InterPore2021

Improving the reliability of phase segmentation by combining 3D imaging and machine learning methods

3 Jun 2021, 20:00

1h

Poster (+) Presentation (MS15) Machine Learning and Big Data in Porous Media Poster +

Speaker

Ms Parisa Asadi

Description

X-ray CT imaging, which provides a 3D view of a sample, is a powerful tool for investigating the internal features of a porous rock. Classifying phases in these images is highly desirable but, like any other digital rock imaging technique, is time-consuming, labor-intensive, and subjective. Combining 3D X-ray CT imaging with machine learning is a promising, powerful method that addresses the key challenges. This method also makes it possible to simultaneously consider several extracted features in addition to the original images for a more reliable phase segmentation. This study investigates the performance of several filtering techniques with three machine learning methods and one deep learning method to assess the potential for reliable feature extraction and pixel-level phase segmentation of a Marcellus and a Mancos shale. The extracted features are produced by applying image filtering techniques, such as Gaussian and Median, to X-ray CT images and evaluated to determine the best ones to use as feature inputs for machine learning algorithms to obtain more accurate phase segmentation. Three machine learning methods including k-means clustering (k-means), Random Forest (RF), and Feed forward Artificial Neuron Networks (FNN), as well as the deep learning U-Net model, are applied to the original images and the stacked extracted features and their performance compared and contrasted. The results show that all classification algorithms deliver high accuracy ranging from 0.87 to 0.96 when considering more dimensionality (i.e., more features). RF demonstrates the best performance among the machine learning models, with an accuracy of 0.96, due to its capability to handle imbalanced datasets and data scarcity. The U-Net model outperforms the RF model when applied to the test images. ML-based phase segmentation of X-ray CT images enables faster data collection and interpretation than traditional methods. Considering more dimensionality (i.e., more features) provides promising and reliable segmentation results that are valuable for analyzing the composition of dense samples, such as shales, which are significant unconventional reservoirs in oil recovery.