Pujin Wang's research while affiliated with Tongji University and other places

Concrete cracking is one of the most significant damage types in reinforced concrete structures due to its potential to adversely affect durability, safety, and serviceability and even reduce the bearing capacity during operation. Thus, damage inspection of damage caused by concrete cracking is important for management, maintenance, and structural assessment for both damaged and undamaged existing buildings but with concrete cracking after a long time of use that needs reconstruction or renovation. This study provides an improved building damage inspection approach by applying Unmanned Aerial Vehicles (UAVs) and state‐of‐the‐art deep learning algorithms to detect concrete cracks on building surfaces. Two distinct architectures for Convolutional Neural Networks (CNNs), namely ResNet50 and YOLOv8 based on classification, and object detection approaches to create a total of 11 models are established, trained, and compared. The classification models attained accuracy levels exceeding 99%, whereas the object detection models achieved approximately 85%. All models effectively identified and accurately located concrete cracks on building surfaces. Besides, the CNN models' capacity to detect cracks is influenced by a variety of model hyperparameters, encompassing factors such as model architecture, the number of network layers, different training dataset sizes, and the quantity of trainable parameters necessary to learn the specific features of detection targets during the training process. The results of this study ultimately demonstrate that the proposed approach can yield accurate detection results and holds high potential for application in crack inspection to advance automatic damage inspection in building structures to a greater extent. In addition, it is important to note that a universal rule cannot be established rule as a larger and more complex model, or an increased number of trainable parameters, necessarily leads to improved detection performance. Models that are trained from scratch using local datasets might not necessarily result in enhanced performance in comparison to the improvements gained through fine‐tuning via transfer learning. Therefore, an appropriate training type, dataset size, task complexity, computational resources, and time demands to achieve a balance between accuracy and efficiency should be considered for specific application scenarios.

Ultra-high-performance concrete (UHPC) is widely used in the field of large-span and ultra-high-rise buildings due to its advantages such as ultra-high strength and durability. However, the large amount of cementitious materials used results in the cost and carbon emission of UHPC being much higher than that of ordinary concrete, limiting the wide application of UHPC. Therefore, optimizing the design of the UHPC mix proportion to meet the basic properties of UHPC with low carbon and low cost at the same time will help to realize the wide application of UHPC in various application scenarios. In this study, the basic properties of UHPC, including the compressive strength, flexural strength, fluidity, and shrinkage properties, were predicted by machine-learning algorithms. It is found that the XGBoost algorithm outperforms others in predicting basic properties, with MAPE lower than 5% and R2 higher than 0.9 in four output properties. To evaluate the comprehensive performance of UHPC, a further analysis was conducted to calculate the cost- and carbon-emissions-per-unit volume for 50,000 UHPC random mixes. Combined with the analytical hierarchy process (AHP) model, the comprehensive performance of UHPC, including basic properties, cost-per-unit volume, and carbon-emissions-per-unit volume, was evaluated. This study proposes an optimized UHPC mix proportion, based on low-cost or low-carbon emission, oriented to comply with the excellent overall performance and obtain its corresponding various properties.

To alleviate the workload in prevailing expert-based onsite inspection, a vision-based method using state-of-the-art deep learning architectures is proposed to automatically detect ceiling damage in large-span structures. The dataset consists of 914 images collected by the Kawaguchi Lab since 1995 with over 7000 learnable damages in the ceilings and is categorized into four typical damage forms (peelings, cracks, distortions, and fall-offs). Twelve detection models are established, trained, and compared by variable hyperparameter analysis. The best performing model reaches a mean average precision (mAP) of 75.28%, which is considerably high for object detection. A comparative study indicates that the model is generally robust to the challenges in ceiling damage detection, including partial occlusion by visual obstructions, the extremely varied aspect ratios, small object detection, and multi-object detection. Another comparative study in the F1 score performance, which combines the precision and recall in to one single metric, shows that the model outperforms the CNN (convolutional neural networks) model using the Saliency-MAP method in our previous research to a remarkable extent. In the case of a large-area ratio with a non-ceiling region, the F1 score of these two models are 0.83 and 0.28, respectively. The findings of this study push automatic ceiling damage detection in large-span structures one step further.

To overcome the limitations of human-based visual onsite inspections, a vision-based method using deep learning with a convolutional neural network (CNN) is proposed to detect and localize the damaged ceiling of large-span structures. The designed CNN model is trained, validated, and tested using 1953 ceiling images, and a prediction accuracy of 86.22% is obtained. The results of a comparative study demonstrate that the saliency map method can accurately localize regions with damaged ceiling and demonstrate the outline shape of the damaged regions. The features visualization using a saliency map reveals that the CNN model is capable of recognizing the overall layout of the inside of a building through images of the intact part of the building and regions with damaged ceiling through images of damaged areas, although, the non-ceiling regions, particularly isolated regions with regular shapes, have a significant influence on the damage prediction probability. Non-ceiling regions and the area ratio are two important factors influencing the prediction accuracy of the CNN model. A statistical analysis indicates that a prediction accuracy of greater than 98% can be obtained in the case of no non-ceiling regions and an area ratio ranging from 20% to 30%. Therefore, photographic method is proposed for capturing ceiling images and improving the prediction accuracy of the CNN model.