Introducing 

The I-Team

The iTeam, under the direction of Dr. Paul Sava, focuses on development of full wavefield methodology for imaging the interior of the Earth and other planetary bodies. The team pursues its goals by blending theoretical analysis, software development and data applications. The work covers imaging and inversion theory, computational tools, and field acquisition instrumentation and methodology. The team emphasizes simple and efficient methods that take advantage of the CWP’s growing High-Performance Computing capabilities, and which can be readily deployed in multiple complex geologic settings.

Paul Sava

Paul Sava

Tiago Cabral

Tiago Cabral

Nicholas Dorogy

Nicholas Dorogy

Werter Silva

Werter Silva

The Problems We Solve

How Can We Remove Ghost Images in Marine Acquisition?

Various methods tackle the ghost reflections problem in marine acquisition. These methods have one thing in common; they target receiver-side ghost reflections. The question is, how can we remove the source-side ghost reflections as well? With the aid of deep convolutional neural networks (CNNs), one might solve the ghost reflections problem once and for all. Our current work investigates the ability of CNNs to recover the missing signal due to ghost reflections. Preliminary results, using the  model, show that deep learning is capable of resolving the ghost reflections problem. 

Convolutional Neural Networks

 

Machine Learning

 

Figure (1a) shows a ghost-free trace in red and the same trace with ghost reflections in green. Figures (1b) and (1c) show the corresponding amplitude and phase spectral, respectively. As a result of ghost reflections, ghost notches are present at 0 Hz, 50 Hz, and 100 Hz. By training a CNN, we have been able to recover the missing frequencies, as shown in Figure (2b). Further, the neural network recovered the phase information. Figure (2a) shows the ghost-free trace in red and the reconstructed trace in blue.

How can we explicitly incorporate petrophysical information into elastic full-waveform inversion? ​

​The main motivation for wavefield-based seismic tomography is to obtain an accurate representation of the subsurface properties. This allows us to ensure correct image interpretation, which is a crucial step for seismic exploration.  However, the quality of elastic models derived from full waveform inversion (FWI) is often hampered because the inversion is non-linear and ill-posed, and the objective function is non-convex. Thus, inversion may converge to a local minimum due to the lack of access to good starting models, data noise and modeling errors, the absence of low frequencies in the data, incomplete acquisition coverage and the dimensionality of the model space. Additionally, simultaneous determination of multiple physical parameters using FWI suffers from interparameter trade-off difficulties and model updates provided by this methodology may not represent real lithologies in unconstrained multiparameter inversion. 

To generate geologically feasible and accurate inverted models, we incorporate petrophysical information into the inversion, by explicitly imposing the petrophysical data as penalties to guide models toward realistic lithology. The method uses probability density functions estimated without user-defined parameters, which avoids assuming explicit and potentially inaccurate petrophysical relations among the parameters. This technique addresses many relevant problems of elastic FWI. For more details, check this paper.

Full-Waveform Inversion (FWI)

 

Petrophysical Probabilistic Model Constraints

 

The Techniques and Tools We Use

 

Seismic Anisotropy

Elastic and acoustic reverse-time migration (RTM)

Microseismic imaging, inversion, and monitoring

Velocity analysis and model-building

Elastic and acoustic full-waveform inversion (FWI)

Distributed acoustic sensing

Machine learning

Angle-domain wavefield imaging

Autonomous/dense seismic data acquisition

The Fun We Have

 

Join the I-Team

If you are interested in I-Team or CWP, mention us in your statement of interest in your application. 

Recent Awards and Honors

 

Top 25 papers at SEG, 2019, Odette Aragao, Paul Sava
Best Student Poster Paper at SEG, 2016, Ivan Lim