Transformer for data-driven determination of material parameters from optical deformation analysis

The research project aims to investigate a novel method for characterising sheet metal materials. By combining finite element simulations, realistic rendering, and modern transformer-based deep learning architectures, the project will develop and analyse a framework for identifying parameters of flow curve and flow location models.The project's core concept is to replace experimental measurement series entirely with synthetically generated image data on which to train a neural network. The network should be capable of reconstructing the underlying material parameters from strain fields or image data directly. To this end, a comprehensive virtual laboratory is being created to realistically simulate lighting conditions, camera parameters, and surface textures. A ViViT transformer will be used as a data-driven model to analyse these datasets and infer the flow location and flow curve parameters.The project comprises four central work packages: (1) construction of the virtual laboratory; (2) development of a suitable transformer architecture; (3) systematic analysis using synthetic data; and (4) systematic analysis using real test data. The aim is to investigate the performance and limits of this approach, thereby contributing to data-driven, resource-efficient material characterisation.