Artificial Neural Network Applications for Predicting Drag Coefficient in Flexible Vegetated Channels Muhammad Mujahid Muhammad1, 2, Khamaruzaman Wan Yusof 1, Muhammad Raza Ul Mustafa1, Nor Azazi Zakaria3, and Aminuddin Ab. Ghani3 1

Department of Civil and Environmental Engineering, Universiti Teknologi Petronas, Seri Iskandar 32610, Perak, Malaysia. 2 Department of Water Resources and Environmental Engineering, Ahmadu Bello University, Zaria, Nigeria. 3 River Engineering and Urban Drainage Research Center, Universiti Sains Malaysia, Pulau Pinang, Malaysia. [email protected]

AbstractโPreviously numerous equations were developed using conventional methods to estimate vegetal drag coefficient by treating submerged and emergent vegetation independently, there is need to derive a generalized relationship that can be applied irrespective of the vegetation submergence with respect to flow depth. In this regard, the present study uses artificial neural network (ANN) as an advanced tool for prediction of drag coefficient in flexible vegetated channels. The training and testing patterns of the proposed ANN model were based on experimental results from the field and laboratory studies that combined both the submerged and emergent grass. A functional relation based on flow parameters and vegetation properties was derived through the use of dimensional analysis. The ANN model developed herein showed significantly better results in several model performance criteria when applied for verification. Index TermsโArtificial Neural Network; Dimensional Analysis; Drag Coefficient; Flexible Vegetated Channels.

I. INTRODUCTION The flow resistance in open channel are usually derived from the viscous and drag force over the wetted perimeter [1]. This drag is categorized into three comprising โ soil grain roughness, form roughness, and vegetative roughness. The vegetation drag is the most significant compared to others, as it has the utmost flow resistance that eventually decreases the average flow in vegetated channels [2, 3]. Thus, this will lead to rise in flow depth and residence time of sediments in the channel due to drag of vegetation [4]. Drag coefficient is an important parameter, this is because the drag enhanced the tendency for trapping, deposition, and stabilization of sediments, and its increases flow resistance and decreases the bed shear stress, which results in reduction of bed load transport capacity and erosion rate. Several procedures were proposed by researchers to model flow - vegetation interactions and sediment transport in open-channel [5-10]. Also, vegetation-induced drag force has been systematically studied by Kothyari, Tang, Wu and others [1, 11, 12]. However, with these investigations, it was deduced that there are discrepancies for the derivations of vegetation drag which necessitate for a general formula in evaluation of drag coefficient [13]. In addition, conventional regression has been used to develop mathematical equations in estimating hydraulic variables, such as drag coefficient, in vegetated channels. This approach of regression was found to either

over or under-estimate the hydraulic variables. Thus, for a better and accurate predictions, soft computing techniques such as artificial neural network (ANN) are nowadays employed in water resources engineering to estimate hydraulic and hydrologic variables [14-16]. Therefore, the main objective of this paper is to develop ANN model that will compute the vegetal drag coefficient in natural grassed channels irrespective of the grass submergence in relation to flow depth. Thus, field and laboratory experiments were conducted to collect hydraulic data to establish the ANN model. II. FUNCTIONAL RELATIONSHIP A functional relationship was formulated based on the criteria of Wu et al. [1], Tang et al. [11], Kothyari et al. [12] and Wilson et al. [17] that the drag coefficient CD is depended on Reynolds number Rh, vegetation density, ฮป measured as per unit meter; submergence ratio S, and length of vegetation reach, Lv. Also, theoretically, CD depends on the channel slope, So. Thus, CD could be expressed as follows: ๐ถ๐ท = ๐(๐ โ , ๐, ๐ฟ๐ฃ , ๐, ๐๐ )

(1)

It follows that (1) can be expressed as (2) based on dimensional homogeneity: ๐ถ๐ท = ๐ โฒ (R โ , ๐๐ฟ๐ฃ , ๐โ๐๐ )

(2)

The value of Rh , can be determined based on the vegetation height, hv [11] as expressed by (3): ๐ โ =

๐โ๐ฃ ๐

(3)

Also, the submergence ratio, S, can be expressed as follows [11]: ๐=

โ โ๐ฃ

(4)

And the vegetation density, ฮป, based on the idea of Xia and Nehal (2013):

e-ISSN: 2289-8131 Vol. 10 No. 1-12

99

Journal of Telecommunication, Electronic and Computer Engineering

๐=

๐ด๐ฃ ๐๐ค

(5)

where, hv, = vegetation height, h is the water depth, Av = Area of vegetation, Av = B*Lv, B = width of the channel and Vw = volume of water, Vw= Av*h. However, CD is determined depending on whether the vegetation is emergent or submerged condition. For the case of emergent vegetation (6) is used, while for the case when the vegetation was submerged (7) was applied: ๐ถ๐ท =

2. ๐. ๐๐ ๐2๐

height in the swale varied from 25 to 40 mm. Axonopus Compressus commonly known as Cow grass was used within the channel bed of the swale Type B, where the grass may be submerged or unsubmerged depending on the flow depth and the grass height. This grass was chosen been commonly available in Malaysia which is currently adopted in the ecological swale of USM for the management of runoff. The technical details of swale components have been discussed by Ghani et al. [20]. Also, using the techniques of surveying the average cross-sectional area of the grassed swale was determined as presented in Figure 2.

(6)

๐ป 2. ๐. ๐๐ ๐ถ๐ท = ( ) 2 โ๐ฃ ๐ ๐

(7)

where, U is the mean velocity of flow (m/s), g = gravity constant (m2/s), H = depth of water above the vegetation. III. DEVELOPING THE ARTIFICIAL NEURAL NETWORK ANN is a powerful mathematical modelling tool that has the ability to process complex input-output relationships, similar to the human brain [17]. This means ANNs are based on the concepts of biological nervous system [18]. They are mostly applied to predict or forecasting the value of an output (depended variable) based on known values of independent variables in an input layer, particularly where the relationships between these variables are not simple linear. Levenberg - Marquardt (LM) back propagation algorithm was used in training the network. This is because LM is an effective training algorithm for training smaller networks [19]. The algorithm uses Newton method that approximate the network error using second order relationship. To execute the process, program algorithm of the LM was developed using Matlab version 2015a, where the values of CD, was set as the target, while Rh, ฮป.Lv and S/So were set as the input as expressed by (2). The ANN model for predicting the value of CD, the optimum number of hidden neuron was selected to be 11 for best performance. 70% of the experimental data set was randomly selected for training the ANN model network. The remaining 30% of the data set were used for model validation and testing. The performances of the regression equation and ANN were evaluated using the statistical parameters like coefficient of determination (R2) and mean square error (MSE). IV. EXPERIMENTAL PROCEDURE A. Field Data Acquisition Figure 1 shows the study area which is located at Universiti Sains Malaysia (USM) Engineering Campus, Nibong Tebal, Penang, Malaysia. In USM three basic types of swale can be distinguished as Type A, Type B and Type C swales with single, double and triple subsurface modules respectively [20]. However, hydraulic and hydrologic data were obtained in grassed swale Type B in this study. All hydraulic data were obtained using an installed automatic flow meter (American Sigma InSight 4.200), this flow meter recorded the flow level (depth), velocity and discharge at every 15 minutes interval. The swale has a bed slope of 1 in 500. The average grass

100

Figure 1: Flow through the Grassed Swale Type B

Figure 2: Cross Section of Swale Type B

B. Laboratory Experiments Experiments were conducted in a concrete channel of working dimension 12 m x 1.5 m x 1 m, at the physical modelling laboratory of River Engineering and Urban Drainage Research Center (REDAC), Universiti Sains Malaysia. The overall length of the channel is 16 m and comprises inlet sump, test flume and the outlet sump. Figure 3 shows test channel with Cow grass planted over a length of 10 m, under a bed slope of 1 in 1000. A grass height of 50 mm was maintained throughout the experiments. Three (3) different flow depths of y = 0.15 m, y = 0.20 m and y = 0.40 m, were used to create flow over the grass under submerged condition. Flow velocity was measured using Acoustic Doppler Velocimeter (ADV). Velocity distributions were determined at five (5) different vertical points, measured along 7 โ cross sections, starting from the inlet 3.0 m, 4.5 m, 5.5 m, 6.0 m, 6.5 m, 8.5 m and up to 11.5 m, respectively. At each vertical point, 8 different depths were measured in fractions of the flow depths, that is, 0.2y, 0.25y, 0.3y, 0.4y, 0.5y, 0.6y, 0.7y and 0.8y, respectively. Hence, average vertical velocities were calculated and used for developing the ANN model.

e-ISSN: 2289-8131 Vol. 10 No. 1-12

Drag Coefficient, CD

Artificial Neural Network Applications for Predicting Drag Coefficient in Flexible Vegetated Channels

250 200 150 100 50

Rยฒ = 0.0634

0 0.00

0.20

0.40

0.60

0.80

1.00

Reynolds number, Rh (x 10^4) Figure 6: Variation of Drag Coefficient with Reynolds Number for Grassed Swale

250 200

V. RESULTS AND DISCUSSION Figure 4 and 5 show the respective readings obtained in grassed swale Type B for the flow depth, and velocity variations due to different rainfall events in September 2009. The readings were taken continuously at 15 minutes intervals using the automatic flow meter.

Drag Coefficient, CD

Figure 3. Laboratory test channel

150

Rยฒ = 0.8912

100 50 0 0.0

5.0

10.0

15.0

20.0

25.0

Submergence Ratio, S

Figure 7: Variation of Drag Coefficient with Submergence Ratio for Grassed Swale

Figure 4: Variation of Flow Depth with Rainfall Events for September 2009

While Table 1 shows the hydraulic flow parameters obtained for the laboratory experiments. Using this table, Figure 8 was plotted to illustrate the respective variations of CD with Reynolds number, Rh. From this figure, it shows that the drag CD with Rh as the correlation coefficient R2 is more than 80%. These results obtained under laboratory conditions are approximately similar to the earlier results presented for the grassed swale selected for field study that combines both the submerged and unsubmerged vegetation. However, the laboratory experiment was mainly focused only on submerged vegetation. Table 1 Summary of laboratory flow parameters

Figure 5: Variation of Flow Velocity with Rainfall Events for September 2009

Using the above data in Figures 4 to 5, Figures 6 and 7, were produced to indicate the relationship between the CD with Rh and S. From this figure, it shows that the drag CD, has a fair correlation coefficient because it combines both submerged and unsubmerged vegetation. However, the correlation coefficient for CD versus S was quite high, indicating strong relationship.

e-ISSN: 2289-8131 Vol. 10 No. 1-12

101

Drag Coefficient, CD

Journal of Telecommunication, Electronic and Computer Engineering

REFERENCES

8 6

[1]

4 2

[2]

Rยฒ = 0.898

0 0

2

4

6

8

Reynolds number, Rh (x 10^4)

Figure 8: Variation of Drag Coefficient with Reynolds Number for Laboratory Experiment

VI. ANN MODELLING OF VEGETATIVE ROUGHNESS Table 2 illustrates the summary of statistical analysis for the prediction of the ANN model with R2 value close to unity. It follows that ANN model performed very good in predicting the drag coefficient of both submerged and emergent vegetation respectively using published data of Cantalice et al. [21], to serve as verification of the developed ANN model. Table 2 Summary of statistical analysis on for Drag Coefficient Prediction by ANN

[3]

[4]

[5] [6]

[7]

[8]

[9]

[10]

[11]

[12] MSE Mean Square Error; R2 Coefficient of Determination

VII. CONCLUSION

[13]

The results obtained from this study show that the drag coefficient depends on the Reynolds number, vegetation density and submergence ratio. The drag coefficient, CD, generally decreases with increase in vegetation density and Reynolds number, whereas it increases with increase in the grass submergence. The ANN model developed shows excellent performances when applied for verification, irrespective of the grass submergence with the flow depth.

[14]

[15]

[16]

[17]

ACKNOWLEDGEMENT The authors would like to acknowledge the financial assistance from Ministry of Education under HiCOEโs niche area Sustainable Urban Stormwater Management (Grant No.311.PREDAC.4403901), and to as well the grant from Universiti Sains Malaysia (Grant No. 304.PREDAC.60312035 / 1001.PREDAC.814225). The first author would also like to acknowledge Universiti Teknologi PETRONAS for the fellowship scheme to undertake the postgraduate study.

[18]

[19]

[20]

[21]

102

F.-C. Wu, H. W. Shen, and Y.-J. Chou, "Variation of roughness coefficients for unsubmerged and submerged vegetation," Journal of Hydraulic Engineering, 1999. H. Nepf, "Drag, turbulence, and diffusion in flow through emergent vegetation," Water resources research, vol. 35, pp. 479-489, 1999. D. M. Temple, "Velocity distribution coefficients for grass-lined channels," Journal of Hydraulic Engineering, vol. 112, pp. 193-205, 1986. M. M. Muhammad, K. W. Yusuf, M. R. Mustafa, and A. Ab. Ghani, "Vegetated Open Channel Flow for Urban Stormwater Management: A Review," in 36 th IAHR World Congress, The Hague, the Netherlands, 2015. V. T. Chow, Open channel hydraulics. New York: McGraw-Hill, 1959. H. Gotoh, T. Tsujimoto, and H. Nakagawa, "Numerical simulation of transition from saltation to sheet flow in sediment transport," Proc. of Coastal Engrg, vol. 40, pp. 326-330, 1993. Y. Shimizu and T. Tsujimoto, "Numerical analysis of turbulent openchannel flow over a vegetation layer using a k-ฮต turbulence model," Journal of hydroscience and hydraulic engineering, vol. 11, pp. 57-67, 1994. S. E. Darby, "Effect of riparian vegetation on flow resistance and flood potential," Journal of Hydraulic Engineering, vol. 125, pp. 443-454, 1999. F. Lopez and M. H. Garcia, "Risk of sediment erosion and suspension in turbulent flows," Journal of Hydraulic Engineering, vol. 127, pp. 231-235, 2001. W. Wu and S. S. Wang, "A DepthโAveraged TwoโDimensional Numerical Model of Flow and Sediment Transport in Open Channels with Vegetation," Riparian vegetation and fluvial geomorphology, pp. 253-265, 2004. H. Tang, Z. Tian, J. Yan, and S. Yuan, "Determining drag coefficients and their application in modelling of turbulent flow with submerged vegetation," Advances in Water Resources, vol. 69, pp. 134-145, 2014. U. C. Kothyari, K. Hayashi, and H. Hashimoto, "Drag coefficient of unsubmerged rigid vegetation stems in open channel flows," Journal of Hydraulic Research, vol. 47, pp. 691-699, 2009. H. T. Nguyen, "Characteristics of Hydraulic Resistance and Velocity Profile in Vegetated Open-Channel Flows," PhD, School of Civil and Environmental Engineering, Nanyang Technological University, 2012. A. Jha and B. Kumar, "Particle Swarm Optimization Neural Network for Flow Prediction in Vegetative Channel," Journal of Intelligent Systems, vol. 22, 2013. M. Karbasi and H. M. Azamathulla, "Prediction of scour caused by 2D horizontal jets using soft computing techniques," Ain Shams Engineering Journal, 2016. R. Mohammadpour, S. Shaharuddin, N. A. Zakaria, A. A. Ghani, M. Vakili, and N. W. Chan, "Prediction of water quality index in free surface constructed wetlands," Environmental Earth Sciences, vol. 75, 2016. F. Salmasi, G. Yฤฑldฤฑrฤฑm, A. Masoodi, and P. Parsamehr, "Predicting discharge coefficient of compound broad-crested weir by using genetic programming (GP) and artificial neural network (ANN) techniques," Arabian Journal of Geosciences, vol. 6, pp. 2709-2717, 2013. F. Ozkan and T. Kaya, "Using intelligent methods to predict airdemand ratio in venturi weirs," Advances in Engineering Software, vol. 41, pp. 1073-1079, 2010. M. T. Hagan and M. B. Menhaj, "Training feedforward networks with the Marquardt algorithm," Neural Networks, IEEE Transactions on, vol. 5, pp. 989-993, 1994. A. Ab. Ghani, N. A. Zakaria, R. Abdullah, M. F. Yusof, L. Mohd Sidek, A. H. Kassim "Bio-Ecological Drainage System (BIOECODS): Concept, Design and Construction," presented at the lnternariond Conference on HydroScience and Engineering, Brisbane, Australia, 2004. J. R. B. Cantalice, R. O. Melo, Y. J. A. B. Silva, M. Cunha Filho, A. M. Araรบjo, L. P. Vieira "Hydraulic roughness due to submerged, emergent and flexible natural vegetation in a semiarid alluvial channel," Journal of Arid Environments, vol. 114, pp. 1-7, 2015.

e-ISSN: 2289-8131 Vol. 10 No. 1-12

Department of Civil and Environmental Engineering, Universiti Teknologi Petronas, Seri Iskandar 32610, Perak, Malaysia. 2 Department of Water Resources and Environmental Engineering, Ahmadu Bello University, Zaria, Nigeria. 3 River Engineering and Urban Drainage Research Center, Universiti Sains Malaysia, Pulau Pinang, Malaysia. [email protected]

AbstractโPreviously numerous equations were developed using conventional methods to estimate vegetal drag coefficient by treating submerged and emergent vegetation independently, there is need to derive a generalized relationship that can be applied irrespective of the vegetation submergence with respect to flow depth. In this regard, the present study uses artificial neural network (ANN) as an advanced tool for prediction of drag coefficient in flexible vegetated channels. The training and testing patterns of the proposed ANN model were based on experimental results from the field and laboratory studies that combined both the submerged and emergent grass. A functional relation based on flow parameters and vegetation properties was derived through the use of dimensional analysis. The ANN model developed herein showed significantly better results in several model performance criteria when applied for verification. Index TermsโArtificial Neural Network; Dimensional Analysis; Drag Coefficient; Flexible Vegetated Channels.

I. INTRODUCTION The flow resistance in open channel are usually derived from the viscous and drag force over the wetted perimeter [1]. This drag is categorized into three comprising โ soil grain roughness, form roughness, and vegetative roughness. The vegetation drag is the most significant compared to others, as it has the utmost flow resistance that eventually decreases the average flow in vegetated channels [2, 3]. Thus, this will lead to rise in flow depth and residence time of sediments in the channel due to drag of vegetation [4]. Drag coefficient is an important parameter, this is because the drag enhanced the tendency for trapping, deposition, and stabilization of sediments, and its increases flow resistance and decreases the bed shear stress, which results in reduction of bed load transport capacity and erosion rate. Several procedures were proposed by researchers to model flow - vegetation interactions and sediment transport in open-channel [5-10]. Also, vegetation-induced drag force has been systematically studied by Kothyari, Tang, Wu and others [1, 11, 12]. However, with these investigations, it was deduced that there are discrepancies for the derivations of vegetation drag which necessitate for a general formula in evaluation of drag coefficient [13]. In addition, conventional regression has been used to develop mathematical equations in estimating hydraulic variables, such as drag coefficient, in vegetated channels. This approach of regression was found to either

over or under-estimate the hydraulic variables. Thus, for a better and accurate predictions, soft computing techniques such as artificial neural network (ANN) are nowadays employed in water resources engineering to estimate hydraulic and hydrologic variables [14-16]. Therefore, the main objective of this paper is to develop ANN model that will compute the vegetal drag coefficient in natural grassed channels irrespective of the grass submergence in relation to flow depth. Thus, field and laboratory experiments were conducted to collect hydraulic data to establish the ANN model. II. FUNCTIONAL RELATIONSHIP A functional relationship was formulated based on the criteria of Wu et al. [1], Tang et al. [11], Kothyari et al. [12] and Wilson et al. [17] that the drag coefficient CD is depended on Reynolds number Rh, vegetation density, ฮป measured as per unit meter; submergence ratio S, and length of vegetation reach, Lv. Also, theoretically, CD depends on the channel slope, So. Thus, CD could be expressed as follows: ๐ถ๐ท = ๐(๐ โ , ๐, ๐ฟ๐ฃ , ๐, ๐๐ )

(1)

It follows that (1) can be expressed as (2) based on dimensional homogeneity: ๐ถ๐ท = ๐ โฒ (R โ , ๐๐ฟ๐ฃ , ๐โ๐๐ )

(2)

The value of Rh , can be determined based on the vegetation height, hv [11] as expressed by (3): ๐ โ =

๐โ๐ฃ ๐

(3)

Also, the submergence ratio, S, can be expressed as follows [11]: ๐=

โ โ๐ฃ

(4)

And the vegetation density, ฮป, based on the idea of Xia and Nehal (2013):

e-ISSN: 2289-8131 Vol. 10 No. 1-12

99

Journal of Telecommunication, Electronic and Computer Engineering

๐=

๐ด๐ฃ ๐๐ค

(5)

where, hv, = vegetation height, h is the water depth, Av = Area of vegetation, Av = B*Lv, B = width of the channel and Vw = volume of water, Vw= Av*h. However, CD is determined depending on whether the vegetation is emergent or submerged condition. For the case of emergent vegetation (6) is used, while for the case when the vegetation was submerged (7) was applied: ๐ถ๐ท =

2. ๐. ๐๐ ๐2๐

height in the swale varied from 25 to 40 mm. Axonopus Compressus commonly known as Cow grass was used within the channel bed of the swale Type B, where the grass may be submerged or unsubmerged depending on the flow depth and the grass height. This grass was chosen been commonly available in Malaysia which is currently adopted in the ecological swale of USM for the management of runoff. The technical details of swale components have been discussed by Ghani et al. [20]. Also, using the techniques of surveying the average cross-sectional area of the grassed swale was determined as presented in Figure 2.

(6)

๐ป 2. ๐. ๐๐ ๐ถ๐ท = ( ) 2 โ๐ฃ ๐ ๐

(7)

where, U is the mean velocity of flow (m/s), g = gravity constant (m2/s), H = depth of water above the vegetation. III. DEVELOPING THE ARTIFICIAL NEURAL NETWORK ANN is a powerful mathematical modelling tool that has the ability to process complex input-output relationships, similar to the human brain [17]. This means ANNs are based on the concepts of biological nervous system [18]. They are mostly applied to predict or forecasting the value of an output (depended variable) based on known values of independent variables in an input layer, particularly where the relationships between these variables are not simple linear. Levenberg - Marquardt (LM) back propagation algorithm was used in training the network. This is because LM is an effective training algorithm for training smaller networks [19]. The algorithm uses Newton method that approximate the network error using second order relationship. To execute the process, program algorithm of the LM was developed using Matlab version 2015a, where the values of CD, was set as the target, while Rh, ฮป.Lv and S/So were set as the input as expressed by (2). The ANN model for predicting the value of CD, the optimum number of hidden neuron was selected to be 11 for best performance. 70% of the experimental data set was randomly selected for training the ANN model network. The remaining 30% of the data set were used for model validation and testing. The performances of the regression equation and ANN were evaluated using the statistical parameters like coefficient of determination (R2) and mean square error (MSE). IV. EXPERIMENTAL PROCEDURE A. Field Data Acquisition Figure 1 shows the study area which is located at Universiti Sains Malaysia (USM) Engineering Campus, Nibong Tebal, Penang, Malaysia. In USM three basic types of swale can be distinguished as Type A, Type B and Type C swales with single, double and triple subsurface modules respectively [20]. However, hydraulic and hydrologic data were obtained in grassed swale Type B in this study. All hydraulic data were obtained using an installed automatic flow meter (American Sigma InSight 4.200), this flow meter recorded the flow level (depth), velocity and discharge at every 15 minutes interval. The swale has a bed slope of 1 in 500. The average grass

100

Figure 1: Flow through the Grassed Swale Type B

Figure 2: Cross Section of Swale Type B

B. Laboratory Experiments Experiments were conducted in a concrete channel of working dimension 12 m x 1.5 m x 1 m, at the physical modelling laboratory of River Engineering and Urban Drainage Research Center (REDAC), Universiti Sains Malaysia. The overall length of the channel is 16 m and comprises inlet sump, test flume and the outlet sump. Figure 3 shows test channel with Cow grass planted over a length of 10 m, under a bed slope of 1 in 1000. A grass height of 50 mm was maintained throughout the experiments. Three (3) different flow depths of y = 0.15 m, y = 0.20 m and y = 0.40 m, were used to create flow over the grass under submerged condition. Flow velocity was measured using Acoustic Doppler Velocimeter (ADV). Velocity distributions were determined at five (5) different vertical points, measured along 7 โ cross sections, starting from the inlet 3.0 m, 4.5 m, 5.5 m, 6.0 m, 6.5 m, 8.5 m and up to 11.5 m, respectively. At each vertical point, 8 different depths were measured in fractions of the flow depths, that is, 0.2y, 0.25y, 0.3y, 0.4y, 0.5y, 0.6y, 0.7y and 0.8y, respectively. Hence, average vertical velocities were calculated and used for developing the ANN model.

e-ISSN: 2289-8131 Vol. 10 No. 1-12

Drag Coefficient, CD

Artificial Neural Network Applications for Predicting Drag Coefficient in Flexible Vegetated Channels

250 200 150 100 50

Rยฒ = 0.0634

0 0.00

0.20

0.40

0.60

0.80

1.00

Reynolds number, Rh (x 10^4) Figure 6: Variation of Drag Coefficient with Reynolds Number for Grassed Swale

250 200

V. RESULTS AND DISCUSSION Figure 4 and 5 show the respective readings obtained in grassed swale Type B for the flow depth, and velocity variations due to different rainfall events in September 2009. The readings were taken continuously at 15 minutes intervals using the automatic flow meter.

Drag Coefficient, CD

Figure 3. Laboratory test channel

150

Rยฒ = 0.8912

100 50 0 0.0

5.0

10.0

15.0

20.0

25.0

Submergence Ratio, S

Figure 7: Variation of Drag Coefficient with Submergence Ratio for Grassed Swale

Figure 4: Variation of Flow Depth with Rainfall Events for September 2009

While Table 1 shows the hydraulic flow parameters obtained for the laboratory experiments. Using this table, Figure 8 was plotted to illustrate the respective variations of CD with Reynolds number, Rh. From this figure, it shows that the drag CD with Rh as the correlation coefficient R2 is more than 80%. These results obtained under laboratory conditions are approximately similar to the earlier results presented for the grassed swale selected for field study that combines both the submerged and unsubmerged vegetation. However, the laboratory experiment was mainly focused only on submerged vegetation. Table 1 Summary of laboratory flow parameters

Figure 5: Variation of Flow Velocity with Rainfall Events for September 2009

Using the above data in Figures 4 to 5, Figures 6 and 7, were produced to indicate the relationship between the CD with Rh and S. From this figure, it shows that the drag CD, has a fair correlation coefficient because it combines both submerged and unsubmerged vegetation. However, the correlation coefficient for CD versus S was quite high, indicating strong relationship.

e-ISSN: 2289-8131 Vol. 10 No. 1-12

101

Drag Coefficient, CD

Journal of Telecommunication, Electronic and Computer Engineering

REFERENCES

8 6

[1]

4 2

[2]

Rยฒ = 0.898

0 0

2

4

6

8

Reynolds number, Rh (x 10^4)

Figure 8: Variation of Drag Coefficient with Reynolds Number for Laboratory Experiment

VI. ANN MODELLING OF VEGETATIVE ROUGHNESS Table 2 illustrates the summary of statistical analysis for the prediction of the ANN model with R2 value close to unity. It follows that ANN model performed very good in predicting the drag coefficient of both submerged and emergent vegetation respectively using published data of Cantalice et al. [21], to serve as verification of the developed ANN model. Table 2 Summary of statistical analysis on for Drag Coefficient Prediction by ANN

[3]

[4]

[5] [6]

[7]

[8]

[9]

[10]

[11]

[12] MSE Mean Square Error; R2 Coefficient of Determination

VII. CONCLUSION

[13]

The results obtained from this study show that the drag coefficient depends on the Reynolds number, vegetation density and submergence ratio. The drag coefficient, CD, generally decreases with increase in vegetation density and Reynolds number, whereas it increases with increase in the grass submergence. The ANN model developed shows excellent performances when applied for verification, irrespective of the grass submergence with the flow depth.

[14]

[15]

[16]

[17]

ACKNOWLEDGEMENT The authors would like to acknowledge the financial assistance from Ministry of Education under HiCOEโs niche area Sustainable Urban Stormwater Management (Grant No.311.PREDAC.4403901), and to as well the grant from Universiti Sains Malaysia (Grant No. 304.PREDAC.60312035 / 1001.PREDAC.814225). The first author would also like to acknowledge Universiti Teknologi PETRONAS for the fellowship scheme to undertake the postgraduate study.

[18]

[19]

[20]

[21]

102

F.-C. Wu, H. W. Shen, and Y.-J. Chou, "Variation of roughness coefficients for unsubmerged and submerged vegetation," Journal of Hydraulic Engineering, 1999. H. Nepf, "Drag, turbulence, and diffusion in flow through emergent vegetation," Water resources research, vol. 35, pp. 479-489, 1999. D. M. Temple, "Velocity distribution coefficients for grass-lined channels," Journal of Hydraulic Engineering, vol. 112, pp. 193-205, 1986. M. M. Muhammad, K. W. Yusuf, M. R. Mustafa, and A. Ab. Ghani, "Vegetated Open Channel Flow for Urban Stormwater Management: A Review," in 36 th IAHR World Congress, The Hague, the Netherlands, 2015. V. T. Chow, Open channel hydraulics. New York: McGraw-Hill, 1959. H. Gotoh, T. Tsujimoto, and H. Nakagawa, "Numerical simulation of transition from saltation to sheet flow in sediment transport," Proc. of Coastal Engrg, vol. 40, pp. 326-330, 1993. Y. Shimizu and T. Tsujimoto, "Numerical analysis of turbulent openchannel flow over a vegetation layer using a k-ฮต turbulence model," Journal of hydroscience and hydraulic engineering, vol. 11, pp. 57-67, 1994. S. E. Darby, "Effect of riparian vegetation on flow resistance and flood potential," Journal of Hydraulic Engineering, vol. 125, pp. 443-454, 1999. F. Lopez and M. H. Garcia, "Risk of sediment erosion and suspension in turbulent flows," Journal of Hydraulic Engineering, vol. 127, pp. 231-235, 2001. W. Wu and S. S. Wang, "A DepthโAveraged TwoโDimensional Numerical Model of Flow and Sediment Transport in Open Channels with Vegetation," Riparian vegetation and fluvial geomorphology, pp. 253-265, 2004. H. Tang, Z. Tian, J. Yan, and S. Yuan, "Determining drag coefficients and their application in modelling of turbulent flow with submerged vegetation," Advances in Water Resources, vol. 69, pp. 134-145, 2014. U. C. Kothyari, K. Hayashi, and H. Hashimoto, "Drag coefficient of unsubmerged rigid vegetation stems in open channel flows," Journal of Hydraulic Research, vol. 47, pp. 691-699, 2009. H. T. Nguyen, "Characteristics of Hydraulic Resistance and Velocity Profile in Vegetated Open-Channel Flows," PhD, School of Civil and Environmental Engineering, Nanyang Technological University, 2012. A. Jha and B. Kumar, "Particle Swarm Optimization Neural Network for Flow Prediction in Vegetative Channel," Journal of Intelligent Systems, vol. 22, 2013. M. Karbasi and H. M. Azamathulla, "Prediction of scour caused by 2D horizontal jets using soft computing techniques," Ain Shams Engineering Journal, 2016. R. Mohammadpour, S. Shaharuddin, N. A. Zakaria, A. A. Ghani, M. Vakili, and N. W. Chan, "Prediction of water quality index in free surface constructed wetlands," Environmental Earth Sciences, vol. 75, 2016. F. Salmasi, G. Yฤฑldฤฑrฤฑm, A. Masoodi, and P. Parsamehr, "Predicting discharge coefficient of compound broad-crested weir by using genetic programming (GP) and artificial neural network (ANN) techniques," Arabian Journal of Geosciences, vol. 6, pp. 2709-2717, 2013. F. Ozkan and T. Kaya, "Using intelligent methods to predict airdemand ratio in venturi weirs," Advances in Engineering Software, vol. 41, pp. 1073-1079, 2010. M. T. Hagan and M. B. Menhaj, "Training feedforward networks with the Marquardt algorithm," Neural Networks, IEEE Transactions on, vol. 5, pp. 989-993, 1994. A. Ab. Ghani, N. A. Zakaria, R. Abdullah, M. F. Yusof, L. Mohd Sidek, A. H. Kassim "Bio-Ecological Drainage System (BIOECODS): Concept, Design and Construction," presented at the lnternariond Conference on HydroScience and Engineering, Brisbane, Australia, 2004. J. R. B. Cantalice, R. O. Melo, Y. J. A. B. Silva, M. Cunha Filho, A. M. Araรบjo, L. P. Vieira "Hydraulic roughness due to submerged, emergent and flexible natural vegetation in a semiarid alluvial channel," Journal of Arid Environments, vol. 114, pp. 1-7, 2015.

e-ISSN: 2289-8131 Vol. 10 No. 1-12