Prediction of Ultimate Soil Bearing Capacity For Shallow Strip Foundation On Sandy Soils Using (ANN) Technique

Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/325261082

Prediction of Ultimate Soil Bearing Capacity for Shallow Strip Foundation on


Sandy Soils Using (ANN) Technique

Article · May 2018

CITATIONS READS

0 577

2 authors, including:

Abdul Kareem E. Zainal


University of Baghdad
25 PUBLICATIONS   29 CITATIONS   

SEE PROFILE

All content following this page was uploaded by Abdul Kareem E. Zainal on 18 June 2018.

The user has requested enhancement of the downloaded file.


Journal of Thi-Qar University number3 Vol.7 June/2012

Prediction of Ultimate Soil Bearing Capacity


for Shallow Strip Foundation on Sandy Soils
Using (ANN) Technique

Dr. Zainal, Abdul Kareem Esmat, Dr. Al_Saidi, A'amal


Department of Civil Engineering, College of Engineering, University of
Baghdad

ABSTRACT
Bearing capacity of soil is an important factor in designing shallow foundations. It
is directly related to foundation dimensions and consequently its performance.
The calculations for obtaining the bearing capacity of a soil needs many varying
parameters, for example soil type, depth of foundation, unit weight of soil, etc. which
makes these calculation very variable–parameter dependent.
This paper presents the results of comparison between the theoretical equation stated by
Terzaghi and the Artificial Neural Networks (ANN) technique to estimate the ultimate
bearing capacity of the strip shallow footing on sandy soils. The results show a very good
agreement between the theoretical solution and the ANN technique.
Results revealed that using ANN gave a very high correlation factor associated with the
results obtained from Terzagih’s equation, besides little computation time needed
compared with computation time needed when applying Terzagih’s equation.

‫اﻟﺧﻼﺻﺔ‬
‫ﻗﺎﺑﻠﯾﺔ ﺗﺣﻣل اﻟﺗرﺑﺔ ﻟﻸﺣﻣﺎل ﻣـن اﻟﻌواﻣـل اﻟﻣﻬﻣـﺔ اﻟﺗـﻲ ﻧﺣﺗﺎﺟﻬـﺎ ﻓـﻲ ﺗﺻـﻣﯾم اﻷﺳـس اﻟﺿـﺣﻠﺔ ﻟﻣـﺎ ﻟﻬـﺎ ﻣـن ﺗـﺄﺛﯾر ﻋﻠـﻰ أﺑﻌـﺎد‬
.‫اﻟﺗﺻﻣﯾم وﺑﺎﻟﺗﺎﻟﻲ ﻋﻠﻰ أداﺋﻪ ﺑﺷﻛل ﻣﺑﺎﺷر‬
‫ وﺣـدة اﻟـوزن‬،‫ ﻋﻣـق اﻷﺳـﺎس‬،‫ان ﻋﻣﻠﯾﺔ اﺣﺗﺳﺎب ﺗﺣﻣل اﻟﺗرﺑﺔ ﺗﺣﺗﺎج إﻟﻰ ﻋدة ﻋواﻣل وﺗﺷﻣل ﻣﺗﻐﯾرات ﻛﺛﯾرة ﻣﺛل ﻧوع اﻟﺗرﺑﺔ‬
.‫ ﻣﻣﺎ ﯾﺟﻌل اﺣﺗﺳﺎب ﺗﺣﻣل اﻟﺗرﺑﺔ ﻣن اﻟﻣﻘﺎدﯾر اﻟﻣﺗﻐﯾرة ﺑﺷﻛل ﻛﺑﯾر ﺗﺑﻌﺎ ﻟﻠﻌواﻣل اﻟﻣذﻛورة‬.‫اﻟﺦ‬... ،‫ﻟﻠﺗرﺑﺔ‬
‫ﻟﻬذا ﺗم ﺗﺻﻣﯾم ﻣودﯾل ﺑﺎﺳﺗﺧدام اﻟﺷﺑﻛﺎت اﻟﻌﺻـﺑﯾﺔ ﻟﺣﺳـﺎب ﻗﺎﺑﻠﯾـﺔ ﺗﺣﻣـل اﻟﺗرﺑـﺔ ﯾﻐﻧـﻲ ﻋـن اﺟـراء اﻟﺣﺳـﺎﺑﺎت اﻟﻣﻌﻘـدة وﺗﻣـت‬
‫اﻟﻣﻘﺎرﻧﺔ ﺑﯾن ﻧﺗﺎﺋﺟﻬﺎ واﻟﻧﺗﺎﺋﺞ اﻟﻣﺳﺗﺣﺻﻠﺔ ﻣن اﺳﺗﺧدام اﻟﻣﻌﺎدﻻت اﻟﻧظرﯾﺔ ﺣﯾث اظﻬرت اﻟﻧﺗﺎﺋﺞ ﺗواﻓق ﻛﺑﯾـر ﺟـدا ﻓﯾﻣـﺎ ﺑﯾﻧﻬـﺎ‬
‫ﯾﺿﺎف اﻟﻰ ذﻟك اﻟﺗوﻓﯾر اﻟﻛﺑﯾـر ﻓـﻲ اﻟوﻗـت اﻟـﻼزم ﻻﺟـراء اﻟﺣﺳـﺎﺑﺎت ﺑﺎﺳـﺗﺧدام طرﯾﻘـﺔ اﻟﺷـﺑﻛﺎت اﻟﻌﺻـﺑﯾﺔ ﻣﻘﺎرﻧـﺔ ﻣـﻊ اﻟطـرق‬
.‫اﻟﺗﻘﻠﯾدﯾﺔ‬

Keyword: Soil Bearing capacity, Artificial Neural Network, shallow


foundation.

47

PDF created with pdfFactory Pro trial version www.pdffactory.com


Journal of Thi-Qar University number3 Vol.7 June/2012

1. INTRODUCTION
The ultimate bearing capacity for a soil qu is defined as the least pressure which
would cause shear failure of the supporting soil immediately below and adjacent to a
foundation.
The ultimate bearing capacity can be determine either experimentally or by calculations
using analytical and / or empirical formulae.
Artificial Neural Network (ANN) technique became a powerful tool that can be used to
solve the civil engineering problems (Jeng, et al., 2003), and a more effective tool for
engineering applications, thus this study was undertaken in order to predict the ultimate
bearing capacity of shallow strip footing over sandy soil by using artificial neural
networks technique.
A set of varying conditions are studied and the results obtained by implementing the
artificial neural network technique are then compared to the results obtained by
implementing Terzaghi's equation, results revealed a very high correlation factor between
answers obtained from implementing the ANN technique and the answers obtained by
implementing Terzagi’s equation.

2. Theory
The ultimate bearing capacity of the soil under shallow strip footing can be
expressed by the following general equation, Terzaghi (1943).See Figure (1).

qu = c Nc + γ D Nq + 0.5 B γ Nγ ………………… (1)

where c = Cohesion of soil.


γ = Unit weight of soil.
D = Footing depth.
B = Footing width.
Nc, Nq, Nγ = bearing capacity factors depending only on (φ)
Nc = (Nq −1) cotφ ……………………………….… (2)
 φ
Nq = e (π tan φ ) tan 2  45 +  ………………………… (3)
 2
Nγ =2 (Nq+1) tanφ ………………………………….(4)
φ = Angle of internal friction of the soil.

Eq.(2) for Nc was originally derived by Prandtl (1921),and Eq.(3) for N q was presented
by Reissner (1924). Caquot and Kerisel (1953) and Vesic (1973) gave the relation for N γ
(Eq.(4)).

48

PDF created with pdfFactory Pro trial version www.pdffactory.com


Journal of Thi-Qar University number3 Vol.7 June/2012

Figure(1)
Failure surface of shallow foundation

3. Implementation of Neural Network


MatLab version R2008a was used in designing and implementation of the ANN
(Demuth, et al., 2008). To find the most appropriate design and learning algorithm, the
method of trial and error was used by choosing different learning algorithms, layers, and
neurons, as follows (Zurada, 1992):
1.Ten different learning algorithms were used presented in Appendix (A) Table A–1.
2.Three different numbers of layers were used, 1 layer, 2 layers, and 3 layers.
3.Three different numbers of neurons were used, 10, 20, and 30 neurons per each layer.
First, Eq. (1) was used to calculate the ultimate bearing capacity for various soil
properties, considering different values for the parameters needed to solve the equation, as
follows:
1. φ varying from 10° to 45°, steps of 2.5° was used.

2. γ varying from 14 kN/m3 to 23 kN/m3 steps of 1.0 kN/m3 was used.

3. B varying from 0.5 m to 1.5 m , steps of 0.1 m was used.

4. D varying from 0.5 m to 1.5m, steps of 0.25 m was used.

5. c = 0 for sandy soil.

Practically, these values could represent and cover the actual range that may be needed in
the analysis and design of real problems. The sum of 8250 cases were taken into
consideration, each case represents a different design alternative and has a unique
ultimate bearing capacity value. The ultimate bearing capacity was calculated for each
case using Eq.1 as mentioned before.
All these cases and their parameters are considered as the input data for a special ANN
designed to memorize each individual case and its calculated bearing capacity so that it
could predict the ultimate bearing capacity later.

49

PDF created with pdfFactory Pro trial version www.pdffactory.com


Journal of Thi-Qar University number3 Vol.7 June/2012

A procedure of trial and error was used to find the most appropriate number of layers,
number of neurons per layer, and the most efficient learning algorithm among ten
learning algorithms implemented in teaching the Neural Networks.
Another set of random data was prepared to verify the reliability and the consistency of
the Neural Network, the data were totally different from the input data and there values
were never shown in the input data.
This procedure was conducted to obtain the most efficient Neural Network which is
considered to have:
1. maximum correlation ratio between the target data and the output data obtained,
2. maximum correlation ratio between verifying data and the output obtained, and
3. minimum time to reach solution.

4. Results and Discussion


Table A-2 represents a sample of the first 100 input data (Appendix-A) , the total
number of data inputs were 2640. The method of trial and error was used to find the most
appropriate Neural Network that can reflect the most suitable design requirements (i.e.
the correct ultimate bearing capacity qu for the required design parameters, φ, D, B, γ, and
c).
Among ten learning algorithms, ten outputs were obtained, each output was obtained after
teaching the Neural Network with the most representative number of neurons, and
number of layers. A correlation factor was calculated for each output to show the
reliability of the network.
Table(1) shows the algorithm name and the highest correlation factor that can be obtained
after applying the learning rule for a variety of neuron numbers and layers.

Table 1 Algorithm name vs. correlation factor


Neuron numbers
No. Algorithm name Correlation Factor
And Number of Layers
1 GDA 0.995057444 10
2 GDX 0.997960736 10 x 10
3 RP 0.999956575 20
4 CGF 0.999120732 10
5 CGP 0.998901795 10 x 10
6 LM 0.999999993 10
7 BFG 0.997986589 10
8 SCG 0.999295558 10
9 CGB 0.996573746 10
10 OSS 0.997773765 10

50

PDF created with pdfFactory Pro trial version www.pdffactory.com


Journal of Thi-Qar University number3 Vol.7 June/2012

As can be seen from Table 1, the most efficient algorithm that gave the highest
correlation factor is no. 6 (LM learning rule) with 10 neurons (i.e. one layer which
consist of 10 Neurons) with a correlation factor of 0.999999993.
Table 2 shows the verifying data that was used to test each algorithm and its
corresponding Neural Network, the input data were chosen so that they were never
taught to the Neural Network before (they were never shown in the input data that was
used for teaching the network in the first step).

Table 2 Verifying Data Used to Test Reliability of Neural Network


No. B γ D φ qu Output
1 0.75 19.1 1.1 18 139.589 139.5673802
2 0.8 22 0.87 29 484.899 484.7750421
3 1.1 15.86 0.57 41 1803.897 1804.095986
4 1.3 18.2 0.97 38 1786.895 1786.784995
5 1.45 22.5 0.76 14 98.61363 98.64410774
6 1.15 15.73 0.81 32 568.5862 568.6354125
7 0.88 16.6 1.49 19 177.5926 177.4755966
8 1.22 21.5 0.55 27 345.8494 345.9459282
9 1.45 15.66 1.3 17 137.223 137.1086517
10 1.55 19.24 0.73 42 3518.381 3518.516495
11 0.22 14.3 0.56 34 300.3472 299.7223238
12 0.38 20.1 1.44 22 253.5961 253.7477964
13 1.11 17.8 0.61 44 3471.177 3471.39989
14 0.93 20.5 0.88 11 62.64117 62.45822203
15 0.67 21 0.59 37 997.4388 997.6196914
16 0.4 13.5 1.45 19.8 136.9498 137.127496
17 0.3 17.6 1.1 33.2 613.1835 612.9913123
18 1.45 15 0.3 9.5 22.8533 22.84814129
19 1 13 0.68 43.4 2235.155 2235.190448
20 1.4 12 0.25 15.6 36.78086 36.3282789

The output was then compared to the calculated values using the same formula (Eq. 1)
and a correlation factor is evaluated the see the most efficient algorithm that gave the
highest correlation factor for the test data. Results are shown in Table 3.

Table 3 Correlation Factor Obtained for Each Learning Algorithm


No. Algorithm Name Correlation Factor
1 GDA 0.997409464
2 GDX 0.998322484
3 RP 0.999934347
4 CGF 0.999094581
5 CGP 0.999188115

51

PDF created with pdfFactory Pro trial version www.pdffactory.com


Journal of Thi-Qar University number3 Vol.7 June/2012

6 LM 0.999999984
7 BFG 0.998235235
8 SCG 0.997448906
9 CGB 0.993152833
10 OSS 0.997531545

As could be seen from Table 3 that the algorithm that gave the best correlation factor
is no. 6 (LM) with a correlation factor of 0.999999984.
Figure 2 shows the performance of the Neural Network reflected by showing the Mean
Squared error (MSE) of value less than 0.01.

Figure (2)
Performance of the trained Neural Network

Where Figure. 3 shows the regression value obtained after training the Neural
Network which shows a value of (1) which means that the output obtained have a very
strong relation to the target values desired.

52

PDF created with pdfFactory Pro trial version www.pdffactory.com


Journal of Thi-Qar University number3 Vol.7 June/2012

Figure (3)
Regression Value of Neural Network
Between Input Data and Target Data

5. Conclusions and recommendations


The calculation of bearing capacity of shallow foundation is a many parameter
dependant process, and it has many pre calculations till we can implement the
Terzaghi’s equation (Eq. 1), these calculations include the bearing capacity factors N q,
Nγ, and Nc. Another alternative is to use the charts which could lead to some
approximations.
Using an Artificial Neural Network can facilitate these calculations to a great extent.
The Neural Network can remember the parameters that were used as an input (B, D, φ,
c, and γ) and the calculated values of the ultimate bearing capacity qu, and this
operation has to be done only once, then the network can be used to predict the bearing
capacity for any input values and give the bearing capacity value as was done here by
using the verifying data.
The advantage of using the Artificial Neural Network comes mainly from saving
calculation time of the parameters and the ultimate bearing capacity, and once the
network was ready, the same network can be used as many times as desired with no
further need for teaching or modifying, besides, the calculation needed when using the
Neural Network are simple compared to the calculations needed to obtain the results in
the original equation.

53

PDF created with pdfFactory Pro trial version www.pdffactory.com


Journal of Thi-Qar University number3 Vol.7 June/2012

6. References
• Caquot, A. and Kerisel, J. 1953. Surle le Terme de Surface dans le Calcul des
Fondations en Milieu Pulverulent. Proceedings of the 3rd International Conference on
Soil Mechanics and Foundation Engineering. Vol. 1, pp. 336-337 (as cited by Cerato,
A.B. and Lutenegger, A.J., 2007 , "Scale Effects of Shallow Foundations Bearing
Capacity on Granular " , Journal of Geotechnical and Geoenvironmental Engineering,
ASCE, Vol. 133, No. 10, pp. 1192-1202).
• Demuth, H., Mark B., and Martin H., 2008, ” Neural Network Toolbox™ 6 User’s
Guide”, The MathWorks, Inc.
• Jeng, D.S., Cha D. H. and Blumenstein M., 2003, “Application of Neural Network in
Civil Engineering Problems“, Proceedings of the International Conference on
Advances in the Internet, Processing, Systems and Interdisciplinary Research (IPSI-
2003).
• Prandtl, L. (1921). Uber Die Eindringungsfestigkeit (Harte) Plastischer Baustoffe Und
Die Festigkeit Von Schneiden. Zeitschrift fur angewandte Mathematik und Mechanik.
Vol. 1, No. 1. pp. 15-20 ( as cited by Lee, J. H. and Salgado, R., 2005, "Estimation of
Bearing Capacity of Circular Footings on Sands Based on Cone Penetration Test " ,
Journal of Geotechnical and Geoenvironmental Engineering, ASCE, Vol. 131, No. 4,
pp. 442-452).
• Reissner, H. (1924). “Zum Erddruckproblem” (Concerning the earth-pressure
problem), Proc. 1st Int. Congress of Applied Mechanics, Delft, pp. 295-311 ( as cited by
Lee, J. H. and Salgado, R., 2005, "Estimation of Bearing Capacity of Circular Footings
on Sands Based on Cone Penetration Test", Journal of Geotechnical and
Geoenvironmental Engineering, ASCE, Vol. 131, No. 4, pp. 442-452).
• Terzaghi,K., 1943,"Theoretical Soil Mechanics",John Wiley and Sons,New York.
• Vesic, A. S., 1973., "Analysis of Ultimate Loads of Shallow Foundation", Journal of
Soil Mechanics and Foundations Division, American Society of Civil Engineers ,Vol.99
, No.SM1,pp.45-73.
• Zurada, J. M., 1992, “Introduction to Artificial Neural Systems“, PWS Publishing
Company.
Appendix A
Table A–1 Training Algorithms Names and Symbols
Symbol Algorithm Name
GDA Backpropagation training with an adaptive learning rate
GDX adaptive learning rate with momentum training
RP Resilient Backpropagation
CGF Fletcher-Powell Conjugate Gradient
CGP Polak-Ribiére Conjugate Gradient
LM Levenberg-Marquardt
BFG BFGS Quasi-Newton
SCG Scaled Conjugate Gradient
CGB Conjugate Gradient with Powell/Beale Restarts
OSS One Step Secant

54

PDF created with pdfFactory Pro trial version www.pdffactory.com


Journal of Thi-Qar University number3 Vol.7 June/2012

Table A-2 Sample of 100 Input Data and Output Data

B γ D φ qu
output
(m) kN/m3 (m) Degree kN/m2
0.5 14 0.5 10 21.5848 21.22723
0.6 14 0.5 10 22.44175 22.131557
0.7 14 0.5 10 23.29871 23.036521
0.8 14 0.5 10 24.15566 23.942056
0.9 14 0.5 10 25.01261 24.848096
1 14 0.5 10 25.86956 25.754572
1.1 14 0.5 10 26.72651 26.661415
1.2 14 0.5 10 27.58346 27.568553
1.3 14 0.5 10 28.44041 28.475913
1.4 14 0.5 10 29.29736 29.38342
1.5 14 0.5 10 30.15431 30.290999
0.5 15 0.5 10 23.12658 22.701227
0.6 15 0.5 10 24.04474 23.669899
0.7 15 0.5 10 24.9629 24.639469
0.8 15 0.5 10 25.88106 25.609868
0.9 15 0.5 10 26.79922 26.581027
1 15 0.5 10 27.71738 27.552874
1.1 15 0.5 10 28.63555 28.525336
1.2 15 0.5 10 29.55371 29.498339
1.3 15 0.5 10 30.47187 30.471805
1.4 15 0.5 10 31.39003 31.445657
1.5 15 0.5 10 32.30819 32.419815
0.5 16 0.5 10 24.66835 24.174608
0.6 16 0.5 10 25.64772 25.208604
0.7 16 0.5 10 26.62709 26.243726
0.8 16 0.5 10 27.60646 27.279905
0.9 16 0.5 10 28.58584 28.317065
1 16 0.5 10 29.56521 29.355131
1.1 16 0.5 10 30.54458 30.394026
1.2 16 0.5 10 31.52395 31.433671
1.3 16 0.5 10 32.50333 32.473987
1.4 16 0.5 10 33.4827 33.514889
1.5 16 0.5 10 34.46207 34.556294
0.5 17 0.5 10 26.21012 25.677576
0.6 17 0.5 10 27.2507 26.777482
0.7 17 0.5 10 28.29129 27.878705
0.8 17 0.5 10 29.33187 28.981169
0.9 17 0.5 10 30.37245 30.084795
1 17 0.5 10 31.41303 31.189503
1.1 17 0.5 10 32.45362 32.295211
1.2 17 0.5 10 33.4942 33.401837
1.3 17 0.5 10 34.53478 34.509294
1.4 17 0.5 10 35.57537 35.617494
1.5 17 0.5 10 36.61595 36.726348
0.5 18 0.5 10 27.75189 27.218992
0.6 18 0.5 10 28.85368 28.384881
0.7 18 0.5 10 29.95548 29.552236
0.8 18 0.5 10 31.05727 30.720975
0.9 18 0.5 10 32.15907 31.891015
1 18 0.5 10 33.26086 33.062271

55

PDF created with pdfFactory Pro trial version www.pdffactory.com


Journal of Thi-Qar University number3 Vol.7 June/2012

1.1 18 0.5 10 34.36265 34.234655


1.2 18 0.5 10 35.46445 35.408079
1.3 18 0.5 10 36.56624 36.582451
1.4 18 0.5 10 37.66804 37.757677
1.5 18 0.5 10 38.76983 38.933664
0.5 19 0.5 10 29.29366 28.788127
0.6 19 0.5 10 30.45667 30.019713
0.7 19 0.5 10 31.61967 31.252875
0.8 19 0.5 10 32.78268 32.487525
0.9 19 0.5 10 33.94568 33.723576
1 19 0.5 10 35.10869 34.960935
1.1 19 0.5 10 36.27169 36.19951
1.2 19 0.5 10 37.4347 37.439205
1.3 19 0.5 10 38.5977 38.679923
1.4 19 0.5 10 39.7607 39.921564
1.5 19 0.5 10 40.92371 41.164027
0.5 20 0.5 10 30.83543 30.363828
0.6 20 0.5 10 32.05965 31.660667
0.7 20 0.5 10 33.28386 32.959154
0.8 20 0.5 10 34.50808 34.259197
0.9 20 0.5 10 35.7323 35.560699
1 20 0.5 10 36.95651 36.863564
1.1 20 0.5 10 38.18073 38.167692
1.2 20 0.5 10 39.40494 39.472979
1.3 20 0.5 10 40.62916 40.779322
1.4 20 0.5 10 41.85337 42.086614
1.5 20 0.5 10 43.07759 43.394744
0.5 21 0.5 10 32.37721 31.922273
0.6 21 0.5 10 33.66263 33.283877
0.7 21 0.5 10 34.94806 34.647164
0.8 21 0.5 10 36.23348 36.012033
0.9 21 0.5 10 37.51891 37.378382
1 21 0.5 10 38.80434 38.746104
1.1 21 0.5 10 40.08976 40.115094
1.2 21 0.5 10 41.37519 41.48524
1.3 21 0.5 10 42.66062 42.856431
1.4 21 0.5 10 43.94604 44.22855
1.5 21 0.5 10 45.23147 45.60148
0.5 22 0.5 10 33.91898 33.440565
0.6 22 0.5 10 35.26561 34.866431
0.7 22 0.5 10 36.61225 36.293971
0.8 22 0.5 10 37.95889 37.723078
0.9 22 0.5 10 39.30553 39.153639
1 22 0.5 10 40.65216 40.585541
1.1 22 0.5 10 41.9988 42.018669
1.2 22 0.5 10 43.34544 43.452902
1.3 22 0.5 10 44.69207 44.88812
1.4 22 0.5 10 46.03871 46.324198
1.5 22 0.5 10 47.38535 47.76101
0.5 23 0.5 10 35.46075 34.897374

56

PDF created with pdfFactory Pro trial version www.pdffactory.com


View publication stats

You might also like