AI Assistant for CT Analysis of Concrete Beams


Computed tomography (CT) has been increasingly used for the investigation of building materials for about 10 years, for example to analyze cracks and other damage on drill cores and mortar samples, as well as microstructure, crack initiation and propagation spatially and non-destructively. Still, relatively small samples are generally investigated, in some cases even with an edge length of only a few millimeters. Since concrete is a very heterogeneous material, larger volumes have to be imaged and analyzed to obtain representative data. The Rheinland-Pfälzische Technische Universität Kaiserslautern-Landau (RPTU) has applied to the DFG (The Deutsche Forschungsgemeinschaft is the self-governing organisation for science and research in Germany. for a globally unique large-scale device for »Computed Tomography for the Investigation of Structures under Load Increases«.

One of the first and most important application scenarios of this device is 3D imaging of crack development in concrete beams during a 4-point bending test. The extensive 3D image processing and analysis software of the Fraunhofer ITWM is currently being upgraded to be able to efficiently handle the huge amounts of data generated. In parallel, mathematical and statistical methods are being developed in the BMBF-funded project »Detection of Anomalies in Image Data« (DAnoBi) in order to find and completely detect crack structures in concrete based on computed tomographic data in a robust and automatable way.

The PhD project aims to directly link the expertise of civil engineers with 3D image analysis in order to optimally select and parameterize complex algorithms, correctly evaluate intermediate results, and correct errors as early as possible. For this purpose, an AI assistant is being developed that learns the expected workflow and data flow, as well as expected intermediate results and typical error patterns. It is trained for the use case »cracks from 4-point bending test«

  • evaluate the image data quality on the basis of specimen properties (dimensions, material mixture) and CT measurement parameters
  • select and parameterize pre-processing and analysis algorithms on the basis of the image data and the analysis task
  • detect problematic (intermediate) results and suggest solutions.

In the end, the AI will assist in monitoring and controlling the entire experiment from sample insertion to visualization and graphical representation of the results. Initially, the focus is on image analysis, i.e. after tomographic reconstruction of the image data, in order not to go beyond the scope of a PhD.