Damage localization is the process of detecting the location of damage using a structural health monitoring system. However, existing damage localization methods cannot be used for localizing the damage of bridges in real time because of their slow testing speed. Thus, in this study a damage localization approach was developed using a convolutional neural network (CNN). To develop the CNN model for damage localization, simulation data was generated through a numerical model of a reinforced concrete Rahmen bridge with static loading conditions. The proposed CNN-based approach aims to identify 12 single damage locations or no damage. The approach was trained and tested with three different data set generated with three damage severities, and it was possible to estimate the damage location with an accuracy of 87.3% when the damage severity in the bridge is serious. The results showed that the deep learning has the potential to overcome the limitations of existing damage localization techniques.

The most important structural element of prestressed concrete (PSC) bridges is the prestressed tendon, and in order to ensure safety of such bridges, it is very important to determine whether the tendon is damaged. However, it is not easy to detect tendon damage in real time. This study proposes a novelty detection approach for damage to the tendons of PSC bridges based on a convolutional autoencoder (CAE). The proposed method employs simulation data from nine accelerometers. The accuracies of CAEs for multi-vehicle are 79.5%–85.8% for 100% and 75% damage severities with all error levels and 50% damage severity without error. However, the accuracies for 50% damage severity with 5% and 10% error levels drop to 69.4%–73.3%. The accuracies of CAEs for single-vehicle ranges from 90.1%–95.1% for all damage severities and error levels that are satisfactory. The findings indicate that the CAE approach for multi-vehicle can be effective when the damages are severe, but not when moderate. Meanwhile, if acceleration data can be obtained for single-vehicle, then the CAE approach can provide a highly accurate and robust method of tendon damage detection in PSC bridges in use, even if the measurement errors are significant.

In this study, a field experiment was performed for damage detection on a PSC-I bridge based on a convolutional autoencoder using the damage detection approach proposed in a previous study by the authors. The field experiment measured the acceleration and strain data of the PSC-I bridge while a single vehicle passed the bridge; subsequently, these data were used to train and test the convolutional autoencoder–based damage detection model. The results of the test showed that the convolutional autoencoder–based model could perform accurate and robust damage detection. Furthermore, these findings indicate that the convolutional autoencoder–based damage detection could also perform satisfactorily in practice. The results of this study can form the basis to facilitate the adoption of the convolutional autoencoder–based damage detection method to monitor bridges in practice.

Object detection refers to computer vision technology that identifies the class and location of objects in an image. With the development of deep-learning techniques, object detection has been widely studied in the construction industry and has mainly focused on detecting heavy equipment and workers at outdoor sites. Therefore, an object detection model for small tools in indoor sites, where various accidents frequently occur owing to them, is required to enhance safety. In particular, many high-quality image databases are important for developing object detection approaches for small tools. In this study, 12 small tools commonly used in indoor construction were selected, and a database of 34,738 images of the selected tools was created and used to train the object detection models. YOLOv5x and YOLOv5s are selected as object detection models, for which the mean average precision with the 0.5 intersection-over-union value were 69.1% and 64.4%, respectively, and their frame rates were 58.82 and 142.86 frame per second, respectively. The results demonstrate that the established database is appropriate for the development of object detection for small tools and that object detection can be used for real-time safety management of construction indoor sites.