Computer Vision as Key to an Automated Concrete Production Control

The need to reduce CO ₂ emissions from concrete leads to increasingly complex mix designs involving e.g. CO ₂ reduced cements, recycled materials, and various chemical additives. This complexity results in a larger sensitivity of the concrete to unpredictable fluctuations in both, the base material properties and in boundary conditions such as temperature and humidity during the production process. Digital sensor systems and quality control schemes are considered as key to counteract this problem by enabling an automated production control. As contribution towards this goal, this paper investigates the research question whether Computer Vision can be used for the predictive characterisation of raw materials (here: of concrete aggregates) and of the fresh concrete quality during the mixing process. In particular, we propose the usage of imaging sensors for the observation of both, aggregate material and the flow behaviour of fresh concrete during the mixing process, and present deep learning methods for the prediction of granulometric and rheological properties from the image observations, respectively. Incorporating such systems into the concrete production process enables the facilitation of a digital control loop for ready-mixed concrete production by allowing an in-line reaction to raw material fluctuations and to deviations of the concrete from the target properties.