DFG Project : Superresolution of multiscale images from materials sciences using geometrical features
Recent and ongoing developments in imaging techniques and computational analysis deeply modify the way materials sciences and engineering consider their objects of research. Our project will contribute to this direction of research by developing new superresolution methods guided by high-resolution local subimages of 3D materials data.The mathematical methods of choice will be based on local and global Generalized Gaussian Mixture Models as well as Student-t Mixture Models in conjunction with variational methods. Appropriate geometrical features related to the engineering topics have to be established to provide an evaluation platform for the superresolution images, and to be directly involved into the Bayesian and variational models. The mathematical models will be developed, analyzed and appropriate efficient algorithms will be derived, including an examination of their convergence behavior. The models will be extended to multimodal images, where due to the size of the structures of interest, the high resolution image are taken by serial sectioning (FIB-SEM) tomography and the low resolution images by micro computed tomography. This requires to take the special acquisition of FIB-SEM tomographic images, in particular curtaining effects and the anisotropy with respect to the third dimension into account. A numerical evaluation of the relevance and the benefit of the developed superresolution methods will be performed by comparing the effective properties computed for reactive flow in porous media.
Project partner:
Professorin Dr. Gabriele Steidl
Professor Jean-Francois Aujol, Ph.D.
Major-Instrumentation Initiative: CT of structural elements under changing load
Computer Tomography has been used for many years in construction materials technology in order to gain knowledge, using small format samples, of the joint structure of construction materials and how they change under different conditions of exposure. However, there is lack of appropriately configured equipment that would enable by means of X-ray technology a better understanding of bearing and deformation behaviour of entire building parts. Hence, in line with the DFG offer to tender, a completely new type of Tomography Portal (TOP) has been conceived; one that can scan building components of real life dimensions under increasing load with high resolution accuracy. It makes use of a high radiation energy, which is indispensable in order for example to illuminate bar-like reinforced concrete components using cross-section dimensions of at least 30 x 30 cm, and imaging of cracks of 0.1 mm width. In addition, a second radiating source has been integrated, which in the case of smaller dimensions or less demanding materials permits higher resolutions. The design has been understood as an interdisciplinary work. The guiding factors in the design were: the needs of the construction industry, machine, X-ray and measuring technology, calibration procedure, mathematical reconstruction of raw data, with their visual preparation and further processing on high performance computers as well as the availability of the research results.
Project partner:
Professor Dr.-Ing. Jürgen Schnell
Professor Dr.-Ing. Manfred Curbach
Professor Dr.-Ing. Josef Hegger
Professor Dr.-Ing. Harald S. Müller
Professor Dr.-Ing. Ralf Müller
Professor Dr.-Ing. Matthias Pahn
German-French doctoral college
The aim of the research training group is to develop new methods for analyzing, modeling and simulating complex material structures using image data. The focus of the third funding period of the Research Training Group 2017-2020 is as follows:
- Recording and editing multimodal data (i.e. data obtained using different recording modalities),
- multivariate and diverse data (data acquisition with e.g. hyperspectral cameras, DT-MRI, electron backscattered diffusion, catadi-optical cameras or cameras with depth sensors such as Kinect).
On the one hand, this type of data offers an enormous gain in information e.g. compared to grayscale images, but requires a much more complex analysis and the development of completely new models with broad mathematical knowledge. The analysis focuses on difficult image segmentation problems that should be dealt with using variation methods and / or morphological methods.
Project partner:
BMBF Project: poST - Synthetic data for ML segmentation of FIB-SEM nanotomographs of highly porous structures
The nanostructure of complex materials can be imaged in 3D by the FIB-SEM serial sectioning technique. To analyze the material, its components must be reconstructed from the image data. In the case of high porosity, this is difficult because structures behind the current imaging plane are also visible through the pores.
Machine learning techniques have high potential here. However, training data are difficult to obtain. Manual segmentation is hardly possible, since even humans often cannot decide which structures form the foreground of the current section. Synthetic images for which the correct result is known are an attractive way out. The similarity of the simulated to the real structure has a considerable influence on the quality of the result.
Project partner:
Fraunhofer-Institut für Techno- und Wirtschaftsmathematik ITWM
BMBF Project: SynosIs - Synthetic, Optically Realistic Image Data of Surface Structures for AI-Based Inspection Systems
Artificial intelligence (AI) is used very successfully in image recognition, processing and understanding. However, training an AI-based inspection system for industrial quality assurance requires large amounts of representative annotated image data for all defect types. Manual annotation is laborious and error-prone. Many defects, especially safety-critical ones, occur very rarely.
Realistic synthetic image data help circumvent these problems.
We combine physics, mathematics and computer science to generate synthetic images of typical defects on metallic surfaces with unprecedented realism. These defect images that are guaranteed to be correctly and objectively annotated are available for training and validation of AI systems for optical surface inspection after the end of the project.
More information about the project
Project partner:
Fraunhofer-Institut für Techno- und Wirtschaftsmathematik ITWM
Fraunhofer-Institut für Angewandte Optik und Feinmechanik IOF
BMBF Project: DAnoBi
The aim of the research project is to develop methods for the detection of anomalies in large image data. This could be e.g. micro cracks in concrete beams, material densification in textile web goods or local fiber misalignments in components made of fiber-reinforced plastic. For this purpose, methods of machine learning, modeling of structures and imaging as well as statistical methods for the detection of abnormalities can be combined. In all this cases mentioned, the structure varies greatly. Methods and measures must therefore be developed in order to decide objectively, robustly and repeatably what an anomaly, i.e. a significant deviation is. The anomalies to be detected rarely occur both statistically and spatially. For this reason, there is a lack of training data for purely machine learning (ML) based solutions, which annotation is also very difficult. One way out is training based on realistic, simulated data, since in this case enough data can be generated and the basic truth is immediately available. An alternative approach is to adapt statistical methods for structure break detection in time series for the detection of spatial anomalies. The spatial component of the image data gives rise to new tasks, such as recognizing complex anomaly regions, which will also be solved in the project. The planned research work focuses on the detection of cracks in concrete components, since this question is of great practical relevance, but also particularly difficult in a methodological sense. ML and statistical methods for solving this specific problem are expected to be transferable to the other use cases mentioned above. Models for concrete and crack structures, on the other hand, are problem-specific, but are already a major challenge in this thematic limitation.
Project partner:
Otto von Guericke-Universität Magdeburg
Fraunhofer-Institut für Techno- und Wirtschaftsmathematik ITWM
BORAPA Ingenieurgesellschaft mbH
MKT Metall-Kunststoff-Technik GmbH & Co KG
Completed projects
DFG-Projects
Bubble Distributions as Markers for Deformation in Polar Ice
2015-2020DFG Research Training Group 1932
Stochastic Models for Innovations in the Engineering Sciences
2014-2019Stochastic Modeling and Calculation of the Failure Probability of Metal Foams on the Mesoscale 2009-2011
BMBF-Projects
Analysis, Modelling and Simulation of Chromatographic Purification Processes (AMSCHA)
2016-2020Analysis of Low-Dimensional Structures in Three-Dimensional Image Data (AniS)
2013-2016Stochastic Models for the Analysis of Highly Porous Micro- and Nanostructures (AMiNa)
2010-2013
Other Projects
Junior Endowed Professorship
Statistics of Spatial Structures for Innovations in the Engineering Sciences of the "Carl Zeiss Foundation"
2013-2017Project as Part of the Innovation Centre "Applied System Modeling"
Applied System Modeling for Multi Scale Materials
2010-2013(CM)²-Project
Image Processing in Civil Engineering
2008-2013