Long-term dynamic characteristics and cumulative strain prediction modeling of round gravel foundation under cyclic subway loading Authors: Haijun Huang, Shaokun Ma, Cong Peng, Min Ma Published: 2026 ABSTRACT A series of saturated undrained dynamic triaxial tests were carried out by using DYNTTS large-scale triaxial cyclic test system to investigate the long-term dynamic characteristics and cumulative strain evolution of round gravel foundation under cyclic subway loading. The influence mechanisms of relative density, confining pressure and dynamic stress amplitude on the long-term dynamic characteristics and cumulative deformation of round gravel were systematically investigated. A prediction model of cumulative plastic strain considering the coupling effect of 3 parameters was proposed. The results of the study show that for the subway tunnels with shallow burial depth and low relative density of the round gravel foundation, the train operation has potential safety hazards, while for the subway tunnels with deeper burial depths, the operation of the trains is relatively safe. Increasing relative density and confining pressure can effectively improve the resilient modulus of round gravel, reduce its cumulative plastic strain, elastic strain and dynamic pore-pressure ratio, and significantly enhance the foundation’s liquefaction resistance. Therefore, for unfavorable working conditions such as Dr=0.3 and σ3 = 100 kPa, improvement measures such as increasing tunnel depth and implementing high-pressure grouting can be adopted to reduce the settlement of the subway foundation. Based on the time-hardening theory, a cumulative plastic strain prediction model was established with comprehensive consideration of relative density, confining pressure and dynamic stress amplitude. The research results can provide theoretical support and a design basis for the dynamic stability assessment of subway round gravel foundation and its long-term settlement prediction after construction. Introduction With the rapid development of urban rail transit construction in China, it is increasingly common for metro shield tunnels to pass through round gravel strata. For example, many metro lines in cities such as Chengdu, Chongqing, Nanning, and Guangzhou in China need to traverse such strata1,2. Under the long-term cyclic subway loading, the round gravel foundation is prone to excessive cumulative settlement, which leads to uneven subsidence of the subgrade or a serious threat to traffic safety due to settlement exceeding the standard3–5. Therefore, it is of great engineering significance to study the long-term dynamic characteristics and cumulative deformation evolution of round gravel foundations under the action of subway loading. The dynamic triaxial test, as an effective means to investigate the dynamic characteristics of soils, has been widely used in the study of the dynamic characteristics of coarse-grained soils. Thakur et al.6 through conducting extensive dynamic triaxial tests on railway ballast under varying confining pressures, found that the higher the confining pressure, the higher the resilient modulus of the specimens. Indraratna et al.7 conducted a series of cyclic drainage triaxial tests on the ballast of different sizes using a large triaxial test device designed in-house by the University of Wollongong and found that the permanent axial strain and volumetric strain of the ballast decreased and then increased with the increase of the uniformity coefficient. Ma et al.8 carried out dynamic triaxial tests on round gravels with different relative densities under subway loading using a large dynamic triaxial apparatus, and the results showed that the cumulative strain of round gravels decreased with the increase of relative density. However, this study only carried out tests on three relative density samples and lacked a systematic study on the dynamic characteristics of round gravel. Wang and Zhuang9 investigated the dynamic response behavior of high-speed railroad ballast under different confining pressures through a series of triaxial tests, and the results showed that the critical dynamic stresses of coarse-grained soils increased with the increase of effective confining pressure. Yang et al.10 conducted a series of dynamic triaxial tests on gravel with different fine particle contents on heavy railroad subgrade. They found that the cumulative plastic strain of the specimens increased approximately linearly with the increase of the fine particle content. Cui et al.11 conducted a series of cyclic triaxial tests on coarse-grained soils (consisting of pulverized clay, sand and gravel) of passenger railroad subgrade under different initial stress states. They found that the higher the initial stress and the lower the confining pressure, the higher the cumulative strain of the specimens under the same cyclic loading. From the above, the existing studies mainly focus on the exploration of the dynamic characteristics of coarse-grained soils (mostly ballast and gravel) for high-speed and heavy haul railroads. However, there is a lack of systematic research on the long-term dynamic characteristics of round gravel (with (sub) spherical particle shape, smooth surface, and poor gradation) under subway loading (with low dynamic stress amplitude, long running time, and high number of runs). Therefore, it is necessary to explore this issue further. In addition, it is especially important to construct a prediction model that can accurately predict the development trend of cumulative plastic strain of foundation materials according to the results of dynamic characteristic tests. Scholars have proposed empirical models for the prediction of cumulative plastic strain considering the number of cycles, dynamic stress amplitude, cyclic stress ratio, water content, etc. for materials such as clay, ballast, and crushed stone of railroad subgrade. For example, Zhou and Tang12 developed a hyperbolic empirical model for cumulative plastic strain prediction considering cyclic stress ratio and water content for soft clay after freeze-thawing of subway foundations. Xu et al.13 proposed a cumulative strain model considering water content, confining pressure, dynamic stress amplitude, and static destructive deviator stress for coarse-grained soils (gravels, sands, and fine-grained mixtures) for heavy railroad foundations. He et al.14 developed a cumulative strain model considering the number of cycles, deviator stresses, loading frequency and water content for coarse-grained soils (gravel, sand and soil composition) of highway subgrade. However, little research has been reported on the cumulative plastic strain prediction model considering the coupling of relative density, confining pressure and dynamic stress amplitude under cyclic subway loading with round gravel as the research object, and further research is still needed. Therefore, the saturated undrained dynamic triaxial test under different conditions of relative density, confining pressure and dynamic stress amplitude is carried out by utilizing the DYNTTS large-scale triaxial cyclic test system with the round gravel of the subway foundation in Nanning as the research object. The influence mechanism of each factor on the long-term dynamic properties of round gravel is systematically investigated. The cumulative plastic strain prediction model with 3-parameter coupling is established based on the time hardening theory. The research results can provide theoretical support and a design basis for the dynamic stability assessment of the subway gravel foundation and its long-term settlement prediction after construction. Experiment materials and methods Experiment apparatus and materials The test apparatus uses the coarse-grained soil DYNTTS large-scale triaxial cyclic test system, which offers high stability and testing accuracy, as shown in Fig. 1. The system primarily consists of an axial exciter, a pressure cover, a confining pressure/back pressure controller, displacement/pore water pressure sensors, and a data acquisition system. The system allows applying 1.0 MPa maximum circumferential pressure, 60 kN maximum axial stress, 2 Hz maximum loading frequency, and 75 mm maximum axial displacement. This test applies dynamic loads to the samples using vertical vibrations generated by an axial exciter to simulate the impact of subway operations on the foundation. Fig. 1DYNTTS large-scale triaxial cyclic test system. DYNTTS large-scale triaxial cyclic test system. The materials used in the test were taken from the construction site of Nanning Metro, and the sampling location and soil samples are shown in Fig. 2a. As seen in Fig. 2a, the coarse-grained soil particles are round and subround in shape, with large particle sizes and smooth surfaces. The gradation curve of the round gravel is shown in Fig. 1, and its basic physical parameters are listed in Table 1. Fig. 2Sampling location and grading curve of soil samples. The map in (a) was generated using Microsoft PowerPoint version 2016 (https://www.microsoft.com/zh-cn/microsoft-365). Sampling location and grading curve of soil samples. The map in (a) was generated using Microsoft PowerPoint version 2016 (https://www.microsoft.com/zh-cn/microsoft-365). Table 1Basic physical parameters of round gravel.Specific gravity GsBulk unit weight γd (kN·m− 3)Maximum void ratio emaxMinimum void ratio emin2.6720.7350.2240.352 Table 1Basic physical parameters of round gravel.Specific gravity GsBulk unit weight γd (kN·m− 3)Maximum void ratio emaxMinimum void ratio emin2.6720.7350.2240.352 Basic physical parameters of round gravel. Experiment scheme and loading condition setting A series of dynamic triaxial tests with different relative density, confining pressure and dynamic stress amplitude were carried out by using DYNTTS test system to investigate the long-term dynamic characteristics of round gravel under cyclic subway loading. According to the survey data of the Nanning subway construction site, the round gravel layer in the subway area is mostly slightly dense to medium dense state, locally dense state, so the three typical relative densities of 0.3, 0.5 and 0.7 are selected. The depth of the Nanning subway tunnel is about 10 m and mostly below the normal water level, so three confining pressure conditions of 100 kPa, 200 kPa and 300 kPa are set, and the samples are set in a saturated state. The dynamic stresses on the foundation soil under subway loading are roughly in the range of 10–40 kPa15,16, so the dynamic stress amplitudes are taken as 10 kPa, 20 kPa, 30 kPa and 40 kPa. The experiment scheme is shown in Table 2. Table 2Experiment scheme.Working conditionRelative density DrConfining pressure σ3/kPaDynamic stress amplitude σamp d/kPaS1-10.310010, 20, 30, 40S1-220010, 20, 30, 40S1-330010, 20, 30, 40S2-10.510010, 20, 30, 40S2-220010, 20, 30, 40S2-330010, 20, 30, 40S3-10.710010, 20, 30, 40S3-220010, 20, 30, 40S3-330010, 20, 30, 40 Table 2Experiment scheme.Working conditionRelative density DrConfining pressure σ3/kPaDynamic stress amplitude σamp d/kPaS1-10.310010, 20, 30, 40S1-220010, 20, 30, 40S1-330010, 20, 30, 40S2-10.510010, 20, 30, 40S2-220010, 20, 30, 40S2-330010, 20, 30, 40S3-10.710010, 20, 30, 40S3-220010, 20, 30, 40S3-330010, 20, 30, 40 Experiment scheme. The test was carried out under the consolidation undrained condition in view of the instantaneity of subway loading and the deeply buried characteristics of subway tunnels. A sine wave was used for the test loading waveform according to the dynamic response waveform of the foundation soil during subway operation17, as shown in Fig. 3. Multi-stage loading method18–21 was used for the test loading method to reduce the test workload. The loading number is set to 10 000 times (single-stage dynamic stress amplitude). The loading frequency is calculated by the formula f = V/L22(V is the hourly speed of the subway, L is the length of the carriage), Nanning subway V = 72 km/h, L = 20 m, so the loading frequency is taken as 1 Hz. Fig. 3Loading waveform. Loading waveform. Experiment process The sample preparation is carried out by the Code for Soil Test of Railway Engineering23. The preparation steps mainly include two steps: (1) Sample preparation: the sample is prepared by layered vibration method to ensure the homogeneity of the sample, and it is compacted in 10 layers one by one. That is, weigh the mass of a single layer of the round gravel, stirring uniformly into the sample bucket, compacted to a predetermined height, and then each layer of soil samples prepared according to this step one by one, the final size of the sample into 300 mm (diameter) × 600 mm (height). (2) Sample saturation: After sample preparation is complete, CO₂ is injected from the bottom of the sample for 3 h to displace the air inside the sample. Subsequently, 12 h of air-free water is introduced, followed by the application of 16 h of back pressure. Back pressure adopts a step-by-step pressurization method, increasing 50 kPa at each level and stabilizing for 4 h, finally reaching 200 kPa. Finally, the saturation degree of the sample is assessed. If the pore pressure coefficient B is greater than 0.95, the sample is considered essentially saturated and ready for loading. The experiment process is shown in Fig. 4. Fig. 4Experiment process illustration. Experiment process illustration. Experiment results and analysis Accumulated plastic strain Figure 5 shows a typical axial strain curve under cyclic subway loading. As can be seen from the figure, the axial strain ε1 of the specimen consists of reversible elastic strain εe and irreversible cumulative plastic strain εp. The lowest point on the axial strain curve represents the specimen’s cumulative plastic strain, while the difference between the highest and lowest points on the axial strain curve represents the specimen’s elastic strain. Fig. 5Typical axial strain curve. Typical axial strain curve. The relationship curves between the cumulative plastic strain εp and the number of cycles N of the round gravel under cyclic subway loading with different relative densities, confining pressures and dynamic stress amplitudes are shown in Fig. 6. As shown in Fig. 6, the εp-N curves of the three relative density samples under different confining pressure conditions show different development patterns with different dynamic stress amplitudes. The εp-N curves of round gravel under cyclic subway loading can be categorized into three types A/B/C according to the shakedown theory of Werkmeister24. (As shown in Eq. (1) and Fig. 7): (1) Plastic stabilized type (A): the cumulative plastic strain under cyclic loading grows rapidly at the initial stage of loading. Then, it slows down until relatively stable, and the strain value is small, at this time the sample is in an elastic state. (2) Plastic creep type (B): the cumulative plastic strain under cyclic loading in the initial stage of loading rapid growth, with the continuation of the loading, the strain growth rate slows down but remains in a state of continuous growth, this time the sample is in the plastic creep state. (3) Incremental damage type (C): the cumulative plastic strain under cyclic loading grows sharply, and the specimen has reached the damage state within a small range of loading numbers. The cumulative plastic deformation patterns of the samples in Fig. 6 were classified based on the above classification criteria. When the confining pressure was low (σ3 = 100 kPa), found that all three developmental patterns of cumulative plastic deformation were presented: (1) The cumulative plastic strain curves for sample Dr=0.3 at σamp d = 10 and 20 kPa, Dr=0.5 at σamp d = 10, 20, and 30 kPa and Dr=0.7 at σamp d = 10, 20, 30, and 40 kPa exhibit a plastic stabilized type. (2) The cumulative plastic strain curves for sample Dr=0.3 at σamp d = 30 kPa and sample Dr=0.5 at σamp d = 40 kPa exhibit plastic creep type. (3) The cumulative plastic strain curve for sample Dr=0.3 at σamp d = 40 kPa exhibits an incremental damage type. Therefore, for shallow subway tunnels and the small relative density of round gravel foundations, there are safety hazards for train operation, and the strength of the round gravel foundations should be upgraded. The cumulative plastic strain curves of samples Dr=0.3, 0.5 and 0.7 for higher confining pressures σ3 = 200 kPa and σ3 = 300 kPa at all levels of dynamic stress amplitude (σamp d = 10, 20, 30 and 40 kPa) exhibit a plastic stabilized type. This is mainly because the high confining pressure limits the slip between particles within the sample and improves the ability of the sample to withstand external loads. This indicates that the operation of trains in deeper subway tunnels is relatively safe, and there is no need to treat the round gravel foundation. Fig. 6Relationship curves between cumulative plastic strain εp and the number of cycles N for round gravels. 1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\{ \begin{gathered} R{\text{ange A}}:{\text{ }}(\varepsilon _{p}^{{5000}} - \varepsilon _{p}^{{3000}})<0.045 \times {10^{ - 3}} \hfill \\ R{\text{ange B}}:{\text{ }}0.045 \times {10^{ - 3}}{\mathrm{<}}(\varepsilon _{p}^{{5000}} - \varepsilon _{p}^{{3000}})<0.4 \times {10^{ - 3}} \hfill \\ R{\text{ange C}}:{\text{ }}0.4 \times {10^{ - 3}}{\mathrm{<}}(\varepsilon _{p}^{{5000}} - \varepsilon _{p}^{{3000}}) \hfill \\ \end{gathered} \right.$$\end{document} Relationship curves between cumulative plastic strain εp and the number of cycles N for round gravels. where ε5000 p and ε3000 p are the cumulative plastic strains corresponding to the sample at N = 5 000 and N = 3 000, respectively. Fig. 7Classification method of plastic deformation behavior in soil based on shakedown theory. Classification method of plastic deformation behavior in soil based on shakedown theory. A three-dimensional relationship between the final cumulative plastic strain of the sample at the end of the four loading stages versus the relative density, confining pressure and dynamic stress amplitude is given in Fig. 8. As shown in Fig. 8, the whole graph shows a left-high-right-low trend, i.e., the higher the relative density and confining pressure, the lower the dynamic stress amplitude, while the cumulative plastic strain of the sample is smaller. Specifically, when the confining pressure and dynamic stress amplitude are constant, as the relative density increases from 0.3 to 0.7, the maximum reduction rate of accumulated plastic strain is 77.05%, while the average reduction rate is 20.03%. When the relative density and dynamic stress amplitude are constant, as the confining pressure increases from 100 kPa to 300 kPa, the maximum reduction rate of cumulative plastic strain is 80.33%, while the average reduction rate is 28.77%. This indicates that increasing both confining pressure and relative density can significantly reduce the plastic strain. Moreover, when the relative density and confining pressure are constant, the dynamic stress amplitude decreases from 40 kPa to 10 kPa with a maximum reduction rate in cumulative plastic strain of 97.3%, while the average reduction rate is 88.68%. Meanwhile, comparing the results of the analysis showed that the degree of influence of each parameter on the cumulative plastic strain is, in descending order, the dynamic stress amplitude, confining pressure and relative density. The above analysis shows that for the subway round gravel foundation project with low relative density and low confining pressure such as Dr=0.3 and σ3 = 100 kPa, engineering measures such as increasing the tunnel burial depth (to improve the confining pressure) and implementing high-pressure grouting (to enhance the degree of densification) are recommended to effectively reduce tunnel settlement under the train cyclic loading and ensure the long-term stability of the project. Fig. 8Three-dimensional relationship between the final cumulative plastic strain εp of round gravel and relative density, confining pressure, and dynamic stress amplitude. Three-dimensional relationship between the final cumulative plastic strain εp of round gravel and relative density, confining pressure, and dynamic stress amplitude. Elastic strain The relationship curves between the elastic strain εe and the number of cycles N of the round gravel under cyclic subway loading with different relative densities, confining pressures and dynamic stress amplitudes are shown in Fig. 9. As shown in Fig. 9, the elastic strains of most of the samples exhibit a rapid increase in the initial stage of loading for each dynamic stress amplitude, and the elastic strains gradually become constant as the loading continues. This is due to the rearrangement of particles within the sample at the beginning of loading, resulting in a rapid increase in the sample’s elastic strain. As the loading continues to be applied, the rearrangement of particles is limited, and eventually, the elastic strain of the sample stabilizes. However, there are a small number of samples for which the εe-N curves develop in a different pattern. At a small confining pressure (σ3 = 100 kPa), the elastic strain of the Dr=0.3 sample under σamp d = 40 kPa increases sharply with the number of cycles, and the sample undergoes destruction. Meanwhile, comparing Figs. 6 and 9, the elastic strain of the sample can be found to be roughly similar to the trend of plastic strain change. Fig. 9Relationship curves between elastic strain εe and the number of cycles N for round gravels. Relationship curves between elastic strain εe and the number of cycles N for round gravels. The three-dimensional relationship between the elastic strain of the sample at the end of the four loading stages versus relative density, confining pressure and dynamic stress amplitude is given in Fig. 10. As shown in Fig. 10, the whole graph shows a left-high-right-low trend, i.e., the higher the relative density and confining pressure, the lower the dynamic stress amplitude, while the lower the elastic strain of the specimen. Specifically, when the confining pressure and dynamic stress amplitude are constant, as the relative density increases from 0.3 to 0.7, the maximum reduction rate of elastic strain is 56.79%, while the average reduction rate is 13.1%. When the relative density and dynamic stress amplitude are constant, as the confining pressure increases from 100 kPa to 300 kPa, the maximum reduction rate of elastic strain is 66.67%, while the average reduction rate is 34.12%. This indicates that high relative density and high confining pressure can effectively inhibit the increase of elastic strain. When the relative density and confining pressure are constant, as the dynamic stress amplitude decreases from 40 kPa to 10 kPa, the maximum reduction rate of its elastic strain is 88.02%, while the average reduction rate is 74.3%. The analysis above reveals that the three parameters affect the elastic strains from high to low degrees as the dynamic stress amplitude, confining pressure and relative density, respectively. Fig. 10Three-dimensional relationship between the final elastic strain εe of round gravel and relative density, confining pressure, and dynamic stress amplitude. Three-dimensional relationship between the final elastic strain εe of round gravel and relative density, confining pressure, and dynamic stress amplitude. Resilient modulus The resilient modulus (Mr) is considered to be an effective indicator for checking the stiffness of the subgrade and roadbed25,26. Its definition is shown in Eq. (2) and Fig. 11.2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${M_{\mathrm{r}}}=\frac{{\sigma _{{\mathrm{d}}}^{{{\mathrm{amp}}}}}}{{{\varepsilon _{1{\mathrm{e}}}}}}$$\end{document} where σamp d is the dynamic (deviator) stress amplitude, kPa; ε1e is the elastic strain, %. Fig. 11Definition of resilient modulus Mr. Definition of resilient modulus Mr. The relationship curves between the resilient modulus Mr and the number of cycle N of the round gravel under cyclic subway loading with different relative densities, confining pressures and dynamic stress amplitudes are shown in Fig. 12. As shown in Fig. 12, the trend of the resilient modulus with the number of cycles is closely related to the dynamic deformation behavior of the sample. For example, when the sample is in plastic stabilized type (e.g., Dr=0.7, σ3 = 100 kPa, σamp d = 30 kPa), the resilient modulus exhibits a sharp decrease at the beginning of the loading and then stabilizes with the number of cycles. This is mainly because at the beginning of loading, relative slip occurs between the particles inside the sample, the elastic strain increases rapidly, and the corresponding resilient modulus decreases rapidly. As the loading continues, the particle slip is limited, the elastic strain stabilizes, and the corresponding resilient modulus stabilizes. When the sample is in plastic creep type (e.g. Dr=0.5, σ3 = 100 kPa, σamp d = 40 kPa), the resilient modulus also shows a sharp decrease at the beginning of the loading and then slowly decreases until stabilized as the loading continues. When the sample is in incremental damage type (e.g., Dr=0.3, σ3 = 100 kPa, σamp d = 40 kPa), the resilient modulus decreases sharply with the number of cycles. This is due to the sharp increase in elastic modulus as the sample undergoes damage. Fig. 12Relationship curves between resilient modulus Mr and the number of cycles N for round gravels. Relationship curves between resilient modulus Mr and the number of cycles N for round gravels. A three-dimensional relationship between the final resilient modulus of the sample at the end of the four loading stages versus relative density, confining pressure and dynamic stress amplitude is given in Fig. 13. As shown in Fig. 13, the figure shows an overall trend of low-left-high-right, i.e., the higher the relative density and confining pressure, the higher the resilient modulus of the sample. Specifically, when the relative density is increased from 0.3 to 0.7 with constant confining pressure and dynamic stress amplitude, the maximum increase rate of resilient modulus is 130.1%, while the average increase rate is 22.82%. When the confining pressure is increased from 100 kPa to 300 kPa with constant relative density and dynamic stress amplitude, the maximum increase rate of resilient modulus is 203.15%, while the average increase rate is 63.67%. This indicates that the confining pressure and relative density increase can effectively enhance the resilience properties of the samples. Meanwhile, it can be found that the degree of influence of confining pressure on resilient modulus is stronger than that of relative density on it. There are some differences in the variation of resilient modulus under different dynamic stress amplitudes. When the confining pressure and relative density are low, the resilient modulus decreases with the increase of dynamic stress amplitude, while when both are high, the resilient modulus increases with the increase of dynamic stress amplitude. This is because at low confining pressures and relative densities, the elastic strain of the samples is more sensitive to the effects of dynamic stresses. Fig. 13Three-dimensional relationship between the final resilient modulus Mr of round gravel and relative density, confining pressure, and dynamic stress amplitude. Three-dimensional relationship between the final resilient modulus Mr of round gravel and relative density, confining pressure, and dynamic stress amplitude. Dynamic pore pressure ratio The dynamic pore water pressure was normalized and transformed into the corresponding dynamic pore pressure ratio27 ru to facilitate the comparison and understanding of the experiment results under different working conditions.3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${r_{\mathrm{u}}}={u_{N\hbox{min} }}/{\sigma ^{\prime}_c}$$\end{document} where uNmin is the irrecoverable pore water pressure of the sample at the Nth cycle of loading, kPa; and σc′ is the effective confining pressure, kPa. The relationship curves between the dynamic pore pressure ratio ru and the number of cycle N of the round gravel under cyclic subway loading with different relative densities, confining pressures and dynamic stress amplitudes are shown in Fig. 14. As shown in Fig. 14, the dynamic pore pressure ratios of most samples show the development trend of a sharp increase first and then a gradual stabilization as the number of cycles increases, and the values are all less than 0.14. However, the dynamic pore pressures of some individual specimens show a different trend. For example, the sample σ3 = 100 kPa, Dr=0.3, under the action of dynamic stress amplitude σamp d = 40 kPa, its dynamic pore pressure ratio increases sharply with the increase in the number of cycles and gradually reaches about 0.8, at which time the sample begins to show initial liquefaction. Fig. 14Relationship curves between dynamic pore pressure ratio ru and the number of cycles N for round gravels. Relationship curves between dynamic pore pressure ratio ru and the number of cycles N for round gravels. A three-dimensional relationship between the final dynamic pore pressure ratio of the sample at the end of the four loading stages versus relative density, confining pressure and dynamic stress amplitude is given in Fig. 15. As shown in Fig. 15, the overall trend of the graph is left-high-right-low, i.e., the higher the relative density and confining pressure, the lower the dynamic stress amplitude, the smaller the dynamic pore-pressure ratio, and the stronger the liquefaction resistance of the sample. Specifically, when the confining pressure and dynamic stress amplitude are constant, as the relative density increases from 0.3 to 0.7, the maximum reduction rate of the dynamic pore pressure ratio is 81.45%, while the average reduction rate is 57.91%. When the relative density and dynamic stress amplitude are constant, as the confining pressure is increased from 100 kPa to 300 kPa, the maximum reduction rate of the dynamic pore pressure ratio is 90.24%, while the average reduction rate is 59.77%. When the relative density and confining pressure are constant, as the dynamic stress amplitude decreases from 40 kPa to 10 kPa, the maximum reduction rate of the dynamic pore pressure ratio is 96.38%, while the average reduction rate is 86.2%. The above analysis reveals that the increase of confining pressure and relative density can effectively enhance the liquefaction resistance of the samples. Meanwhile, the degree of influence of the three parameters on the dynamic pore pressure ratio from high to low is the dynamic stress amplitude, confining pressure and relative density, respectively. Fig. 15Three-dimensional relationship between the final dynamic pore pressure ratio ru of round gravel and relative density, confining pressure, and dynamic stress amplitude. Three-dimensional relationship between the final dynamic pore pressure ratio ru of round gravel and relative density, confining pressure, and dynamic stress amplitude. Cumulative plastic strain prediction model Currently, the more classical empirical models mainly include the semi-logarithmic prediction empirical model, exponential prediction empirical model and hyperbolic prediction empirical model. Zhou and Tang12, by conducting cyclic triaxial tests on frozen soft clay under cyclic subway loading, found that the developed law of cumulative plastic strain of soft clay with the number of cycles is similar to the developed law of Hardin-Drnevich28 hyperbolic model (shown in Fig. 16). And in this test, it is found that the εp-N curve of the sample is also highly consistent with the hyperbolic model by fitting the εp-N curve of the sample under a single dynamic stress amplitude (as shown in Fig. 17). Therefore, this study proposes to reconstruct the cumulative plastic strain model of round gravel based on the Hardin-Drnevich (Eq. 4) and Zhou hyperbolic model (Eq. 5).4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\sigma _{\mathrm{d}}}=\frac{{{\varepsilon _{\mathrm{d}}}}}{{A+B{\varepsilon _{\mathrm{d}}}}}$$\end{document}5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\varepsilon _{\mathrm{p}}}=\frac{N}{{A+BN}}$$\end{document} where εp is the cumulative plastic strain, %; N is the number of cycles; A and B are hyperbolic model parameters. In particular, when the number of cycles N tends to 0, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\frac{1}{A}={\left.{\dot{\epsilon\:}}_{{\mathrm{p}}_{\mathrm{i}\mathrm{n}\mathrm{i}\mathrm{t}}}\right|}_{N\to\:+\infty\:}$$\end{document}, the parameter A can be found to be related to the initial rate of cumulative plastic strain, while when the number of cycles N tends to +∞, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\frac{1}{B}={\left.{\epsilon\:}_{\mathrm{p}\:\mathrm{m}\mathrm{a}\mathrm{x}}\right|}_{N\to\:+\infty\:}$$\end{document}, the parameter B can be found to be related to the maximum cumulative plastic strain, indicating that the parameters A and B in the model have clear physical significance. Fig. 16Schematic of Hardin-Drnevich and Zhou hyperbolic models. Schematic of Hardin-Drnevich and Zhou hyperbolic models. Fig. 17Typical εp-N fitting curves for a single dynamic stress amplitude. Typical εp-N fitting curves for a single dynamic stress amplitude. In multistage loading tests, all loadings prior to each loading stage can be considered as stress history. The time-hardening method17,29 was used to evaluate the stress history in order to analyze its effect on the plastic strains. The method introduces the concept of “equivalent loading times (Neq i)”, where Neq i is the number of cycles required to reach εp at the end of the previous dynamic stress under the current dynamic stress, as shown in Fig. 18. With the help of Neq i, the εp corresponding to the Nth loading number can be transformed into the εp corresponding to the N-Ni−1+Neq i-1 loading number at the current dynamic stress. The relevant equations are shown in Eq. (6)–(8).6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_{{i - 1}}^{{eq}}=\frac{{{B_i}}}{{\frac{1}{{{\varepsilon _{{p_{i - 1}}}}}} - {A_i}}}$$\end{document}7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${N^{\prime}_i}=N - {N_{i - 1}}+N_{{i - 1}}^{{eq}}$$\end{document}8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\varepsilon _{{p_i}}}=\frac{{{{N^{\prime}}_i}}}{{{A_i}{{N^{\prime}}_i}+{B_i}}}$$\end{document} where Ai and Bi are the model parameter values of the ith loading stage, respectively; Ni−1 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\epsilon\:}_{{p}_{i-1}}$$\end{document} are the total number of cycles and cumulative plastic strain at the end of the ith-1 loading stage, respectively, %; and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\epsilon\:}_{{p}_{i}}$$\end{document} is the cumulative plastic strain of the ith loading stage, %. Fig. 18Time-hardening method. Time-hardening method. The variable cyclic stress ratio CSR (Eq. 9) is defined first to facilitate the establishment of the cumulative plastic strain prediction model30.9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$CSR=\frac{{\sigma _{{\mathrm{d}}}^{{{\mathrm{amp}}}}}}{{2{\sigma _3}}}$$\end{document} Based on the above analysis, the cumulative plastic strain prediction model of round gravel under multi-parameter coupling is constructed with the following steps: (1) The cumulative plastic strain fitting equations and parameters of the sample under each loading stage were obtained using a modified hyperbolic model based on the time-hardening method. (2) The variation of the fitted parameters with relative density, confining pressure and dynamic stress amplitude is analyzed. A prediction model is established and verified for accuracy to predict the cumulative plastic strain under multistage loading conditions. The relationship between the parameter Ar and the cyclic stress ratio CSR is shown in Fig. 19. As shown in Fig. 19, the values of Ar for different relative density samples at the same confining pressure are very close to each other. Meanwhile, the final cumulative plastic strain is less affected by the value of Ar in the hyperbolic model. The parameter Ar can be considered to be little affected by the relative density. Therefore, parameter Ar is analyzed by taking its average value \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{-}{A}$$\end{document}r for different relative densities at the same confining pressure to simplify the model. Fig. 19Relationship between parameter Ar and cyclic stress ratio CSR. Relationship between parameter Ar and cyclic stress ratio CSR. The relationship between the average value \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{-}{A}$$\end{document}r of the parameter Ar and the cyclic stress ratio CSR is shown in Fig. 20. Fig. 20Relationship between the average value \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{-}{A}$$\end{document}r of the parameter Ar and the cyclic stress ratio CSR. Relationship between the average value \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{-}{A}$$\end{document}r of the parameter Ar and the cyclic stress ratio CSR. The parameter \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{-}{A}$$\end{document}r (for easy expression, notation \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{-}{A}$$\end{document}r in Ar) in Fig. 20 is further processed to facilitate the analysis. The relationship between parameter 1/Ar and the cyclic stress ratio CSR at different confining pressures is obtained, as shown in Fig. 21. As shown in Fig. 21, the parameter 1/Ar shows a good linear relationship with the cyclic stress ratio CSR for a certain confining pressure, and the slope of the 1/Ar-CSR curve increases with the increase of confining pressure. This indicates that both the cyclic stress ratio and confining pressure affect the parameter 1/Ar. The relationship between the parameter Ar, the cyclic stress ratio CSR and the confining pressure σ3 can be obtained by analyzing the figures, as shown in Eq. (10).10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A_{\mathrm{r}}}=\frac{1}{{{\mathrm{(0}}{\mathrm{.0023}}\frac{{{\sigma _3}}}{{{P_{\mathrm{a}}}}}{\mathrm{+0}}{\mathrm{.0003)}}CSR{\mathrm{+7}} \times {\mathrm{1}}{{\mathrm{0}}^{{\mathrm{-5}}}}{{(\frac{{{\sigma _3}}}{{{P_{\mathrm{a}}}}}{\mathrm{)}}}^{{\mathrm{-0}}{\mathrm{.07}}}}}}$$\end{document} where Ar is a model parameter, dimensionless; Pa is the standard atmospheric pressure, 101 kPa, which serves to adjust the magnitude. Fig. 21Relationship between parameter 1/Ar and cyclic stress ratio CSR. Relationship between parameter 1/Ar and cyclic stress ratio CSR. The relationship between the parameter Br and the cyclic stress ratio CSR is shown in Fig. 22. A reciprocal function (Reciprocal) relationship between the parameter Br and the cyclic stress ratio CSR at a certain relative density can be seen from the figure, the expression of which is shown in Eq. (11).11\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${B_{\mathrm{r}}}=\frac{1}{{{H_{B{\mathrm{r}}}}+{T_{B{\mathrm{r}}}}CSR}}$$\end{document} where Br, HBr and TBr are model parameters, dimensionless. Fig. 22Relationship between parameter Br and cyclic stress ratio CSR. Relationship between parameter Br and cyclic stress ratio CSR. The parameters HBr and TBr in Eq. (11) are summarized and generalized for all working conditions. The relationship between the parameters HBr and TBr and the relative density Dr and confining pressure σ3 is obtained, as shown in Fig. 23. As shown in Fig. 23, there is a good linear relationship between parameters HBr and TBr and relative density at the same confining pressure. At the same relative density, HBr and TBr increase with the increase of confining pressure. This indicates that both relative density and confining pressure affect the parameters HBr and TBr. The relationship between the parameters HBr, TBr and the relative density Dr, confining pressure σ3 can be obtained by analyzing the data in the figure, as shown in Eqs. (12) and (13).12\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${H_{B{\mathrm{r}}}}=0.0094{D_{\mathrm{r}}} - 0.0329{(\frac{{{\sigma _3}}}{{{P_{\mathrm{a}}}}})^{ - 0.231}}$$\end{document}13\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${T_{B{\mathrm{r}}}}= - 0.1812{{\mathrm{e}}^{0.52\frac{{{\sigma _3}}}{{{P_{\mathrm{a}}}}}}}{D_{\mathrm{r}}}+0.7621\frac{{{\sigma _3}}}{{{P_{\mathrm{a}}}}}+0.2983$$\end{document} Fig. 23Relationship between parameters HBr and TBr and relative density Dr. Relationship between parameters HBr and TBr and relative density Dr. Finally, by combining Eqs. (5) and (10)–(13), the hyperbolic prediction model of cumulative plastic strain of round gravel under cyclic subway loading can be obtained by considering the effects of relative density, confining pressure, cyclic stress ratio, and number of cycles, as shown in Eq. (14). 14\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\varepsilon _{p} = \frac{N}{{A_{r} + B_{r} N}},\,\left\{ \begin{gathered} A_{r} = \frac{1}{{(0.0023\frac{{\sigma _{3} }}{{P_{a} }} + 0.0003)CSR + 7 \times 10^{{ - 5}} (\frac{{\sigma _{3} }}{{P_{a} }})^{{ - 0.07}} }} \hfill \\ B_{r} = \frac{1}{{H_{{Br}} + T_{{Br}} CSR}} \hfill \\ H_{{Br}} = 0.0094D_{r} - 0.0329(\frac{{\sigma _{3} }}{{P_{a} }})^{{ - 0.231}} \hfill \\ T_{{Br}} = - 0.1812e^{{0.52\frac{{\sigma _{3} }}{{P_{a} }}}} D_{r} + 0.7621\frac{{\sigma _{3} }}{{P_{a} }} + 0.2983 \hfill \\ \end{gathered} \right.$$\end{document} To verify the accuracy of Eq. (14), The experimental values of the cumulative plastic strain of the sample are compared with the predicted values obtained from the cumulative plastic strain prediction model, and the results are shown in Fig. 24. As seen in Fig. 24, the predicted values of cumulative plastic strains of the sample are in high agreement with the test values, indicating that the cumulative plastic strain prediction model proposed in this study has high accuracy. Fig. 24Prediction results of the cumulative plastic strain prediction model for round gravel under multi-level loading. Prediction results of the cumulative plastic strain prediction model for round gravel under multi-level loading. Conclusion A series of saturated undrained dynamic triaxial tests were carried out in this study using the DYNTTS large-scale triaxial cyclic test system with the subway round gravel foundation as the research object. The effects of relative density, confining pressure and dynamic stress amplitude on the long-term dynamic properties and cumulative deformation of round gravel were systematically analyzed. A prediction model of cumulative plastic strain with 3-parameter coupling was established. The following conclusions were obtained: The accumulated plastic strain of the round gravel is large (> 0.61%) and shows a rapid development trend for subway tunnels with shallow burial depth (σ3 = 100 kPa) and low relative density of the foundation (Dr=0.3). Train operation is a safety hazard, and foundation reinforcement measures are required. However, for subway tunnels with deeper burial depths (σ3 = 200 and 300 kPa), the cumulative plastic strain is small (less than 0.13%), and the foundation is well stabilized so that no special treatment can be done.The sample Dr=0.3, σ3 = 100 kPa under the action of σamp d = 40 kPa, its dynamic pore pressure ratio reaches about 0.8, and the sample begins to show initial liquefaction. The increase in confining pressure and relative density can effectively inhibit the development of dynamic pore pressure and enhance the liquefaction resistance. The influence degree of each parameter on the long-term dynamic characteristics of round gravel from high to low is dynamic stress amplitude, confining pressure and relative density.Increases in both relative density and confining pressure can significantly reduce the cumulative plastic (maximum decreases of about 77.05% and 80.33%) and elastic strains (maximum decreases of about 56.79% and 66.67%) of the round gravels, and improve their resilient modulus (maximum increases of about 130.1% and 203.15%). Therefore, for unfavorable working conditions such as Dr=0.3 and σ3 = 100 kPa, improvement measures such as increasing tunnel depth and implementing high-pressure grouting can be adopted to reduce the settlement of the subway foundation.A cumulative plastic strain prediction model considering relative density, confining pressure and dynamic stress amplitude was developed using a hyperbolic improvement model based on the time-hardening method. The model has high accuracy and can be used to assess the stability of round gravel foundations under cyclic subway loading. It should be noted that this model is based on the conventional dynamic triaxial test under fixed axial cyclic load, and does not consider the rotation effect of the principal stress axis caused by subway cyclic load. The accumulated plastic strain of the round gravel is large (> 0.61%) and shows a rapid development trend for subway tunnels with shallow burial depth (σ3 = 100 kPa) and low relative density of the foundation (Dr=0.3). Train operation is a safety hazard, and foundation reinforcement measures are required. However, for subway tunnels with deeper burial depths (σ3 = 200 and 300 kPa), the cumulative plastic strain is small (less than 0.13%), and the foundation is well stabilized so that no special treatment can be done. The accumulated plastic strain of the round gravel is large (> 0.61%) and shows a rapid development trend for subway tunnels with shallow burial depth (σ3 = 100 kPa) and low relative density of the foundation (Dr=0.3). Train operation is a safety hazard, and foundation reinforcement measures are required. However, for subway tunnels with deeper burial depths (σ3 = 200 and 300 kPa), the cumulative plastic strain is small (less than 0.13%), and the foundation is well stabilized so that no special treatment can be done. The sample Dr=0.3, σ3 = 100 kPa under the action of σamp d = 40 kPa, its dynamic pore pressure ratio reaches about 0.8, and the sample begins to show initial liquefaction. The increase in confining pressure and relative density can effectively inhibit the development of dynamic pore pressure and enhance the liquefaction resistance. The influence degree of each parameter on the long-term dynamic characteristics of round gravel from high to low is dynamic stress amplitude, confining pressure and relative density. The sample Dr=0.3, σ3 = 100 kPa under the action of σamp d = 40 kPa, its dynamic pore pressure ratio reaches about 0.8, and the sample begins to show initial liquefaction. The increase in confining pressure and relative density can effectively inhibit the development of dynamic pore pressure and enhance the liquefaction resistance. The influence degree of each parameter on the long-term dynamic characteristics of round gravel from high to low is dynamic stress amplitude, confining pressure and relative density. Increases in both relative density and confining pressure can significantly reduce the cumulative plastic (maximum decreases of about 77.05% and 80.33%) and elastic strains (maximum decreases of about 56.79% and 66.67%) of the round gravels, and improve their resilient modulus (maximum increases of about 130.1% and 203.15%). Therefore, for unfavorable working conditions such as Dr=0.3 and σ3 = 100 kPa, improvement measures such as increasing tunnel depth and implementing high-pressure grouting can be adopted to reduce the settlement of the subway foundation. Increases in both relative density and confining pressure can significantly reduce the cumulative plastic (maximum decreases of about 77.05% and 80.33%) and elastic strains (maximum decreases of about 56.79% and 66.67%) of the round gravels, and improve their resilient modulus (maximum increases of about 130.1% and 203.15%). Therefore, for unfavorable working conditions such as Dr=0.3 and σ3 = 100 kPa, improvement measures such as increasing tunnel depth and implementing high-pressure grouting can be adopted to reduce the settlement of the subway foundation. A cumulative plastic strain prediction model considering relative density, confining pressure and dynamic stress amplitude was developed using a hyperbolic improvement model based on the time-hardening method. The model has high accuracy and can be used to assess the stability of round gravel foundations under cyclic subway loading. It should be noted that this model is based on the conventional dynamic triaxial test under fixed axial cyclic load, and does not consider the rotation effect of the principal stress axis caused by subway cyclic load. A cumulative plastic strain prediction model considering relative density, confining pressure and dynamic stress amplitude was developed using a hyperbolic improvement model based on the time-hardening method. The model has high accuracy and can be used to assess the stability of round gravel foundations under cyclic subway loading. It should be noted that this model is based on the conventional dynamic triaxial test under fixed axial cyclic load, and does not consider the rotation effect of the principal stress axis caused by subway cyclic load.