The mechanical properties of large-scale columnar jointed rock masses (CJRMs) after excavation would be reduced due to the unloading relaxation, in addition to its influence on the stability of underground caverns. Moreover, the cutting of the interlayer shear weakness zones through the cavern complicates the overall stability of CJRM, and brings challenges to the stability control of project. In view of such problems encountered during the construction of underground caverns, we reconstructed the large-scale CJRMs containing interlayer shear weakness zones to conduct the 3D rock failure process analysis. The numerical models of surge tank caverns during excavation were established and the mechanical properties, failure modes and acoustic emission (AE) characteristics were analyzed. The simulation results indicate that the large-scale columnar joints and C2 interlayer shear weakness zones influence the mechanical properties and failure modes significantly. Three types of damage modes in numerical models containing crush-shear failure along columnar joints, tensile-slip failure along C2 interlayer shear weakness zone and crush failure inside and around caverns were summarized. The concentration zones of tensile stress were found around each cavern and gradually shrunk from sector shape to linear shape. The caverns were divided into two parts along C2 shear weakness zone in the process of X stress loading and two sector damage zones occurring on both horizontal sides of each cavern. In the process of Y stress loading, some crack networks consisting of large-scale columnar joints, intralayer joints and C2 shear weakness zone were formed which led to the complete failure of the numerical models. A new method based on BP neural network was proposed to evaluate the stability of rock mass with large-scale columnar joints and interlayer shear weakness zones. The training results show that the method can distinguish the stability of rock mass well, while its accuracy reaches 97.6%.

大型柱状节理岩体(CJRMs)开挖后的力学性能会因卸荷松弛而降低，同时影响地下洞室的稳定性。此外，通过洞室切割层间剪切薄弱区使CJRM的整体稳定性复杂化，给工程的稳定性控制带来挑战。针对地下洞室施工中遇到的此类问题，我们对包含层间剪切薄弱区的大型CJRMs进行重构，进行3D岩石破坏过程分析。建立了开挖过程中调压罐洞室的数值模型，分析了其力学性能、破坏模式和分析了声发射（AE）特性。模拟结果表明，大型柱状节理和C2层间剪切薄弱区对力学性能和破坏模式有显着影响。总结了数值模型中柱状节理挤压-剪切破坏、C2层间剪切薄弱区拉-滑破坏和洞内及洞周挤压破坏三种破坏模式。各洞室周围均出现拉应力集中区，并由扇形逐渐缩小为线形。洞室在X应力加载过程中沿C2剪切弱化带分为两部分，每个洞室水平两侧出现两个扇形破坏区。在 Y 应力加载过程中，形成了一些由大尺度柱状节理、层内节理和C2剪切薄弱区组成的裂缝网络，导致数值模型完全失效。提出了一种基于BP神经网络的大尺度柱状节理和层间剪切薄弱区岩体稳定性评价方法。训练结果表明，该方法能较好地判别岩体的稳定性，准确率达到97.6%。