Laboratory Earthquakes

Laboratory earthquakes can predict future fault friction

An artificial intelligence approach derived from natural language processing like language translation and auto fill for text on your smartphone. 

Laboratory earthquakes can predict future fault friction and subsequent failure time with high resolution. The technique, which applies artificial intelligence to fault acoustic signals, develops and exceeds previous work by predicting aspects of the future state of the fault's physical system.

In the words of Chris Johnson, co-lead author of a paper on the findings in Geophysical Research Letters, we are simply predicting a future collision. That has never happened, and it offers a potential way to near-term prediction of earthquake times on Earth. The acoustic signals emitted by the laboratory fault contain advance information about the future fundamental physics of the system through the entire earthquake cycle and beyond. It has never been seen before as we are now showing.

Paul Johnson, a geophysicist and laboratory fellow, leads a team that has made steady progress in applying various machine learning techniques to the challenge of predicting earthquakes in the laboratory and in the field.

The laboratory applied a deep learning transformer model to the acoustic emissions transmitted from the fault to predict the collision condition. The Los Alamos team.

The deep-learning transformer model we used is similar to a language translation model like Google Translate that uses a codebook to translate a sentence into another language. You can think of this as writing an email in English and having the AI ​​translate English into Japanese and auto fill the end of the sentence by guessing your words. Artificial intelligence takes the data of what is currently happening and predicts what will happen next on a fault. Chris Johnson says

The Los Alamos team previously estimated fault failure times in laboratory earthquakes, using several machine learning techniques on historical slow-slip Earth data. Applying machine learning to data from laboratory erosion experiments demonstrated that the fault emission was imprinted with information on its current state and slip cycle.

In fact the statistical properties of the continuous seismic signal emanating from the fault and identified by machine learning have the Los Alamos researchers evolving in the immediate but not the future. Allowed estimation of fault friction, displacement and other properties.

In previous work, the waveform, or acoustic emission, is the input to the data model to estimate the current state of the fault system. That estimate includes a countdown estimate for the next slip event or time to failure with some degree of uncertainty. It is not a forecast of the future but a description of the current state of the system.

Now we are predicting the future from past data which is no more than 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, rather than using only past information. Using any data from a phase of interest in the future. Chris Johnson says that the model is not constrained by the physics but rather it uses the physics to predict the actual behavior of the system.

The next challenge is whether we can do this in the ground to predict future fault displacement, for example. This is an open question because we don't have as many data sets for model training as we do in the laboratory. Says Paul Johnson.

Story Source:

Materials provided by DOE/Los Alamos National Laboratory.

Note: Content has also been changed for vogue and length.

Google Search Engine

Oxygen Level

Know Carbonate Dissolve

Visit Official Home Page