PhD project: Automation of seismic processing with machine learning

Seismic imaging is a powerful technology used by geologists and engineers to image soils and rocs. Like X-ray imaging that allows physicians to detect fractured bones, subsurface images obtained with seismic waves allow geologists and engineers to detect oil reservoirs, to estimate the roc quality or to understand groundwater flow, among other applications. However, seismic imaging technologies are far from being automated: they require extensive manual processing to separate noise from useful signals that can be correlated to geologic and physical properties of rocks. This manual labour is complex and represents a significant cost for the oil and gas industry. It additionally hinders the use of seismic imaging for near-surface applications, particularly for geotechnical and ground water studies.

This PhD project aims to facilitate the use of seismic imaging by automating part of the data processing workflow with machine learning. By using deep learning, we can replace and simplify many of the complex processing steps required to perform imaging. Particularly, velocity model building is a crucial step for migration and full waveform inversion. A neural net capable of detecting and interpreting reflection events will be designed to automatically estimate velocities from raw seismic data. Training will be based on simulated seismic data, which touches a fundamental question: how can we bridge the gap between simulation and real data applications? Unsupervised learning will be used to study how neural nets can be designed to understand seismic events and separate them from the noise. A better comprehension of how neural nets build a representation of seismic data is indeed necessary to optimize their design for signal processing and analysis. Practical applications with industry partners in the oil and gas sector or with civil engineering consulting firms are an integral part of this project.

Applicants from physics, computer sciences, applied mathematics or geology are favored, but anyone interested in the topic is encouraged to apply. Experience with programming is required.