You are on page 1of 11

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

net/publication/305725330

Modeling of California Bearing Ratio using Basic Engineering Properties

Conference Paper · November 2015

CITATIONS READS
2 1,363

5 authors, including:

Samar Taha Sherif El-Badawy


Mansoura University Mansoura University
3 PUBLICATIONS   7 CITATIONS    72 PUBLICATIONS   280 CITATIONS   

SEE PROFILE SEE PROFILE

Alaa Gabr Abdelhalim Moawad Azam


Mansoura University Mansoura University
38 PUBLICATIONS   192 CITATIONS    26 PUBLICATIONS   155 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Characterization of unbound granular materials and subgrade soils in Egypt View project

Unsaturated Unbound Granular Materials from Construction and Demolition Waste View project

All content following this page was uploaded by Alaa Gabr on 30 July 2016.

The user has requested enhancement of the downloaded file.


Modeling of California Bearing Ratio using Basic Engineering Properties
S. Taha1, S. El-Badawy2, A. Gabr3, A. Azam3, and U. Shahdah3
1Graduate Student, Public Works Engineering Department, Faculty of Engineering, Mansoura University
2Associate Professor, Public Works Engineering Department, Faculty of Engineering, Mansoura University, Egypt; PH (+201000183519);

Email: sbadawy@mans.edu.eg
3Assistant Professor, Public Works Engineering Department, Faculty of Engineering, Mansoura University.

ABSTRACT
One of the main inputs for flexible pavement design is the strength of the Unbound Granular Materials (UGMs)
and subgrade soils. In Egypt, California Bearing Ratio (CBR) is used to characterize the strength of UGMs and
subgrade soils for pavement structural design. The CBR can be measured either in the laboratory or in the field. The
CBR test is tedious, laborious and time consuming. This paper presents the applications of Multiple Linear Regression
Analysis (MLR) and Artificial Neural Networks (ANNs) to develop CBR models through soil index properties.
Database containing 120 measured CBR values along with the index properties was gathered from the quality control
reports prepared at the Highway and Airport Engineering Laboratory, Mansoura University. The index properties were
gradation, Atterberg limits, American Association of State Highways and Transportation Officials (AASHTO) soil
classification and compaction parameters according to the modified Proctor test. CBR values ranged between 12% and
95% for unbound granular materials and subgrade soils. A CBR model was proposed as a function of Maximum Dry
Density (MDD), grading modulus (GM) and percentage of retained on sieve No.10 (R#10). The GM is the summation
of the passing percentage of the sieves, No. 40, No. 10 and No. 200 divided by 100. The coefficient of determination
(R2) was found to be 0.92 for MLR and 0.88 for the ANNs.
I. INTRODUCTION

CBR test is an empirical test performed on subgrade soils and UGMs for material quality control and flexible
pavement design all over the world. CBR is a measure of the relative resistance of the unbound granular base/subbase
materials and subgrade soils to uniaxial penetration. The CBR can be measured either in the laboratory or in the field.
This test was developed by the California Division of Highways around 1930s (AASHTO, (2003); ASTM, (1999)). In
Egypt, CBR is the major input parameter used to characterize the strength of the UGMs and subgrade soils for flexible
pavement structural design. The test is tedious, laborious, time consuming and sometimes it yields inaccurate results.
Thus, many attempts have been conducted to estimate the CBR from the physical and compaction properties of the
materials as summarized in TABLE 1.

BLACK (1962) proved that the CBR values can be estimated from plasticity index. AGARWAL & GHANEKAR (1970)
proposed a relationship between the CBR and the Optimum Moisture Content (OMC) and liquid Limit for 48 soils
in India. Two models, one for plastic and the other one for non-plastic soils correlating CBR to routine material
properties were implemented in the AASHTO Mechanistic-Empirical Pavement Design Guide (MEPDG) since its
first release (NCHRP, 2004). Other attempts were found in literature e.g., (RASHMI (2010); SINGH ET AL . (2011); YADAV
ET AL . (2014); TALUKDAR (2014); BLACK (1961); FAIRBROTHER (2011); HARISON ET AL . (1989); SATHAWARA ET AL (2013);
HAKARI AND NADGAUDA (2013); ZUMRAWI (2013); BELLO (2012); ALAWI & RAJAB (2013); TAHA ET AL. (2014)).
In recent years from 2011 to 2014, there were many attempts to use Artificial Neural Networks (ANNs) as an
application to develop models for predicting the CBR from other physical properties of soil (Venkatasubramanian,
C., & Dhinakaran 2011), YILDIRIM & GUNAYDIN (2011) , HARINI HN (2014) and BHATT ET AL . (2014)). Most of the
literature models were developed for the fine grained and plastic soils for subgrade layers. The number of data points
used for the development of these models ranged from only 7 to 611 data points.

1
TABLE 1. SUMMARY OF SOME LITERATURE MODELS CORRELATING CBR AND INDEX MATERIAL PROPERTIES.

No. of Used Parameters


No. Equation R2 Type of soil Country Reference
Materials
Regression analysis: Non-Plastic
7 0.84 Coarse Grained D60 (mm)

(NCHRP
0.358
1 CBR = 28.09 (D60)

2004)
Soils wPI = P#200 * PI
USA
Plastic Where P#200 (in decimal&
75
2 CBR  11 0.67 Fine Grained PI (%)
1  0.728( wPI )
Soils
Regression analysis: -
3 CBR = -0.0892 FG + 7.9851 0.132
4 CBR = 0.1996 G – 0.6365 0.184
5 CBR = 21.101 MDD -30.56 0.624
CBR = -0.3776G-0.4528S-0.4094 FG + FG = (28.18 – 74.73)%

(B HATT ET AL. 2014)


6 0.882 G = (2.75 – 31.14)%
0.3487 OMC + 24.7518 MDD
S = (12.61 – 51.5)%
Artificial Neural Networks: Structure -
L.L = (29-47.16)%
G, S, FG, L.L, P.L, OMC, 124 - PL = (10.67-26.56)% India
7 8-2-1 0.957
MDD. PI = (11.85-30)%
G, S, PI, OMC, MDD. OMC = (10-21)%
8 5-4-1 0.962
MDD = (1.42 -2.03)gm/cc
G, S, FG, OMC, MDD. CBR = (1.55 – 22.4)%
9 5-3-1 0.950
G, S, OMC, MDD.
10 4-4-1 0.979
11 OMC, MDD. 2-3-1 0.887
Artificial Neural Networks: Structure -
12 LL, PI, γd, Wopt, (C+S), S, G. 7-4-1 0.918

(T ASKIRAN 2010)
LL, PI,
13 LL, PI, γd, Wopt, (C+S), S. 6-5-1 0.882 γd, Wopt,
Base and
14 PI, γd, Wopt, (C+S), S. 5-4-1 0.812 (C+S),
151 Subbase Turkey
15 PI, γd, Wopt, (C+S). 4-4-1 0.821 S,
materials
G.
16 PI, γd, (C+M), S. 4-5-1 0.911
17 PI, γd, (C+M). 3-4-1 0.879
18 γd, Wopt, (C+M). 3-5-1 0.617
Regression analysis: P#10 = (54.8 – 88.2) %
19 CBRu = 0.031(LL) + 83.19 P#40 = (32.2-65.6) %
20 CBRu = 0.8 (PL) + 65.31 P#200 = (10.5 – 24.7) %

(BELLO 2012)
21 CBRu = 10.43 (SG) + 56.19 L.L = (28.6 – 45) %
P.L = (17.4 – 25) %
22 CBRu = 8.66 (MMD) + 65.88 lateritic soil Osogbo,
8 N.A P.I= (11.2 – 20.7)
23 CBR = 0.22 (LL) + 28.87 (A-2-6) Nigeria
Gs= (2.66-2.77)
24 CBR = 1.04 (PL) + 13.56 CBR un = (64 -85) %
CBRs = (26-34) %
25 CBR = 9.42 (SG) + 10.91 MDD = (1.98 – 2.21)gm/cc
26 CBR = 50.28 (MMD) -70.22 OMC= (10.81 – 12.52) %
Regression analysis:

(T AHA ET AL.
P200 = (0.08 – 6.6)%

2014)
53 0.785 Non-plastic soil MDD = (1.47– 2.23) gm/cc Egypt
27 CBR= 0.025(P200)4+30.130(MDD)-25.813
CBR = (10.80 – 95)%

Regression analysis:
R AJAB 2013)

LA = (17 -32)%
(ALAWI &

MSE = OMC = (6.4 – 8.1)% Saudi


CBR = -112.4335 – 0.2856 LA – 4.7280 19
9.874%
Subbase
Density =(1.99 -2.3) gm/cc Arabia
28
OMC + 98.4613 Density CBR = (49 -69)%

Regression analysis:
(BREYTENBACH
ET AL . 2010)

Fine grained OMC


48 - India
29 CBR = 2- 16 log (OMC) + 0.07 L.L soil L.L

CBR = California Bearing Ratio (soaked), %; D60 = Diameter at 60% passing from grain size distribution, mm; wPI = Weighted Plasticity index; MDD or
Density or γd =Maximum dry density; OMC or W opt=Optimum Moisture Content; CBRu = unsoaked CBR; LL = Liquid limit, %; PL = Plastic limit, %;
PI=Plasticity index, %; SG = Specific gravity, gm/cc; P200 = passing No. 200 U.S. sieve, %; ( C+S) = percent of passing sieve no.200 (clay+silt), %; S =sand
percent, %; G= gravel percent, %; N.A = Not Available.

2
II. OBJECTIVES

This paper presents the applications of Multiple Linear Regression Analysis (MLR) and ANNs to develop CBR models
through physical and compaction properties of UGMs used for road construction projects in the Delta region in Egypt.
III. DATA COLLECTION
In order to obtain the correlation between soaked CBR values and physical properties for base and subbase soils,
database including different materials for pavement construction projects in the Delta region in Egypt were collected.
The database was collected from technical reports prepared at Mansoura University Highway and Airport Engineering
Laboratory (H&AE-LAB) during the period of 1992 to 2015. The locations of the materials are shown in Figure 1.
The majority of the materials contained in the database were non-plastic cohesionless materials, which were used as
fill materials for road embankments as well as subbase and base courses.

Figure 1: The collected data locations in the delta region

This historical database contains 120 measured CBR values under soaking conditions along with index material
properties and compaction characteristics as follows:

 Particle size distribution and D60,


 Atterberg Limits (liquid limits, LL, plastic limit, PL, and plasticity index, PI),
 Soil classification according to the AASHTO system,
 Modified Proctor Parameters ( MDD, and OMC),
 Soaked CBR values.
The descriptive statistics of the parameters contained in the database are summarized in TABLE 2. The distribution of
the soaked CBR values for the entire database is shown by the histogram in Figure 2. The Figure indicates that the
largest portion of the data had CBR values ranged from 25% to 50%, which was suitable for embankments and subbase
layers according to the Egyptian specifications (ECP, 2008). Furthermore, the other portion of the data had CBR values
in the range of more than 80%, which is suitable for base courses.

TABLE 2. Descriptive Statistics of the Database Parameters


MDD OMC LL P4 P200 D60
Parameter CBR GM
(t/m3) (%) (%) (%) (%) (mm)
Mean 42.56 1.86 10.66 23.51 84.84 2.71 1.22 4.31
Median 31.10 1.81 11.00 23.70 95.35 1.20 1.09 0.90
Mode 30.00 1.79 12.00 24.00 100 0.10 1.08 1.00
Standard Deviation 25.19 0.204 2.39 3.74 24.12 3.18 0.51 7.42
Range 83.00 0.86 10.80 30.60 38.90 19.40 1.62 24.91
Minimum 12.00 1.47 6.00 2.60 31.10 0.00 0.422 0.09
Maximum 95.00 2.33 16.80 33.20 100 19.40 2.037 25
No. of Data Points 120 120 120 104 120 120 120 120

3
50
45
40
35

Frequency
30
25
20
15
10
5
0
0 10 20 30 40 50 60 70 80 90 More
CBR Values

Figure 2. Histogram of California Bearing Ratio values in the database

IV. MODEL DEVELOPMENT

In order to develop a model correlating the soaked CBR with the soil index properties, the literature models
were first tried with the collected database. However, results showed very poor correlation and significant bias.
Thus, the scatter diagrams of the soaked CBR and each of the index soil properties were drawn and presented
in Figures 3 through 8. Figures 3 to 6 show strong linear correlations in terms of coefficient of determination,
R2 between the CBR, MDD, GM, percentage of retained on sieve No. 10, and percentage of passing sieve
No.4, respectively. On the other hand, Figures 7 and 8 show very high scatter and very poor correlation
between the laboratory measured soaked CBR values and P200 (R2 = 0.177) and D60 (R2 = 0.218).

100
100
80 y = 114.7x - 170.57 y = -41.531x + 93.179
R² = 0.8481 80
CBR (%)

R² = 0.6932
CBR (%)

60
60
40
40
20
20
0 0
1 1.5 2 2.5 0 0.5 1 1.5 2 2.5
Maximum Dry Density (MDD), t/m3
Gradimg Modulus (GM), %
Figure 3: Relationship between CBR and Maximum Dry Figure 4: Relationship between CBR and grading
Density modulus
80
120
y = 0.9828x - 21.354 y = -0.9829x + 126.26
60 100
CBR (%)

R² = 0.9004 R² = 0.8785
CBR (%)

80
40
60
20 40
0 20
0 20 40 60 80 100 0
Retained on sieve No.10 (%) 0 50 100
Passing No.4 Sieve (%)
Figure 5: Relationship between CBR and percent of Figure 6: Relationship between CBR and percent of
retained on sieve no.10 passing sieve no.4

4
120 100
100 y = 3.3489x + 34.004 80 y = 0.5871x + 39.73
R² = 0.1772
CBR (%)

CBR (%)
80 60 R² = 0.2177
60
40
40
20
20
0
0
0 0.5 1 1.5 2
0 2 4 6 8 10
D60 (mm)
Passing No.200 sieve (%)
Figure 8: Relationship between CBR and diameter @ 60%
Figure 7 : Relationship between CBR and percent of passing passing (D60)
sieve no.200

V. STATISTICAL ANALYSIS
MLR analysis was performed to establish a relationship between soaked CBR values and other variables. The
MLR analysis was carried out using the Solver function in Microsoft Excel Software. ANNs method was also
applied on the database.

A. Multiple Linear Regression (MLR) Modeling

In order to develop a reliable model with minimal bias and high accuracy, MLR modeling was tried using the
solver tool in Microsoft Excel. The goodness of fit statistics was used to select the best model according to the
conceptual criteria proposed by (Pellinen, 2001) and shown in Table 3. This criteria is based on R2 and the
standard error divided by the standard deviation of the measured CBR values about the mean measured (Se/Sy).
The R2 is simply the square of the correlation coefficient between the measured and predicted CBR. The Se/Sy
is an indicator for the relative improvement in accuracy. Higher R2 values and smaller Se/Sy indicate higher
accuracy (less scatter). These statistical parameters are calculated with respect to the line of equality which is
a line with a slope of unity and intercept of zero. Equations (1) through (4) were used to calculate the goodness
of fit statistical parameters to judge the model accuracy:

TABLE 3. CRITERIA FOR GOODNESS-OF-FIT STATISTICAL PARAMETE (Pellinen 2001).

Criteria R2 Se/Sy
Excellent  0.90 ≤0.35
Good 0.70 – 0.89 0.36 – 0.55
Fair 0.40 – 0.69 0.56 - 0.75
Poor 0.20 – 0.39 0.76 – 0.89
Very Poor ≤0.19  0.90

 y  yi 
n

 ŷ  yi 
2 2
i i
(1) Sy  i 1 (2)
Se  i 1

np n 1
2
 n  1   S e 

S 
2
R 2
 1   
 n  p   S y 
adj
R  1  e
2  (3)  (4)
S 
 y 

5
where, Se = standard error of estimate; Sy = standard deviation; yi = measured value; ŷ i = predicted value; y i =
average of yi,; n = number of data points; p=number of regression coefficients, R2 = coefficient of determination; and
R2adj = adjusted coefficient of determination.
Many attempts were conducted to get the most reliable and accurate model with minimal bias. The best formula was
found to be as follows:
CBR = 21.21 MDD - 4.34 GM + 0.69 R#10 - 6.013 (5)

where, CBR = soaked California Bearing Ratio, %; MDD = maximum dry density according to modified Proctor test,
(t/m3); GM = grading modulus, GM = (P10 + P40 + P200)/100, R#10 = percentage retained on sieve No.10;
P10, P40, P200 = percentage passing U.S. sieve No. 10, 40, and 200, respectively.

Figure 9 presents the measured CBR values against the predicted ones using Equation (5). It can be seen from the
Figure that the model has an excellent accuracy (R2 of 0.928, R2adj of 0.926, and Se/Sy of 0.268). The unconstrained
line of equality and its equation is also shown in Figure 9. The slope of this line is very close to the line of equality and
the intercept is very close to zero. This indicates an excellent model with very minimum bias in the predictions.

100
90 Goodness of fit
y = 0.9224x + 3.329
80 R2 0.928 R² = 0.9224
Predicted CBR, %

70 R2adj 0.926
60 Se/Sy 0.268
50
40
30
20
10 Linear (line of equality)
0
0 20 40 60 80 100
Laboratory measured CBR, %
Figure 9. Comparison between the laboratory measured and predicted
CBR values using MLR.

To evaluate the adequacy of the proposed model, an analysis of variance (ANOVA) was performed on the regression
results. This analysis was conducted with the null hypothesis, which is that the CBR is not related to the MDD, GM
and R10, while the alternate hypothesis is that the CBR is related to the three variables. The results of the ANOVA
test are summarized in Table 4.

TABLE 4. ANOVA RESULTS

SS df MS F P-value
Regression 31421.3145 2 15710.66 68.0623897 9.09E-26
Residuals 82405.34401 357 230.8273
Total 113826.6585 359
df = degrees of freedom; SS=sum of squares; MS=mean square; F = F statistic.

Since the P-value is 9.09E-26, which is significantly lower than the significance value of 0.05, the null hypothesis is
rejected. In other words, there is a good relation between the CBR and the proposed variables in the model.

6
B. Artificial Neural Networks (ANNs)
An Artificial neural network is an information processing system that has been developed as a generalization
of the mathematical model of human cognition Neural network is the study of networks consisting of nodes
connected by adaptable weights, which store experimental knowledge from task examples through a process
of learning (Sivanandam et al., 2009).
An ANN is an abstract simulation of a real nervous system and its study corresponds to a growing
interdisciplinary field, which considers the systems as adaptive, distributed and mostly nonlinear. The
biological neural network, sometimes called a neural pathway, (Fernandez de Canete et al. 2011) is a series of
interconnected neurons, whose activation defines a recognizable linear pathway (Figure 10). The interface
through which neurons interact with their neighbors usually consists of several axon terminals connected via
synapses to dendrites on other neurons. The artificial neural network as shown in Figure 10 contains three
layers, the first layer consists of number of inputs, the second layer contain a number of hidden layers, which
contains number of neurons in each one, finally the third layer contains the outputs.

Figure 10. Illustration of Biological and Artificial Neural Networks (Negishi 1998).

In this paper, ANNs method was applied for the prediction of CBR values using Neurosolutions V.5 software
(Jose C. Principe 2000). NeuroSolutions software provides an easy-to-use and intuitive user interface for
Microsoft Excel. To develop the ANN model, the available data set was rearranged randomly because the
software doesn’t deal with data arranged regularly.
For developing the optimum model, the available data was divided into two separated data sets; the training
data set and the testing data set. Among 120 data sets, only 80% of the total data sets were used in the training
process and the rest (20%) were used for the testing stage. The network architecture involved the selection of
input parameters, input layers, the number of hidden layers, a combination of transfer functions between the
layers and the learning rule for the training stage.
In this study, the gradation parameters, Atterberg limits and compaction parameters in different combinations
forming different sets were the input layer, while the soaked CBR values were selected as the output layer.
Many trails were carried out to develop an ANN model by changing the network type, number of hidden layers
and number of neurons in each hidden layer. The multilayer perceptron network models were developed using
one hidden layer containing 20 neurons.
To judge the performance of the developed models, there are three main parameters; mean square error (MSE),
Normalized Mean Square Error (NMSE) and Correlation Coefficient (R). The smaller the errors, the higher
the correlation coefficient, and hence the higher the R2. In this study, a sigmoid axon function is used for the

7
hidden layer neurons and the output neuron (CBR). The performance of the developed models for the testing
data set is shown in Table 5.
Table 5: Performance of multilayer perceptron neural networks
Performance of testing data set (27 data points)
Dependent Correlation Coefficient of
Model Independent variables Structure
variable MSE NMSE MAE coefficient determination
(r) (R2)

Gradation, AASHTO Class,


I 15-20-1 0.01013 0.14110 0.07814 0.932 0.868
Compaction parameters

Soaked CBR
II P200, MDD 2-20-1 0.01169 0.16278 0.08188 0.928 0.861

AASHTO Class., MDD,


III 6-20-1 0.01989 0.27697 0.11830 0.911 0.829
OMC, GM

IV MDD, GM, R10 3-20-1 0.01123 0.15642 0.09506 0.939 0.882

In these four models, the number of independent input variables changed from fifteen to three and the output dependent
variable was the soaked CBR. Several networks with different number of hidden layers and neurons were trained and
tested to predict the desired values. As seen from Table 5, Model IV seems to be the best model in terms of MSE,
NMSE, MAE, r and R2 for the testing data set. To develop a relationship between desired and predicted output, the
software provides an option to apply the network to the training data that tagged as production as shown in Figure 11.
However, it is clear from this Figure that the predictions are biased and the ANNs model overpredicts the CBR values.

100
y = 0.9015x + 9.712
80 R² = 0.92
Predicted CBR (%)

60

40

20 Linear (line of quality)


Regression line of measured
and predicted
0
0 20 40 60 80 100
Measured CBR (%)
Figure 11. Comparison between laboratory measured CBR and predicted CBR values
(Model IV) using ANNs

VI. Summary and Conclusions

In this paper an attempt has been made to estimate the value of soaked CBR using other parameters like
gradation, Atterberg limits and compaction parameters of non-plastic soils. The collected database with 120
data points was used to develop MLR and ANN models. In MLR analysis a simple but a reliable and
relatively accurate regression model predicting the CBR as a function of (R#10), GM and MDD, was
proposed. The goodness of fit statistics of the measured and predicted CBR values using the proposed model
indicates excellent prediction accuracy and minimal bias (R2 of 0.928, R2adj of 0.926, and Se/Sy of 0.268).

8
The ANN analysis showed excellent accuracy, however, it overestimated the CBR values compared to the
regression model.

VII. REFERENCES:

Agarwal, K.B. & Ghanekar, K.D., 1970. Prediction of CBR from plasticity characteristics of soil. In Proceeding of 2nd
south-east Asian conference on soil engineering, pp. 11–15, Singapore.

Alawi, M.H. & Rajab, M.I., 2013. Prediction of California bearing ratio of subbase layer using multiple linear regression
models. Road Materials and Pavement Design, 14 (1), pp.211–219.

American Association of State Highway and Transportation Officials (AASHTO), 2003. The California Bearing Ratio.
AASHTO T 193-99.

ASTM, 1999. Standard Test Method for CBR ( California Bearing Ratio ) of Soils in Place. In Annual Book of ASTM
Standards.

Bello, A.A., 2012. Regression Analysis between Properties of Subgrade Lateritic Soil. Leonardo Journal of Sciences, ISSN
1583-0233, Issue 21, pp.99–108.

Bhatt, S., Jain, P.K. & Pradesh, M., 2014. Prediction of California Bearing Ratio of Soils Using Artificial Neural Network.
American International Journal of Research in Science, Technology, Engineering & Mathematics, pp.156–161.

Black, W.P.M., 1962. A Method of Estimating the California Bearing Ratio of Cohesive Soils from Plasticity Data.
Géotechnique, 12(4), pp.271–282.

Black, W.P.M., 1961. The Calculation of Laboratory and In-situ Values of California Bearing Ratio from Bearing Capacity
Data. Géotechnique, 11(1), pp.14–21.

Breytenbach, I.J., Paige-Green, P. & Van Rooy, J. L. 2010. The relationship between index testing and California Bearing
Ratio values for natural road construction materials in South Africa. Journal of the South African Institution of Civil
Engineering, 52 (2), pp.65–69.

Egyptian Code of Practice for rural and urban roads, ECP, 2008, "Part 4: Roads materials and its tests", First Edition, PP
24-30, 104/4.

Fairbrother, S., 2011. Estimating forest road aggregate strength by measuring fundamental aggregate properties, 34th
Council on Forest Engineering, Quebec, Canada.

Fernandez de Canete, J., Gonzalez-Perez, S. & Ramos-Diaz, J.C., 2011. Artificial neural networks for closed loop control
of in silico and ad hoc type 1 diabetes. Computer methods and programs in biomedicine, pp.1–48.

Hakari, U.D., & Nadgauda, K. D., 2013. Estimation and evaluation of california bearing ratio by indirect methods,
Proceedings of Indian Geotechnical Conference December 22-24, Roorkee.

Harini HN, S.N., 2014. Prediction CBR of Fine grained soils by Artificial Neural network and Multiple Linear Regression.
International Journal of civil engineering and technology (IJCIET), 5(2), pp.119–126.

Harison, Ja., Woodman, GR., Nataatmadja, A.& Grant, Mc ., 1989. Technical note. Correlation between california bearing
ratio and dynamic cone penetrometer strength measurement of soils. ICE Proceedings, 87(1), pp.119–125.

Jose C. Principe, Euliano N.R.& Lefebvre W.C., 2000. Neural and adaptive systems: fundementals through simulations,
ISBN: 978-0-471-35167-2, Wiley.

9
Sathawara, K.J. & A.K.Patel, 2013. Comparison between soaked and unsoaked CBR. International Journal of Advanced
Engineering Research and Studies, 2(3), pp.132–135.

NCHRP, 2004. Guide for Mechanistic -Empirical Design of new and rehailated pavement structures, NCHRP 1-37A Final
Report, Appendix CC-1: Correlation of CBR values with soil index properties",

Negishi, M. 1998. Everything that Linguists have Always Wanted to Know about Connectionism.

Pellinen, T.K., 2001. Investigation of the use of dynamic modulus as an indicator of hot-mix asphalt performance. Arizona
State University.

Rashmi, S., 2010. CBR Predicted by Index Properties for Alluvial Soils of South Gujarat. Indian Geotechnical Conference,
pp. 79-82

Singh, D., Reddy, K.S. & Yadu, L., 2011. Moisture and Compaction Based Statistical Model for Estimating CBR of Fine
Grained Subgrade Soils. International Journal of Earth Sciences and Engineering, 04(06), pp.100–103.

Sivanandam, M. & others, 2009. Introduction to artificial neural networks, vikas publishing House PVT LTD.

Taha, S.A., El-Badawy, S.M. & Ali, A.M., 2014. Determination of California Bearing Ratio through soil index properties.
In 4th Jordan International Conference and Exhibition for Roads and Transport ( JITC4). Jordan.

Talukdar, D.K., 2014. A Study of Correlation Between California Bearing Ratio (CBR) Value With Other Properties of
Soil. International Journal of Emerging Technology and Advanced Engineering, 4(1), pp.559–562.

Taskiran, T., 2010. Prediction of California bearing ratio (CBR) of fine grained soils by AI methods. Advances in
Engineering Software, 41(6), pp.886–892..

Venkatasubramanian, C., & Dhinakaran, G., 2011. ANN model for predicting CBR from index properties of soils.
International Journal of Civil and Structural Engineering, 2(2), pp.605–611.

Yadav D., Jain P. K. & Kumar R.., 2014. Prediction of soaked cbr of fine grained soils. International Journal of advanced
Engineering Research and Studies, pp.119–121.

Yildirim, B. & Gunaydin, O., 2011. Estimation of California bearing ratio by using soft computing systems. Expert Systems
with Applications, 38(5), pp.6381–6391.

Zumrawi, M.M.E., 2013. Prediction of CBR Value from Index Properties of Cohesive Soils. Advances in Civil Engineering
and Building Materials, pp.561–565.

10

View publication stats

You might also like