In a significant advancement for the construction sector, researchers have unveiled a groundbreaking approach to detecting cracks in concrete surfaces using deep learning techniques. The study, led by Faris Elghaish from the School of Natural and Built Environment at Queen’s University Belfast, demonstrates a multi-level optimisation strategy that enhances the accuracy of two prominent deep learning models, ResNet101 and Xception. This research, published in ‘Developments in the Built Environment’, promises to revolutionize structural health assessments and maintenance strategies.
With the construction industry increasingly relying on technology to ensure safety and durability, the ability to accurately identify and classify distress in concrete surfaces is paramount. The study utilized an extensive dataset of 40,000 images, showcasing various types of cracks, to train and test the models. The results are impressive: ResNet101 achieved an accuracy of 98.9%, while Xception reached an astounding 99.2%. “Our approach not only improves detection rates but also enhances the efficiency of structural health surveys, which can be a game changer for large buildings,” Elghaish stated.
The research employed four innovative optimisation algorithms: Sequential Motion Optimisation (SMO), Shuffled Frog-Leaping Algorithm (SFLA), Grey Wolf Optimisation (GWO), and Walrus Optimisation (WO). By strategically grouping these algorithms into two layers, the researchers were able to significantly boost the models’ performance. Before optimisation, the accuracy rates were considerably lower, with Xception at 87.4% and ResNet101 at 83.1%. The optimisation process, therefore, represents a substantial leap forward in the field.
The implications of this research extend beyond mere academic interest. For construction firms, the ability to detect cracks with over 99% accuracy translates to reduced costs and improved safety measures. Structural engineers can now rely on precise data to make informed decisions regarding maintenance and repairs, ultimately extending the lifespan of buildings and ensuring public safety. “This technology not only saves time and resources but also helps prevent catastrophic failures,” added Elghaish.
As the construction industry continues to evolve, embracing artificial intelligence and machine learning will likely become essential in maintaining infrastructure integrity. This study sets a precedent for future research and applications in the field, paving the way for smarter, more efficient construction practices.
For more insights into this pioneering research, you can visit the School of Natural and Built Environment at Queen’s University Belfast. The findings underscore the growing intersection of technology and construction, highlighting a future where data-driven decisions enhance the safety and longevity of our built environment.
