XWITM Revolutionary Earth Model Building on the Cloud
What is FWI?
FWI is a data-fitting optimisation feedback loop applied to seismic recorded data which:
- models the field experiment with a system of partial differential equations (PDE) which honour the true physics of wave propagation
- computes the misfit between generated predictions and real world sensor measurements
- solves a PDE constrained optimisation to find the model update direction which adjusts the existing model to reduce the misfit
- uses correctly constructed cost functions for global minimum convergence that optimally matches simulations to reality
The XWITM optimisation framework
XWITM is the industry's leading Full Waveform Inversion (FWI) velocity optimisation algorithm due to:
- Combining proprietary Adaptive Waveform Inversion (AWITM) and generic Full Waveform Inversion (FWI) cost functions for robustness against cycle skipping without low frequencies in the field data
- Using least-squares Reverse Time Migration (RTM) to generate deep reflectors with which to accurately update the macro-velocity model below the diving wave zone with Reflection Waveform Inversion (RWI)
- Applying total variation and asymmetric total variation regularisation for Constrained Waveform Inversion (CWI)
Why use XWITM on the Cloud?
The S-Cube Cloud hosted on Amazon Web Services (AWS) enables you to access the industry's leading Full Waveform Inversion algorithm on demand in a managed compute environment configured for data parallel wave equation velocity optimisation. It consists of the following components:
- XWITM: S-Cube’s cloud-native fork of FULLWAVE3D FWI + additional methodologies such as AWI, CWI, RWI ...
- S-Cube Cloud: distributed fault tolerant architecture for large-scale data-parallel parameter optimisation on the AWS spot market
- S-Cube Cloud Dashboard: web-based dashboard for real-time result visualisation and hyper-parameter tuning
It enables you to:
- Upload your raw field shot gathers to a dedicated AWS S3 bucket with a few commands
- Parameterise a new job and calculate the computational workload per shot
- Select the number of shots per iteration and number of iterations to run
- Monitor and QC the live job on the S-Cube Cloud dashboard accessed via a URL secured utilising Auth0
Some Concrete Numbers
In this NW Australia deployment, the shot throughput is 4 per hour.
- 5186 total shots (9 saillines)
- Grid size = 25m, nx = 113, ny =1041, nz=161
- 20% shots per iterations
- 550 x 16 core EC2 instances
- 2 shots per worker instance
- Average runtime per shot = 14 mins
- Runtime per iteration = 28 mins
S-Cube Cloud Architecture
We're experts in making FWI work in the cloud. We make heavy use of commodity cloud hardware fleets to minimise cost and maximise burst availability, while adapting a variety of robust cloud-native AWS services to fill some capabilities that traditional HPC systems served.
The stability and performance of cloud-native data storage systems allows acceptable replacement of broadcasting operations for the FWI orchestrator with minimal code change through smartly architected adaptor layers. This means we keep our scientific code doing what it does best, but also raise the bar on all operational aspects around it as well, massively increasing the stability, capability, observability, and burst capacity without needing heavy capital investment.
Run FWI on the Cloud
Discover an unprecedented increase in the resolution of your velocity model.