Register now After registration you will be able to apply for this opportunity online.
Fracture detection from geological datasets using machine learning
Borehole image logs serve as essential datasets for extracting a wide range of geological features. Traditionally, analysing these logs has involved a tedious labelling process by geological experts. Recently, machine learning has demonstrated remarkable capabilities in image processing. This project focuses on utilizing machine learning model to improve the efficiency and accuracy of detecting and classifying discontinuity features within borehole wall imagery.
Clay rock is being investigated as a potential host rock for the storage of nuclear waste. Therefore, detecting and differentiating various discontinuities in clay rocks are of great important. Discontinuities, such as natural and artificial fractures, are critical indicators of rock mass movements and can significantly influence the structural integrity assessments required for safe waste storage.
Leveraging machine learning can significantly streamline this identification process, enhancing the efficiency of feature extraction and potentially uncovering valuable hidden information. This project is in collaboration with Mont Terri Lab. Plenty of high-resolution borehole wall imagery will be available for the experiment. Since the provided dataset will only include limited annotation, data labelling would be a necessary part in the project.
The initial phase of the project will involve data preprocessing and labelling, where student need to identify and map various types of discontinuities feature according to provided guideline. After constructing the complete dataset, student can apply advanced deep learning techniques, such as semantic segmentation or object detection. The U-Net architecture has previously shown promising results for image segmentation with our data, however, there is still a lot of flexibility to explore other machine learning approaches.
**Your Task:**
• Literature review;
• Data preprocessing and labelling;
• Implement machine learning model on the dataset.
**Your Profile:**
• Proficiency in QGIS or ArcGIS;
• Experience in Python, and familiar with PyTorch would be a plus;
• Knowledge of geology would be a plus;
• Interest in machine learning.
Clay rock is being investigated as a potential host rock for the storage of nuclear waste. Therefore, detecting and differentiating various discontinuities in clay rocks are of great important. Discontinuities, such as natural and artificial fractures, are critical indicators of rock mass movements and can significantly influence the structural integrity assessments required for safe waste storage. Leveraging machine learning can significantly streamline this identification process, enhancing the efficiency of feature extraction and potentially uncovering valuable hidden information. This project is in collaboration with Mont Terri Lab. Plenty of high-resolution borehole wall imagery will be available for the experiment. Since the provided dataset will only include limited annotation, data labelling would be a necessary part in the project. The initial phase of the project will involve data preprocessing and labelling, where student need to identify and map various types of discontinuities feature according to provided guideline. After constructing the complete dataset, student can apply advanced deep learning techniques, such as semantic segmentation or object detection. The U-Net architecture has previously shown promising results for image segmentation with our data, however, there is still a lot of flexibility to explore other machine learning approaches.
**Your Task:**
• Literature review;
• Data preprocessing and labelling;
• Implement machine learning model on the dataset.
**Your Profile:**
• Proficiency in QGIS or ArcGIS;
• Experience in Python, and familiar with PyTorch would be a plus;
• Knowledge of geology would be a plus;
• Interest in machine learning.
Create a viable dataset from the available borehole wall imagery; Train a machine learning model to achieve efficiency and accuracy in extracting fracture features in borehole imagery.
Create a viable dataset from the available borehole wall imagery; Train a machine learning model to achieve efficiency and accuracy in extracting fracture features in borehole imagery.
For inquiries and applications, please contact Rushan Wang (rushan.wang@slf.ch) and PD Dr. Andrea Manconi (andrea.manconi@slf.ch).
For inquiries and applications, please contact Rushan Wang (rushan.wang@slf.ch) and PD Dr. Andrea Manconi (andrea.manconi@slf.ch).