In the relentless pursuit of structural integrity and safety, a groundbreaking study has emerged from the School of Civil Engineering and Transportation at Guangzhou University, led by Liujie Chen. The research, published in the esteemed journal *Frontiers in Built Environment* (translated as “前沿建筑环境”), evaluates the robustness and generalization ability of five cutting-edge Convolutional Neural Networks (CNNs) for crack recognition and classification. The findings could revolutionize how we maintain and inspect infrastructure, particularly in the energy sector where structural health is paramount.
The study, which compared VGG16, GoogLeNet, MobileNetV3-Large, EfficientNetB0, and EfficientNetV2-S, revealed that MobileNetV3-Large stood out as the most robust and efficient model. “The results indicate that MobileNetV3-Large has the best performance,” Chen noted. “For low-resolution building crack images, the accuracy of crack recognition reached an impressive 99.58%, with an F1-score of 99.60%. For higher resolution bridge crack images, the classification accuracy was 95.70%, with a Macro-F1 of 95.67%.”
The implications for the energy sector are profound. Infrastructure such as oil rigs, pipelines, and power plants often operate in harsh environments where cracks and structural damage can lead to catastrophic failures. Traditional inspection methods are time-consuming and labor-intensive, but with the advent of AI-driven crack recognition, these processes could become faster, more accurate, and more cost-effective.
The study employed semantic segmentation based on VGG16-U-Net to address background noise in images and utilized transfer learning with fine-tuning to enhance the performance of the CNNs. This approach not only improved accuracy but also reduced training time, making it a practical solution for real-world applications.
“The results show that MobileNetV3-Large has the best robustness and generalization ability with a small CNN size and the shortest training time,” Chen explained. This efficiency is crucial for commercial applications, where rapid deployment and minimal computational resources are often required.
As the energy sector increasingly adopts digital transformation strategies, the integration of AI-driven crack recognition could become a standard practice. The ability to predict and prevent structural failures before they occur can save millions in maintenance costs and, more importantly, ensure the safety of personnel and the environment.
This research not only highlights the potential of MobileNetV3-Large but also sets the stage for future developments in AI-driven structural health monitoring. As Chen’s work demonstrates, the intersection of civil engineering and artificial intelligence is unlocking new possibilities for maintaining and extending the life of critical infrastructure. The energy sector, in particular, stands to benefit greatly from these advancements, paving the way for a safer and more efficient future.