Deformation Pattern and Failure Mechanism of Railway Embankment Caused by Lake Water Fluctuation Using Earth Observation and On-Site Monitoring Techniques

Abstract

The prediction of railway embankment failure is still a global challenge for the railway industry due to the complexity of embankment failure mechanisms. In this work, the pre-failure deformation and the settlement from abnormal deformation to the final failure were investigated based on earth observation and on-site monitoring with a focus on the deformation stage and failure mechanism of railway embankments. Some new viewpoints are suggested: (1) the differential settlement of ~19 mm revealed via InSAR at the failure region of the embankment may have been caused by internal erosion after rapid drawdown. The cumulative settlement was found to increase with the decline of the lake water level. (2) The railway embankment experienced three phases of primary, secondary, and accelerated creep phases, similar to the evolution of most landslide or dam failures. However, the train loading and seepage force may have aggravated the secondary consolidation, promoting the embankment to enter the accelerated creep phase quickly. The deformation pattern was presented as an exponential curve trend. (3) The formation mechanism of embankment collapse can be summarized as “seepage failure-creep-shear slip-collapse” failure under repeated train loading and rapid drawdown. This work provides some clues for early warnings and for the development of maintenance plans.

1. Introduction

The railway embankment is the main structure alongside a river or lake. These railway embankments not only serve for transportation, but also have water conservation functions. It functions via the gravity of the track structure and the traffic loading compared to the embankment dam serving to hold the water resources. However, the railway embankment is susceptible to seepage failure or sliding after rapid drawdown, which may threaten the operation of trains and even cause great human and material losses. For example, in 2001, the embankment collapse was caused by the drawdown of the reservoir water level on the Long-Hai railway. In 2003, the right side of the embankment slid over 3 m due to the river water level dropping at the Hunan-Guizhou railway [1]. About 170 m of embankment collapse was also triggered by the lake water level dropping on the Beijing-Guangzhou railway in 2020. Furthermore, the rapid drawdown of water levels causing structural damage to railway embankments is one of the dominant causes of derailment according to the Transportation Safety Board Canada (TSBC), as reported over the past decades [2].

Many failures of nature slope or earth-rock-fill dams after rapid drawdown have been widely reported [3,4]. The evolution period of soil slope or artificial fill dams has been proven to be a long time. The slow creep process from creep deformation to failure, according to a larger number of deformation monitoring results, is also termed as the slope creep phenomenon [5]. Some scholars explained this changing law of displacement-time, including Saito laws [6], exponential laws [7], power laws [8], and stepwise patterns [9]. Generally, soil slope failure undergoes a three-stage deformation process [10], including (1) initial deformation, (2) constant deformation, and (3) an accelerated deformation stage. Moreover, the accelerated deformation stage can also be divided into initial acceleration, medium-term acceleration, and a critical failure sub-phase. Combining the initial and constant deformation stages, a five-stage deformation is proposed [5]. The early warning of slope instability is carried out by judging the stage of creep slope and identifying abnormal pre-slip precursor information [11]. This is the beginning of the early identification of a slope failure, although this displacement-time law of landslides is obtained under gravity [5]. Although the material, construction technique and stress state of railway embankments are different from nature slope and artificial fill dams, railway embankments also undergo a process from creep to collapse during their service life after the post-construction period. The prediction of railway embankment failure is still a global challenge for the railway industry due to the complexity of embankment failure mechanisms. Therefore, if the whole process of railway embankment failure can be represented via the displacement-time law with observation of the deformation trends and stages, it can be useful for early warning, the development of maintenance plans and the implementation of fault recovery. However, few studies focus on and record the progressive failure process of railway embankments.

The rapid drawdown of the water level is detrimental to the stability of the embankment slope in the reservoirs. Usually, the deformation processes of embankment and degradation of construction are observed via in situ and remote monitoring techniques [12,13,14,15]. Inclinometers, crack test instruments, strain sensors, displacement meters, and leveling are often used as conventional ground monitoring techniques to monitor the deformation caused by long-term consolidation and shear failure. These in situ monitoring methods can only obtain the monitoring data on a few sites, which is not usually enough to represent the overall deformation of an observation object, even if the high precision deformation information (up to millimeter level) can be obtained at different periods. In addition, a lot of time and labor costs are required for manual measuring techniques.

The ground-based remote measure (e.g., laser scanning) and satellite remote sensing techniques provide a non-contact measurement method. In particular, the interferometric synthetic aperture radar (InSAR) technique has the advantage of wide coverage and remote monitoring of the Earth’s surface, which is often used to monitor slow-moving objects. Currently, multi-temporal InSAR techniques have been widely used to monitor the deterioration and deformation of slopes or artificial structures such as reservoir landslides [16,17]), tailings dams [18,19], embankment dams [20,21], and railway infrastructure [22,23,24]. The open-access Sentinel-1A images with high temporal resolution and a short revisiting period (12 days) can be adopted for frequent ground acquisitions, but the radar cannot capture deformation data with rapid changes due to the limitation of the radar wavelength [25]. The PS (permanent scatterer) InSAR and the SBAS (small baseline subset) InSAR techniques were developed to obtain a series of deformation data and to identify potential creep deformation of the railway embankment before the critical failure [26]. Therefore, the InSAR technique can observe the pre-failure evolution of the railway embankment. If a warning of an anomalous deformation is given in advance before the collapse, it will provide sufficient time for engineering intervention to ensure the safe operation of trains.

Previous work has paid attention to the time series analysis of deformation for dams and the surrounding areas based on multi-temporal InSAR techniques. However, few studies focus on the failure process of railway embankments and the identification of failure portentous during their service life. The function of “turn back time” can review historical deformation based on InSAR techniques. The ground-based monitoring device can measure the rapid deformation of railway embankments in a very short time, which can help to compensate for the time constraints of the satellites’ revisiting period. Therefore, the main objectives of this work were as follows: (1) InSAR and precise level were used to monitor the displacement time series from creep deformation to failure during the rapid drawdown of water level. (2) The progressive failure stage and the abnormal pre-slip precursor information were determined for railway embankments. (3) The failure mechanism of a railway embankment was described considering the seepage and dynamic loading. This work provides some suggestions for early warning and for developing maintenance plans to ensure the safe operation and prevention of disaster on railways.

2. Study Area

Nanhu railway embankment is located in Yueyang City, Hunan Province (Figure 1a), and belongs to the Beijing-Guangzhou mixed passenger and freight railway with a design speed of 160 km/h. This study area has a subtropical monsoon climate with about 1289.8~1556.2 mm average annual rainfall [27]. The study area faces the South Lake to the east and is adjacent to the Dongting Lake to the west. The study area was near the lake coast with a well-developed water network, leading to the soft soil being widely distributed. The Nanhu railway embankment was built in 1999. It is composed of 16 m of artificial filling soil, silty clay and a muddy clay layer. The earth fill of the embankment is very high, forming the artificial high slope (Figure 1c).

The east side of the Nanhu railway embankment was constructed to protect the slope with pulp and stone. Another side of the embankment was exposed to the Dongting Lake, where the stability of the embankment was vulnerable to water level fluctuation. Unfortunately, about 170 m of the embankment collapsed during the period of water level decline. The sliding surface is displayed in Figure 1c. The Dongting Lake water submerged the top of the Nanhu embankment for two months after 13 July 2020, and then dropped continuously, resulting in the occurrence of this accident on 1 November. The maximum vertical displacement exceeded 1.2 m within 112 h after 1 November. It even resulted in a loss of 70 million yuan and railway disruption of up to 41 h.

3. Data

Sentine-1A operated on a radar carrier frequency of 5405 MHz (C band) with a 5.5 cm wavelength and a revisiting cycle of 12 days. Single look complex (SLC) data in vertical transmission and vertical receiving (VV) mode was used for InSAR processing. A total of 17 ascending orbit images were selected from 8 April 2020 to 29 October 2020, the incidence angle of these images was about 39.08°. The temporal baseline and the perpendicular baseline were limited to within 24 days and 200 m, respectively. Digital elevation model (DEM) (30 m × 30 m) data was prepared from the shuttle radar topography mission (SRTM). Significantly, the phenomenon of train shaking occurred at 8:20 on 1 November. It was indicated that a large settlement difference appeared between the left and right sides of the Nanhu railway embankment. Then, the settlement of the embankment was monitored every 2 h via precise level until the embankment collapsed on 6 November. An on-site monitoring technique can meet the requirement of continuous observation in a short time.

4. Time-Series InSAR Data Processing Method

The small baselines subset InSAR (SBAS-InSAR) technique with the weighted least squares estimation method was adopted to obtain the high precision time-series deformation based on the coherent targets with high coherence [28,29] because it has good adaptability in countryside vegetation areas [30]. One image was selected as the primary image from N + 1 images. Other images were registered to the primary image and the spatial and temporal baseline maps were drawn (Figure 2). Then, SBAS-InSAR created M interferences. M must be met to the following formula [31]:

$$\frac{N+1}{2}\le M\le N\times \left(\frac{N+1}{2}\right)$$

(1)

Assuming that the imaging times of master and slave images were t_a and t_b in the interferogram i, then, the phase signal of the unwrapping interference of the coherent pixel is presented as follows:

$$\Delta {\phi }_i=\Delta {\phi }_{t_b}-\Delta {\phi }_{t_a}\approx {\phi }_{orb}+{\phi }_{topo}+{\phi }_{atm}+{\phi }_{defo}+{\phi }_{noi}$$

(2)

where $\Delta {\phi }_{t_b}$ and $\Delta {\phi }_{t_a}$ were the phase error at time t_b and t_a with respect to the t₀, respectively. $\Delta {\phi }_{orb}$, $\Delta {\phi }_{atm}$, $\Delta {\phi }_{noi}$, $\Delta {\phi }_{topo}$ and $\Delta {\phi }_{defo}$ expressed the phase error of satellite orbit, atmospheric delay, residual noise, topography, and deformation, respectively. The deformation interferometric phase was caused by the deformation of the target object. $\Delta {\phi }_{defo}$ can be achieved via removing the residual components. Then, M equations with N unknowns were established, which are shown as follows:

$$A{\left[\phi \left(t_1\right),\phi \left(t_2\right),\dots ,\phi \left(t_N\right)\right]}^T=\delta {\left[\phi \left(t_1\right),\phi \left(t_2\right),\dots ,\phi \left(t_N\right)\right]}^T$$

(3)

where A was a M × N design matrix related to the interferometric atlas obtained from the SAR data. ${\left[\phi \left(t_1\right),\phi \left(t_2\right),\dots ,\phi \left(t_N\right)\right]}^T$ was a N × 1 order vector based on an unknown deformation phase of the measuring point. $\delta {\left[\phi \left(t_1\right),\phi \left(t_2\right),\dots ,\phi \left(t_N\right)\right]}^T$ represented the unwrapped phase value. The unknown deformation phase can be replaced by the average phase velocity v. It is expressed as follows:

$$v={\left[v_1,v_2\dots ,v_n\right]}^T={\left[\frac{{\phi }_1}{t_1-t_0},\frac{{\phi }_2-{\phi }_1}{t_2-t_1}\dots ,\frac{{\phi }_N-{\phi }_{N-1}}{t_N-t_{N-1}}\right]}^T$$

(4)

Equation (3) can then be expressed as:

$$Bv=\delta {\left[\phi \left(t_1\right),\phi \left(t_2\right),\dots ,\phi \left(t_N\right)\right]}^T$$

(5)

where B was a coefficient matrix (M × N), composed of 0, 1 and −1. Each row of the B matrix was an unwrapped differential interference phase and each column corresponded to a scene image. Finally, the surface deformation rate was solved via the least squares method and the singular value decomposition (SVD) method.

The external DEM data and precise orbit ephemerides data were adopted to conduct the geometry co-registration for each image. The temporal and spatial baselines of each interferogram pairs were calculated via SBAS-InSAR technology. The time interval for each interferogram pair (temporal baselines) and the space distance of the satellite at different shooting times (spatial baselines) were observed and the spatial and temporal baseline plots were then created (Figure 2). Then, the Goldstein filter and minimum cost flow (MCF) phase unwrapping were performed to obtain a series of unwrapped differential interferograms [32]. The ground control points (GCPs) were selected for refinement and re-flattening to remove the residual constant phase and residual phase ramp, which were used to improve the accuracy of deformation monitoring. The unwrapping coherence threshold was set as 0.2 to mask pixels with coherence less than this threshold. Finally, the deformation velocity was obtained after the interferogram inversion and geocoding procedure. The flow chart of SBAS-InSAR used in this work is presented in Figure 3.

5. Results

5.1. Deformation Detected Based on SABS-InSAR

To quantitatively analyze the progressive failure process of the railway embankment and the difference in deformation trends between the railway embankment and nature slope or artificial fill dams, the deformation of the line-of-sight (LOS) direction was first converted to the vertical direction. Thus, a profile of vertical displacement rates was prepared along the railway corridors based on the SBAS-InSAR method from 8 April 2020 to 29 October 2020 (Figure 4). However, a limitation in the experimental site was that the dense vegetation or water level fluctuation affected the interference results of InSAR.

As shown in Figure 4, the deformation occurs mainly on the embankment top and the upstream face. Three representative sections (white oval dotted lines A, B and C in Figure 4) were regarded as the key observation regions of the embankment deformation. The displacement rates measured in area A of the left embankment were as high as −38 mm yr⁻¹ (negative value indicates subsidence). This part of the embankment was exposed to Dongting Lake, and the stability of the embankment in this area was easily affected by water level fluctuation. Vertical deformation rates indicated rapid settlement (>37 mm yr⁻¹) in section B, although the embankment in this area was covered with mortar rubble. The protecting slope structure should have prevented the infiltration of water and the sliding of the embankment although the mortar rubble was damaged (Figure 4). There were a few points with deformation rates ranging from −30 to −35 mm yr⁻¹ in the top of the railway embankment in section C. Section C was located on the embankment side of the bridge-subgrade transition section, and there were apparent differential settlements in this region. The soft soil settlement observed near the bridge side was not harmful to the health of the bridge structure. Here, deformation was most probably related to the consolidation of soil after dissipation of pore water pressure when the lake water ebbed, instead of the deformation of the bridge. The railway line in this section was mainly composed of bridges with 12.2 m width prestressed concrete box beams and 50 m piers. The on-site investigation indicated that the bridge was well-structured and without cracks, offsets, or deformation.

The high density of measurement points provided deformation information on the activity of three regions traversed by the railway such as clear subsidence zone A, with deformation rates as high as −38 mm yr⁻¹. Here, the embankment had not been preserved with mortar rubble and was exposed to the lake water. Unfortunately, this part of area A slipped after 1 November 2020, and the potential deformation signs were observed in area A. The state of embankments in sections B and C were suggested as well-structured.

5.2. Displacement Time Series of the Railway Embankment

Deformation was an inevitable process in the service life of the railway embankment, which was often aggravated by rapid drawdown and train loading. The time series analysis of the railway embankment deformation helped to discern whether the settlement was normal or abnormal. To understand the movement trend, the vertical settlement time series of two sites (Location P₁ and P₂) in section A was displayed. It was noted that the deformation of the railway embankments was not obvious when the water level rose (from 26 May 2020 to 1 July 2020) (Figure 5). Therefore, the period (13 July 2020 to 29 October 2020) from the initial creep to the embankment pre-failure during the rapid drawdown was focused on, with the cumulative displacement being obtained via the SBAS-InSAR method. However, the process from critical failure to collapse of the embankment was a relatively short period: 1 November to 6 November. This time was in the period between two satellites returning to the same position. Thus, when the phenomenon of train shaking was perceived by the train driver at 8:20 on 1 November, the precise level was used to monitor the deformation in this period until the embankment collapsed.

Site P₁ was located at the top of the embankment failure section. P₂ was on the transition section between the culverts and embankment. The deformation time series of the two sites suggested a similar change trend but of different magnitudes. The cumulative settlements of P₁ and P₂ reach −19 and −17 mm in the final time, respectively. The cumulative settlements were found to increase with the decrease in lake water level from 13 July 2020 to 29 October 2020. It was indicated that this embankment slope exhibited creeping deformation before the pre-failure stage on 1 November. The mortar rubble was found to be damaged due to differential settlement in the transition section between the culverts and embankment via field observation (Figure 4).

Creep deformation of the embankment did not affect driving comfort before 8:20 on 1 November. The phenomenon of train shaking was first perceived by the train driver. Subsequently, the deformation was continuously monitored using the precise level (per 2 h) until the embankment collapsed. The observation point was near to site P₁. As shown in Figure 6, the large deformation began at 8:20 on 1 November and gradually escalated until the full sliding of the embankment (positive value indicates accumulated settlement). The deformation was 35 mm within 30 h after 8:20 on 1 November, which exceeded the cumulative deformation during the creep process from 13 July 2020 to 29 October 2020. The deformation increased sharply at 2:00 p.m. on 4 November (after 102 h), being 232 mm. This embankment was closed to traffic at this time due to the excessive subsidence. Finally, the cumulative subsidence of the embankment reached more than 1400 mm within the following 16 h, and the rate of deformation increased abruptly.

6. Discussion

6.1. Displacement Time Series of the Railway Embankment

Generally, nature slope and artificial fill dams will undergo a certain period of evolution from deformation to final failure under gravity. However, the deformation law of railway embankment was still not clear under the traffic loading. Therefore, it was important to describe the failure process and the movement patterns of the railway embankment.

The macroscopic deformation of the embankment failure was recorded via SBAS-InSAR and leveling during the rapid drawdown. The creep process was very long (108 days, 13 July 2020 to 29 October 2020) and the rapid deformation process was relatively short. To clearly describe the deformation-time law, the deformation trend was concerned before embankment failure from 29 October to 6 November 2020 (within 200 h) (Figure 7). The velocity-time and acceleration-time history were obtained via differential calculation of the measured displacement data (Figure 8). In addition, a tangential angle-time curve was calculated to confirm the sub-stage in the accelerated deformation stage via drawing a T-t (relative time T-monitoring time t) [5] in Figure 7. The formula is expressed as follows:

$$T\left(i\right)=\frac{s\left(i\right)}{v}$$

(6)

where s(i) refers the cumulative displacement within a specific time t_i, T(i) was the relative time corresponding to monitoring time. v presents the average velocity of the secondary creep phase.

$${\alpha }_i=\arctan \frac{T\left(i\right)-T\left(i-1\right)}{t_i-t_{i-1}}=\frac{\Delta T}{\Delta t}$$

(7)

where α_i was the tangential angle, t_i refers to the ith monitoring time (i = 1, 2, 3, …, n), Δt was the time interval, i.e., one day or one week. ΔT refers the change of T(i) during a specific period.

The anomaly deformation behavior was defined and deformation phases were distinguished based on the macro-deformation signs in field observation and the inflection point of the kinematic characteristic curve. Then, the movement progress of the embankment from the creeping to the final failure was divided into three stages:

(1)

Primary creep phase (decelerated stage), the velocity and acceleration increased from zero and then decreased (0 to 88 h) (Figure 8). The tangential angle was less than 20° (Figure 7). The settlement was very small in this phase, which was the consolidation of the embankment after the dissipation of the pore pressure under train loading. In this phase, the embankment was in a healthy state and did not affect train operation.

(2)

Secondary creep phase (steady-state stage). The velocity increased slowly, and the acceleration varied around zero (88 to 155 h), even becoming a negative value (Figure 8). The tangential angle was less than 60° (Figure 7). The cumulative settlement increased gradually, larger than 70 mm (average value of settlement in this stage). Cracks were observed in the shoulder structure of the embankment via visual interpretation. The phenomenon of shaking and an uncomfortable driving feel were perceived by train drivers at this moment.

(3)

Accelerated creep phase (accelerated stage), the velocity and settlement show a trend of increasing non-linearity with time. The initial (lower than 82°), medium-term acceleration (82~85°) and imminent failure (greater than 85°) sub-phase were divided based on the inflection point at the tangential angle-time curve (Figure 7). The velocity increased abruptly in the imminent failure sub-phase and continued to increase until the embankment collapse (Figure 8). The majority of deformation (about 90% of total) occurred in this stage within a short time, resulting in the failure of the railway infrastructure. Since the slip surface developed progressively, obvious deformation and huge pavement cracks were caused successively. At this time, the embankment was not suitable for trains to pass. Nevertheless, when the embankment failed completely, the acceleration dropped sharply, becoming a negative value.

Identifying the deformation stage and abnormal pre-slip behavior of the embankment can guide the implementation of maintenance plans and ensure the safety of train operations. It means that the changes in kinematic parameters (velocity, acceleration, deformation, and tangential angle) should be used to determine whether an embankment has entered the secondary creep, or the accelerated creep phase. For this study, the settlement-time and velocity-time relationships were initially linear, but began to show curvature after entering the accelerated creep phase, turning the curve nearly straight up after the imminent failure sub-phase. Therefore, the deformation time series should be carried out based on multi-temporal InSAR techniques during the wet season or dry season, and the macroscopic deformation of the embankment recorded from the primary creep phase. The signs of fractures and deformation should be investigated by increasing manual patrol time and frequency, when the abnormal deformation and on-site phenomena can be observed. Nevertheless, it is alarming that if huge cracks and deformation appear in the infrastructure of the embankment, and even the phenomenon of train shaking perceived, it indicates that the degree of embankment deterioration is very serious and that this deformation has entered the accelerated creep phase. Then, engineering measures must be carried out to prevent further sliding, e.g., pile foundation reinforcement and loading berm.

6.2. Comparison of Deformation Patterns of Various Slope

The failure process of the Nanhu embankment caused by rapid drawdown presents a progressive failure process, which includes primary creep (decelerated stage), secondary creep (steady-state stage) and accelerated creep phase (accelerated stage). This work proves that railway embankments also go through three stages like most nature slopes and earth-rock-fill dams, although the railway embankment was subjected to train loading. However, it was interesting to see whether the deformation pattern of railway embankments was different from natural slopes or artificial fill dams influenced by the changing water level. To find the unique deformation features of railway embankments, the deformation patterns of many landslides and slides at dams are summarized in this work and divided into three modes [33,34] including: steady type, step type, and sudden failure type.

Figure 9 displays the difference in deformation patterns between various slopes. Steady type: the accumulated displacement changes slowly and remains stable, such as a gentle slope with a higher safety margin. The slope is still in a stable state under the natural conditions. However, this type of slope easily loses stability and sliding is affected by external adverse factors. Lianziya landslide presents this deformation pattern due to the impact of changes in reservoir water level [35]. Step type: this type of slope is affected by a cyclical adverse factor. For example, the deformation pattern of landslides always exhibits a step-like rise trend due to the fluctuation of the reservoir water level in the Three Gorges region [36,37]. Sudden failure type: the accumulated displacement-time curve shows a trend that the deformation rate is gradually increasing, also termed “Saito law.” The sudden failure means a very short time for landslide collapse after entering the accelerated creep phase. For example, the failures at the Waco Dam, Carsington Dam, and the Xintan landslide show this deformation pattern from the creep to the final failure under the influence of changes in the reservoir water level [38,39]. However, the deformation curve of the railway embankment shows an exponential curve, which tends to increase nonlinearly. It was concluded that the time of the secondary creep phase at the railway embankment was shortened. Furthermore, the train loading aggravated the failure process of the embankment, leading to a short time from abnormal deformation to embankment sliding. Because the train to work has a high requirement for the settlement of the embankment, the large deformation of the embankment may lead to a traffic accident. Thus, only the correct interpretation of deformation behavior for the railway embankment can improve the response time of safety decisions or recovery.

6.3. Failure Mechanism of Embankment Caused by Rapid Drawdown

Many soil slope failure accidents during rapid drawdown have been widely reported. For example, in the United States, 30% of landslides occurred under rapid drawdown conditions in Roosevelt Lake from 1941 to 1953 [40]. The same was encountered in Japan. The drawdown of the reservoir caused ~60% of landslides in recorded events [41]. In China, the discharge of the Three Gorges Reservoir triggered group-occurring landslides, which encouraged people and local governments to pay attention to the failure mechanism of these reservoir-induced landslides [42]. Some researchers have begun to pay attention to the failure mechanism of soil structure under seepage [43,44]. Unlike most landslides or dams, the deformation of railway embankments is exacerbated under the additional stress of train loading and the permeation effect [45]. Therefore, this work aimed to interpret the unique mechanism of embankment failure considering the coupling of seepage and stress.

(1)

The saturated surface inside the embankment appeared after a rapid drawdown of the water level, generating the hydraulic gradient and penetration force. Some particles were forced to migrate out of the materials caused by flowing water, which changed the grain grading of embankment materials. The finer particles were washed out, resulting in the increase of the soil pore. The rearrangement of soil particles and the disruption of the soil skeleton caused small deformation under gravity and traffic loading (Figure 10). Thus, the settlement was almost imperceptible in the primary creep phase.

(2)

With the continuous falling of lake water, the irregular seepage paths were formed and gradually expanded with the loss of fine particles, forming the erosion tunnels. The hydrostatic pressure acting on the embankment decreased, while a greater hydraulic gradient was formed inside the embankment. Together, the seepage force pointing to the outside of the embankment and the train loading induced the secondary consolidation of soil, leading to significant deformation and a minor fracture around the embankment surface. The shear stress concentrated at the toe of the embankment caused the plastic zone. The deformation velocity and settlement significantly increased, which caused differential settlement in the secondary creep phase.

(3)

As the lake water completely ebbed, the hydrostatic pressure acting on the embankment vanished. however, the pore water was not completely drained from the embankment soil. Because of the increase in excessive pore water pressure under repeated train loading, the effective stress and the shear strength of soils became smaller. The plastic-yielding area first appeared at the foot of the embankment. The shear plastic and tensile plastic zone developed gradually upward from the foot and downward from the top of the embankment. The sliding surface formed when the two plastic zones coalesced, and the embankment slid downwards until it collapsed. Thus, the velocity and settlement of the embankment increased abruptly in the accelerated creep phase.

7. Conclusions

Nanhu embankment failure was a traffic accident that caused economic losses. Thus, it was urgently necessary to find the main factors and to prevent similar incidents from happening again. The time series InSAR technique was adopted to retrieve the pre-failure deformation of the embankment. The settlement monitoring data was recorded in the field from abnormal deformation to the final failure. This work also drew attention to the failure process and deformation stage of the embankment. The failure mechanism of the railway embankments is discussed and the following conclusions are suggested:

(1)

The differential settlement was revealed via InSAR at section A of the embankment exposed to the lake water before the accident, which exceeded −38 mm yr⁻¹ deformation rates. The cumulative settlements of two sites were found to increase with the decrease in the lake water level from 13 July to 29 October 2020 which was about 19 mm before the failure. This may be an indication of internal erosion caused by rapid drawdown. Although the magnitude of deformation was relatively small, the differential settlement area should have aroused close attention.

(2)

The settlement was recorded via leveling every 2 h from abnormal deformation to the final failure after the occurrence of train shaking. It was indicated that the settlement-time curve was initially linear, reaching 232 mm within 102 h. Afterward, the settlement began to show a curvature increase, turning the curve nearly straight up before the failure. The settlement reached more than 1400 mm within the following 16 h, and the embankment completely collapsed at 6:00 a.m. on 6 November 2020.

(3)

The railway embankment experienced three phases of primary creep, secondary creep and accelerated creep phase, like the evolution of most landslide or dam failures. Train loading and seepage force may have aggravated the secondary consolidation of soil and promoted the embankment to enter the accelerated creep phase quickly. The majority of the deformation (about 90% of the total) took place in the accelerated creep phase within a short time. Immediate engineering measures should be adopted to prevent future deformation at this stage.

(4)

The deformation patterns of many landslides and slides at dams were summarized, which seem to be divided into three modes. However, this work proved that the deformation pattern of the railway embankment was presented as an exponential curve trend after rapid drawdown, unlike the three common modes. The time of the secondary creep phase was shortened due to train loading and the permeation.

(5)

The failure mechanism of the embankment considering the coupling of seepage and stress was discussed in various deformation stages. It was indicated that the formation mechanism of the embankment collapse can be summarized as “seepage failure-creep-shear slip-collapse.” This work provides some clues for early warnings and for developing maintenance plans to ensure the safe operation of railways.