The effects of cement dosage, compaction coefficient, molding method (vertical vibration method and static pressure method), and dry–wet and freeze–thaw cycles on the mechanical strength of cement-improved loess (CIL) were studied to reveal its strength degradation law under dry–wet and freeze–thaw cycles. Results show that when using the vertical vibration molding method, the strength degradation effect of CIL can be improved by increasing the cement dosage and compaction coefficient; however, it is not obvious. Under the action of dry–wet cycle, damages, such as voids and cracks of CIL, develop continuously. Further, the strength deteriorates continuously and does not decrease after more than 15 dry–wet cycles. Therefore, the dry–wet cycle degradation system is selected by considering the most unfavorable conditions. In the process of freeze–thaw alternation, the pores and fissures of CIL develop and evolve continuously and the strength deteriorates continuously under the joint influence of water and low temperature. The strength tends to become stable after more than 12 freeze–thaw cycles. According to the safety principle, the deterioration coefficient of the freeze–thaw cycles is 0.3.

A numerical model of the dynamic triaxial test of graded crushed stone was established based on DEM (Discrete Element Method) to study its dynamic characteristics. The influence of test conditions on simulation results was analysed through numerical simulation. A method for determining the test conditions was proposed, and the reliability of the simulation was verified. We studied the accumulation rule and failure standard of permanent deformation. We then determined the critical failure stress level. The prediction model of permanent deformation cumulative failure was then established. When the calculation time step is greater than 1E−4 s per step, the stability of the dynamic triaxial numerical test of graded crushed stone is good. The plastic deformation of the simulated specimen tends to be stable under 10,000 dynamic loading cycles. When the specimen height and diameter are greater than 20 and 10 cm, respectively, the specimen size has little influence on the simulated axial strain. The recommended specimen size is 10 cm (Ф) × 20 cm (h). The action time curve results are consistent with indoor measurement results, which proves the simulation’s reliability. The critical failure stress is approximately linearly correlated with the confining pressure, and the cumulative failure equation of plastic deformation is established.