Automating Image Analysis to Learn from Hazard Events

After a natural hazard event, such as a hurricane or earthquake, engineers are deployed to collect perishable visual data to document the consequences (the failures and damage) of the event. Many natural hazard studies are in need of new knowledge from these visual data to mitigate the risks (e.g., damage and loss of life) in future events. However, this is now entirely done manually, requiring costly and time-consuming tasks. Thus, unfortunately, only a very small portion of these images are shared, curated and actually used, leading to minimal use of the expensive data. This reality often prevents the extraction of critical discovery that can further strengthen the science and engineering to save human lives in future events. We aim to enable real-world automation in data analysis and organization through the use of technologies to collect data and powerful GPUs to classify such data for use to examine building failures during a natural hazard event. In our research group, we have gathered over 300,000 images from various recent natural hazard events, and spent a great deal of time labeling this data based on a schema intended to support the scientific use of this data. Our ground-truth damage database is the first-of-its-kind, and has the potential to advance scene classification for such complex and unstructured data to the point where it can support field applications within just a few years.

Graduate Students: Jongseong Choi (Brad), Ali Lenjani, Xiaoyu Liu

Structural Health Monitoring (SHM) and the Application of Wireless Sensor Networks

Health monitoring allows the engineer to use sensing of the structural responses in conjunction with appropriate data analysis and modeling techniques to monitor the condition of a structure. This approach has advantages which will lead to the ability to continuously observe the structural responses, as well as an improved understanding of the behavior of civil engineering structures. Furthermore, in many structures the inspection process is hindered by the fact that the columns, beams, and connections are typically covered by non-structural elements, and significant efforts are required to access these areas for visual inspection. However, the use of health monitoring techniques would simplify such procedures, allowing the engineer to ascertain the degree of structural damage without requiring visual inspection of the structural connections and components. The purpose of this research is to develop such techniques. In particular, the SHM system based on Wireless Sensor Networks (WSNs) has shown considerable promise. It has several advantages over most traditional SHM systems: low production cost, low installation and maintenance expenses, fast installation, reprogrammable software, convenient reconfiguration, and so on. Using the WSN, a dense deployment of measurement points in a SHM system is possible, which helps obtaining better damage detection results

Graduate Students: Wei Song, Zhuoxiong (Charlie) Sun, Sriram Krishnan