that occur in the real inspection run and can influence the image result. At the same time, cameras and lighting can be optimally positioned in the virtual space.

The effort required before the start of series production generally pays off: without the camera eye having seen a real workpiece beforehand, the synthetic image generation provides realistic good and bad examples. The quality of the image data is so high that it is ideal for training the AI algorithms. The real images captured during ongoing testing optimise the system and are added to the virtual training data set.

Customisable image processing solutions from VisionTools will also become standard in other production and assembly areas in the future. With the VisionCockpit training environment and the decentralised VoE-AIBox edge device, VisionTools is adapting its range of AI-based image processing solutions to individual customer requirements.

contact us