Author ORCID Identifier



Date of Graduation


Document Type


Degree Type



Statler College of Engineering and Mineral Resources


Petroleum and Natural Gas Engineering

Committee Chair

Ebrahim Fathi

Committee Member

Samuel Ameri

Committee Member

Kashy Aminian

Committee Member

Mohamed Nouri Salem El Sgher

Committee Member

Timothy Carr

Committee Member

Elias Pirayesh


The oil and gas industry has historically spent significant amount of capital to acquire large volumes of analog and digital data often left unused due to lack of digital awareness. It has instead relied on individual expertise and numerical modelling for reservoir development, characterization, and simulation, which is extremely time consuming and expensive and inevitably invites significant human bias and error into the equation. One of the major questions that has significant impact in unconventional reservoir development (e.g., completion design, production, and well spacing optimization), CO2 sequestration in geological formations (e.g., well and reservoir integrity), and engineered geothermal systems (e.g., maximizing the fluid flow and capacity of the wells) is to be able to quantify and map the subsurface natural fracture systems. This needs to be done both locally, i.e., near the wellbore and generally in the scale of the wellpad, or region. In this study, the conventional near wellbore natural fracture mapping techniques is first discussed and integrated with more advanced technologies such as application of fiber optics, specifically Distributed Acoustic Sensing (DAS) and Distributed Strain Sensing (DSS), to upscale the fracture mapping in the region. Next, a physics-based automated machine learning (AutoML) workflow is developed that incorporates the advanced data acquisition system that collects high-resolution drilling acceleration data to infer the near well bore natural fracture intensities. The new AutoML workflow aims to minimize human bias and accelerate the near wellbore natural fracture mapping in real time. The new AutoML workflow shows great promise by reducing the fracture mapping time and cost by 10-fold and producing more accurate, robust, reproducible, and measurable results. Finally, to completely remove human intervention and consequently accelerate the process of fracture mapping while drilling, the application of computer vision and deep learning techniques in new workflows to automate the process of identifying natural fractures and other lithological features using borehole image logs were integrated. Different structures and workflows have been tested and two specific workflows are designed for this purpose. In the first workflow, the fracture footprints on actual acoustic image logs (i.e., full, or partial sigmoidal signatures with a range of amplitude and vertical and horizontal displacement) is detected and classified in different categories with varying success. The second workflow implements the actual amplitude values recorded by the borehole image log and the binary representation of the produced images to detect and quantify the major fractures and beddings. The first workflow is more detailed and capable of identifying different classes of fractures albeit computationally more expensive. The second workflow is faster in detecting the major fractures and beddings. In conclusion, regional subsurface natural fracture mapping technique using an integration of conventional logging, microseismic, and fiber optic data is presented. A new AutoML workflow designed and tested in a Marcellus Shale gas reservoir was used to predict near wellbore fracture intensities using high frequency drilling acceleration data. Two integrated workflows were designed and validated using 3 wells in Marcellus Shale to extract natural fractures from acoustic image logs and amplitude recordings obtained during logging while drilling. The new workflows have: i) minimized human bias in different aspects of fracture mapping from image log analysis to machine learning model selection and hyper parameter optimization; ii) generated and quantified more accurate fracture predictions using different score matrices; iii) decreased the time and cost of the fracture interpretation by tenfold, and iv) presented more robust and reproducible results.