Particle physics laboratory

AI predicts the physics of future fault slip in earthquakes in the lab

October 10, 2022 – An artificial intelligence approach borrowed from natural language processing – much like language translation and auto-filling text on your smartphone – can predict future fault friction and the next moment of failure with a high resolution in laboratory earthquakes. The technique, which applies AI to acoustic signals from the fault, advances previous work and goes beyond predicting aspects of the future state of the fault’s physical system.

“In simple terms, we predict future friction. This has never been done, and it offers a potential route to short-term prediction of the timing of earthquakes on Earth,” said Chris Johnsonco-lead author of an article on the findings in Geophysical Research Letters.

Paul JohnsonThe paper’s corresponding author, a geophysicist and researcher at Los Alamos National Laboratory, leads a team that has made steady progress in applying various machine learning techniques to the challenge of predicting earthquakes in the lab and on field.

“The acoustic signals emitted from the lab fault contain information foreshadowing the future fundamental physics of the system throughout the seismic cycle and beyond, as we now show,” said Paul Johnson. “This has never been seen before.”

As a model of language translation

In a novel approach, the Los Alamos team applied a deep learning transformer model to acoustic emissions broadcast from the lab fault to predict frictional state.

“The deep learning transformer model we used is synonymous with a language translation model, such as Google Translate, using a codebook to translate a sentence into another language,” said Chris Johnson. “You can think of this as writing an email in English and having the AI ​​translate English to Japanese while anticipating your words and auto-filling the end of the sentence.”

Chris Johnson said the AI ​​”takes data on what’s happening right now and says what’s happening next on the rift.”

The Los Alamos team had previously predicted the failure time of faults in laboratory earthquakes and historical slow-slip Earth data using a number of machine learning techniques. Applying machine learning to data from laboratory shear experiments has demonstrated that defect emissions are imprinted with information regarding its current state and location in the slip cycle.

Indeed, the statistical characteristics of the continuous seismic signal emitted by the fault and identified by machine learning allowed the Los Alamos researchers to predict the instantaneous – but not future – evolution of friction, displacement and other characteristics. of the fault, as well as the timing of the next lab. earthquake.

In this previous work, waveform, or acoustic emission, data is entered into a model to predict the current state of the fault system. This prediction includes an estimate of the countdown, or time to failure, for the next slip event, with some degree of uncertainty, which is not a future prediction but a description of the current state of the system.

“Now we make a future prediction from past data, which goes beyond describing the instantaneous state of the system. The model learns from the waveforms to predict future fault friction and when the next slip event will occur using only past information, not using any data from the future of no interest,” Chris said. Johnson.

“The model is not constrained by physics, but it predicts physics, the actual behavior of the system,” Chris Johnson said.

“The next challenge is whether we can do it on Earth to predict the future movement of faults, for example,” said Paul Johnson. “It’s an open question, because we don’t have long datasets for model training like we do in the lab.”

The method could be applied to other disciplines, such as non-destructive testing of materials, where it could provide information on progressive damage and imminent damage to, for example, a bridge.

The paper: “Predicting Future Laboratory Fault Friction Through Deep Learning Transformer Models,” by Kun Wang, Christopher W. Johnson1, Kane C. Bennett, and Paul A. Johnson in Geophysical Research Letters. DOI: 10.1029/2022GL098233

Funding : US Department of Energy, Office of Science, Office of Basic Energy Sciences and Laboratory Directed Research and Development at Los Alamos National Laboratory.


Source: Los Alamos National Laboratory