


Surface inspection of components in series production
Generative AI for training camera-based inspection systems
In series production, such as in metal and plastic processing industries, the reliable and complete inspection of components is an important prerequisite for quality assurance. To this end, camera-based inspection systems based on AI models offer an efficient and cost-effective solution. However, a challenge is to detect defects for which only a few sample parts are available.
The project aims to solve this issue by developing a generative AI model that can synthetically generate the image data of the defect parts needed for training on its own. The generative AI model is trained on the basis of a few real images of defective parts. It learns to recognize the characteristics of the defects and – this is the crucial step – to generate new, plausible images of defects. This extends the training data set of the inspection model and improves the quality of recognition and classification.
In order to verify and ensure the transferability of the approach to different industries and applications, flexible optical inspection systems are used to acquire the image data. Most notably, a tunnel inspection system is being developed that can be used in two different applications. The approach is being validated in various industry-related application areas, making a direct transfer highly likely upon successful project completion.