Computer Vision And ML Defect Detection In Concrete AM

Key Takeaways

• RGB cameras cheap but struggle with dust, glare, and variable lighting.
• Depth and thermal sensors improve geometry and internal flaw detection.
• Small, site‑specific datasets hinder model generalization across printers.
• Linking visual defects to compressive strength is critical for safety.
• Real‑time detection could auto‑adjust extrusion parameters during printing.

Pulse Analysis

Construction additive manufacturing (AM) is moving beyond laboratory demos toward full‑scale building projects, but quality assurance remains a bottleneck. Unlike metal or polymer printers that operate in controlled environments, concrete 3‑D printers contend with dust, rain, fluctuating sunlight, and rapidly changing material properties. Without reliable inspection, hidden voids or mis‑aligned layers can compromise structural integrity, driving up repair costs and eroding stakeholder confidence. The industry therefore needs robust, on‑site monitoring that can keep pace with the pace of construction.

Researchers are exploring a spectrum of sensors to feed visual and acoustic data into defect‑detection algorithms. Simple RGB cameras offer low cost but suffer from shadows and glare, while depth scanners add geometric insight and thermal imagers can expose cold joints or internal voids. Acoustic and ultrasonic probes promise to reveal subsurface flaws but are harder to deploy on rugged sites. On the software side, pipelines have evolved from edge‑detection filters to supervised classifiers and, more recently, deep‑learning models capable of segmenting complex defect patterns. However, most studies train on limited, lab‑based datasets, raising concerns about model transferability to diverse printers, mixes, and weather conditions.

The path forward hinges on standardization and data sharing. Public, well‑annotated datasets that combine RGB, depth, and thermal streams would enable fair benchmarking and accelerate algorithmic improvements. Crucially, future work must correlate detected anomalies with compressive strength, durability, and serviceability metrics, turning visual alerts into actionable engineering decisions. When integrated with real‑time control loops, defect detection could automatically adjust extrusion rates, travel speeds, or pause prints, mirroring the closed‑loop quality systems already common in metal AM. Such capabilities would unlock the commercial viability of concrete AM for housing, infrastructure, and disaster‑relief projects.