Argillaceous rocks are of interest to engineers due to their wide distribution, sensitivity to environmental factors (e.g., water and external loads), and complex mechanical properties and failure characteristics (Corkum and Martin, 2007). Common in China, red-bed soft rock is considered to be a special kind of argillaceous rock characterized by weak permeability, strong hydrophilicity, and water softening capacity. Argillaceous interlayers (mudstone), are common in Chinese red-bed regions and have a significant influence on engineering stability. According to the statistics from Chai et al. (2014), approximately 18 % of landslides in the Three Gorges Reservoir area occur in these mudstones. With increased engineering activity (e.g., slope stepped excavation), engineering problems (e.g., landslides) related to red-bed soft rock are increasing. Thus, a systematic investigation of the mechanical behavior and failure mechanisms of red-bed soft rock considering factors such as water and loading stress is urgently required.
Numerous studies have examined the mechanical properties of rock. For example, Geertsema (2002) found that the shear strength of mudstone was far lower than expected, which should be considered when designing a dam on argillaceous rock mass with planar joints. Furthermore, loading tests revealed that mudstone deformation and failure under compression was mainly dependent on axial loading and dilatancy in the lateral direction (Wen et al., 2018; Zheng et al., 2018). He et al. (2010) conducted a true triaxial single-face dynamic unloading test on limestone to investigate its acoustic emission (AE) characteristics and found AE signals with a high amplitude and low frequency before the test sample was broken. A direct shear test was also used to study bedding planes formed by purplish-red mudstone and argillaceous limestone from the Three Gorges Reservoir Area (Wu et al., 2018). Results illustrated that bedding planes between intensively weathered mudstone and slightly weathered limestone were the most dangerous potential sliding planes. Environmental scanning electron microscopy (SEM) and digital image correlation techniques have also been used to study the mechanical behavior of mudstone at a microscopic level (Wang et al., 2015). However, none of the aforementioned studies have considered the effect of water on rock degradation.
Water-rock interaction has a considerable impact on rock mechanical strength and failure behavior (Liu et al., 2008a, 2008b; Deng et al., 2016), and studies generally agree that water reduces rock strength (Lin et al., 2013; Liu et al., 2018; Liu et al., 2020; Lu et al., 2017; Wasantha and Ranjith 2014). Water-induced weakening was more apparent in rocks rich in clay compared to those rich in quartz (Grant, 1982; Poulsen et al., 2014). Furthermore, rocks with different clay content would have varying degrees of dilatancy and softening following water absorption, resulting in mechanical strength decline (Erguler and Ulusay, 2009). The strength of mudstone is related to its saturation (water content); that is, the uniaxial compressive strength of rock is either negatively linear or negatively exponentially correlated with water content (Ramos da Silva et al., 2008; Yilmaz, 2010). Jiang et al. (2014) conducted water-mudstone interaction tests and found that weakened mechanical strength was attributed to the physical expansion force generated by the dilatancy of clay minerals following water absorption. Gautam and Shakoor (2013) found that the clay content was a key factor affecting mudstone disintegration and failure. Furthermore, mudstone strength and slope stability was significantly reduced by the drying-wetting cycle (Xu et al., 2018). In recent years, advanced techniques such as SEM and computed tomography (CT) have been used to study microscopic water-induced weakening mechanisms in mudstone, providing a strong basis for researchers to directly observe the mechanical properties of water-weakened rocks (Jiang et al., 2014; Wang et al., 2015; Xiao et al., 2016; Zhang et al., 2012). However, these techniques are not capable of studying the evolution of rock cracks.
Rock failure is a consequence of the initiation, propagation, and connection of cracks. Furthermore, research shows that the combined action of external load and water is the main reason for the mechanical deterioration of red-bed mudstone. However, the evolution of rock cracks is difficult to monitor and observe. Furthermore, evaluating the impact of water flow in cracks on rock mechanical properties remains challenging. The circulatory system of the human body is analogous to the rock internal fracture network. Advanced digital radiography (DR) techniques have provided technical support for understanding changes in human body structure, and enhanced angiography has become an effective means to directly observe and calibrate targets (Hirono et al., 2003). The DR image, which is similar to a CT image, is composed of pixels with different gray levels from black to white, which reflect the X-ray absorption coefficient of the associated voxel (Sheikhzadeh et al., 2019). However, it can be difficult to obtain accurate results from image segmentation methods based on threshold or machine learning techniques due to complex geometric topology, small grayscale differences, and the need to manually design a task-related recognition process (Shi et al., 2016). Consequently, the accurate detection of crack distribution remains challenging at the pixel scale. In recent years, use of the convolutional neural network (CNN) at the pixel scale has been prominent in various deep neural networks because it is capable of self-learning from image samples, showing considerable advantages in terms of image recognition, segmentation, and classification (Ai et al., 2020; Polsinelli et al., 2020; Xu et al., 2019).
Stepped excavation disturbance and water-induced weakening would inevitably lead to deformation and failure, accelerating the instability of the red-bed slope. In this study, a visual experimental platform based on DR and a multi-level loading device was designed and constructed to realize the synchronous visualization of rock fracture processes and fissure water flow. Multilevel loading was applied to red-bed cubic mudstone samples for the collection of AE signals, stress, flow rate, and DR images during the failure process. An original image processing method based on Hough transform (HT) and a CNN was proposed to segment and extract cracks from DR imagery to reveal water-induced damage characteristics and the failure mechanisms of red-bed mudstone under the action of water-rock hydro-mechanical coupling. Finally, we discuss our research findings in the context of early warning index selection for rainfall-type landslides at an engineering scale. The original experimental platform proposed and evaluated in this study provides a new and powerful tool to investigate the mechanical behavior of different rock types under the action of water-rock hydro-mechanical coupling at a laboratory scale. The research findings will help to improve disaster prevention for red-bed geological bodies at an engineering scale.