Abstract: Leakage of water and sand in underground structures can cause seepage erosion of the surrounding soil, leading to ground subsidence that poses a significant threat to urban safety. Previous research has mainly focused on seepage erosion in sandy soils, and little is known about the development process and mechanism of this phenomenon in different stratum conditions. To explore the failure patterns and mechanisms of seepage erosion in underground structures within strata containing different clay contents, seepage erosion model tests were designed and conducted, and the differences in phenomena of seepage erosion development under varying clay content were analyzed. Additionally, a Finite Difference Method-Discrete Element Method fluid-solid coupling numerical analysis model considering clay seepage erosion degradation was established. This model was used to compare the microscopic mechanisms of different seepage erosion development patterns, thereby exploring the mechanisms underlying the different erosion failure modes. The results show that: (1) the soil can be categorized into three types: No soil cave soil, Unstable soil cave soil, and Stable soil cave soil, according to the differences in the development mode of the soil when seepage erosion occurs; (2) The macroscopic phenomena of the three erosion development modes are different, which are characterized by the erosion development process, flow velocity distribution, and ground deformation; (3) The microscopic mechanisms of the three seepage erosion modes are different, which are characterize d by the erosion area, soil arch effect, and load distribution; (4) The coupled flow-solid Finite Difference Method-Discrete Element Method model, which takes into account the seepage erosion degradation of the clays, is able to simulate the development process of the soils in different seepage erosion modes efficiently; (5) The computation of the ultimate tensile height is proposed, which explains the differences of the three erosion modes in terms of the mechanism. The study results provide references for risk assessment and mitigation strategies in the event of leakage disasters.