A photometric stereo and Vision Transformer-based framework for automated air void analysis in hardened concrete

Abstract

Air void analysis in hardened concrete typically requires the Linear Traverse Method (Procedure A) or the Modified Point-Count Method (Procedure B), as described in ASTM C457, to assess the content and distribution of entrapped and entrained air voids. These parameters are critical for assessing freeze–thaw resistance, workability, and mechanical properties. Conventional approaches rely on manual sampling and inspection by trained technicians, which are not only time-consuming and labor-intensive but also inherently subjective due to operator-dependent interpretations. This study proposes a novel framework integrating Photometric Stereo and Vision Transformer models to automate air void analysis with enhanced speed, accuracy, and generalizability. By leveraging a newly designed photometric stereo system and a computer vision-based image processing pipeline, the proposed method addresses key limitations in existing techniques, such as the inability to detect shallow or deep voids, incomplete boundary identification, and challenges in distinguishing air voids from aggregates. The framework seamlessly integrates image acquisition, recognition, extraction, and analysis, completing the entire process within approximately 10 min, excluding initial specimen preparation. When compared to current automated petrographic analysis methods, this approach achieves state-of-the-art precision, with mean accuracies of 97.970% for air void content and 97.176% for spacing factor, while eliminating the need for surface treatments or extensive model training. Furthermore, the study reports a comprehensive sensitivity analysis of parameter selection in Procedures A and B, offering deeper insights into ASTM C457 specifications. The proposed solution significantly reduces time and labor costs associated with air void analysis, enhances result reliability, and demonstrates potential for broader applications in micro-scale 3D surface reconstruction and property evaluation.