Estimating attached mortar paste on the surface of recycled aggregates based on deep learning and mineralogical models

Recycled aggregates, obtained from construction and demolition waste (C&DW), are currently underutilized in the production of new concrete given the incidence of widespread leftover cement paste adhering to the surface. C&DW sorting facilities based on optical technology can be developed and applied on an industrial scale, improving the overall quality of this secondary raw material. In this study, we present a novel approach based on image analysis and mineralogical laboratory methods to determine the residual attached mortar volume. Through clustering analysis, we classify C&DW samples with a comparable cement content determined by the image analysis. The leftover cement paste from these C&DW classes is mechanically extracted and examined using X-ray Powder Diffraction and Rietveld refinement. To estimate the attached mortar volume and the carbonation of the cement paste, we present a novel mathematical model based on the mineralogical data. To overcome the bottleneck associate with the image analysis, we further incorporate a deep learning model to automate the determination of the mortar volume, which enables high-throughput screening of C&DW in real production.