Detection and modelling crack structures in concrete based on spatial image data

Description

During the last 10 years, computed tomography (CT) has become more and more popular for studying building materials. Crack initiation and crack growth have been studied as well as damage and micro-structure. The imaged samples are still rather small, sometimes just a couple of millimeters edge length. Larger samples have to be scanned and analyzed as concrete is a highly heterogeneous material. The complex spatial crack morphology can be studied very well based on 3D images. However, analysis is often restricted to visualization and analysis of 2D slices due to difficulties to detect and segment cracks in the reconstructed CT data.

In this PhD project, detecting, segmenting, analyzing, and modelling cracks in concrete beams shall be attacked from several points of view - statistically, via image processing and machine learning.

Cracks will be modelled by using stochastic geometry models like fracture surfaces consisting of facets and edges of random tessellations or random graphs. Based on realizations of these models, synthetic CT images will be generated in order to have realistic images with ground truth available.

The thus derived synthetic image data can be used for training machine/statistical learning methods for detecting cracks. Nevertheless, classical crack detection methods in 3D will be exploited, too, and promising algorithms from 2D will be generalized to 3D. Additionally, the statistical method of [Dokládal] based on morphological path openings will be generalized from 1D to 2D structures.

This PhD project builds on previous and running work [Jung, Muesebeck]. It will contribute to developing new and efficient methods for finding and analyzing anomalies in large, spatially sparse data.

[Dokládal] P. Dokládal: Statistical Threshold Selection for Path Openings to Detect Cracks, Mathematical Morphology and Its Applications to Signal and Image Processing, 13th International Symposium, ISMM 2017, LNCS 10225, 2017, pp. 369-380

[Jung] C. Jung, Bildanalytische Erkennung von Rissen in Asphalt basierend auf dem Dijkstra-Algorithmus. BSc, TU Kaiserslautern und Fraunhofer ITWM, 2017

[Muesebeck] F. Müsebeck, Crack detection in 3D concrete images. MSc, TU Kaiserslautern und Fraunhofer ITWM, 2020

Status

current