We propose a mechanical learning method that can be used to predict stability coefficients for slopes where slopes with predetermined shear planes are subjected to cyclic seismic loads under undrained conditions. Firstly, shear tests with cyclic loading of different parameters were simulated on designated slip zone soil specimens, in which the strain softening process leading to landslide occurrence was closely observed. At the same time, based on the limit equilibrium analysis of the Sarma method, the variation of slope stability coefficients under different cyclic loads was investigated. Finally, a Box-Jenkins' modeling approach is used to predict the data from the time series of slope stability coefficients using a mechanical learning approach. The simulation results show that (1) reduction in coordination number can be an accurate indicator of the level of strain softening and evolutionary processes; (2) the gradual reduction of shear stress facilitates the soil strain softening process, while different cyclic loading stress amplitudes will result in rapid penetration or non-penetration of the fracture zone by means of particulate flow. Although the confining pressure of the slip zone soil can inhibit the increase of fractures, it has a limited inhibitory effect on strain softening; (3) based on field observations of the slope stability factor and stress field, two possible landslide triggering mechanisms are described. (4) Mechanical learning of time series can accurately predict the changing pattern of stability coefficients of slopes without loading. This study establishes a potential bridge between the geological investigation of landslides and the theoretical background of landslide stability coefficient prediction.
Keywords: cyclic loading; fracture zone; landslide triggering mechanisms; mechanical learning; slip zone soils; strain softening.