Infrastructure Automated Defect Detection with Machine Learning: A Systematic Review

Infrastructure defects pose significant public safety risks and, if undetected, can lead to costly repairs. While machine learning (ML) technologies have significantly enhanced the capabilities for inspecting infrastructure, a comprehensive synthesis of these advancements and their practical application across various infrastructures is lacking. This study addresses this gap by providing a literature review, offering a consolidated view of current ML methodologies in Infrastructure Automated Defect Detection (IADD). This research employs a systematic literature review (SLR) approach to analyse 123 papers on ML methodologies applied to IADD. The analysis reveals the wide use of deep learning architectures like Convolutional Neural Network and its variants, which perform well in defect detection across various infrastructures, including roads, bridges, and sewers. However, standardised, comprehensive datasets are critical to train and test these models more effectively. The study also highlights the importance of developing ML approaches that can accurately assess the severity of defects, an area currently underexplored but with significant implications for risk management in infrastructure. This SLR provides a consolidated perspective on ML technologies’ advancements and practical applications in IADD, and it offers substantial value to researchers, engineers, and policymakers engaged in infrastructure asset management.