Vision system guarantees quality at BMW

The latest PC-based image processing technology is helping BMW guarantee the quality of head-up displays on its 5- and 6-series models. Dean Palmer reports

BMW is using the latest vision system and image processing technology to guarantee the quality of its new head-up displays (HUDs) for its 5- and 6-series models. The vision system not only copes with the speed of BMW's production assembly lines, but also maintains strict safety standards. The HUD is a driver information device developed and manufactured by Siemens VDO Automotive, which projects virtual images onto the windscreen of the vehicle by using an array of optical mirror systems. The image is projected in the viewing field of the driver at a distance of around 2m. So the driver can perceive important information about the car and navigation within his normal field of viewing. If these HUDs are slightly misaligned during assembly of the car, this interferes with the driver's perception of the surrounding environment, such as street markings, vertical lines of houses or garage entry. BMW's department for Technical Integration at the Research and Innovation Center in Munich had been looking at suitable image processing system for integration into the company's assembly process in the laboratory but needed to source a solution for the production line. In mid-2002, the department decided to work with German company Gefasoft, a systems integrator of Cognex equipment. Measurement parameters had to be determined and then checked quickly and efficiently without disrupting production of vehicles. The system also had to cope with the high volume of cars and be suitable for both right- and left-hand drive vehicles. The testing system of the HUDs also needed to be fully traceable and embedded into the information infrastructure of the entire corporate network. In BMW's own feasibility studies, it was found that a PC-based high performance image processing system was required. Efficient hardware was also needed, along with the vision software algorithms. No jams or standstills are tolerated on BMW's two-shift assembly lines, so the complete image acquisition and evaluation process, which included 50 acquired test images, had to be completed within a time frame of less than two minutes. The complete testing equipment had to be swung into the vehicle as production is in progress, precisely positioned, secured in place, then removed from the car again. This meant a 30-second time slot for image acquisition and image processing calculations in the test cycle. Vision tools from the CVL (Cognex Vision Library) proved successful here. They provided BMW with interactive access to different programming levels, in-depth vision software, including an option for more precise detailed optimisation. The software can separately identify important features within the object image, including edges, measurements, shapes, angles, bends and shades. The spatial conditions and relation between these central features of the acquired image are then compared with the real time image. The position of the object is then precisely determined from this information. The influence of changing conditions of illumination and contrast are eliminated by a simultaneous investigation of the contour and the structure of the object image. These qualities proved to be of advantage in detecting deformations and distortions of the HUDs. The functional safety of the vision tools also satisfied the standards specified by BMW, which required it to be above 99%. The image processing system is fully integrated into BMW's plant diagnosis system. By 2004, the first HUD testing systems were installed. The success of this operation led to BMW installing the HUD testing systems in normal series production lines at its factory in Dingolfing. Today, BMW is using around 12 systems, 24 hours a day. The test results are stored in a database on a central server, therefore providing total traceability. The test images are archived, and in case of requests from customers, the flawless condition of the HUD at the time of delivery can be proved, even years later.