Digital Core 3D Reconstruction Based on Micro-CT images via a Deep Learning Method

摘要

Digital core technology is the cutting-edge technology of petroleum engineering. Digital core 3D reconstruction is a key issue in digital core technology. At present, the most accurate 3D reconstruction method of digital core is X-Ray Computed Tomography. However, this method is too expensive to afford even for the big oil companies. Especially with the current low oil prices, oil companies are looking to reduce costs. There are two reasons for the high cost of X-ray computed tomography equipment: first, as the accuracy of the equipment improves, the cost of the sensor and the supporting circuit board rises linearly; second, the traditional method of establishing a digital core requires scan the core layer by layer continuously, which allows the sensor to process only a limited number of cores throughout the entire life cycle, increasing the cost of rebuilding a single core. In response to this problem, we propose a new method to make the sensor no longer scan rock core layer by layer. The workflow is as follows: the unscanned layers are interpolated through the values of scanned layers, and then use deep learning algorithm to improve the interpolation quality. The feasibility experiments show that the to-be-established digital core can be obtained with 10 scanned layers instead of 73 scanned layers. This is actually equivalent to using more calculations to release the pressure on the x-ray sensor and its circuit board. Although this method requires the device to do additional calculations, it greatly reduces the workload of the x-ray sensor and its supporting circuits. Since computing equipment such as CPU and GPU are much cheaper than them, our proposed method has practical value.