AI-based image segmentation for systematic characterization of parent concrete in selective demolition

The concrete recycling industry faces significant challenges due to the uncontrolled mixing of parent concretes with varying properties during demolition, resulting in inconsistent recycled concrete aggregate (RCA) quality and limiting its potential for use in new concrete production. Existing literature typically characterizes parent concrete solely based on compressive strength, neglecting other critical parameters. Consequently, RCA is often labelled as inherently heterogeneous, without fully considering the variability introduced by mixed-source demolition. This study introduces an novel protocol for systematic parent concrete characterization, combining an Artificial Intelligence (AI)-based segmentation approach with complementary techniques like polarized light and fluorescence microscopy (PFM). The proposed methodology quantifies critical properties of parent concrete, including water-to-cement (W/C) ratio, cement and aggregate content, air void content, and aggregate gradation. This enables a detailed evaluation of parent concrete variability, providing the basis for selective demolition strategies that reduce RCA heterogeneity and enhance the predictability of its properties. To illustrate its applicability, the protocol was applied to structural components of a Dutch viaduct, including prestressed beams, heavily reinforced columns, foundations, and abutment-wall. Results revealed significant differences in estimated water-to-cement ratios, ranging from 0.29 ± 0.03 in beams to 0.38 ± 0.03 in abutment walls, while hydration degrees varied between 0.80 ± 0.08 and 0.93 ± 0.02, indicating differences in cement maturity. Cement content ranged from 316 ± 11 kg/m³ in foundations to 390 ± 10 kg/m³ in beams, and air void content ranged significantly from 0.9% in abutment walls to 4.3% in foundations. Microstructural composition also differed substantially: paste volume ranged from 17% to 27%, and coarse aggregate content from 37% to 52% depending on the component. These quantitative differences confirm that structural concretes, even within the same structure, exhibit substantial internal variation. As such, uncontrolled demolition leads to the mixing of materials with fundamentally different properties—supporting the argument that RCA heterogeneity is not intrinsic, but largely a result of conventional demolition. This reinforces the value of the proposed protocol in identifying material variability and informing targeted demolition to enhance the predictability and uniformity of RCA.