Bosch is piloting generative AI and foundation models in manufacturing. In initial projects in two Bosch plants in Germany, generative AI creates synthetic images in order to develop and scale AI solutions for optical inspection and optimize existing AI models. Bosch expects that this will reduce the time needed for planning, launching, and ramping up AI applications from the current six to twelve months to just a few weeks. Following successful piloting, this service for generating synthetic data is to be offered to all Bosch locations.
Bosch pilot plants are already using AI in production scheduling, monitoring, and control. At the plant in Hildesheim, for example, AI-based data analysis has helped reduce cycle times during the production ramp-up of new lines by 15 percent. At the plant in Stuttgart-Feuerbach, new algorithms cut component-testing processes from three and a half to three minutes. The software models for generative AI were developed by Bosch research and are now being implemented in the field by Bosch factories. One plant uses an AI method of synthetically generated images to reliably inspect welds of copper wires in electric motor production, while another focuses on the quality assurance of high-pressure pumps.
For years, the Feuerbach plant inspected fuel-injection components manually. The nature and complexity of the products, as well as differences in the structure of the production lines, meant that neither rule-based nor AI-assisted optical inspection was possible. The new approach is a scalable generative AI that recognizes variants of a product and error patterns and takes into account different arrangements and sequences in the production process. This is based on a foundation model developed by Bosch research and fed by large data sets from the Bosch manufacturing network. Synthetically generated data is used to refine the model and customize it for on-site applications. This is expected to make the AI capable of inspecting components independently, only submitting cases to visual inspectors where it is unsure. At the Hildesheim plant, synthetically generated images have already been successfully used for training purposes in the first standard systems in electric motor production. The human eye cannot distinguish the artificially generated images from real ones. The plant expects that project duration will be six months shorter with the new approach than with conventional methods, leading to annual productivity increases in the six figure euro range. Bosch plans to expand the AI approach to other locations.
