To evaluate the stability and long-term structural safety of hard rock roadways,this study analyzes the long-term deformation characteristics of surrounding rock induced by TBM excavation and performs high-precision trend predictions.However,traditional prediction methods are often impeded by idealized assumptions and the limitations inherent in single-model frameworks,thereby failing to achieve the high-precision predictions required for surrounding rock deformation under hard rock field conditions.Using the TBM excavation project at the-280 m experimental level of the Beishan underground research laboratory in China as a case study,this research employed real-time in-situ deformation data captured by an embedded multipoint displacement meter system to construct a hybrid prediction model,termed CPO-CLA.This model integrates the crested porcupine optimizer(CPO),convolutional neural network(CNN),long short-term memory(LSTM),and Attention mechanism,utilizing the monitoring data to conduct precise predictions of the long-term deformation of tunnel surrounding rock.The results indicate that during TBM excavation,the internal displacement of the surrounding rock undergoes two distinct phases:an excavation phase and a stable convergence phase,with the maximum cumulative displacement reaching 0.359 8 mm.The CPO-CLA model exhibited superior predictive performance in TBM hard rock tunnel engineering,achieving a mean coefficient of determination R2 of 0.953 on the test set,thereby validating its effectiveness.Comparative analysis against various mainstream meta-heuristic algorithms reveals that the CPO algorithm significantly outperforms them in terms of convergence speed,computational efficiency,and optimization accuracy.
A composite cemented backfill system incorporating magnesia,fly ash,and tailings was developed to address the challenge of low-cost utilization of solid-liquid waste generated from underground Sylvinite mining.An orthogonal experimental design with three factors at three levels was employed,focusing on binder content,the mass ratio of brine to tailings,and fly ash content.A series of tests were conducted on the cemented specimens,including uniaxial compression tests,acoustic emission monitoring(AE),scanning electron microscopy(SEM),and energy dispersive spectroscopy(EDS).The variations in compressive strength and elastic modulus of specimens at different factor levels were investigated.An optimal formulation of the composite cemented material was established.The energy dissipation characteristics during various loading stages were analyzed,and the micromorphology of hydration products within the specimens was examined to elucidate the composite cementation mechanism of the solid-liquid tailings.Additionally,the contribution weights of different factors to the damage variable were quantified,leading to the establishment of a damage constitutive model for the magnesia-fly ash-tailings composite under varying factors and levels.The results demonstrate that the optimal specimen,with a binder content ratio of 1:5,a brine-to-tailings ratio of 1:5,and a fly ash content of 25%,achieves a peak compressive strength of 4.51 MPa.The brine-to-tailings ratio was identified as the primary factor influencing mechanical properties.The failure mode was governed by the fly ash content,transitioning from tensile splitting to shear failure while exhibiting significant plastic residual deformation as the fly ash content increased.The AE activity exhibited three distinct stages:initial quietness,rapid growth,and post-peak quietness.Higher levels of binder and fly ash content,combined with a lower brine-to-tailings ratio,enhanced the energy release per AE event,as well as both the cumulative event count and the total cumulative AE energy.These changes significantly accelerated micro-crack development and improved the connectivity of macroscopic fractures.Based on cumulative AE energy and factor contribution weights,a segmented damage constitutive model for the composite backfill was developed.The theoretical calculations from this model closely matched the experimental data,effectively capturing the stress-strain relationships and damage evolution processes across specimens with different mix proportions.This study provides a foundation for the development of cemented backfill technology utilizing solid-liquid tailings from underground potash mines.
Nano-scale coal-rock CT imaging offers the advantages of high resolution and three-dimensional structural reconstruction;however,its acquisition is expensive and its axial resolution is limited,which compromises digital core modeling and seepage simulation.To enhance axial resolution while preserving the true topological structure,this study proposes a generative adversarial network-based super-resolution reconstruction method,termed SRGAN-CT.The proposed method integrates residual blocks and a perceptual loss,and develops a specialized discriminator to strengthen the discrimination of structural details.In addition,a transverse-slice mapping strategy is adopted to construct the training dataset,and orthogonal experiments are conducted to evaluate the robustness and generalization performance of the model.The results show that SRGAN-CT effectively preserves both fine details and topological structures of nano-CT images under multiple upsampling factors and random perturbations,outperforming conventional super-resolution methods in terms of structural-metric deviations and topological errors.Overall,this study verifies the high-fidelity reconstruction capability of SRGAN-CT for 3D topological and geometric information,providing a new technical pathway for efficiently obtaining high-resolution digital cores.
