Vision and Attention-Based Models for Process Control in 3D Earth Construction Printing

Summarise with VIRAT
Authors: Yunfan Li, Elodie Paquet, Sébastien Levilly, Mathieu Riand
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Abstract. Anomaly detection is a fundamental step in the control of 3Dprinting processes, particularly in the field of earthen construction. Thepresent paper proposes a novel vision-based methodology for anomalydetection. The pipeline is composed of three main modules: (1) the imageacquisition, (2) the segmentation of interlayer lines and the printedlayers themselves, and (3) the characterization of the layers’ local geometryand texture to identify anomalies. Image acquisition is carried outusing two cameras: one mounted vertically on the robot’s printing nozzleand another positioned to observe the structure from the side. Thesegmentation method aims to recognize and localize the interlayer lineswhich are necessary to isolate patches. The objective of the final moduleis to assess the layer characteristics using an attention-based neuralnetwork originally designed for 3DCP (3D Concrete Printing). This approachenables the analysis of the material’s texture in patches, with thegoal of identifying local anomalies.The experimental results demonstrate the relevance and effectiveness ofthe proposed methodology for the automatic detection and localizationof anomalies in images acquired during the 3D printing of earthen structures.Moreover, the proposed model’s inference time suggests promisingopportunities toward closed-loop control strategies.

Fields of Study & Subjects

Innovations in Concrete and Construction Materials Industrial Vision Systems and Defect Detection Infrastructure Maintenance and Monitoring

Publication details

Journal / source
HAL
Publication date
8 July 2026
Language
EN