In-Person
Where: The Petroleum Club of Shreveport, 15th floor
Cost: $20, Children 10 and under $8
If you’d like a seat, kindly use the form below to make your reservation by the preceding Friday.
We encourage members to invite guests, spouses, and friends to any of our meetings.
Virtual
Where: Zoom
Cost: $10
If you’d like to attend virtually, please pay with a credit card below and use the email address where you would like Zoom login details sent. Please make your reservation to attend virtually by noon the day before the meeting.
Biography
Scott Comegys
Scott G. Comegys is a professional geoscientist in Shreveport, LA and is currently the manager and co-founder of LAS PRO, LLC, owner of IntelliLAS. His background, both educationally and professionally, is subsurface investigations relating to the exploration for hydrocarbons. He has worked offshore Gulf of Mexico and onshore Gulf Coast USA primarily. He has broad experience utilizing well logs in a variety of geoscience applications and wanted to automate the process of digital data capture using new technology and to automate workflows he currently uses, thereby modernizing and industry. His job at LAS is to direct software development by scoping projects and working with third-party developers. He has also created a training database consisting of more than 20 mega pixles in length (70,000 images equiv.) for future research.
Abstract
In the current state of the oil and gas industry, increasing focus is being directed to efficiency through adoption of new technologies. This allows for companies to do more with less and to better leverage available data for informed decision making at all levels of the E&P process. Furthermore, the exploitation of shale reservoirs has reinvigorated domestic basins across the USA, highlighting the importance of legacy, vertical well log data. Vast amounts of 3D seismic have been acquired and are being used in conjunction with well data to model reservoir characteristics through processes such as inversion. The need for a more advanced log digitizing process has emerged to help facilitate these endeavors. While most logs are available digitally, they are provided as scanned raster images. For more advanced analysis, and greater end-user value, the curve data need to be manually traced directly from the image. The process is cumbersome and the underlying technology that guides it has changed little in the prior two decades. Typically, curves are traced individually in a time-consuming process which has no automated alternative. The existing auto tracing functionalities can be helpful in expediting the workflow, but this more the exception than the rule when dealing with poor image quality, non-solid line types, etc. Recent advancements in image classification technology have opened the door to a fully automated solution, specifically, convolutional neural networks. These have shown impressive results in extracting features from raster images once provided some ground truths, i.e. training. Such supervised learning technology not only allows for a program to identify features on unseen images but also to improve their performance over time. Our research has proven the efficacy of this technology to identifying and extracting individual curve data from raster log images with limited human input. Importantly, no seeding or manual starting points need be given. The technology can differentiate between different line types, such as solid or dashed, and can successfully choose correctly between multiple curves in one image. These results include a high percentage of accuracy on the identified curve. Results to date indicate the possibility of tremendous cost savings to organizations in both time and equipment; most individual curves are traced within two minutes and no local hardware requirements are necessary. With other steps added into the process, such as image preparation and post processing, the solution for a fully automated digitizing solution is well within reach.
