Max Coenen

Concrete production is highly resource-intensive, with natural aggregates comprising roughly 70% of its volume. Replacing these natural materials with recycled aggregates like crushed concrete and masonry offers a sustainable alternative. However, the utilization of recycled aggregates requires rigorous quality control. Currently, this inspection remains a time-consuming, manual process with human error, creating a bottleneck for broader industry adoption. This paper proposes a deep learning solution using transfer learning to identify the composition of recycled aggregates in real-time and in an automated system, which effectively translates the lab-based classification of recycled aggregates in an automated production process. This solution relies on a new labeled database of recycled aggregate images classified in nine categories. In this study, recycled aggregates are obtained firstly by crushing demolition waste. Then, nine different combinations of the obtained material were prepared in the lab. In the next step, three pre-trained CNN models were examined, including EfficientNetB0, ResNet50, and VGG16, resulting in validation accuracy of 87%, 81%, and 73%. Accordingly, the EfficientNetB0 model was selected for further performance validation/improvement through implementing cross-validation for generalizability and fine-tuning the top 30 layers of the core model for better feature learning. Finally, we gained a validation accuracy of 91% in recycled aggregates classification. Moreover, EfficientNetB0 had a small size of 23.47 MB and an inference speed of 87.76 ms. Hence, the model was run successfully on an edge device that consisted of a Raspberry Pi 5 for real-time classification. Therefore, automated aggregate classification can be implemented in the concrete production plant.

A key requirement onto the concrete properties is that during the placement process the fresh concrete needs to completely fill the formwork while staying homogeneous. Whereas substantial research onto the flow and sedimentation behavior of fresh concrete has been presented regarding highly flowable self-compacting concretes, research regarding these properties on ordinary, vibrated concretes has been missing. In the presented research thus the sedimentation behavior of ordinary concrete as well as model systems was investigated. Here it could be shown, that even concretes which are seen as stable show significant changes in their homogeneity due to vibration compaction and that these local variations in aggregate content and particle size distribution have pronounced influence on the local rheological properties of the mix. The results could be closly linked to the nanoscopic particle inventory. Further, a novel measurement technique for the rheological properties of fresh concrete was developed which relies on the inverse analysis of the flow behavior of the concrete during the discharge process of a mixing truck. Using so-called Computer Vision Methods, video images of the concrete flow process can be used to derive rheological parameters such as the yield stress of the plastic viscosity.

Increasing the degree of digitisation and automation in the concrete production process can play a crucial role in reducing the CO$_2$ emissions that are associated with the production of concrete. In this paper, a method is presented that makes it possible to predict the properties of fresh concrete during the mixing process based on stereoscopic image sequences of the concretes flow behaviour. A Convolutional Neural Network (CNN) is used for the prediction, which receives the images supported by information on the mix design as input. In addition, the network receives temporal information in the form of the time difference between the time at which the images are taken and the time at which the reference values of the concretes are carried out. With this temporal information, the network implicitly learns the time-dependent behaviour of the concretes properties. The network predicts the slump flow diameter, the yield stress and the plastic viscosity. The time-dependent prediction potentially opens up the pathway to determine the temporal development of the fresh concrete properties already during mixing. This provides a huge advantage for the concrete industry. As a result, countermeasures can be taken in a timely manner. It is shown that an approach based on depth and optical flow images, supported by information of the mix design, achieves the best results.

Homogenität – d.h. die Gleichmäßigkeit der Zusammensetzung und Eigenschaften des Betons im frischen und festen Zustand – ist eine zentrale Anforderung an den Baustoff Beton. Variationen der Homogenität äußern sich in der Regel in lokalen Veränderungen z.B. des E‐Moduls, der Festigkeit oder Dauerhaftigkeit. Insbesondere bei hochfließfähigen (≥F5) und selbstverdichtenden Betonen (SVB), aber z.B. auch bei LP‐Betonen ist die Homogenität eine zeitabhängige Größe und ändert sich durch Entmischungsprozesse wie z.B. Sedimentation der groben Gesteinskörnung und Wasserabsonderung oder infolge einer ungeeigneten Verarbeitung. Eine Bewertung der Homogenität des frischen Betons ist mit den gegenwärtigen Prüfverfahren auf der Baustelle nur augenscheinlich möglich, da insbesondere die räumliche Verteilung der Betonausgangsstoffe (und damit die Homogenität) bislang nicht quantifiziert werden kann. Aufbauend auf den bereits in den ersten beiden Teilen dieses Beitrags vorgestellten bewertbaren Eigenschaften wird in diesem Beitrag das Eigenschaftsspektrum der bildbasierten Frischbetonprüfung um die Homogenität des Frischbetons erweitert. Dazu werden visuell erfassbare Eigenschaften des ausgebreiteten Frischbetons im Rahmen der Ausbreitmaßprüfung – Leimabsonderungen im Randbereich sowie geometrische Verteilung der Gesteinskörnung an der Oberfläche – herangezogen.