Applying Deep Learning and Physics Modeling to Structures Subjected to Earthquakes
CEE Assistant Professor Hao Sun, ECE Assistant Professor Yanzhi Wang, and CEE Professor & Chair Jerome Hajjar were awarded a $600K NSF grant for “Physics-Reinforced Deep Learning for Structural Metamodeling.”
Critical to the design of any building in seismic regions is understanding how its structural systems behave under seismic duress. Earthquakes pose a major threat to structures in these regions, and engineers must understand how their designs will respond to these threats. The challenge is compounded when you consider that cities are diverse places, filled with hundreds of buildings of all shapes, sizes, designs, and materials. Providing owners and public officials with an accurate picture of how their region would respond to a variety of earthquakes holds great promise for enhancing seismic resilience. In order to address the computational efficiency of regional seismic hazard modeling, three Northeastern professors are looking to transform the way engineers predict damage to structures subjected to seismic hazards.
Deep Learning and Structural Engineering
Civil and Environmental Engineering Assistant Professor Hao Sun, CDM Smith Professor and Department Chair Jerome Hajjar, and Electrical and Computer Engineering Assistant Professor Yanzhi Wang were awarded a National Science Foundation grant to develop a physics-reinforced deep learning metamodels for structural systems. Sun will serve as Principal Investigator.
“There is a traditional approach that we take in structural engineering to make estimations of what damage is across a region due to a hazardous event like an earthquake,” said Jerome Hajjar. “But it is expensive to do a detailed simulation of each structure for the impact of all types of hazardous events.” Given the time-consuming process of analyzing each building, structural engineers have relied on quicker methods to estimate the damage for individual structures, and therefore aggregate that damage across a region to understand the seismic risk across a metropolis. The grant looks to transform structural engineering practices by integrating them with computer science approaches related to deep learning. “Structural engineers and computer scientists can collaborate to develop groundbreaking new methodologies to facilitate and transform the traditional design and analysis of our structures,” said Hao Sun. Deep learning methods allow the traditional analysis of a building to be conducted in mere seconds, as opposed to the hours or days sometimes required for complex structures.
The professors’ new deep learning modeling technique is reinforced by traditional physics principles to train it for better modeling and understanding of nonlinear structural behavior under seismic loads. Additionally, Assistant Professor Wang’s research into deep learning will help make the method quicker and leaner. Deep learning models require many complex layers and nodes. Wang’s work involves “pruning” such a network, getting rid of unnecessary nodes. “Thus, it makes the metamodel much faster, and computational efficiency is greatly improved,” explained Sun. “The resulting approach will scale up well for analysis from individual buildings to urban applications.”
The grant highlights the department’s focus on urban engineering, which integrates civil and environmental engineering with the emerging technologies of informatics, data science, simulation, and smart sensing. This work also coincides with the launching of Northeastern’s new MS in Civil Engineering concentration in Data and Systems.