Artificial intelligence powered real-time quality monitoring for additive manufacturing in construction

Abstract

In the manufacturing process of 3D Concrete Printing (3DCP), defects and anomalies have a significant impact on both the success rate and the quality of the final products, underscoring the need for real-time monitoring. Currently, monitoring is primarily based on manual observation and existing automated methods are limited in real-time performance and accuracy. This study introduced a real-time and highly accurate defect detection and measurement system for using deep learning (DL) and computer vision (CV) techniques. A range of improvement methods were applied in YOLOv7, showing better capacities of accuracy and speed for detecting defects in 3DCP than current cutting-edge detectors such as YOLOv8. Notably, the virtual high-fidelity data were produced by DL based data augmentation strategy and their effects were assessed. Replacing real data as the training dataset, the generated virtual data were used in the models to improve measurement accuracy. Applying the proposed method, the comprehensive insights into 3DCP defects were obtained. Consequently, the relationship formula between defect frequency and printer parameters was investigated by the proposed method, guiding operators in effectively controlling printer parameters and preventing breakpoint defects during the printing process.