In this paper, the dilatancy stress and mechanical characterization of sandstone were evaluated under uniaxial loading at different elastic modulus and porosity conditions. The prediction model of dilatancy stress was established using a regression equation and an artificial neural network based on a multilayer perceptron (ANN–MLP). The results indicate that: (1) The rock crack initiation stress, dilatancy stress and its elastic modulus are a power function relationship, while porosity is linearly negatively correlated. (2) σ_ci/σ_max hardly changes with the change of elastic modulus (E) and porosity (n); its value is about 0.443. σ_cd/σ_max increase with the increase in the elastic modulus, but decrease with the increase in the porosity. (3) Most of the rock samples are observed as a tensile failure when the porosity is low, while they are a shear failure at medium porosity and tensile shear composite failure at high porosity. (4) The optimum value from the ANN–MLP model for dilatancy stress with architecture 6-5-1 having coefficient correlation (R², 0.96%) was obtained at mean absolute error (MAE, 0.18981) and root mean square error (RMSE, 0.17016). It is worth mentioning that the research results will help and provide a reference for the related to rock mechanics test, rock engineering deformation and failure mechanism, and will also give specific guidelines significance for the efficient design of excavation and support in deep rock engineering.

本文在不同弹性模量和孔隙率条件下，在单轴载荷下评估了砂岩的膨胀应力和力学性能。使用回归方程和基于多层感知器（ANN–MLP）的人工神经网络，建立了剪胀应力预测模型。结果表明：（1）岩石的裂纹萌生应力，剪胀应力及其弹性模量为幂函数关系，而孔隙度呈线性负相关。（2）σ _CI / σ_最大与弹性模量（E）和孔隙率（n）的变化几乎不变化; 它的值约为0.443。σ _CD / σ_最大随着弹性模量的增加而增加，但是随着孔隙率的增加而减少。（3）当孔隙度低时，大多数岩石样品被视为拉伸破坏，而在中等孔隙度时为剪切破坏，而在高孔隙率时为拉伸剪切复合破坏。（4）基于ANN–MLP模型的剪胀应力的最佳值，其结构6-5-1具有系数相关性（R ²的平均绝对误差（MAE，0.18981）和均方根误差（RMSE，0.17016）获得0.96％。值得一提的是，该研究成果将为岩石力学测试，岩石工程变形与破坏机理的相关研究提供参考，并为深层岩石工程的有效开挖与支护设计提供具体的指导意义。