Learning viscoelasticity with neural networks

Phoebe DeVries, Thomas B. Thompson, & Brendan J. Meade

Published August 15, 2016, SCEC Contribution #6886, 2016 SCEC Annual Meeting Poster #018

Viscoelastic models have been widely used to explain geodetic observations of the earthquake cycle, time-dependent stress transfer, and delayed earthquake triggering. The calculations involved in these modeling efforts are often computationally intensive; as a result, studies tend to adopt a few fixed rheological structures and model geometries, and examine the associated predictions of stress evolution over short (<10 yr) time periods at a given depth or specific location after a large earthquake. Training a deep neural network to accurately approximate these viscoelastic solutions – at any time, location, and for a large range of rheological structures – allows these calculations to be done quickly, for geometrically complex faults and over many earthquake cycles, with arbitrarily high spatial and temporal resolution. Preliminary tests suggest this method could accelerate these calculations by a factor of 500 and perhaps orders of magnitude more; this kind of acceleration will facilitate an understanding the potential viscoelastic effects of large earthquakes across wider ranges of model parameters and at larger spatial and temporal scales than have previously been possible.

DeVries, P., Thompson, T. B., & Meade, B. J. (2016, 08). Learning viscoelasticity with neural networks. Poster Presentation at 2016 SCEC Annual Meeting.

Related Projects & Working Groups
Stress and Deformation Over Time (SDOT)