Implications and Benefits of Deep Learning (CNN) in Borehole Image Interpretation: Cost Savings and Improved Accuracy

Abstract Objectives/Scope saving cost and increasing accuracy of the data interpretation are considered a serious challenge within the oil and gas industry. These challenges come to the surface when there are a critical discission on the drilling of new wells inside the geological units created with the normal procedures of the sedimentological studies inside any area. the main focus of this study is the application of the Convolution neural networks (CNN) techniques which outstanding performance in pattern recognition and classification to predict the borehole image facies in an efficient and accurate way inside the Qawasim Formation which was deposited during late Messinian time. Methods, Procedures, Process The main focus of this study is the application of the Convolution neural networks (CNN) techniques which outstanding performance in pattern recognition and classification to predict the borehole image facies in an efficient and accurate way inside the Qawasim Formation which was deposited during late Messinian time. This study presents the application of CNN workflow into five major steps including data collection, preprocessing, CNN model learning testing and evaluation. And For performance analysis. The dataset used to train and evaluate the model consists of 1350 images from three types of labeled facies (cross laminated, laminated and massive facies). The trained labeled mages will pass inside a tunnel of convolution and max pooling feature extraction filters and finally a fully connected layers neural network applied as a final stage of the classification results from the model Results, Observations, Conclusions The produced model demonstrates high efficiency and scalability for automatic facies classification with a reasonable accuracy reached to 82%. This model particularly useful in when quick facies prediction is necessary to support real-time decision making and for cost reduction scenarios during performing a numerous number of borehole images. The produced model is easily implementable and expandable to other clastic reservoirs in order to create a quick and accurate geological model and be implemented for the future field development plane and production enhancement from a specific zone. Novel/Additive Information the application of deep learning, as demonstrated in this study, will kill two birds with one stone, it increases the efficiency and accuracy Borehole image interpretations, decreasing the cost impact of the geological studies and minimize the risk by increase the accuracy of geological model for any reservoir.