[ad_1]
Austin, Texas – Our homes and offices are only as solid as the ground beneath them. When this solid ground turns to liquid – as sometimes happens during earthquakes – buildings and bridges can collapse. This phenomenon is known as liquefaction, and was a major feature of the 2011 Christchurch, New Zealand earthquake, which measured 6.3 and killed 185 people and destroyed thousands of homes.
One of the positive aspects of the Christchurch earthquake was that it was one of the most well-documented earthquakes in history. Since New Zealand is seismically active, the city had many seismic sensors. The survey after the event provided a wealth of additional data on how soils were responding across the city.
Two researchers from the University of Texas at Austin developed a machine-learning model that predicted the amount of lateral movement that occurred when the Christchurch earthquake caused the soil to lose strength and shift relative to its surroundings.
The results were published online in Earthquake Spectra in April 2021.
“It is one of the first studies of machine learning in our geotechnical engineering field,” said postdoctoral researcher Maria Giovanna Durante, a former Marie Sklodowska-Curie Fellow at the University of Austin. “It’s a huge amount of data in our field. If we have thousands of data points, maybe we can find a trend.”
Durante and Elaine Rathjee, Janet S. Cockrell Centennial Chair in Engineering at the University of Austin and principal investigator for the DesignSafe electronic infrastructure funded by the National Science Foundation, first used the Random Forest approach with binary classification to predict whether lateral propagation motions occurred in a specific location. Then they applied a multi-layer classification approach to predict the amount of displacement, from none to more than one meter.
“It was important to identify specific input features that correspond to the phenomenon we are studying,” Durante said. “We’re not using the model as a black box – we’re trying to integrate our scientific knowledge as much as possible.”
Durante and Rathje trained the model using data on the experienced ground shaking peak (a liquefaction inducer), water table depth, topographic slope, and other factors. In total, more than 7,000 data points from a small area of the city were used for the training data – a huge improvement, as previous geotechnical machine learning studies used only 200 data points.
They tested their model citywide at 2.5 million locations around the epicenter to determine displacement. Their model predicted whether liquefaction had occurred with 80% accuracy. It was 70% accurate in determining the amount of offset.
The researchers used the Frontera supercomputer at the Texas Advanced Computing Center (TACC), one of the world’s fastest, to train and test the model. TACC is a principal partner in the DesignSafe project, providing computing, software, and storage resources to the natural hazard engineering community.
Access to Frontera provided Durante and Rathje machine learning capabilities on a scale previously unavailable in the field. Deriving the final machine learning model required testing 2,400 possible models.
“It would have taken years to do this research elsewhere,” Durante said. “If you want to do a parametric study or do a comprehensive analysis, you need to have computational power.”
[ad_2]
Picture Credit!