Rockbursts in surrounding rocks containing structural planes present significant challenges to engineering construction. Therefore, the spatiotemporal evolution law is crucial for the prevention and control of deep underground rockbursts. To address the deficiencies in the study of the spatiotemporal evolution of structural plane-induced rockbursts, granite specimens with non-closed structural planes at various angles were prepared. Utilizing the concept of stress relaxation, uniaxial compression stress relaxation tests were conducted to analyze and discuss the evolution of rockburst fractures and the differences between immediate and time-delayed occurrences influenced by structural planes of differing angles. The results indicate that: (1) when the stress relaxation or attenuation process exhibits three stages—rapid attenuation, gradual attenuation, and stable attenuation—it is classified as a stress relaxation process, characterized by the absence of rockbursts or the occurrence of local weak rockbursts with a long time delay. Conversely, if the process shows two stages—gradual attenuation and rapid attenuation—it is classified as a stress attenuation process, characterized by immediate or short time-delayed rockbursts. (2) For a given structural plane, as the load increases, the state of the rock mass containing structural planes transitions from a stress relaxation process (without rockbursts) to a process where stress relaxation transforms into stress attenuation (with rockbursts, characterized by long time-delayed rockbursts), and eventually to a stress attenuation process (with rockbursts, characterized by short time-delayed or immediate rockbursts). (3) As the angle of the structural plane increases, the failure mechanism of rockbursts shifts from being controlled by the structural plane to being governed by strain, with the failure of the structural plane progressing from the tip to the entire structural plane. (4) Under the same peak load, rockbursts are more likely to occur in structural planes with smaller angles, and the time required for these occurrences is shorter. For structural planes with the same angle, higher peak loads result in shorter durations before rockbursts occur. The findings of this research provide a foundational understanding and guidance regarding the mechanisms, prevention, control, monitoring, and prediction of structural plane-induced rockbursts in deep underground engineering.

Accurate measurement of the elastic modulus of rock is a key component in geotechnical engineering construction design and stability evaluation. However, in deep high-stress environments and brittle formations with well-developed weak structures, obtaining intact and standardized rock samples for direct measurement of elastic modulus is often challenging. This study proposes a method for predicting elastic modulus by leveraging artificial intelligence to build a bridge between microstructural characteristics and elastic modulus, requiring only microstructure images as input to efficiently infer the modulus. The implementation process consists of two stages: first, an improved DeepLabv3+ model is employed to perform semantic segmentation experiments on a multi-source image dataset, generating microstructural characterization images; second, a RockNet-E model is constructed to establish a mapping relationship between the characterization images and the elastic modulus, enabling accurate prediction. Research results show that the improved DeepLabv3+ model achieves mean intersection over union (IoU) of 87.7%, 97.5% and 87.9% for quartz, feldspar, and mica, respectively, outperforming mainstream models such as DeepLabv3+, PSPNet, and UNet. On the validation set, the RockNet-E model achieves mean absolute error (MAE), root mean square error (RMSE), mean absolute percentage error (MAPE) and R-squared (R2) values of 1.20 GPa, 1.51 GPa, 1.83% and 0.61, respectively. The elastic modulus prediction errors are all below 5%, and the predicted value distribution matches the experimental results. This study provides a low-cost and highly efficient approach for intelligent, image-driven prediction of rock mechanical parameters.