This Faculty Early Career Development (CAREER) Program grant pioneers a change-based risk analysis method that uses diverse spatiotemporal data for predictive defect detection of aging civil infrastructure systems and early warning of structural collapse. Many failures of bridges, marine structures, and other large-scale facilities often can be forecast by gradual changes in their appearances. Civil engineers have found that certain changes across multiple structures always occur together and form collective patterns that predict collapses. For example, decaying steel structures often have collective crack developments around components under cyclic loads. Maintenance agencies can collect various spatiotemporal data, such as images, for monitoring structures and predicting behaviors of structural systems. Unfortunately, current change analysis practices are manual and tedious, so they cannot recover collective patterns from heterogeneous data. This award is for fundamental research in the development of an automated process for detecting changes across heterogeneous data and predicting structural collapse. The developed procedure will enable automated, data-driven analysis of the decays in civil infrastructure systems. Detailed data-driven change analyses of decaying structures have applications in the domains of infrastructure management, aerospace engineering, biomedical engineering, and material science. Therefore, results from this research will broadly benefit the American economy and society. In addition, findings from this project enable wider, global efforts related to change-based risk analysis of civil infrastructures through domestic and international collaborations. This research will aid workforces capable of handling diverse spatiotemporal data for proactive infrastructure management. Finally, change analysis games developed in this effort are to be integrated into the engineering curriculum, K-12 summer workshops, and industry outreach activities. A focus of this award is to engage people from underrepresented groups, especially Hispanics and Native Americans.
Existing change analysis algorithms calculate deviations of data points from their nearest neighbors in other data sources. Such neighborhood searching cannot reliably track both global displacements and local deterioration of objects across heterogeneous data. That limitation impedes engineers from characterizing change patterns for predictive defect analysis of structural systems. This research resolves this difficulty through matching isomorphic spatiotemporal patterns across data sources to track both global and local changes of objects, and then correlating these changes across multiple structures for collapse prediction. Specifically, the research effort: 1) examines algorithms that hierarchically match similar spatiotemporal patterns in order to identify corresponding parts of data; 2) investigates algorithms that use corresponding parts of data to detect and classify changes; 3) uses spatial statistical algorithms to discover patterns of correlated changes of multiple structures; and 4) identifies patterns of changes that have strong correlations with collapses. The contributions include: 1) a new method that predicts structural failures through analysis of spatiotemporal change patterns of multiple structures and 2) computationally efficient algorithms for reliably detecting, classifying, and correlating changes across heterogeneous data. Developed techniques will be validated using field data collected from twenty-five structures within the United States, China, Korea, and Hong Kong, along with data from the literature and civil infrastructure data repositories.
