Concrete defect identification and measurement in buildings

Abstract

Ensuring the durability of civil buildings is critical for preventing long-term structural failure. Traditional damage assessment methods, based on visual inspections, are labour-intensive and subjective. The adoption of digital platforms emerges as a solution for complete monitoring of a building during its life cycle. This study, carried out as part of the European BIM2TWIN project, focuses on the monitoring of concrete surface quality using vision techniques and deep learning algorithms. Four neural network models are employed for cracks, honeycombing, pitting, and exposed bars, collectively analysing the images to identify and quantify defects. Project managers can assess the criticality of defects based on accurate pixel counts and geometric features, converting measurements from pixels to millimetres. Data is then stored in the digital platform, providing a historical record for future reference and decision-making by project managers.