skip to main content


This content will become publicly available on July 1, 2024

Title: Learned multiphysics inversion with differentiable programming and machine learning
We present the Seismic Laboratory for Imaging and Modeling/Monitoring open-source software framework for computational geophysics and, more generally, inverse problems involving the wave equation (e.g., seismic and medical ultrasound), regularization with learned priors, and learned neural surrogates for multiphase flow simulations. By integrating multiple layers of abstraction, the software is designed to be both readable and scalable, allowing researchers to easily formulate problems in an abstract fashion while exploiting the latest developments in high-performance computing. The design principles and their benefits are illustrated and demonstrated by means of building a scalable prototype for permeability inversion from time-lapse crosswell seismic data, which, aside from coupling of wave physics and multiphase flow, involves machine learning.  more » « less
Award ID(s):
2203821
NSF-PAR ID:
10436336
Author(s) / Creator(s):
; ; ; ; ; ; ; ; ;
Date Published:
Journal Name:
The Leading Edge
Volume:
42
Issue:
7
ISSN:
1070-485X
Page Range / eLocation ID:
474 to 486
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Abstract

    Solving multiphysics-based inverse problems for geological carbon storage monitoring can be challenging when multimodal time-lapse data are expensive to collect and costly to simulate numerically. We overcome these challenges by combining computationally cheap learned surrogates with learned constraints. Not only does this combination lead to vastly improved inversions for the important fluid-flow property, permeability, it also provides a natural platform for inverting multimodal data including well measurements and active-source time-lapse seismic data. By adding a learned constraint, we arrive at a computationally feasible inversion approach that remains accurate. This is accomplished by including a trained deep neural network, known as a normalizing flow, which forces the model iterates to remain in-distribution, thereby safeguarding the accuracy of trained Fourier neural operators that act as surrogates for the computationally expensive multiphase flow simulations involving partial differential equation solves. By means of carefully selected experiments, centered around the problem of geological carbon storage, we demonstrate the efficacy of the proposed constrained optimization method on two different data modalities, namely time-lapse well and time-lapse seismic data. While permeability inversions from both these two modalities have their pluses and minuses, their joint inversion benefits from either, yielding valuable superior permeability inversions and CO2plume predictions near, and far away, from the monitoring wells.

     
    more » « less
  2. The recent application of neural network algorithms to problems in gravitational-wave physics invites the study of how best to build production-ready applications on top of them. By viewing neural networks not as standalone models, but as components or functions in larger data processing pipelines, we can apply lessons learned from both traditional software development practices as well as successful deep learning applications from the private sector. This paper highlights challenges presented by straightforward but naïve deployment strategies for deep learning models, and identifies solutions to them gleaned from these sources. It then presents HERMES, a library of tools for implementing these solutions, and describes how HERMES is being used to develop a particular deep learning application which will be deployed during the next data collection run of the International Gravitational-Wave Observatories. 
    more » « less
  3. null (Ed.)
    International Ocean Discovery Program (IODP) Expedition 360 was the first leg of Phase I of the SloMo (shorthand for “The nature of the lower crust and Moho at slower spreading ridges”) Project, a multiphase drilling program that proposes to drill through the outermost of the global seismic velocity discontinuities, the Mohorovičić seismic discontinuity (Moho). The Moho corresponds to a compressional wave velocity increase, typically at ~7 km beneath the oceans, and has generally been regarded as the boundary between crust and mantle. An alternative model, that the Moho is a hydration front in the mantle, has recently gained credence upon the discovery of abundant partially serpentinized peridotite on the seafloor and on the walls of fracture zones, such as at Atlantis Bank, an 11–13 My old elevated oceanic core complex massif adjacent to the Atlantis II Transform on the Southwest Indian Ridge. Hole U1473A was drilled on the summit of Atlantis Bank during Expedition 360, 1–2 km away from two previous Ocean Drilling Program (ODP) holes: Hole 735B (drilled during ODP Leg 118 in 1987 and ODP Leg 176 in 1997) and Hole 1105A (drilled during ODP Leg 179 in 1998). A mantle peridotite/gabbro contact has been traced by dredging and diving along the transform wall for 40 km. The contact is located at ~4200 m depth on the transform wall below the drill sites but shoals considerably 20 km to the south, where it was observed in outcrop at 2563 m depth. Moho reflections, however, have been found at ~5–6 km beneath Atlantis Bank and <4 km beneath the transform wall, leading to the suggestion that the seismic discontinuity may not represent the crust/mantle boundary but rather an alteration (serpentinization) front. This in turn raises the interesting possibility that methanogenesis associated with serpentinization could support a whole new planetary biosphere deep in the oceanic basement. The SloMo Project seeks to test these hypotheses at Atlantis Bank and evaluate the processes of natural carbon sequestration in the lower crust and uppermost mantle. A primary objective of SloMo Leg 1 was to explore the lateral variability of the stratigraphy established in Hole 735B. Comparison of Hole U1473A with Holes 735B and 1105A allows us to demonstrate a continuity of process and complex interplay of magmatic accretion and steady-state detachment faulting over a time period of ~128 ky. Preliminary assessment indicates that these sections of lower crust are constructed by repeated cycles of intrusion, represented in Hole U1473A by approximately three upwardly differentiated hundreds of meter–scale bodies of olivine gabbro broadly similar to those encountered in the deeper parts of Hole 735B. Specific aims of Expedition 360 focused on gaining an understanding of how magmatism and tectonism interact in accommodating seafloor spreading, how magnetic reversal boundaries are expressed in the lower crust, assessing the role of the lower crust and shallow mantle in the global carbon cycle, and constraining the extent and nature of life at deep levels within the ocean lithosphere. 
    more » « less
  4. International Ocean Discovery Program (IODP) Expedition 360 was the first leg of Phase I of the SloMo (shorthand for “The nature of the lower crust and Moho at slower spreading ridges”) Project, a multiphase drilling program that proposes to drill through the outermost of the global seismic velocity discontinuities, the Mohorovičić seismic discontinuity (Moho). The Moho corresponds to a compressional wave velocity increase, typically at ~7 km beneath the oceans, and has generally been regarded as the boundary between crust and mantle. An alternative model, that the Moho is a hydration front in the mantle, has recently gained credence upon the discovery of abundant partially serpentinized peridotite on the seafloor and on the walls of fracture zones, such as at Atlantis Bank, an 11–13 My old elevated oceanic core complex massif adjacent to the Atlantis II Transform on the Southwest Indian Ridge. Hole U1473A was drilled on the summit of Atlantis Bank during IODP Expedition 360, 1–2 km away from two previous Ocean Drilling Program (ODP) holes: Hole 735B (drilled during ODP Leg 118 in 1987 and ODP Leg 176 in 1997) and Hole 1105A (drilled during ODP Leg 179 in 1998). A mantle peridotite/gabbro contact has been traced by dredging and diving along the transform wall for 40 km. The contact is located at ~4200 m depth at the drill sites but shoals considerably 20 km to the south, where it was observed in outcrop at 2563 m depth. Moho reflections have, however, been found at ~5–6 km beneath Atlantis Bank and <4 km beneath the transform wall, leading to the suggestion that the seismic discontinuity may not represent the crust/mantle boundary but rather an alteration (serpentinization) front. This then raises the interesting possibility that a whole new planetary biosphere may thrive due to methanogenesis associated with serpentinization. The SloMo Project seeks to test these two hypotheses at Atlantis Bank and evaluate carbon sequestration in the lower crust and uppermost mantle. A primary objective of SloMo Leg 1 was to explore the lateral variability of the stratigraphy established in Hole 735B. Comparison of Hole U1473A with Holes 735B and 1105A allows us to demonstrate a continuity of process and complex interplay of magmatic accretion and steady-state detachment faulting over a time period of ~128 ky. Preliminary assessment indicates that these sections of lower crust are constructed by repeated cycles of intrusion, represented in Hole U1473A by approximately three upwardly differentiated hundreds of meter–scale bodies of olivine gabbro broadly similar to those encountered in the deeper parts of Hole 735B. Specific aims of Expedition 360 focused on gaining an understanding of how magmatism and tectonism interact in accommodating seafloor spreading, how magnetic reversal boundaries are expressed in the lower crust, assessing the role of the lower crust and shallow mantle in the global carbon cycle, and constraining the extent and nature of life at deep levels within the ocean lithosphere. 
    more » « less
  5. We experimentally study the transition from droplet to wave regimes in microfluidic liquid–liquid multiphase flows having large differences in viscosity. A unified approach based on periodic pattern analysis is employed to study relationships between dispersed and separated flow regimes, including dripping, jetting, capillary waves, inertial waves and core–annular flows over a wide range of flow rates and viscosity contrasts. We examine the morphology and dynamics of each flow regime based on wavelength, frequency and velocity of repeating unit cells to elucidate their connections and to develop predictive capabilities based on dimensionless control parameters. We demonstrate in particular that pattern selection is contingent upon the propagation velocity of droplets and waves at the transition. We also investigate microfluidic wave breaking phenomena with the formation of ligaments and droplets from wave crests in both capillary and inertial wave regimes. This work expands conventional multiphase flow regimes observed in microchannels and shows new routes to disperse highly viscous materials using interfacial waves dynamics in confined microsystems. 
    more » « less