Ensuring industrial quality through new research
The ML-SYNTHOM project is developing an AI-based test procedure for quality assurance for new types of components in industrial production. The innovative processes and tools used in the project make it possible to synthetically generate training data of components for machine learning for quality assurance. At the centre of this is the concept of generative models, i.e. parameterised, three-dimensional scene graphs that can generate training data through photorealistic image synthesis. Using these techniques to improve generalisation, synthetic data generation represents a significant further development of established data augmentation. The use of synthetic data makes the training of deep learning affordable, accelerates production-related development and further improves quality control.
To demonstrate the process, a diaphragm spring provided by automotive supplier ZF was tested using a specific use case. Image-based inspections during the production of the spring allow defects to be recognised earlier and more reliably, thus reducing inspection and defect costs.
Following an initial application of the technology in vehicle construction, the process can be transferred to other areas, such as plastics processing, and is thus able to improve the competitiveness of the Saarland automotive industry and other sectors. It can also reduce production costs in companies and increase product quality.
Follow-up initiatives and collaborations
After completion of the project, specialised workshops as well as consulting and services are planned to present the results to the industrial sector and make them accessible. In addition, the research results will also be incorporated into the DEAI-Hub Saarland (Digital Economy & Artificial Intelligence Hub), which has been active since the beginning of January and is part of the EDIH network (European Digital Innovation Hub).
Partners
ZF Friedrichshafen AG