Fractures play a crucial role in the hydromechanical behavior of rocks. To investigate the fundamental fracturing mechanism, the imagery of hydraulic fracture evolution is captured in laboratory testing of rock specimens. Conventionally, temporal-spatial characteristics of rock fractures must be identified and extracted manually or by image processing techniques (IPTs) for interpretation, requiring enormous time and labor with low accuracy. This paper develops a deep learning-based method that quickly and automatically identifies and extracts hydraulically induced fractures in rock specimens at the pixel level.
The applicability of this method is validated through image datasets from hydraulic fracturing tests. This method shows better effectiveness and efficiency than previous IPTs. The accuracy of the deep learning method reaches 99 percent and the average processing speed is only 389 ms per image when adopting an NVIDIA Tesla T4 GPU, saving a large amount of time compared to human work.
Hydraulic fracturing has been widely used in geo-engineering areas including hydrocarbon extraction, the disposal of waste drill cuttings underground, heat production from geothermal reservoirs and fault reactivation in mining (Adachi et al., 2007). To study the mechanism of the fracturing process, lots of physical experiments were performed to capture the temporal evolution of fractures (AlDajani, 2017; AlDajani, 2022; Li, 2019; Roshankhas et al., 2018). The temporal evolution of the fracturing is done by taking high-resolution and high-speed images and analyze them. Up to now this analysis has usually been done "manually" (AlDajani, 2022; Wang et al., 2021). Such manual analysis is a very time-consuming process, even though benefitting from the expertise of the researchers. Considering these challenges, this paper aims to test the capability of machine learning in the fracture extraction from the images of rock specimens in a hydraulic fracturing test. As a first step, we will consider a single fracture image taken toward the end of the fracturing process.
