Physical domain knowledge based image processing algorithm for in-situ surface porosity inspection and digitalization for 3D concrete printing structures

Concrete 3D printing holds transformative potential for construction because it can enable rapid, design-flexible structures from terrestrial housing to extraterrestrial habitats. However, its scalability is limited by challenges in quality control which governs structural integrity and durability, among which the volume fraction and size distribution of macro-porosity are especially critical. This study presents a novel scientific in-process monitoring technique for real-time macro-porosity analysis in 3D-printed concrete extrudates. Using a microscopy system with a side-mounted light source, the monitoring system captures surface microstructure without labor-intensive petrographic post-processing. A self-calibrating computer vision algorithm, leveraging a Perimeter-to-Area Ratio (PAR) metric and concrete domain knowledge, accurately distinguishes air voids from aggregates, overcoming conventional image processing misclassification. Validated against expert annotations, the algorithm achieved a 98% porosity prediction accuracy relative to manual petrographic baselines. Furthermore, the bimodal thresholding successfully mathematically isolates small-entrained spherical pores (1.25 mm) without manual input. This precise pore geometry and distribution analysis provides the actionable data required to optimize mixture designs on-the-fly, forecast a material's resistance to environmental stressors like freeze-thaw cycles, and enable the creation of digital twins for predictive maintenance. While challenges remain in adapting to diverse concrete compositions and variable lighting, this method provides a scalable, quantitative foundation for reliable 3D printing, advancing additive manufacturing toward sustainable, industry-standard construction.