Abstract: The effect of particle breakage on the critical state of granular soils is of great significance. The existing studies have shown that the critical state line (CSL) of granular soils in the e-lnp space shifts downward as a result of particle breakage. However, it remains a big challenge for capturing the degree of particle breakage and the movement of CSL. In this study, the UH model for sands is modified by introducing the particle breakage parameter e_B and embedded in the real number encoding immune genetic algorithm (RIGA) to establish the RIGA-MUH model, which proposes a new method that can obtain the CSLs for the sands with varying particle-size distributions. The model is optimized and improved to obtain more accurate critical state parameters by adjusting the weight ratio in the error function under the critical state of granular soils. The stability, rationality and accuracy of the model are verified through the results of conventional drainage triaxial compression tests on the Toyoura sand and Cambria sand. The results show that the proposed model can be used to obtain the CSLs with high accuracy under a certain amount of particle breakage, which provides new insight into the constitutive modeling of crushable sands.