CNN-based model for automated anomaly recognition in facade execution to support Quality Management

Abstract

Unmanned Aerial Systems (UAS) are frequently used to inspect building envelopes. Computer vision technologies and Convolutional Neural Networks (CNN) have emerged as promising solutions for automating image-based inspections. However, some anomalies still occur during construction due to the lack of an efficient Quality Management System (QMS). To address this gap, this study proposes an automated CNN-based recognition model to detect and classify four types of anomalies in cast-in-place concrete facades during construction, aiming to support decision-making within the QMS. The research strategy adopted was a Case Study, in which eight CNN models were developed for training and testing using images of cast-in-place concrete wall facades collected by UASs during construction. The model achieved 51.80% precision, 68.50% recall, 65.00% mAP, and an F1-score of 58.99% during training, making it the most accurate among the eight models. Future studies will focus on fully integrating the proposed method’s workflow, enabling automated image analysis and the automatic generation of reports.