Assessment of Relationship Between Static

and Dynamic Load Using Regression Analysis

and Artificial Neural Network Model

Ahmed H. Abulkareem(&)

Civil Engineering Department, Engineering College,

University of Anbar, Ramadi, Iraq
[email protected]

Abstract. The Light Falling Weight Deflectometer (LFWD) load test has been
developed to directly estimate the in-situ elastic modulus of near surface profiles
as foundation, and subgrade layers is presented in this paper. For this purpose,
field tests were conducted on selected sections from landfill project within
Anbar province. In addition, forty test sections were constructed and tested at
the Civil Engineering Department- University of Anbar. All sections were tested
using the ZFG 3000 model - LFWD in companion with the Plate Load Test
(PLT) that were used as reference measurements. Regression analyses were
conducted to determine the best correlations between the elastic modulus
obtained from LFWD and PLT tests., Evd. ANN model is used to calculate
dynamic deformation modulus, Evd and comparing with the regression statistical
model. The results indicate that ANN model have the capability of predicting
dynamic deformation modulus, Evd with a high degree of accuracy. Good
correlations were obtained, which demonstrated that the LFWD can be reliably
used to predict the modules obtained from plate load test and degree of com-
paction values, and hence can be used to evaluate the stiffness/strength
parameters of shallow subgrade layers.

Keywords: Plate load test
LFWD
Dynamic modulus
ANN
Regression

analysis

1 Introduction

The light falling weight Deflectometer (LFWD) is used in landfill project in Iraq. The
German device (ZFG 3000) from Zorn is a compaction control device used in this
research. It is able to measure the dynamic load bearing capacity of subgrades, subsoils,
embankment layers and backfills.
Extensive application of these apparatus still has not been achieved since the
dynamic modulus in not accepted in the quality assessment and quality control process
of embankments and subgrade layers. Only marginal use of these devices can be
noticed, mainly on areas of low importance (e.g. road shoulders) or trenches where
performing a static plate load test could be complicated (Zoltán 2008).

© Springer International Publishing AG 2018

W. Frikha et al. (eds.), Soil Testing, Soil Stability and Ground Improvement,
Sustainable Civil Infrastructures, DOI 10.1007/978-3-319-61902-6_21
270 A.H. Abulkareem

For being able to use these dynamic devices on subgrade and embankment layers
for landfill project, the research for converting the measured dynamic modulus into
static modulus has been initiated.
The main objective was to determine the correlation between static and dynamic
modules. Since direct conversion formulas are not frequently used in practice, intro-
duction of an easy to use table with the required static and dynamic target values has
been aimed. This is achieved by conducting field tests on constructed layers using the
investigated device LFWD along with standard in-situ test device Plate Load Test
(PLT). Otherwise new quality assessment based on dynamic modulus might be able to
substitute the exclusive usage of the slow and complicated static plate load test in the
near future. With the help of these results, new dynamic design methods can be worked
out and applied.

2 Light Falling Weight Deflectometer (LFWD)

Light Falling Weight Deflectometer (LFWD) is a portable falling weight deflectometer

that has been developed in Germany as an alternative in-situ testing device to the plate
load test. Different types of LFWD exist in the market. All types exhibit many simi-
larities in their mechanics of operation although there are many differences in design
and mode of operation, which lead to variations in the measured results. Generally, the
LFWD consist of a loading device that produces a defined load pulse, a loading plate,
and one center geophone sensor (electric deflection data device) to measure the center
surface deflection (Garcia and Thompson 2004).
The ZFG 3000 LFWD from Zorn Instruments was used in this study Fig. 1 is a
compaction control device, according to ASTM E2835-11 and ASTM E2583-07
(2011). Dynamic modulus of deformation, Evd is the most accurate and independent
means for judging deformation (stiffness) and, thus, a material’s level of compaction
(www.ticservicegroup.com.au, 2013).
A 10 kg falling weight is dropped onto a 300 mm diameter plate from a height of
72 cm through guide rod; the vertical displacement of the plate (dc) is recorded by an

Fig. 1. Light Falling Weight Deflectometer (LFWD) (TIC Service Group 2013).
 Assessment of Relationship Between Static and Dynamic Load 271

accelerometer built in a steel case on the top of the plate. The drop weight, drop height
and plate diameter are constants. The plate coefficient (c) and the Poisson’s ratio (µ) are
also set constant, therefore the dynamic subgrade modulus, (Evd) is calculated by a
simplified Boussinesq equation (Zoltán 2008):

Evd ¼ 22:5=dc ð1Þ

To estimate dynamic modulus of deformation, Evd for each test in this study, eight
position (nearly 45° apart) surrounding the location of the static plate load test were
selected and dynamic plate load test were performed. Each test was repeated three
times, in each time three preconsolidation tests (pre-compaction to remove any bedding
errors, and are ignored) were conducted as shown in Fig. 1.

3 Static Plate Load Test (PLT)

The Static Plate Load Test (PLT) has been a useful site investigation tool for many
years and has been used for proof testing of pavement structure layers in many
countries. Currently it is used for both rigid and flexible pavements. The test was
conducted by the procedure recommended by ASTM D1196-93as shown in Fig. 2. The
test consists of a circular plate (450 mm diameter) that is in close contact with the layer
to be tested and measuring deflections under load increments. The load increments
were applied via a hydraulic jack with a suitable load capacity. The load was applied in
increments up to a final value of externally applied stress of 700 kPa. The corre-
sponding settlement was monitored and recorded, by using three suitable dial gauges
(120° apart), for each increment until the settlement has ceased.

Fig. 2. Static Plate Load Test (PLT).

Plate loading tests can be used to estimate the modulus of subgrade reaction (k).
Determination of the modulus of subgrade reaction is made in the field on the selected
subgrade soil at its natural moisture content. This test is conducted by subjecting the
272 A.H. Abulkareem

subgrade to a known stress at a predetermined rate of speed using a loading system, and
recording the resulting settlement. The modulus of subgrade reaction, k, can be cal-
culated using the following relation (Yoder and Witczak 1975):

k ¼ p=d ð2Þ

Where
P = unit load on plate (kN)
d = Settlement of the plate (mm)
The value of Young’s modulus was obtained from the well-known relationship that
correlates the young’s modulus to the modulus of subgrade reaction (Bowels 1997):


Es ¼ kB 1  l2 ð3Þ

Where B in this case represents the diameter of the testing plate and µ is the
Poisson’s ratio and its value can be 0.5 for the assumption of flexible base plate (Yoder
and Witczak 1975).

4 Existing Relationships

The Institute for Transport Sciences in Hungarian (KTI) launched a research program
in 1995 aiming to convert the dynamic modulus obtained by that device (Evd) into the
well-known static plate load test modulus (E2) obtained by conventional measurements
(research report, ÁKMI Kht 1995). After collecting 64 measurement results performed
on different subgrade and subsoil materials, a general conversion formula was sug-
gested as following:

Evd ¼ 0:52
E2 þ 9:1 ð4Þ

Several correlation results between E2 and Evd are available in the international
literature. The most relevant results are summarized in Fig. 3.
Figure 3 shows that the value of the static plate load test modulus clearly exceeds at
least two times that of the Evd modulus. Some of the results show even higher ratios.
Only two publications give a ratio less than two, but both of them are based on
modulus values measured only at few points and within small intervals (Zoltán 2008).
Nazzal (2003) present a correlation study between the PLT and the LFWD on
cement treated soils, lime treated soils, unstabilizsed fine-grained soils and granular
soils. The relationship for modulus (surface modulus for LFWD) thus developed is as
follows:

EPLT ¼ 20:9 þ 0:69ðELFWD Þ ðR2 ¼ 0:94; for 2:5 MPa\ ELFWD \865 MPa ð5Þ

These regression models are illustrated in Fig. 4.

 Assessment of Relationship Between Static and Dynamic Load 273

Fig. 3. Correlation results between E2 and Evd (Zoltán 2008).

Fig. 4. Relationship between EPLT(R2) and ELFWD (Nazzal 2003).

274 A.H. Abulkareem

5 Analysis of Results
5.1 Linear Regression Model
In this study, collecting 40 measurement results for plate load test (PLT) performed on
subgrade crushed limestone material in excess of (38 mm) in diameter by the field
laboratory for civil engineering department at University of Anbar. After the division of
project area as strips and establish the location of each point by mark, it has performed
LFWD testing as shown in Fig. 5.

Fig. 5. Profile of the project area.

In order to characterize the variation of Evd used as an independent value,

descriptive statistics such as; minimum, maximum, mean, mode, median, variance,
standard deviation, skewness and kurtosis etc. were calculated using the SPSS Version
20 (2012) package. Table 1 shows that the independent statistics value. Depending on
the kind of soil and the degree of compaction, it can possible exceeding the limits of the
lower and upper Evd values as (Nazzal 2003).
In this study, simple regression analysis was performed in the analysis. The rela-
tions between the measured Evd, moduli values with conventional static Es, moduli
values were analyzed employing linear, power, logarithmic and exponential functions.
Statistically significant and strong correlations were found to be linear.
The objective of this regression analysis is to determine the parameters in the least
square error models, which is used to predict the Evd moduli from Es moduli, with their
corresponding coefficient of determination, R2. In the linear regression models, the
dependent variable is assumed to be a linear function of one or more independent
 Assessment of Relationship Between Static and Dynamic Load 275

Table 1. Descriptive statistics for Evd as an independent value.

N 40
Mean 49.87
Std. error of mean 4.7394
Median 46.28
Mode 27.61
Std. deviation 29.97452
Variance 898.472
Skewness 3.957
Std. error of skewness 0.374
Kurtosis 20.714
Std. error of kurtosis 0.733
Range 187.7
Minimum 21.3
Maximum 209
Sum 1994.91

variables plus an error introduced to account for all other factors, a typical form of a
regression linear model is as follow:

Yi ¼ bo þ b1 xi1 þ . . .. . .bk xik ð6Þ

Where Yi is the dependent variable, and xi1, …, xik are the independent or
explanatory variables, and bo is the disturbance or error term (SPSS Inc. 2009).
The coefficient of determination, R2, represents the proportion of variation in the
dependant variable that is accounted by the regression model and has values from 0 to
1. If it is equal to one, the entire observed points lie on the suggested least square line,
which means a perfect correlation exists. Significance level is the result of the statistic
test with null hypothesis b1 = 0; it is expressed in percent. The greater the significance
level the more supportive the model to alternative hypothesis (b1 6¼ 0), which indicates
that a relation does exist between the dependent and independent variable. Finally, the
standard error is the square root of the mean square errors (MSE).
After collecting forty measurement results performed on subgrade layer, a general
conversion formula was suggested as follows:

Evd ¼ 7:384 þ 0:527 Es ð7Þ

This formula can be used to convert the measured Evd dynamic moduli values into
conventional static Es moduli values. The coefficient of correlation gives a value of
R2 = 0.801 and standard error = 13.56 for 40 sites, which seems to be acceptable in
geotechnical testing. This regression model is illustrated in Fig. 6. It can be seen that
there are some values of Evd and Es outlier range caused by the low correlation
coefficient (R2).
276 A.H. Abulkareem

0 40 80 120 160 200 240 280 320 360

240 240

200 200

160 160
Evd(N/mm )
2

120 120

80 80

Evd=7.384+0.527Es
2
40 R =0.801 40
No. of Data=40

0 0
0 40 80 120 160 200 240 280 320 360
2
Es(N/mm )

Fig. 6. Relationship between Evd and Es.

5.2 Multiple Regression Model

In this study, 13 test results were selected from 40 sites. These sites have been tested by
PLT and LFWD and have conducted field density test by sand cone method according
to ASTM D1556-00. The modified proctor compaction test according to ASTM
D1557-12 also showed that the maximum dry unit weight and the optimum moisture
content for crushed limestone material in excess of (38 mm) in diameter were
17.6 kN/m3 and 5.6%, respectively. Summary of results for this section is presented in
Table 2.

Table 2. Summary of test results for crushed limestone layer.

Strips Evd MPa Es MPa MC (field) % cdry (field) kN/m3 Dpr, Modified %
6–2 37.08 35.21 7.59 16.99 96.56
8–1 59.63 73.93 13.57 17.27 98.12
9–5 21.3 34.01 16.2 16.25 92.36
10–8 44.2 63.84 14.08 16.54 93.96
9–1 68.8 127.90 15.6 16.42 93.30
10–3 24.53 50.30 15.7 16.23 92.20
10–4 46.33 41.91 13.08 16.72 94.99
10–5 26.55 43.05 13.8 16.13 91.64
10–6 52.63 79.56 14.08 16.89 95.95
10–7 47.53 123.00 13.8 16.74 95.12
9–2 64 101.57 16.11 16.22 92.18
9–6 24.6 59.97 15.32 16.23 92.21
9–7 86.2 143.12 16.2 16.92 96.14
 Assessment of Relationship Between Static and Dynamic Load 277

Multiple regression, a time-honored technique going back to Pearson’s 1908 use of

it, is employed to account for (predict) the variance in an interval dependent, based on
linear combinations of interval, dichotomous, or dummy independent variables. The
general purpose of multiple regression is to learn more about the relationship between
several independent or predictor variables and a dependent or criterion variable. The
typical form of the multiple regression model is as follow (SPSS Inc. 2009):

y ¼ b1 x1 þ b2 x2 þ . . .: þ bn xn þ c ð8Þ

Where b1, b2, …, bn are the regression coefficients, representing the amount the
dependent variable y changes when the corresponding independent changes 1 unit. c is
a constant, where the regression line intercepts the y axis, representing the amount the
dependent y will be when all the independent variables are 0. The standardized versions
of the b coefficients are the beta weights, and the ratio of the beta coefficients is the
ratio of the relative predictive power of the independent variables. The major con-
ceptual limitation of all regression techniques is that one can only ascertain relation-
ships, but never be sure about underlying causal mechanism (SPSS Inc. 2009).

Table 3. Model summaries of multiple regressions for prediction of Evd.

Independent variables Coefficient Std. error t-Value Sig. level
Constant −345.76 193.484 −1.787 0.108
Es, MPa 0.375 0.1 3.733 0.005
MC, % 0.754 1.855 0.406 0.694
Dpr, % 3.75 1.902 1.971 0.08

Multiple regression analysis was carried out to correlate the measured Evd to three
parameters, namely, static modulus (Es), moisture content (MC) and degree of com-
paction (Dpr) (Table 3). Multiple regression model to predict Evd is given:


Evd ¼ 0:375ðEs Þ þ 0:754ðMC%Þ þ 3:75 Dpr % 345:76 ð9Þ

The coefficient of correlation between the measured and predicted values is a good
indicator to check the prediction performance of the model. Figure 7 shows the rela-
tionships between measured and predicted values obtained from the Multiple regres-
sion model for Evd, which good correlation coefficient with R2 = 0.804.

5.3 Target Values for Dynamic Models

Direct conversion between dynamic, static models and degree of compaction is not
frequently used in practice. Generally target values are given for different embankment
and subgrade layers, more often depending on the required degree of compaction of the
tested layer. Evd modulus target values are fixed in Germany, and some other countries.
278 A.H. Abulkareem

0 20 40 60 80 100
100 100

80 80
Evd, Predicted (N/mm )
2

60 60

40 40

20 20
No. of Data=13
2
R =0.804
0 0
0 20 40 60 80 100
2
Evd, Measured (N/mm )

Fig. 7. Relationship of predicted and measured values of Evd for Multiple regression model.

Based on the results above, a table of target values can be introduced. Different Es
and degree of compaction values are given for required Evd values in Table 4. It can be
seen that Evd values reduce in access 10% compared with Es due to kind of soil layers
and degree of compaction. Interpolation between given values is acceptable (Zoltán
2008).

Table 4. Target values for crushed limestone subgrade layers.

Es MPa Evd MPa Dpr % Modified Es/Evd ratio
Crushed limestone subgrade layers
180 100  100 2
150 85  99 2
120 70  98 2
100 60  98 2
80 50  97 2
60 35  95 2
45 30  95 2
20 15  94 1.5

5.4 Artificial Neural Network (ANN) Modeling of Overview

When the materials are natural, there is always an uncertainty sourced from the nature
of the materials. This could be the main reason why soft computing methodologies
such as artificial neural networks, Fuzzy systems, genetic algorithms have been
 Assessment of Relationship Between Static and Dynamic Load 279

developed in recent years and adopted but used for some other application. These
techniques attract more and more attention in several research fields because they
tolerate a wide range of uncertainty (Jin and Jiang 1999). Multilayer feed-forward
neural network model is the most widely used network for its efficient generalization
capabilities. Figure 7 presents typical multi-layer feed-forward neural networks. This
type of neural network consists of an input layer, one or more hidden layer(s) and an
output layer. Layers are fully connected by arrows, and comprise a number of pro-
cessing units, the so-called nodes or neurons. The strength of connections between
neurons is represented by numerical values called weights. Each neuron has an acti-
vation value that is a function of the sum of inputs received from other neurons through
the weighted connections (Demuth et al. 2006). The optimum number of hidden layers
and the number of neurons in each hidden layer is problem specific. Therefore, trial and
error should be carried out to choose an adequate number of hidden layers and the
number of neurons in each hidden layer.
Back propagation is the most successful and widely used in neural network
applications. In this method, the input is propagated from the input layer through the
hidden layers to the output layer. The network input is connected to every neuron in the
first hidden layer while each network output is connected to each neuron in the last
hidden layer. In this case, this would call full connection ANN. The network weights
were originally set to random values and new values of the network parameters
(weights) are computed during the network training phase. The neurons output are
calculated using (Abdul-Razzak and Yousif, 2007):
X
Oi ¼ Fð I
j j
 Wij þ bi Þ ð10Þ

Where
Oi = the output of the neuron i, Ij are the input of j neurons of the previous layer,
Wij = the neuron weights, bi is the bias for the modeling, and
F = the activation function.
The activation function is the portion of the neural network where all the computing
is performed. The activation function maps the input domain (infinite) to an output
domain (finite). The range to which most activation functions map their output is either
in the interval [0, 1] or the interval [−1, 1]. The most common activation functions
belong to five families as follows: (1) linear activation function; (2) step activation
function; (3) ramp activation function; (4) sigmoid activation function; and (5) Gaus-
sian activation function. The network error is then back propagated from the output
layer to the input layer in which the connection weights are adjusted. This process is
repeated until the error is minimized to a preference level. The error incurred during the
learning can be expressed as Mean Squared Error and is calculate using (Abdul Razzak
and Yousif 2007):

1 Xn Xm  2
MSE ¼ t ij  y ij ð11Þ
nm i¼1 j¼1

Where t is the target value, y is the output value.

280 A.H. Abulkareem

5.5 ANN Model for Prediction of Evd

The use of ANN provides an alternative way to estimate dynamic modulus of defor-
mation, Evd. In this work a multi-layered feed-forward neural network with a
back-propagation algorithm was adopted. (MATLAB 7.1 2005) software was used in
neural network analyses having a three-layer feed-forward network (Alvarez and
Babuska 1999). Forty cases of actual measured were extracted from experimental tests
used in this study. The databases is randomly divided into three sets such as; training
(70% of all data), test (15% of all data), and verification (15% of all data). The training
set is used to adjust the connection weights, whereas the testing set is used to check the
performance of the model at various stages of training and to determine when to stop
training to avoid over-fitting. The validation set is used to estimate the performance of
the trained network in the deployed environment (Shahin et al. 2008). A typical
structure of artificial neural networks consists of many processing elements that are
arranged in layers: an input layer, an output layer, and one or more layers in between,
called intermediate or hidden layers (Fig. 8). Each processing element in a specific
layer is interconnected to all the processing elements in the next layer via weighted
connections. The scalar weights determine the strength of the connection between
interconnected neurons. A zero weight refers to no connection between two neurons
and a negative weight refers to a prohibitive relationship (Shahin et al. 2008). The
model has one input parameter and one output parameter. The model has two hidden
layers with nine nodes each (MATLAB software uses for determining the optimal
number of hidden nodes rather than assuming a fixed number of hidden nodes in
advance), and output layer with one node giving dynamic modulus of deformation, Evd.

Fig. 8. Structure of Multi-layered feed-forward network (Michael 2015)

 Assessment of Relationship Between Static and Dynamic Load 281

By increasing the number of hidden neurons they can typically get a better approxi-
mation and do still better by further increasing the number of hidden neurons. In the
analyses network of learning parameter, momentum parameters and networks training
function, which is an activation (transfer) function for all layers, have typical values of
0.01, 0.9, trainLm (training Levenberg-Marquardt function) and tansig (transfer func-
tion) respectively. As in many other network training methods, models and parameters
were used in order to reach the minimum Mean Square Error (MSE) values.

0 40 80 120 160 200 240

240 240

200 200
Evd,Predicted(N/mm )

160 160
2

120 120

80 80

40 40
No. of Data=40
2
R =0.9862
0 0
0 40 80 120 160 200 240
2
Evd,Measured(N/mm )

Fig. 9. Relationship of predicted and measured values of Evd for ANN Model.

In fact, the coefficient of determination between the measured and predicted Evd
values is a good indicator to check the prediction performance of the model. Figure 9
shows the relationships between measured and predicted values obtained from the
models for Evd, with good coefficient of determinations. In this study, variance account
for VAF (Eq. 12) and root mean square error (RMSE) (Eq. 13), indices were also
calculated to control the performance of the prediction capacity of predictive models
developed in the study as employed by (Alvarez and Babuska 1999):
 
var ðy  y0 Þ
VAF ¼ 1   100 ð12Þ
var ð yÞ
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 XN
RMSE ¼ ð y  y0 Þ 2 ð13Þ
N i¼1

Where y and y’ = the measured and predicted values, respectively.

If the VAF is 100 and RMSE is 0, then the model will be excellent. The obtained
values of VAF and RMSE given in Table 5 indicated a high prediction performance.
282 A.H. Abulkareem

Table 5. RMSE, VAF and R2 values used to predict Evd.

Model Predictive model RMSE VAF% R2
LRM Evd = 7.384 + 0.527 Es 9.25 75.09 0.81
ANN 2.57 98.31 0.98

6 Conclusions

The LFWD proved to be a truly light weight FWD and is highly transportable. It is very
easy to operate and changes from the 10 kg to the 20 kg drop weights or loading plates
(200 mm and 300 mm) are quick and easy to do. All tests are saved automatically on a
chip card, simply insert the card and the data can be transferred to the PC immediately.
The basic information provided by the LFWD clearly proved to be very useful for
construction quality control and assurance purposes.
The objective of this study is to evaluate the potential use of non-destructive testing
device to measure the stiffness/strength parameters of highway materials and
embankment soils during and after construction for landfill project. The field testing
program included conducting tests using the investigated devices, in addition to
standard tests, which included the static Plate Load Test (PLT), field density test by the
sand cone method, and modified proctor compaction test). The field testing performed
on subgrade crushed limestone material in excess of (38 mm) in diameter.
The results of the statistical analysis show that a good correlation do exist between
the device under evaluation LFWD and the standard tests PLT, and degree compaction
depending on standard tests. The relations obtained from statistical analysis, were linear
regression to model and multiple regression for another. All regression models had an
adjusted, correlation coefficient (R2) greater than 0.8.
A multi-layered feed-forward neural network with a back-propagation algorithm
was used to demonstrate the feasibility of ANNs to predict the dynamic modulus of
deformation, Evd. Forty cases of actual field measurements were used for model
development and verification. The predicted Evd obtained by utilising ANNs were
compared with the measured Evd. The results indicate that ANN model have the
capability of predicting Evd with a high degree of accuracy. From VAF, RMSE indi-
cators and correlation coefficient (R2) results, it can be seen that the ANN model is
more accurately than regression analysis to predict Evd as in Table 5.
The result of this study suggests that LFWD can be reliably used to predict the
modules obtained from PLT, and degree of compaction values, and hence can be used
to evaluate the stiffness/strength parameters of subgrade layers.
The new dynamic target values could open up the opportunity to perform the
quality control and assess the bearing strengths of the tested layer, not only by static
plate load test, which proved to be time-consuming and labour intensive, but by
dynamic devices too.
The widespread use of mentioned dynamic devices referred to above, may facilitate
for contractors, laboratories and engineers in the highway and railway construction
industry to perform quick and continuous quality control of embankments, subgrade
and subsoil layers and backfills.
 Assessment of Relationship Between Static and Dynamic Load 283

Acknowledgment. The authors express his sincere thanks to the staff of the field laboratory for
civil engineering department at University of Anbar for help in the conduct the field tests.

References
Zoltán, T.: Conversion between static and dynamic load bearing capacity moduli and
introduction of dynamic target values. Periodica Polytech. Civil Eng. 52(2), 97–102 (2008)
Garcia, G., Thompson, M.R.: Subgrade strength/stiffness evaluation. University of Illinois, Dept
of Civil and Environmental Engineering, Technical Note, 1215NCEL, MC-250, Urbana, IL
61801 (2004)
TIC Service Group: Test Equipment for Geotechnics and Pavements (2013). www.
ticservicegroup.com.au
Yoder, E.J., Witczak, M.: Principles of Pavement Design. Wiley, New York (1975)
Bowels, J.E.: Foundation Analysis and Design, 5th edn. The McGraw-Hill Companies,
New York (1997)
Institute for Transport Sciences: Measurement of load bearing capacity with lightweight
deflectometers (Wemex/ ZFG). Research report, ÁKMI Kht (1995)
Nazzal, M.D.: Field evaluation of in-situ test technology for QC/QA during construction of
pavement layers and embankments. M.Sc. thesis in civil engineering, Louisiana State
University and Agricultural and Mechanical College (2003)
SPSS Inc.: PASW Statistics for Windows, Version 18.0. Chicago: SPSS Inc. (2009)
Jin, Y., Jiang, J.: Techniques in neural-network based fuzzy system identification and their
application to control of complex systems. In: Leondes, C.T. (ed.) Fuzzy Theory Systems,
pp. 112–128. Techniques and Applications, Academic Press, New York (1999)
Demuth, H.B., Beale, M., Hagan, M.: Neural network toolbox for use with matlab User’s guide.
MathWorks, Incorporated (2006)
Abdul-Razzak, A.A., Yousif, S.T.: Artificial neural network model for predicting nonlinear
response of uniformly loaded fixed plates. Eng. Technol. 25(3), 334–348 (2007)
MATLAB 7.1 Software for technical computing and Model-Based Design. The MathWorks Inc.
(2005)
Alvarez, G.M., Babuska, R.: Fuzzy model for the prediction of unconfined compressive strength
of rock samples. Int. J. Rock Mech. Min. Sci. 36, 339–349 (1999)
Shahin, M.A., Jaksa, M.B., Maier, H.R.: State of the art of artificial neural networks in
geotechnical engineering. Electron. J. Geotech. Eng. 8, 1–26 (2008)
Michael, N.: Neural Networks and deep learning. online book (2015) http://
neuralnetworksanddeeplearning.com/chap4.html