Researchers at the Los Alamos National Laboratory used machine learning to identify sounds indicating that a fault will soon rupture.
Earthquakes happen when blocks of earth near the junctions between tectonic plates suddenly slip along fractures. Built-up friction causes this dramatic “slipping,” and the energy is then released via seismic waves.
The research team’s machine learning algorithm collected large quantities of data from previous earthquakes and identified a distinct sound pattern preceding earthquakes.
Using the Lab to Save Lives
The team created “lab earthquakes” using steel blocks and rocky materials to recreate the “slipping” preceding seismic sounds. They trained a computer to analyze the seismic and acoustic signals emitted during movements along a fault.
They undertook to determine whether the experimental fault’s seismic signal contained information about its current frictional state, leading to an understanding of when the quake might occur and at which magnitude. Using machine learning and AI to study the seismic data, they uncovered a quantitative link between the fault’s frictional state and the signal’s strength.
This method is now being tested in real-world scenarios to ascertain whether the pattern follows for real earthquakes.
Source: https://www.lanl.gov/discover/science-briefs/2018/March/0320-earthquake-fault-behavior.php