PI: Shamim Pakzad
Co-PI(s): Martin Takac
University: Lehigh University
The emergence of dense instrumentation techniques and the ubiquitous nature of data in our society have provided an exciting set of opportunities and challenges in health monitoring of structures for the engineering community. Whereas only a few important structures were instrumented with sporadic sensor networks for a very high cost just 20 years ago, sensor networks today provide the opportunity to collect an enormous amount of data from any structure at low cost, which due to its nature is posing a BIG DATA problem. These datasets cannot be processed to extract their information using the existing analytical methods.

The density of information in these datasets is relatively low, and uncertainty exists in data. When scaled, the existing methods would require memory and computational capacity that is very costly to apply. Furthermore, the existing methods of analysis rely on estimating carefully crafted features that often are limited in what they can do and are not automated in nature, thus not appropriate for a broad range of big data applications.

The objective of this proposal is to develop a deep learning platform to analyze the temporally and spatially dense data collected from digital image correlation (DIC) towards condition assessment and monitoring of the structural systems. DIC data will be used both to develop and calibrate damage an identification network, and test the hypothesis that deep learning is a more computationally feasible approach for extracting damaged data. The data will also be used to verify the sensitivity of the damage detection, localization, and severity estimation to uncertainty in DIC environment. This project will be done in close collaboration with the Trilion Quality Systems and Engineering Services, Lehigh Faculty from Civil and Environmental and Industrial and Systems Engineering, and with the participation of at least one graduate student.