AICC

Artificial Intelligence for Smart Design and Testing of Cement and Concrete

Principal Investigators 

  • Cremers, Daniel, Prof. Dr. rer. nat.
    Chair of Computer Vision and Artificial Intelligence, Departments of Informatics and Mathematics

  • Gehlen, Christoph, Prof. Dr.-Ing. Prof. h.c.
    Chair of Materials Science and Testing, Department of Materials Engineering

  • Machner, Alisa, Prof. Dr.
    Professorship for Mineral Construction Materials, Department of Materials Engineering

Summary

Cement and concrete belong to the most common used materials in our built environment. Their usage has a tremendous impact on CO2 emissions, not only due to the large amount of yearly processed materials, but also in terms of their high energy consumption during production. Therefore, it is of utmost importance to create efficient design and testing processes ensuring durability and best possible sustainability of structures made from these materials. We want to propose machine-learning-based methods to characterize the air pore system in concrete, which affects degradation and deterioration processes, mainly due to gas and moisture transport mechanisms within the solid phase. Our objective is to investigate different input/output variables from currently standardized methods in order to increase their accuracy and reliability. We will specifically focus on deriving the characteristic variables of the air pore system from light images of the concrete surface only. Moreover, we will create an image analysis scheme to predict the reaction degree of supplementary cementitious materials (SCM) in cementitious systems. By predicting the performance of SCM-containing composite cements, we try to facilitate the application of such environmentally friendly materials in the future. From the AI point of view, these data constellations will benefit from advanced techniques beyond supervised learning, such as domain adaptation and anomaly detection, in combination with advanced neural-network architectures such as rotation-equivariant ones. We will use and combine existing methods from these areas, and develop novel ones in a systematic way in order to further improve the quality of results. We will also publish our software and establish a public database that others can contribute to, so that researchers worldwide can easily build upon our research, benchmark their AI methods, and/or contribute their data.