DETERMINING THE PROPERTIES OF ROCK SAMPLES USING DEEP MACHINE LEARNING

Abstract

Porosity, absolute permeability, and diffusion coefficient are crucial characteristics governing fluid flow in the porous media of geological formations. Determining these properties traditionally involves resource-intensive and time-consuming processes. However, with the advancement of deep learning methods in the last 3–4 years, artificial neural networks have gained significant traction in predicting the transport properties of the fluid-porous medium system and the geometric characteristics of porous samples based on their images. This approach allows for the rapid determination of these properties with acceptable accuracy. Consequently, questions arise regarding the effectiveness and adequacy of deep learning methods for these purposes.

The aim of this article is to conduct a scientific review of literature from open sources on the determination of absolute permeability, diffusion coefficient, and porosity from their images acquired through various scanning methods. As data for the review, scientific articles from various open sources were used. Additionally, this article incorporates proprietary data, specifically images from four carbonate samples. Convolutional neural networks were examined as the method of choice.

The results of this article comprise a scientific review of moderate depth regarding the effectiveness and applicability of the approach for determining important characteristics of porous media using deep machine learning methods based on sample images. In this article, we also present the results of predicting the open porosity of four carbonate samples based on their X-ray images using the convolutional neural network model we constructed. The conducted review has demonstrated that images (scans) of geological rock samples obtained through various scanning methods allow for the calculation of their transport properties with a high degree of accuracy using deep machine learning algorithms, and this can be achieved within a significantly short timeframe. This implies that deep machine learning can serve as a valuable alternative tool for estimating the properties of geological rock samples based on their images. The convolutional neural network model we constructed exhibited predictive capability for the porosity of three carbonate samples with a coefficient of determination ranging from 0.936 to 0.976.