02 dez 2024
14:00 Master's Defense Room 53 of IC2
Theme
DASF: a high-performance framework for large volumes of seismic data
Student
Júlio Cesar Faracco
Advisor / Teacher
Edson Borin
Brief summary
In Seismic Facies Analysis, Facies Classification is a crucial step in understanding the characteristics and geological composition of a given area. Geologists and oil and gas companies use this information to determine the possibility of exploiting that area to obtain oil and gas, for example. In general, this type of study requires time, precision and involves high costs. Today, with the most varied machine learning techniques, geologists can quickly classify facies using any of these types of algorithms instead of spending hours or days studying the data captured from rock formations. In fact, for this type of classification, there are several well-established unsupervised machine learning techniques, such as K-Means, Kohonen's Self-Organizing Map (SOM) and Gaussian Mixture Models (GMM). However, seismic data are composed of large files and this process of identifying facies can demand a large amount of calculations, making it necessary to use high-performance solutions. With the massive use of graphics processing units (GPUs) to solve problems in the area of artificial intelligence, this work proposes to develop a framework containing the main machine learning techniques for seismic facies classification with support for execution on supercomputers with GPUs. This included the development of some distributed processing techniques such as late loading of seismic data, late processing of seismic attributes for large volumes of data, the SOM algorithm using multiple GPUs, among other techniques relevant to the area. In the end, we will present the result as an open and high-performance seismic facies classification framework so that any researcher or specialist can use it in an easy and simplified way to build seismic models. We will also show some use cases, performance metrics and benchmarks to evaluate the techniques offered by this work.
Examination Board
Headlines:
| Edson Borin | IC / UNICAMP |
| Hervé Cédric Yviquel | IC / UNICAMP |
| Antônio Tadeu Azevedo Gomes | LNCC |
| Maicon Melo Alves | Petrobras |
Substitutes:
| Sandra Eliza Fontes de Avila | IC / UNICAMP |
| Samuel Xavier de Souza | CT / UFRN |