Real-time health monitoring of civil structures by online hybrid learning techniques using remote sensing and small displacement data

Structural Health Monitoring (SHM) has become an essential topic in most of the modern societies due to critical importance of economic and human losses caused by occurring damage. To avoid any catastrophic event such as partial and global collapse in vital civil structures, many data-based methods under the concepts of artificial intelligence and machine learning using various types of sensing technologies have been proposed and developed. The main objective of such techniques is to learn a computational or statistical model/machine by using sufficient training data during a fixed training period in an offline or batch learning manner. On this basis, in the learning process, one assumes that the training period is sufficiently long to capture all possible environmental and/or operational variations and no damage occurs. However, these assumptions may be problematics due to some major limitations. First, there is no guarantee that new measured data out of the training time has the same variability condition during the learning process. Second, it is illogical to suppose that the structure does not suffer from damage within a long monitoring process. On the other hand, SHM via Synthetic Aperture Radar (SAR) images from the remote sensing technology has become popular among civil engineers. However, an SAR-based SHM strategy itself contains major challenges that should be dealt with. Due to some limitations in SAR images, most of the SHM projects via this technology incorporate a few images, even in long-term monitoring. On the other hand, the majority of research studies in this category is based on directly analyzing displacement samples extracted from SAR images. Under such circum- stances, the process of health monitoring of a complex and vital civil structure may not lead to reasonable consequences owing to considering insufficient small displacement data from SAR images and direct data analysis. To deal with the aforementioned challenges and limitations, this article proposes online hybrid learning methods to detect damage via small displacement data from SAR images. A small set of displacement samples extracted from satellite images of TerraSar-X is then used to assess accuracy and performance of the proposed online hybrid method and compare them with some existing state-of-the-art techniques. Results show that the proposed method is successful in pre-collapse prediction of the bridge even under strong EOV conditions.