An Approach in Caving Recognition by An Integrated Model of Computer Vision and Machine Learning for Any Drilling Environment

Abstract

Cavings are a valuable source of information when drilling which indicate that a failure has occurred downhole. This project is an integrated study of Machine Learning, Computer Vision and Geology to recognize the presence of cavings in the shakers and how to link the morphology of cavings with borehole problems. The analysis can be transferred and applied to Midland-Delaware (Permian) Basin and Fort Worth Basin or any unconventional reservoir. The methodology involved developing a structured picture database of cavings from the Norwegian Continental Shelf which it is used to extract features such as Shape, Roundness, Color and Size. Each picture will be converted into vectors that can be used to establish an array of features and labels; Algorithms written in Python can recognize the different characteristics of the caving and causal mechanisms. Also, MxN pixels of images will be the data points to cluster using k-means algorithm for the color feature which is used to correlate with the formation type. To extract the size feature, calibration was first done using a reference object, and OpenCV library is used to extract length and width of each object present in the picture. Caving recognition was achieved by taking a picture of a new sample, this will enable a faster assessment of the drilling problem. A set of 25,536 pictures of cavings is being used for training and testing with 9 different supervised Machine Learning algorithms and 3 different architectures of Neural Networks. Besides, features extracted from caving images can be successfully linked to cavings causal mechanisms; as a result, an integrated model between features, drilling parameters and causal mechanisms, will suggest and implement remedial action in order to solve wellbore stability problems. This process is done automatically by the algorithm.

AAPG Datapages/Search and Discovery Article #90343 ©2019 AAPG Southwest Section Annual Convention, Dallas, Texas, April 6-9, 2019