Visualization of the Onset of Scour under a Pipeline in Waves

This paper visualizes the onset of scour under a pipeline in regular waves and studies the mechanism in this process. After the test started, the sand particles near the downstream mudline began to oscillate periodically, probably due to the distribution pattern of the seepage hydraulic gradient under the pipeline as well as asymmetric waves. As the wave height gradually increased, the sand under the pipeline began to oscillate at increasing amplitude and the coverage of oscillation extended upstream. Some sand rushed out from the bottom of the pipeline, and the sand oscillated almost symmetrically around the bottom line of the pipeline thereafter. The oscillation amplitude of the sand particles continued to rise, probably due to a decrease in the seepage path under the pipeline and loosened sediment in the oscillation. The scour onset occurred after more sand rushed out from under the pipeline. The visualization results reproduced the delay in the scour onset reported previously, and related the delay to the aforementioned increase in sediment oscillation amplitude. This visualization can improve the understanding of the scour onset mechanism, and can serve as a guide to further investigate the critical conditions of scour onset and scour control countermeasures for offshore structures.


Introduction
Submarine pipelines are vital channels which transport crude oil, oil products, and water between the oil refineries and offshore platforms. Scour beneath a pipeline is a key concern for pipeline stability and safety. In marine hydrodynamic environments, scour onset can occur due to waves and currents, eventually causing a local scour hole to form under a pipeline. As the scour hole develops along the pipeline, the pipeline may be threatened by vortex induced vibration due to pipeline spanning and vortex shedding, which can eventually cause pipeline failure [1]. These failures can trigger both environmental and ecological disasters.
The onset of scour is the start of scour hole development under a pipeline, and has been investigated extensively. Chiew [2] analyzed the pressure difference on two sides of the pipeline in steady currents, revealing that piping, also known as seepage failure, is the dominant cause of scour onset under a pipeline. Sumer et al. [3] investigated the temporal variation of the average hydraulic gradient under the pipeline in an accelerating current and in regular waves. It was pointed out that excessive seepage flow beneath the pipeline encourages scour onset. An equation on the critical conditions of scour onset was also proposed for the steady current condition. Zang et al. [20] studied scour onset with numerical simulation. Systematic analysis was performed on the effects of hydrodynamic parameters on the normalized pressure difference on two sides of the pipeline. There is other literature which reports on the onset of scour [21][22][23][24].
However, reports on the mechanism of the scour onset are limited, especially in wave scenarios [3]. Sumer et al. [3] measured the hydraulic gradient across the pipeline during the scour onset in waves. Some important phenomena were found in the scour onset in waves, including a delay of piping failure after the hydraulic gradient exceeded the critical value of the sediment. The detailed mechanism of the scour onset in waves has not been comprehensively revealed, including the attribution of this onset to the delayed piping. The scour onset of a pipeline is closely associated with the behavior of sediment particles under the pipeline, which was not observed in previous studies. Further investigation is needed to improve understanding of this important phenomenon.
This study investigates the elementary mechanism of the scour onset in regular waves using visualization flume experiments ( Figure 1). A miniature camera enclosed in a transparent model pipeline was used to record full view videos of the sediment movement in the scour onset in waves, which has not been previously observed. The behavior of sediment in the scour onset was analyzed using the videos. The details in the scour onset mechanism were discussed based on the phenomena analysis, including the potential causes of the phenomena in previous studies. Direct observation of the scour onset process provides clues for further mechanism investigation of the scour onset in regular waves. The mechanical analysis in this research can be used in future studies on the critical conditions for the scour onset and countermeasures which may be used in offshore and marine structures, such as the ones studied in [25] and others.

Experiment Setup
The experiments were performed in a glass-sided hydraulic flume at the Laboratory of Hydraulic and Harbor Engineering, Tongji University. The flume ( Figure 2) is 50 m in length, 0.8 m in width and 1.2 m in depth. A piston-type wave maker capable of producing regular and irregular waves with a period between 0.5 s and 5.0 s was installed at the upstream end of the flume. The maximum wave height was 0.25 m at 0.5 m water depth. The error of the wave maker in the regular waves was lower than 0.1% in wave period and lower than 2% in wave height. A wave absorber was deployed at the downstream end of the flume. A sediment recess, 2.7 m long and 1.0 m deep, was located in the central part of the flume. The side walls and the bottom of the flume were made of impermeable concrete. Medium sand with a medium grain size d 50 = 0.283 mm and geometric standard deviation σ g = (d 84 /d 16 ) 1/2 = 1.8 was used as the bed sediment in the study.
Appl. Sci. 2020, 10, x FOR PEER REVIEW 3 of 11 The experiments were performed in a glass-sided hydraulic flume at the Laboratory of Hydraulic and Harbor Engineering, Tongji University. The flume ( Figure 2) is 50 m in length, 0.8 m in width and 1.2 m in depth. A piston-type wave maker capable of producing regular and irregular waves with a period between 0.5 s and 5.0 s was installed at the upstream end of the flume. The maximum wave height was 0.25 m at 0.5 m water depth. The error of the wave maker in the regular waves was lower than 0.1% in wave period and lower than 2% in wave height. A wave absorber was deployed at the downstream end of the flume. A sediment recess, 2.7 m long and 1.0 m deep, was located in the central part of the flume. The side walls and the bottom of the flume were made of impermeable concrete. Medium sand with a medium grain size d50 = 0.283 mm and geometric standard deviation g = (d84/d16) 1/2 = 1.8 was used as the bed sediment in the study. A transparent model pipeline made of acrylic was used in the experiments ( Figure 3). The diameter of the model pipeline was D = 0.11 m and the length was 0.80 m. The pipeline was partially buried, with an embedment ratio e/D = 0.045. Two jack screws were used to fix the pipeline in the flume, and thus pipeline movement was considered to be negligible. A Logitech C920 camera was installed inside the pipeline to observe the scour process beneath the pipeline. The resolution of the camera was 1920 × 1080 pixels and the average frame rate of the video was 23.94 fps with a maximum of 30 fps. A 5 mm-sized mesh was printed on the pipeline surface, covering the majority of the camera view. Two ballast blocks were enclosed in the sealed pipeline to improve stability. The setup of the model pipeline was identical to that previously presented in Zhu et al. [9]. Figure 4 is a typical view of the camera, where lines AB and CD indicate the intersection of the bed surface and the pipeline, also called the mudline. The thin yellow dot-dash line in the center of the view illustrates the bottom line of the pipeline; the buried part of the pipeline is symmetrical around this line. A Nortek Vectrino high-resolution acoustic digital velocimeter (ADV) was installed 3.0 m upstream of the sediment recess to monitor the orbital velocity at 1.0D above the bed ( Figure 2). The range of the ADV was ±1.0 m/s, and the accuracy was ±0.5% of measured value ±1 mm/s. The sampling frequency was 100 Hz. A wave gauge was set 1.0 m upstream of the ADV. A transparent model pipeline made of acrylic was used in the experiments ( Figure 3). The diameter of the model pipeline was D = 0.11 m and the length was 0.80 m. The pipeline was partially buried, with an embedment ratio e/D = 0.045. Two jack screws were used to fix the pipeline in the flume, and thus pipeline movement was considered to be negligible. A Logitech C920 camera was installed inside the pipeline to observe the scour process beneath the pipeline. The resolution of the camera was 1920 × 1080 pixels and the average frame rate of the video was 23.94 fps with a maximum of 30 fps. A 5 mm-sized mesh was printed on the pipeline surface, covering the majority of the camera view. Two ballast blocks were enclosed in the sealed pipeline to improve stability. The setup of the model pipeline was identical to that previously presented in Zhu et al. [9]. Figure 4 is a typical view of the camera, where lines AB and CD indicate the intersection of the bed surface and the pipeline, also called the mudline. The thin yellow dot-dash line in the center of the view illustrates the bottom line of the pipeline; the buried part of the pipeline is symmetrical around this line. A Nortek Vectrino high-resolution acoustic digital velocimeter (ADV) was installed 3.0 m upstream of the sediment recess to monitor the orbital velocity at 1.0D above the bed ( Figure 2). The range of the ADV was ±1.0 m/s, and the accuracy was ±0.5% of measured value ±1 mm/s. The sampling frequency was 100 Hz. A wave gauge was set 1.0 m upstream of the ADV.  In this study, the scour onset under a pipeline in waves was recorded using a camera. Before each test case, the sediment underneath the pipeline within the coverage of the mesh was removed and replaced gently. Then the bed surface was carefully leveled. This process ensured that scour onset occurs within the field of view of the camera. A similar arrangement was used in Sumer et al. [3]. In some cases where the onset of scour failed to appear inside the camera view, the test case was repeated, and the video recording was abandoned.
The test parameters were selected based on Zhou et al. [26] as follows: The water depth was h0 = 0.4 m; the wave period was T = 1.3 s and the wave height was H = 0.08 m. Thus, the Keulegan-Carpenter number KC = 2.46, calculated by where Um is the maximum undisturbed orbital velocity near the bed; T is the wave period and D is the pipeline diameter. The Shields number of the wave was θ = 0.13, which was calculated by the equations used by Cheng et al. [8] derived from Soulsby [27]. The Shields number was larger than the critical value of the sediment, and thus test condition was in live-bed condition. In this study, the emphasis was qualitative analysis on the elementary mechanism of the scour onset in waves. The potential scale effects on the results are acceptable for a qualitative mechanism study. Thus, hydraulic similitude was regarded less important.

Test Results
After the experiment, frames were extracted from the video recordings. Figure 5 shows a typical process of the scour onset underneath a pipeline. The size of the mesh in Figure 5 is 5 mm × 5 mm. The corresponding video clip is Video S1 in the supplemental content. Progressive waves started to   In this study, the scour onset under a pipeline in waves was recorded using a camera. Before each test case, the sediment underneath the pipeline within the coverage of the mesh was removed and replaced gently. Then the bed surface was carefully leveled. This process ensured that scour onset occurs within the field of view of the camera. A similar arrangement was used in Sumer et al. [3]. In some cases where the onset of scour failed to appear inside the camera view, the test case was repeated, and the video recording was abandoned.
The test parameters were selected based on Zhou et al. [26] as follows: The water depth was h0 = 0.4 m; the wave period was T = 1.3 s and the wave height was H = 0.08 m. Thus, the Keulegan-Carpenter number KC = 2.46, calculated by where Um is the maximum undisturbed orbital velocity near the bed; T is the wave period and D is the pipeline diameter. The Shields number of the wave was θ = 0.13, which was calculated by the equations used by Cheng et al. [8] derived from Soulsby [27]. The Shields number was larger than the critical value of the sediment, and thus test condition was in live-bed condition. In this study, the emphasis was qualitative analysis on the elementary mechanism of the scour onset in waves. The potential scale effects on the results are acceptable for a qualitative mechanism study. Thus, hydraulic similitude was regarded less important.

Test Results
After the experiment, frames were extracted from the video recordings. Figure 5 shows a typical process of the scour onset underneath a pipeline. The size of the mesh in Figure 5 is 5 mm × 5 mm. The corresponding video clip is Video S1 in the supplemental content. Progressive waves started to In this study, the scour onset under a pipeline in waves was recorded using a camera. Before each test case, the sediment underneath the pipeline within the coverage of the mesh was removed and replaced gently. Then the bed surface was carefully leveled. This process ensured that scour onset occurs within the field of view of the camera. A similar arrangement was used in Sumer et al. [3]. In some cases where the onset of scour failed to appear inside the camera view, the test case was repeated, and the video recording was abandoned.
The test parameters were selected based on Zhou et al. [26] as follows: The water depth was h 0 = 0.4 m; the wave period was T = 1.3 s and the wave height was H = 0.08 m. Thus, the Keulegan-Carpenter number KC = 2.46, calculated by where U m is the maximum undisturbed orbital velocity near the bed; T is the wave period and D is the pipeline diameter. The Shields number of the wave was θ = 0.13, which was calculated by the equations used by Cheng et al. [8] derived from Soulsby [27]. The Shields number was larger than the critical value of the sediment, and thus test condition was in live-bed condition. In this study, the emphasis was qualitative analysis on the elementary mechanism of the scour onset in waves. The potential scale effects on the results are acceptable for a qualitative mechanism study. Thus, hydraulic similitude was regarded less important.

Test Results
After the experiment, frames were extracted from the video recordings. Figure 5 shows a typical process of the scour onset underneath a pipeline. The size of the mesh in Figure 5 is 5 mm × 5 mm. The corresponding video clip is Video S1 in the supplemental content. Progressive waves started to affect the sediment under the pipeline at t = 31.100 s (Figure 5a). The sand particles near the downstream mudline CD moved downstream with a small displacement (Figure 5b), and moved back in approximately 0.5T (Figure 5c). Then, the sand particles continued oscillating (Figure 5d-g). The period of the oscillation was close to the wave period. The involved area gradually extended upstream and finally covered most of the buried pipeline (Figure 5f,g, the area marked with red). The amplitude of the oscillation also increased remarkably from about 2 mm to over 10 mm. At t = t 0 + 2.481T, some sand particles moving downstream traversed the downstream mudline and were deposited on the bed surface downstream of the pipeline (Figure 5f).
Appl. Sci. 2020, 10, x FOR PEER REVIEW 5 of 11 affect the sediment under the pipeline at t = 31.100 s (Figure 5a). The sand particles near the downstream mudline CD moved downstream with a small displacement (Figure 5b), and moved back in approximately 0.5T (Figure 5c). Then, the sand particles continued oscillating (Figure 5d-g). The period of the oscillation was close to the wave period. The involved area gradually extended upstream and finally covered most of the buried pipeline (Figure 5f,g, the area marked with red). The amplitude of the oscillation also increased remarkably from about 2 mm to over 10 mm. At t = t0 + 2.481T, some sand particles moving downstream traversed the downstream mudline and were deposited on the bed surface downstream of the pipeline (Figure 5f). The oscillation of the sand particles continued thereafter (Figure 5h). The affected area was located between curves A1B1 and C1D1, which was nearly symmetrical around the bottom line of the pipeline. At t = t0 + 8.508T, the oscillatory area began to expand and deform. The downstream edge of the oscillation area moved to curve C2D2, which was close to the downstream mudline CD ( Figure  5i). No remarkable change was observed in the upstream edge. In the next wave period, the changes in the oscillation area were more remarkable (Figure 5j). The downstream edge of the affected area (curve C3D3) extended further downstream. Some sand particles rushed out from the bottom of the pipeline and fell on the downstream bed surface, so curve C3D3 extended beyond the downstream mudline where the sand particles rushed out (Figure 5j). The upstream edge was almost unchanged.
The onset of scour occurred within the next wave period. At t = t0 + 10.526T, a jet flow with a high concentration of sediment load broke the downstream mudline CD at point E1 (Figure 5k, in the red circle), and a remarkable cavity formed on this cross section. When the wave trough approached, the remaining sand on the upstream side of the pipeline at point E2 on line AB was also scoured (Figure 5l). An initial scour hole formed. After the onset of scour, the scour hole expanded swiftly in the spanwise direction (Figure 5m,n). The oscillation of the sand particles continued thereafter (Figure 5h). The affected area was located between curves A 1 B 1 and C 1 D 1 , which was nearly symmetrical around the bottom line of the pipeline. At t = t 0 + 8.508T, the oscillatory area began to expand and deform. The downstream edge of the oscillation area moved to curve C 2 D 2 , which was close to the downstream mudline CD (Figure 5i). No remarkable change was observed in the upstream edge. In the next wave period, the changes in the oscillation area were more remarkable (Figure 5j). The downstream edge of the affected area (curve C 3 D 3 ) extended further downstream. Some sand particles rushed out from the bottom of the pipeline and fell on the downstream bed surface, so curve C 3 D 3 extended beyond the downstream mudline where the sand particles rushed out (Figure 5j). The upstream edge was almost unchanged.
The onset of scour occurred within the next wave period. At t = t 0 + 10.526T, a jet flow with a high concentration of sediment load broke the downstream mudline CD at point E 1 (Figure 5k, in the red circle), and a remarkable cavity formed on this cross section. When the wave trough approached, the remaining sand on the upstream side of the pipeline at point E 2 on line AB was also scoured (Figure 5l). An initial scour hole formed. After the onset of scour, the scour hole expanded swiftly in the spanwise direction (Figure 5m,n).

Discussion
In this study, the elementary mechanism of the scour onset in waves is studied based on the full view visualization of the sediment behavior under a pipeline. The phenomena in the tests are discussed and analyzed in this section.
At the beginning of the test, the sediment motion began adjacent to the downstream mudline (Figure 5b). The sediment incipient motion can be expounded by the distribution pattern of seepage hydraulic gradient under the pipeline. The distribution of the seepage hydraulic gradient under a pipeline is not even. In steady currents, the maximum seepage hydraulic gradient appears at the downstream side below the pipeline, close to the mudline [21,22]. In regular waves, the near-bottom flow is bidirectional, so the maximum hydraulic gradient can appear near the upstream and downstream mudlines. The waves in this study were non-linear, so the downstream bottom flow and seepage were more powerful than their upstream counterparts. A similar variation mode of the seepage hydraulic gradient can also be seen in Sumer et al. [3], where the wave was also nonlinear. As a result, the maximum seepage hydraulic gradient was located near the downstream mudline (Figure 6a). When the test started, the wave height gradually increased to the specified value and the local seepage hydraulic gradient overwhelmed the critical value of the sediment first near the downstream mudline, where the incipient motion of sediment under a pipeline occurred (Figure 6b). Herein, the critical seepage hydraulic gradient indicates the hydraulic gradient in the sediment at the critical point of the seepage failure, which is a function of the buoyant unit weight of soil.

Discussion
In this study, the elementary mechanism of the scour onset in waves is studied based on the full view visualization of the sediment behavior under a pipeline. The phenomena in the tests are discussed and analyzed in this section.
At the beginning of the test, the sediment motion began adjacent to the downstream mudline (Figure 5b). The sediment incipient motion can be expounded by the distribution pattern of seepage hydraulic gradient under the pipeline. The distribution of the seepage hydraulic gradient under a pipeline is not even. In steady currents, the maximum seepage hydraulic gradient appears at the downstream side below the pipeline, close to the mudline [21,22]. In regular waves, the near-bottom flow is bidirectional, so the maximum hydraulic gradient can appear near the upstream and downstream mudlines. The waves in this study were non-linear, so the downstream bottom flow and seepage were more powerful than their upstream counterparts. A similar variation mode of the seepage hydraulic gradient can also be seen in Sumer et al. [3], where the wave was also nonlinear. As a result, the maximum seepage hydraulic gradient was located near the downstream mudline ( Figure 6a). When the test started, the wave height gradually increased to the specified value and the local seepage hydraulic gradient overwhelmed the critical value of the sediment first near the downstream mudline, where the incipient motion of sediment under a pipeline occurred (Figure 6b). Herein, the critical seepage hydraulic gradient indicates the hydraulic gradient in the sediment at the critical point of the seepage failure, which is a function of the buoyant unit weight of soil. After the sediment incipient motion under the pipeline, more sediment began to move, and the oscillating amplitude of the sediment increased (Figure 5d-g). The expansion of the area affected by sediment movement can be attributed to the increase in the rising wave height (Figure 6c). With the wave height gradually increasing to the specified value, the variation amplitude of the seepage hydraulic gradient increased accordingly. Similar variation of hydraulic gradient can be seen in Sumer et al. [3]. The area where the local hydraulic gradient exceeded the critical value extended upstream, so the area experiencing sand particle oscillation extended upstream as well. The increasing oscillating amplitude of the sediment can also be explained by rise in the amplitude of the hydraulic gradient variation.
The sand oscillation turned to be almost symmetrical around the bottom line of the pipeline after t = t0 + 6.522T (Figure 5h-j). This change in the sediment oscillation pattern is analyzed below. At t = t0 + 2.481T, some sediment particles rushed out from the bottom of the pipeline due to the remarkable rise in sediment oscillation amplitude (Figures 5f and 7a). Thus, the remaining sand fell almost After the sediment incipient motion under the pipeline, more sediment began to move, and the oscillating amplitude of the sediment increased (Figure 5d-g). The expansion of the area affected by sediment movement can be attributed to the increase in the rising wave height (Figure 6c). With the wave height gradually increasing to the specified value, the variation amplitude of the seepage hydraulic gradient increased accordingly. Similar variation of hydraulic gradient can be seen in Sumer et al. [3]. The area where the local hydraulic gradient exceeded the critical value extended upstream, so the area experiencing sand particle oscillation extended upstream as well. The increasing oscillating amplitude of the sediment can also be explained by rise in the amplitude of the hydraulic gradient variation.
The sand oscillation turned to be almost symmetrical around the bottom line of the pipeline after t = t 0 + 6.522T (Figure 5h-j). This change in the sediment oscillation pattern is analyzed below. At t = t 0 + 2.481T, some sediment particles rushed out from the bottom of the pipeline due to the remarkable rise in sediment oscillation amplitude (Figures 5f and 7a). Thus, the remaining sand fell almost symmetrically around the bottom line of the pipeline due to gravity, and continued moving back and forth between curves A 1 B 1 and C 1 D 1 (Figures 5h and 7b).
Appl. Sci. 2020, 10, x FOR PEER REVIEW 9 of 11 symmetrically around the bottom line of the pipeline due to gravity, and continued moving back and forth between curves A1B1 and C1D1 (Figures 5h and 7b). In this period of time, the wave height may have reached the specified value, but the oscillatory amplitude of the sand particles continued to rise (Figure 5i,j). The rise in the oscillatory amplitude may be attributed to two factors. On one hand, the seepage path under the pipeline was shortened due to the loss of the sand under the pipeline (Figure 7c). Thus, the variation amplitude of the average hydraulic gradient under the pipeline increased with the wave height unchanged. On the other hand, the sand below the pipeline was continuously loosened due to oscillation. The loosened sediment was more vulnerable to the seepage-induced movement due to the decrease in the relative compaction. As a result, the impact of the wave induced hydraulic gradient on the sand beneath the pipeline was increasingly significant, and thus the oscillation amplitude of the sand particles continued to rise. Later, more sand was ejected from the bottom of the pipeline (Figure 5j) due to the rising oscillatory amplitude, which further intensified the sediment oscillation.
Sumer et al. [3] found a delay of the scour onset after the average hydraulic gradient first exceeds the critical value. The delay of scour onset was reproduced in the visualization of this study. The majority of sediment started to move at t = t0 + 2.899T, indicating that the hydraulic gradient overwhelmed the critical value in the majority area under the pipeline. The average hydraulic gradient was supposed to exceed the critical value at this point or even before. The scour onset occurred at t = t0 + 10.526T, which was about 8T later. The longer delay of piping in this study than that in Sumer et al. [3] can be attributed to more frequent changes of bottom flow direction due to the smaller KC number in this study [1]. The reproduction of the delay in the scour onset indicates that the basic mechanism of the scour onset analyzed in this study is similar to that investigated by Sumer et al. [3], where the KC number (KC = 2.2 ~ 29.0) was higher than that in this study (KC = 2.46).
Sumer et al. [3] attributed the delay to the fact that the exposure of sediment in the seepage hydraulic gradient for scour onset was not long enough for each crest half period. In this study, the visualization analysis further revealed that the delay in the scour onset under a pipeline in waves involves the increasing amplitude of sediment oscillation due to the increase of average seepage hydraulic gradient and the loosened sediment during the continuous sediment oscillation. The increase of average seepage hydraulic gradient was attributed to the decreasing seepage path due to the sand rushing out beneath the pipeline. The loosened sand under the pipeline due to oscillation made the sediment more vulnerable to the seepage induced movement. In this period of time, the wave height may have reached the specified value, but the oscillatory amplitude of the sand particles continued to rise (Figure 5i,j). The rise in the oscillatory amplitude may be attributed to two factors. On one hand, the seepage path under the pipeline was shortened due to the loss of the sand under the pipeline (Figure 7c). Thus, the variation amplitude of the average hydraulic gradient under the pipeline increased with the wave height unchanged. On the other hand, the sand below the pipeline was continuously loosened due to oscillation. The loosened sediment was more vulnerable to the seepage-induced movement due to the decrease in the relative compaction. As a result, the impact of the wave induced hydraulic gradient on the sand beneath the pipeline was increasingly significant, and thus the oscillation amplitude of the sand particles continued to rise. Later, more sand was ejected from the bottom of the pipeline (Figure 5j) due to the rising oscillatory amplitude, which further intensified the sediment oscillation.
Sumer et al. [3] found a delay of the scour onset after the average hydraulic gradient first exceeds the critical value. The delay of scour onset was reproduced in the visualization of this study. The majority of sediment started to move at t = t 0 + 2.899T, indicating that the hydraulic gradient overwhelmed the critical value in the majority area under the pipeline. The average hydraulic gradient was supposed to exceed the critical value at this point or even before. The scour onset occurred at t = t 0 + 10.526T, which was about 8T later. The longer delay of piping in this study than that in Sumer et al. [3] can be attributed to more frequent changes of bottom flow direction due to the smaller KC number in this study [1]. The reproduction of the delay in the scour onset indicates that the basic mechanism of the scour onset analyzed in this study is similar to that investigated by Sumer et al. [3], where the KC number (KC = 2.2~29.0) was higher than that in this study (KC = 2.46).
Sumer et al. [3] attributed the delay to the fact that the exposure of sediment in the seepage hydraulic gradient for scour onset was not long enough for each crest half period. In this study, the visualization analysis further revealed that the delay in the scour onset under a pipeline in waves involves the increasing amplitude of sediment oscillation due to the increase of average seepage hydraulic gradient and the loosened sediment during the continuous sediment oscillation. The increase of average seepage hydraulic gradient was attributed to the decreasing seepage path due to the sand rushing out beneath the pipeline. The loosened sand under the pipeline due to oscillation made the sediment more vulnerable to the seepage induced movement.

Conclusions
Visualization testing was performed to better understand the mechanism of scour onset beneath a pipeline in regular waves. A miniature camera was enclosed in a transparent pipeline to observe the sediment in the onset of scour below a pipeline. At the beginning of the test, the sediment motion began near the downstream mudline, probably due to the distribution pattern of the seepage hydraulic gradient under the pipeline and the asymmetric waves adopted in this study. With the wave height gradually increasing to the specified value, the sand under the pipeline began to oscillate at increasing amplitude and the coverage of oscillation extended upstream. Some sand rushed out from the bottom of the pipeline in this process. After that, the sand oscillation became almost symmetrical around the bottom line of the pipeline. This change was related to the loss of sediment that rushed out. The loss of sand under the pipeline reduced the length of the seepage path, and thus increased the variation amplitude of the hydraulic gradient. The oscillatory amplitude continued to rise, probably because the increase in the variation amplitude of the hydraulic gradient and loosened sediment in the oscillation. Later, more sand rushed out before the onset of scour because of the increasing oscillatory amplitude. The onset of scour occurred when a jet flow with high concentration of sediment load was ejected from the bottom of the pipeline. The remaining sediment was scoured when the bottom flow reversed at the wave trough. The initial scour hole formed and extended rapidly.
It was found in previous reports that the onset of scour in waves does not occur until the average seepage hydraulic gradient exceeded the critical value several times. This phenomenon was reproduced in the visualization of this study. In this study, the visualization analysis further revealed that the delay in the scour onset under a pipeline involves the increasing amplitude of sediment oscillation due to the increase in the variation amplitude of the hydraulic gradient and the loosened sediment during the continuous sediment oscillation.