Professional Documents
Culture Documents
Abstract: A block ramp is an environmentally friendly stream restoration structure consisting of a base material of stone with minimal
negative environmental impact. This not only decreases riverbed shear stress but also increases the flow resistance, thus reducing excess
kinetic energy of the flow. This study used two mathematical models that were developed in a dimensionless form for smooth and rough-bed
ramps. The models were calibrated and validated by two sets of experimental tests. A total of 308 experimental runs were done with eight
different ramp slopes and six different gravel material sizes. The experimental results indicated that an increase in the relative roughness
enhances energy dissipation. Additionally, energy dissipation increases along with an increase of ramp slope. Using a gene-expression
programming soft computing model, two mathematical models were developed to estimate energy dissipation on smooth and rough-
bed ramps. Additionally, the energy-loss performance of two configurations of interlocked and dumped blocks ramps was compared.
DOI: 10.1061/(ASCE)IR.1943-4774.0001442. © 2019 American Society of Civil Engineers.
Author keywords: Energy dissipation; Block ramp; Gene-expression programming; Dimensional analysis.
Because critical flow depth yc ¼ ðq2 =gÞ1=3 , using the Over the last 2 decades, soft computing techniques such as fuzzy
Buckingham Π theorem (Barenblatt 1987), the variables in Eq. (2) logic (FL), artificial neural networks (ANNs), evolutionary compu-
can be expressed in a nondimensional form tation (EC), support vector machine (SVM), genetic programming
(GP), and GEP, among others, have been successfully applied in
ΔE y y waterworks engineering (Azamathulla et al. 2009; Azamathulla
ΔEr ¼ ¼ f c ; c ; S; R; W ð3Þ and Ghani 2010). In this study, to extract the relationships by stan-
E0 d50 L
dard regression method, frequency relationships were tested by
where f = functional symbol; yc =d50 = relative critical flow depth; SPSS version 16 software. Considering that in a standard regres-
yc =L = ratio of critical depth to total ramp length; R = Reynolds sion model, selection of mathematical function type is performed
number; and W = Weber number. by the user at each step and only constant ratios are determined by
In all experiments, the head over the ramps was larger than 3 cm the model, this method is very time-intensive. Further, it cannot be
in order to eliminate the effect of surface tension; thus, the contri- claimed that the best type of mathematical function is fitted due to
bution of the Weber number can be considered negligible (Horton the limitation of possible combinations of the selected mathemati-
1907; Novak and Cabelka 1981). cal functions. In a GEP model, both mathematical functions and
The Reynolds number in this study was between 26,666.67 and constant ratios are intelligently selected by the model. Regarding
150,000; therefore, its value was high enough (R > 104 ) to have a the model’s metaheuristic nature due to the use of a genetic
fully turbulent flow. Hence, the effect of the fluid viscosity is not algorithm, the appropriate function is selected with a high speed.
technique.
In this study, after extracting the dimensionless relationship of After evaluating the initial population generated in the previous
energy dissipation on smooth-bed and rough-bed ramps in the first substep, if the program termination condition is satisfied, then the
step, the final relations are determined using the GEP system in program ends. Otherwise, according to the fourth substep, the
four major substeps in Step 2. chromosomes are reproduced.
The first substep includes the settings performed by the user, During reproduction, the chromosomes are modified by several
such as choosing the appropriate set of functions, number of genetic operators including mutation, inversion, insertion sequence
chromosomes, architecture of chromosomes (i.e., length of the (IS) transposition, root insertion sequence (RIS) transposition,
head and number of genes), and setting the rate of genetic oper- three types of crossover, gene crossover, and gene transposition.
ators as well as the number and range of numerical constants. The In this stage, the selection of chromosomes for reproduction is
authors initially used single gene and five length of heads, then based on their fitness values using selection techniques such as
increased the number of genes and heads, one after another during roulette wheel, tournament, or ranking methods. The better trees
each run, and further monitored the performance of each model. have greater chances of being selected for transitioning to the next
It was observed that the number of genes greater than 2 and iteration. These processes are continued until the maximum number
length of heads greater than 7 did not significantly enhance the of iterations is reached or the specified convergence criterion is
performance of GEP models. The authors then tried addition and achieved. The functional set and the rates assigned to operational
multiplication as linking functions and observed that linking the parameters used in GEP modeling during this study are listed in
sub-expression-trees (sub-ETs) by addition offered better fitness Table 2. Additional details about the reproduction step have been
values. provided by Ferreira (2001a, b).
Secondly, the chromosomes of each individual of the population
are randomly generated using the set of terminals and set of func-
tions. In this problem, the set of terminals consists of independent
dimensionless parameters extracted in Step 1 along with numerical
Experimental Setup and Measurements
constants. The functions and terminals are constructed together to The experimental tests were conducted at the Hydraulics
form a computer model with a root node and branches extending Laboratory of the Department of Hydraulic Structures at
from each function ending in a terminal. Shahid Chamran University of Ahvaz, Ahvaz, Iran. Runs were
In the third major substep, the chromosomes are expressed to carried out for block ramps with eight slopes between
the treelike structure where each program is executed, after which 1V ðVerticalÞ∶3.73H ðHorizontalÞ and 1V∶12H using six different
each individual is evaluated based on a fitness function. The tree uniform materials sizes of 11, 14, 17, 22, 24, and 30 mm. The
structure presents mathematical equations in a treelike form, which ramp was made of plexiglass with a thickness of 10 mm with
includes numerical constant, dimensionless variables, arithmetic an angle bar and iron glued onto the plexiglass to avoid warping.
operators (e.g., addition, subtraction, multiplication, and division), Also, a granular material was fixed over the plexiglass to prevent
mathematical functions (e.g., power, sin, and log, among others), ramp movement and instability. This ramp was placed at the
Boolean operators (e.g., and plus or), conditionals (e.g., if, then, upstream of a laboratory flume 8 m long, 0.3 m wide, and 0.8 m
and else), and other user-defined functions. The selection of the high for the first section with a length of 2 m, and 0.4 m for
appropriate function set is not very straightforward; however, a the second section with a length of 6 m, which was connected
good guess can always be helpful in order to include all the neces- upstream to a supply reservoir. A wave suppresser (installed in
sary functions. In this study, four basic arithmetic operators and the outlet of supply pipe) and a grid wall (installed in the inlet
three basic mathematical functions (exp, power, and 1 − x) were section of the flume) were used to ensure smooth entry of water
utilized. For this problem, the fitness, Fi , of an individual program, over the ramp. A sluice gate installed at the end of the flume al-
i, is measured by lowed adjustment of the downstream water level at any required
X
Ct level in order to create the hydraulic jump. The primary section of
Fi ¼ ðM − jCði;jÞ − T j jÞ ð6Þ the ramp was connected to a broad-crested weir with a length
j¼1 of 0.3 m.
A total of 260 tests were conducted to evaluate energy dissipa-
where M = range of selection; Cði;jÞ = value returned by the indi- tion and derivatives in the relations for both smooth-bed and rough-
vidual chromosome i for fitness case j (out of Ct fitness cases); and bed ramps, with discharges ranging between 8 and 36 L=s with
T i = target value for fitness case j. If jCði;jÞ − T j j (the precision) is an increment of 3 L=s. Furthermore, 48 new runs with different
less than or equal to 0.01, then the precision is equal to zero, and hydraulic and geometry characteristics were carried out to validate
f i ¼ f max ¼ Ct · M. In this case M ¼ 100 was used. the developed relations. Validation experiments for every model
were performed for flow rates ranging from 15 to 45 L=s. The Results and Discussion
physical model test matrix also included one dumped block
(gabionlike model) ramp in order to investigate the effect of boulder Effects of Relative Roughness and Bed Slope on
arrangement on energy dissipation. Details from two sets of the Energy Dissipation
experimental model are reported in Table 3.
The measurement of discharge Q was performed by an electro- As previously described, the extent of relative energy dissipation on
magnetic flow meter with an accuracy of 1%, which was installed block ramps depends on the dimensionless parameters of yc =L and
on the main supply line of the laboratory. In each run, the water yc =d50 , along with bed slope S. Thus, in this section, the effects of
depth was measured upstream of the ramp and downstream these parameters on energy dissipation were investigated applying
of the hydraulic jump via a point gauge (with an accuracy of the first set of experimental data (Table 3). Fig. 2 depicts the varia-
0.1 mm) after stabilizing the water level. Meanwhile, minor tion of ΔEr with yc =L for different ramp slopes for various material
irregularities at the water surface, primarily due to the influence sizes. In addition, the energy dissipation over the smooth-bed ramp
of the side walls, resulted in an overall accuracy with the order in comparison with block ramps with the same slope is also illus-
of 0.1 mm. Flow depth was recorded at three points across each trated in Fig. 2. The results show that energy dissipation grows with
section (at W=4, W=2, and 3W=4, where W is the flume width), an increase in relative roughness. On the other hand, the relative
and the mean of the three values was considered as the depth energy dissipation diminishes with an increase in the yc =L ratio,
for each point. A schematic view of the experimental apparatus which means that for a ramp with the same length and roughness
is displayed in Fig. 1. size, as the flow discharge rises, the relative submergence in-
In the experiments, after stabilizing the flow discharge and the creases. This suggests that the relationship between energy loss and
upstream head over the weir, the upstream head (water surface flow rate is inversely proportional. Comparison of smooth-bed
elevation) was measured upstream of the ramp at a distance of about and rough-bed ramps demonstrates that the energy dissipation of a
four times of the flow depth over the ramp crest. Due to the irregu- rough-bed ramp is 0.16–3.2 times higher than that of a correspond-
larity and turbulence of flow at the toe of the ramp, downstream ing smooth-bed ramp.
water depth (y1 ) was determined as the upstream conjugate of Fig. 3 shows the variation of relative energy dissipation with
the hydraulic jump by applying the Belanger equation. For this, yc =L for different roughness sizes with five different ramp slopes.
the downstream water depth was slowly increased using the It is clear that relative energy dissipation grows with an increase
downstream sluice gate until the upstream section of the hydraulic in the ramp slope. This means that at constant values of discharge
jump was in a cross section where the flow direction was practically and ramp length, the dissipation is directly proportional to the
horizontal. ramp slope, and for constant values of discharge and ramp slope,
Energy dissipation of flow over the ramp was calculated for each it rises as the ramp length increases. This result agrees with the
run using the water surface level measured in the upstream findings of Ahmad and Srisvastava (2014), although it is contrary
approach flow and flow depth at the toe of the ramp, which was to the results of Pagliara and Chiavaccini (2006a). The reason for
calculated based on hydraulic jump and Belanger equation. Hence, such disagreement could be attributed to use of different dimen-
the energy dissipation between the entrance and the toe of a ramp sionless numbers in the data analysis. In the study by Pagliara and
can be written as follows: Chiavaccini (2006a), the ratio yc =P (P = height of the block ramp)
was considered as the dimensionless number where P depends
E0 − E1 on S and can vary as block ramp slopes changes; thus the range
ΔEr ¼ ð7Þ
E0 of energy dissipation observed on the horizontal axis is not the
same.
where E0 ¼ H þ v2 =2g; E1 ¼ y1 þ v21 =2g; v1 ¼ q=y1 is the down- This can be clarified by an example. Fig. 4 reveals the energy
stream velocity; and v ¼ q=H represents the upstream approach dissipation in smooth and rough ramps (with a material size of
velocity. Mean flow velocities can be determined using the continu- 1.3 cm) examined in this study versus the dimensionless number
ity equation. selected by Pagliara and Chiavaccini (2006a) (yc =P). The results
Fig. 1. (a) Section view of laboratory flume; (b) plane view of the laboratory flume; (c) details of the block ramp with hydraulic and geometrical
parameters; and (d) downstream view of the block ramp.
indicated that the conclusion presented by Pagliara and Chiavaccini where correlation coefficient (R2 ) root-mean square error (RMSE),
(2006a) can be true if the dimensionless ratio yc =P is used. Ac- and mean absolute percentage error (MAPE), were found to be
cording to Fig. 3, if the ratio yc =L (as in this study) is applied 0.92, 0.032, and 0.084, respectively.
instead of dimensionless number yc =P, the energy-dissipation Using the GEP model for the effective dimensionless parameters
values will be in the same range corresponding to the horizontal on energy dissipation in rough-bed ramps in Eq. (4), Eq. (9) was
axis for a slope with the same length. Therefore, use of yc =L will developed for estimation of the energy dissipation on block ramps
cause the relative energy dissipation to grow with an increase in the as follows:
ramp slope.
yc
6.14 yc yc yc
L
ΔEr ¼ S þ 2.25 − 0.30 ð9Þ
Derivation of Energy-Dissipation Relation Using GEP L L d50
This section describes the energy dissipation modeled using the
GEP approach. The GEP model was calibrated using the first set where the correlation coefficient R2 is 0.93, RMSE is 0.037, and
of collected experimental data. The optimal relationship between MAPE is 0.044.
dimensionless parameters has been shown in Eq. (5) for calculation The GEP formulations proposed in Eqs. (8) and (9) are valid for
of the relative energy dissipation, where ΔEr , in smooth-bed dimensionless parameters ranging between minimum and maxi-
ramps, is given in Eq. (8) mum values reported in Table 1.
Fig. 5 compares the calculated energy dissipations by Eqs. (8)
and (9) versus the measured values in smooth-bed and rough-bed
y y
ΔEr ¼ exp −49.06 c þ S þ S2 1 − 35.54 c ð8Þ ramps, respectively. This figure confirms agreement between the
L L calculated energy dissipation and measured values for both models
Fig. 2. Variation of ΔEr versus yc =L for different ramp slopes: (a) S ¼ 1V∶4H; (b) S ¼ 1V∶5H; (c) S ¼ 1V∶8H; and (d) S ¼ 1V∶12H.
of smooth-bed and rough-bed ramps with an error line equal to and large-scale roughness (LR). Fig. 6(a) shows that the first set of
15% and 10%, respectively. data collected in this study, which contained all three groups (22%
SR, 64% IR, and 14% LR), as well as the data from Ahmad and
Srisvastava (2014) fall in the range of a large-scale roughness
Verification of Developed Relationships group. Fig. 6(b) illustrates that the agreement between the first
This section describes two sets of data that were used to verify the set of smooth-bed ramp data and data calculated using Eq. (1)
derived relationships. The first set of data evaluates the block is within 35%. Fig. 6(c) indicates the precision of Eq. (1) for pre-
ramp relationship based on the experimental data of Ahmad and dicting the energy dissipation on block ramps based on the first set
Srisvastava (2014). They carried out tests on semifield block ramps of data collected in this study and the data from Ahmad and
with three different slopes (1V∶9.87H, 1V∶6H, and 1V∶3H) using Srisvastava (2014). The accuracy of prediction of the block ramp
large-scale boulders on a 0.83-m-wide rectangular ramp. The energy-dissipation relationship developed in this study was also evalu-
experiments were performed for nine gradations of the boulders ated based on the experimental data from Ahmad and Srisvastava
with an average diameter (d50 ) between 64.0 and 133.8 mm. (2014). The results revealed that Eq. (8) developed here more accu-
The height of the ramps (P) was constant and equal to 0.7 m, and rately predicted the data of Ahmad and Srisvastava (2014).
three different ramp lengths (L) equal to 6.21, 4.2, and 2.1 m were Furthermore, to better evaluate the developed models for
applied. smooth-bed and rough-bed ramps, a second set of experiments was
Similar to the classification provided by Bathurst (1978) as well conducted with new geometry and hydraulic conditions [Table 3
as Bathurst et al. (1981) based on yu =d85 (yu is the uniform flow and Fig. 7(a)]. For this purpose, the developed Eqs. (8) and (9) were
depth), Pagliara and Chiavaccini (2006a), based on the ratio of validated using the experimental tests that were not involved in
yc =d50 , classified block ramp roughness into three different groups the training step. According to Fig. 7(b), similar to the first set of
of small-scale roughness (SR), intermediate-scale roughness (IR), experimental data, the second set of rough-bed ramp data was
Fig. 3. Variation of ΔEr versus yc =L for different roughness sizes: (a) smooth bed; (b) d50 ¼ 11 mm; (c) d50 ¼ 14 mm; and (d) d50 ¼ 30 mm.
Fig. 4. Variation of ΔEr versus yc =P for different roughness sizes: (a) smooth-bed ramp; and (b) block ramp.
Fig. 5. Comparison between calculated and measured energy dissipation: (a) smooth-bed ramp; and (b) block ramp.
Fig. 6. (a) Distribution of rough-bed experimental data in triple roughness size categories; (b) comparison between measured and calculated energy
dissipation on a smooth-bed ramp using Eq. (1); and (c) comparison between measured and calculated energy dissipation using Eqs. (1) and (8) from
Ahmad and Srisvastava’s (2014) data and the first set of rough-bed data.
Fig. 7. (a) Schematic of the second set of experiment; (b) distribution of rough-bed experimental data in triple roughness size categories; and (c and d)
comparison between measured and calculated energy dissipation for smooth-bed and rough-bed ramps, respectively.
distributed across all three categories of rough-bed ramps Rough-Bed Configuration Efficiency
(36% SR, 61% IR, and 3% LR). The results of both relationships In addition to hydraulic parameters and geometric characteristics of
for the smooth-bed and rough-bed ramps satisfactorily determined the ramp and boulders, the energy dissipation over the block ramp
the dissipation of energy [Figs. 7(c and d)] and can be considered also depends on configurations of the boulders on the ramp.
as valid for the design of smooth and rough-bed ramps energy Tamagni et al. (2010) classified block ramps into two main classes,
dissipaters structures. namely block carpet and block cluster, based on morphological and
Acknowledgments
The authors would like to acknowledge the Vice Chancellor for
Research of Shahid Chamran University of Ahvaz for providing
financial support to conduct this study (Grant No. 95/3/2/62356).
Notation
Downloaded from ascelibrary.org by University of Liverpool on 12/06/19. Copyright ASCE. For personal use only; all rights reserved.