Buildings - Special Topics in Structures

Full Session with Abstracts

339370-1 - Multi-Limit State Seismic Performance Assessment of Tall Buildings Using Machine Learning

Saturday, April 21
8:00 AM - 9:30 AM
Location: 202A

Multi-Limit state (functional loss, building closure, demolition and collapse) assessment has been widely used to quantify the level of damage and disruption caused by earthquakes at the individual structure and community scales. This paper presents a Machine Learning approach to probabilistically predict the limit state of a damaged building using the response and damage patterns as predictors. The response patterns are characterized through various engineering demand parameters and the damage patterns are obtained using the simulated visual damage states in both structural and nonstructural components. Non-linear response history analyses are performed using sequential ground motions, where the first record generates the patterns and the second record calibrates the performance of the damaged building with such patterns based on a predefined relative performance criteria. Such analyses are performed in large volumes to achieve a robust dataset and Machine Learning is then employed to establish a relationship between the response and damage patterns and the calibrated limit states. The sensitivity of all the predictors are interactively evaluated for each limit state so that the amount of required information could be significantly reduced while the prediction retains an acceptable performance.
A case study is conducted using a 42-story dual system building with reinforced concrete core-walls and special moment frames. The results show distinct response and damage patterns among the analysis cases being classified into different limit states. Although no clear boundary could be found for any individual predictor, Machine Learning is able to recognize the various response and damage patterns in the high-dimension predictor space. Different predictor subsets are found for multiple limit states given the same performance requirement. More importantly, satisfactory performance in terms of sensitivity, specificity and overall accuracy is achieved in the multi-limit state seismic performance assessment. The proposed methodology could be used for rapid assessment and decision-support immediately after a seismic event as well as for long-term resilience and lifecycle performance evaluation.

Henry Burton

Assistant Professor
University of California, Los Angeles

Dr. Burton joined the University of California, Los Angeles Civil and Environmental Engineering Department after completing his Ph.D. in Structural Engineering at Stanford University where Professor Gregory Deierlein was his thesis advisor. He has significant industry experience and is a registered structural engineer (S.E.) in the state of California. Dr. Burton spent six years in practice at Degenkolb Engineers where he worked on a number of projects involving seismic design, evaluation and retrofit of existing buildings. Current major research projects include (1) developing a post-disaster recovery model for residential communities (NSF Award Number 1538747), (2) utilizing remote sensing to assess the implication of tall building performance on the resilience of urban centers (NSF Award Number 1538866) and (3) stochastic characterization of aftershock building collapse risk (USGS Award Number G16AP00006), (4) developing design and assessment principles for the next generation of resilient and sustainable buildings (NSF Award Number 1554714) and (5) developing performance-based design and assessment methods for self-centering steel braced frame systems with controlled rocking.


Send Email for Henry Burton


339370-1 - Multi-Limit State Seismic Performance Assessment of Tall Buildings Using Machine Learning

Attendees who have favorited this

Please enter your access key

The asset you are trying to access is locked. Please enter your access key to unlock.

Send Email for Multi-Limit State Seismic Performance Assessment of Tall Buildings Using Machine Learning