Combining confocal microscopy and deep learning for concrete microstructure analysis
Year: 2026
Type: article
Abstract: Current standardized methods to characterize the microstructure of concrete regarding its freeze-thaw resistance are time-consuming and prone to uncertainty due to the manual assessment of air void parameters such as their size, distribution, content and spacing. This research presents an advanced approach by combining confocal laser scanning microscopy data (CLSM) with deep learning techniques. U-net segmentation models were trained with varying data combinations to precisely detect cement paste, aggregate and air void locations within CLSM data of polished concrete specimen surfaces. An image processing workflow for the automated evaluation of air void parameters according to DIN EN 480-11 was implemented by utilizing the developed semantic segmentation models and directly compared to human evaluation of a previously conducted round-robin test. While all model types achieve results in air void parameters that agree with human evaluation, the study indicates that utilizing height-data obtained from CLSM improves the accuracy in air void predictions. The findings of this study constitute a promising advancement toward the automation of concrete microstructure evaluation and provide a basis for further methodological development and research of concrete properties.
Source: Journal of Building Engineering
Institution: Technical University of Munich
Language: English
Subfield: Civil and Structural Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $3,940
OpenAlex ID: W7132853003