Deep-Learning-Based Surface Crack Detection with a Mobile Robot for Preventive Conservation of Plastered Timber-Framed Historical Buildings

The detection and management of surface cracks in historical buildings are pivotal in maintaining the physical well-being of these structures. Early identification of cracks through regular monitoring and timely adjustments can diminish significant harm before it happens, thereby circumventing the substantial budgets needed for the restoration of deteriorated structures. Conventional approaches to evaluating surface cracks on historical buildings require laborious and time-intensive evaluation and judgment of skilled experts. Lightweight robots may be used for some of these tasks, particularly on timber-braced structures vulnerable to various damages. This study presents surface crack detection experiments using deep learning techniques and a custom mobile robot, aiming to support routine inspections by enabling early identification and mitigation of potential damage. It focuses on Beypazarı, a town recognized by UNESCO on the tentative list, with the Beypazarı History and Culture Museum serving as the primary case. The detection algorithm, based on deep learning, is trained on visual data collected from the town’s plastered timber-framed buildings and used for validation and inference. Equipped with sensors and imaging, the custom robot surveys the environment and transmits visual data to the central server in real time. Performance assessment through metrics that are training loss, mean average precision (mAP), and average recall (AP) confirms the system’s efficiency for object detection.