Projects

Machine Learning-based Prediction

of Reinforced Concrete Column Structural Performance

Traditional modeling techniques are popularly used to assess the seismic performance of reinforced concrete (RC) columns. However, traditional modeling approaches are limited in how accurate they can be, in addition to being extremely time-consuming. This project is focused on predicting the structural performance of reinforced concrete (RC) columns due to reversed cyclic loading in a rapid manner via a machine learning approach relating input factors such as the material and section properties and designated failure mechanism, with yield and maximum drift and shear values to automatically derive the column's backbone curve -- a relationship which is often used to indicate the full performance of RC columns and useful for identifying the existing capacity.

Overview of backbone curve extraction from experimental test. This is replicated using machine learning in an efficient and accurate manner.

Real-time Condition Evaluation and Prediction

For Response, Repair, and Recovery After Natural Hazards

With the rise of technology, large amounts of data are collected and shared immediately after any natural hazard event. This project focuses on collecting such data and performing an immediate evaluation of the state of an affected community structurally as well as further. In addition to the condition evaluation, which can be used immediately to help prioritize response and relief efforts, similar data extracted over time is used to build a model which can predict the time until repair and recovery and further prioritize repair and recovery processes. Currently, the work is focused on mitigating the effects of Hurricane Harvey as can be seen in the image to the right (Top: graduate student, Asim Bashir, collecting UAS data in Port Aransas; Bottom: Ph.D. student currently working on this project, Samuel leach).

Ph.D. students flying UAS to collect data after Hurricanes Harvey and Irma.

Performance Assessment of Pressurized Fluid-Distribution Networks

Worldwide, water utilities lose an estimated 9.6 billion USD each year due to water leakage. One-third of reporting countries lose more than 40% of clean water pumped through distribution systems due to leaks. There is a need for more rational and systematic strategies for managing this infrastructure. Advanced sensor-based diagnostic methodologies have the potential to provide enhanced management support. A model-based system-identification method for rapid detection of leak regions in water supply networks was developed called error-domain model-falsification. The method accounts for both modeling and measurement errors. The method has been tested on portions of the water supply networks for the City of Lausanne and the Commune of Bagnes.

Automated, Real-time Routine Bridge Inspections

Via Unmanned Aerial Systems (UAS)

Bridge inspections are necessary to maintain the safety, health, and welfare of the public. All bridges in the United States are federally mandated to undergo routine evaluations to confirm their structural integrity throughout their lifetime. The traditional process implements a bridge inspection team to conduct the inspection, heavily relying on visual measurements and subjective estimates of the existing state of the structure. Conducting unmanned automated bridge inspections allows for a more efficient, accurate, and safer alternative to traditional bridge inspection procedures. Optimizing bridge inspections in this manner enables frequent inspections in order to more comprehensively monitor the health of bridges and quickly recognize minor problems easily corrected before turning into critical issues. This project aims to automate routine bridge inspection operations via a hybrid machine vision-machine learning method, identifying and tracking key structural components and damage in UAS video sequences. A RCNN is implemented and video data from UAS-flights around physical highway bridges are used for the training and testing of the proposed algorithms.

Automated Damage Assessment of RC columns

For Post-Earthquake Evaluations

The rapid increase in computational power, coupled with the improvements in camera quality provides a unique opportunity to develop automated damage assessment approaches. My doctoral dissertation focused on creating a computer vision-based means to automatically determine the existing state of RC columns in RC frame buildings after damage from an earthquake. Within the scope of my dissertation, I developed algorithms to automatically detect and quantify damage on RC columns in the form of cracks and spalling. This involved extensive study of computer science and electrical engineering fields such as computer programming and digital image processing. Using an existing column damage database, I was able to correlate the visual damage to the load carrying capacity of the column. Once the load carrying capacity was determined, an analytical model of the results of this work formed the foundation for the ability to automate the damage assessment process following a natural or man-made disaster. This research required expertise in structural and construction engineering, coupled with knowledge in various fields of computer science and electrical engineering. This work has significant implications in terms of the ability to rapidly (and more accurately) assess the damage following a disaster. This work has led to noteworthy advancements in the broader fields of vision-based inspection and pattern recognition.

FUNDED RESEARCH PROJECTS

  • An Intelligent Spatial Decision Support Tool for Flood Emergency Management, T3-Texas A&M University, Z. Zhang (PI), S.G. Paal, and N. Gharaibeh; 03/01/2021 – 02/28/2024.

  • Disaster City Digital Twin: Integrating Machine and Human Intelligence to Augment Flood Resilience, X Grants Program, Texas A&M University, A. Mostafavi (PI), S.G. Paal, S. Brody, M. Meyer, X. Hu, L. Zou; 09/01/2020 – 08/31/2023.

  • Veteran’s Legacy Geographic Information System: Citizen Science Approach to Extended Memorialization Digitally, Department of Veteran’s Affairs, S. Lyle (PI), S.G. Paal, L. Zhou, and L. Foote; 09/01/2020 – 09/30/2021.

  • CAREER: Understanding the performance of future infrastructure under extreme loads via a novel approach leveraging the power of existing knowledge and artificial intelligence, National Science Foundation (NSF), S.G. Paal (PI); 08/2020 – 07/2024.

  • Develop a real-time decision support tool for rural roadway safety improvements, Texas Department of Transportation (TxDOT), S. Das (PI), S.G. Paal, K. Fitzpatrick, L. Wu, S. Geedipally, E.S. Park, P. Koeneman, I. Tsapakis, and L.D. White; 08/2019 – 08/2021.

  • Development of automated post-disaster damage detection algorithms, United Services Automobile Association (USAA), S.G. Paal (PI), A. Wang (ECE), and R. Murphy (ECE); 03/2019 – 02/2020.

  • Damage Assessment of Residential Infrastructure Impacted by Hurricane Harvey and Irma, National Association of Home Builders (NAHB), M. Koliou (PI) and S.G. Paal; 08/2018 – 02/2019.

  • In-situ deficiency detection and characterization for automated structural repair, T3-TAMU, S.G. Paal (PI), A. Birely, A. Behzadan (Construction Science);04/2017 – 03/2020.

  • Development of a Strategy to Address Load Posted Bridges Through Reduction in Uncertainty in Load Ratings, Texas Department of Transportation (TxDOT), M. Hueste (PI), S.G. Paal, S. Hurlebaus, and J. Mander; 08/2017 – 10/31/2019.

  • Workshop on Additive Manufacturing (3D Printing) for Civil Infrastructure Design and Construction: Arlington, Virginia; July 13-14, 2017, National Science Foundation (NSF), Z. Pei (PI), S.G. Paal, J. Mander, L. Zeng, and S. Bukkapatnam; 03/15/2017 – 02/28/2019.

  • Road Markings for Machine Vision, National Cooperative Highway Research Program (NCHRP), A. Pike (PI) and S.G. Paal; 07/13/2016 – 07/12/2017.