Numerical Stress Strain Curve Oedometer

U.
Porto Journal of Engineering, 8:6 (2022) 102-120 Received: 29 March, 2022

ISSN 2183-6493 Accepted: 11 May, 2022
DOI: 10.24840/2183-6493_008.006_0008 Published: 28 November, 2022
Numerical Modelling for Prediction of Compression Index

from Soil Index Properties in Jimma town, Ethiopia
Worku Firomsa Kabeta1, Fekadu Fufa Feyessa2,
Yerosan Feyissa Keneni3,
1Facultyof Civil and Environmental Engineering, Gdansk University of Technology, Poland;
Faculty of Civil and Environmental Engineering, Jimma Institute of Technology, Ethiopia
(worku.kabeta@pg.edu.pl) ORCID 0000-0001-9792-8291; 2Faculty of Civil and Environmental
Engineering, Jimma Institute of Technology, Ethiopia (fekaduff2010@gmail.com) ORCID 0000-
0001-8974-0328; 3Faculty of Civil and Environmental Engineering, Jimma Institute of
Technology, Ethiopia (yerosan2007@gmail.com) ORCID 0000-0002-3530-5272
Abstract
In this study, correlations are developed to predict compression index (Cc) from
index parameters so that one can be able to model Jimma soils with compression
index using simple laboratory tests. Undisturbed and disturbed soil samples from
twelve different locations in Jimma town were collected. Laboratory tests like
specific gravity, grain size analysis, Atterberg limit, and one-dimensional
consolidation test for a total of twenty-four test samples were conducted. From one-
dimensional consolidation tests, compressibility soil parameters (Cc and Cs) are
determined. From the results of limited tests, an indicative good correlation is
observed between Cc and LL, PL, and PI. However, a Poor correlation is developed
between Cc and PL when related to the other parameters. The developed
correlations will be important inputs in modeling Jimma clay soils with regression
model and Artificial neural networks (ANN) analysis using simple index tests. In
addition, the results of this study can serve as a basis for further study of such
correlations on different clay soils in the country. In this study, regression analysis
was used to explore the significance of individual independent (index) soil
properties. Regression model and correlation of compression index for liquid limit,
plastic limit, and plasticity index were obtained from the linear regression analysis
and ANN. This correlation will be helpful for geotechnical engineers in developing
the coefficient of compression (Cc) value of expansive/clay soil from index
properties. Finally, based on the general findings of the study, suitable
recommendations have been forwarded.
Author Keywords. ANN Model. Regression Model. Atterberg Limits. Compression
Index. Correlation.
Type: Research Article
Open Access Peer Reviewed CC BY
1. Introduction
Settlement can occur because of a construction foundation built on a compressible soil layer.
Compression index (Cc) is a compressibility parameter calculated from the oedometer test
and used to estimate the magnitude of settlement. It is important to consider soil compression
purposes. The process of consolidation test proceeds longer durations. Therefore, it is
constructive if the value of the compression index can be interrelated with index properties
such as liquid limit, plastic limit, and plasticity index. The correlations between the engineering
and index properties of soils will reduce the workload of a soil investigation program in case
of urgency. The nature of the soil has always been an important part of civil engineering. Soil
properties, such as plasticity, the capability of existing, or soil strength, always influence the
102
Numerical Modelling for Prediction of Compression Index from Soil Index Properties in Jimma town, Ethiopia
Worku Firomsa, Yerosan Feyissa, Fekadu Fufa
design of the structure (Akayuli and Ofosu 2013). Failure to understand soil characteristics can
result in significant construction errors. Soil applicability for a specific application must be
determined by its engineering characteristics and not only by visual inspection or obvious
similarities with other soil. It takes two weeks to complete the oedometer test, but it is
expensive, stressful, and time-consuming (Jain, Dixit, and Chitra 2015).
Engineers need a lot of maturities and the ability to interpret application results with local
conditions. Due to these factors, several attempts were made in the past to predict the
compression index based on the properties of the soil index, which is rather convenient to
decide and requires much less time in the laboratory. Soil index properties, which include
Atterberg Limit values and moisture content, are physical soil properties. Therefore, the
characteristics of this index are used to determine the soil compression index (Cc).
The objective of the research was to determine the relationship between the compression
index (Cc) and the index properties of the soil. The compression tendencies of expansive soils
are quantified by the compressibility parameters. Determination of compressibility of
expansive soils, namely, recompression index and compression index (Cc), is important for the
design of foundations (Abbasi, Javadi, and Bahramloo 2012). The swell percent or volume
change of soil compression is the percentage of soil load for a particular load with an
additional surcharge load. The external pressure that has been applied to the expanded soil
to prevent an increase in volume is called soil pressure (Zumrawi 2012). A major concern of
the foundation engineer is to predict the behavior of changes in the volume of soil stress when
exposed to changes in a stressful environment. Geotechnical engineers practicing in such
areas are involved in a better understanding of relationships between the physical and
chemical properties of active clay. The soil is mostly selected fine-grained and clayey/silt
inherited with compressibility. Selected samples have been checked for index and engineering
properties.
Numerical modeling is an approach or method that is used to estimate, correlate, model, and
analyze the relationships between dependent and independent variables (Namdarvand,
Jafarnejadi, and Sayyad 2013). This study focuses on the correlation between compression
index and Atterberg limit (i.e., liquid limit, plasticity index) by using numerical modeling
(regression analysis and Artificial neural networks approaches). Regression analysis uses
extremely complicated equations to evaluate large datasets and interprets them into
coordinates on a line or curve, whereas artificial neural networks (ANN) are computational
methodologies that carry out multifactorial attempts.
In the MATLAB software, 70 percent of both normalized output and input data were entered
into the network as training and the remaining 30 percent entered as a test (the training is an
observer one). If the value of predicated output is close to the actual one entered into the
output part, indicate the ideal prediction of the network and ensure that the MSE is low and
R is high. Therefore, in this study, the predicted output of the network is close to the target.
The predicted values of Cc depend on the test results. One-dimensional consolidation was
done on twenty-four test samples at loading intensities of 50 kPa to 800 kPa. The compression
index (Cc) is one of the very important compressibility parameters in settlement estimation
for engineering design purposes. There were a lot of formulas that were developed by
different scholars (Ibrahim et al. 2012; Rashed, Salih, and Abdalla 2017; Giasi, Cherubini, and
Paccapelo 2003; Ng, Chew, and Lazim 2018) for the correlation between Cc and other soil
properties. However, most of the researchers recommended Cc = 0.009 (LL-20) (Vinod and
Bindu 2010); but the best correlation of compression index with Atterberg limits was
suggested by Giri (2019). The compression index (Cc) values of medium to soft soil are in the
U.Porto Journal of Engineering, 8:6 (2022) 102-120 103

ranges of 0.15 to 1.0 (Hermans and Irving 2017). The correlation results of the compression
index (Cc) for this study are between 0.227 and 0.33. Coefficient of compressibility (Cc) values
are found to be relatively almost similar to those obtained for undisturbed soil samples by
Jayalekshmi and Elamathi (2020) and according to Alptekin and Taga (2019), the soils are soft
clay that is highly compressible when comparison with the current study.
Hence, many researchers have used these approaches to determine the value of Cc from basic
soil properties (Yoon, Kim, and Jeon 2004; Dwivedi, Kumar, and Jain 2016; Onyejekwe, Kang,
and Ge 2015). They used ANN and regression analysis to predict the compression behavior of
normally consolidated fine-grained soil (Işik 2009). They concluded that, for predicting the
compression index from index soil properties: Liquid limit (LL), water content (w), and void
ratio (e), which are given as input data to provide good correlation models for different soil
and also a comparison with the existing Cc empirical equation (Tiwari and Ajmera 2012).
2. Materials and Methods
The study was conducted in Jimma town, which is located in southwest Ethiopia at a distance
of 335 km from the capital of Ethiopia, Addis Ababa. Jimma town is one of the special towns
of the Oromiya National Regional State and is surrounded by Jimma Zone. The representative
and purposive sampling techniques were used by selecting particular parameters to make sure
that the settings have specific characteristics as indicated for this study. The size of soil
collected is specified in the sample collection procedure according to the ASTM Standard Test
Manual. The sample collection procedure has been according to the ASTM Standard Test
Manual for all different types of laboratory tests. A series of Geotechnical laboratory tests,
including Atterberg Limits, Natural moisture content, particle size distribution (wet sieving),
and one-dimensional consolidation tests, have been conducted. The data processing and
analysis to be carried out in this study were presented and explained using tables, charts, and
graphs. The results of laboratory tests were analyzed and compared to the standards and
current specifications (Legget 1964) proposed by ASTM. The result obtained was organized
and interpreted using MS Excel and numerical modeling according to the established objective
and presented as a chart, table, and graph. There are two different approaches to Numerical
Modeling, namely Regression analysis and Artificial Neural Network (ANN), to obtain the
correlation between compression indexes (Cc) and liquid limit (LL). Similar approaches can
also be used to predict the correlation between compression index (Cc) and plasticity index
(PI).
In order to examine the data based on regression, storing and normalization of the data was
done by SPSS 17 (Namdarvand, Jafarnejadi, and Sayyad 2013) packages. The relation between
the compression coefficient and the other properties was carried out by stepwise regression.
To investigate being interdependent of the factors, the Pearson correlation coefficient was
measured and the properties which had a significant correlation were specified.
In neural network modeling, MATLAB software, as well as the multilayer perceptron (MLP)
model, was used. These networks have the potential to be evolved by input vectors and
include a series of sensory units (base neuron) comprising an input layer, one or more latent
layers and an output layer. The input signals are released as layer by layer through network
and follow a forward route (Sarmadian, Taghizadeh Mehrjardi, and Akbarzadeh 2009). It is
through trial-and-error method, that is, replacement of transfer functions and modification of
the number of neurons that the results can be obtained at best, having the highest correlation
coefficient (R) and the least mean square error (MSE).

3. Index Properties Test Results

3.1. Natural moisture content
For the correlation of soil characteristics within index properties, the natural moisture content
is the significant index soil property. The natural moisture content of the soil examined was
determined according to ASTM D2216-98 (ASTM 1998). The natural moisture content of the
soils in the study area is between 39.14% and 68.78%. Table 1 shows a test specimen
dehydrated at a temperature of 110° ± 5°C to a constant mass using oven-dried.
3.2. Specific gravity
The specific gravity of solids in a soil particle was defined as the ratio of the unit weight of
solid matter to the unit weight of water. The specific gravity of the solid is a measure of and a
means of expressing the heaviness (weight) of the material. By ASTM -D 854-06 (ASTM 2006),
two procedures were provided to perform specific gravity. These are Method A, procedures
for oven-dried samples, and Method B, procedures for wet samples. In this study, the specific
gravities are determined using Method-A procedures. Specific gravity test results are shown
in Table 1. The specific gravity of the solid of the light-colored sand, which is composed mainly
of quartz, can be estimated at 2.65: for clayey and silty soil, it can vary from 2.6-2.9
(Onyejekwe, Kang, and Ge 2015). Table 1 summarizes the specific gravity of the soil samples.
The results indicate that all soil samples conform to specifications. The specific gravity of clay
is between 2.60-2.76 and for silty it varies from 2.68-2.73. The specific gravity values showed
a variation within a limited range at different depths and different locations. The specific
gravity should be a lower value when a high organic content exists”, while the presence of
heavy minerals can lead to higher values.
Sample Moisture Specific Sample Moisture Specific Sample Moisture Specific
No content (%) Gravity No content (%) Gravity No content (%) Gravity
(Gs) (Gs) (Gs)
1 43.90 2.72 9 45.91 2.69 17 43.26 2.69
2 47.41 2.7 10 66.62 2.66 18 44.78 2.65
3 57.00 2.62 11 39.14 2.7 19 45.51 2.65
4 44.45 2.6 12 45.70 2.68 20 47.50 2.61
5 43.96 2.76 13 40.89 2.69 21 43.57 2.61
6 47.50 2.73 14 57.07 2.65 22 39.98 2.65
7 47.39 2.73 15 43.00 2.64 23 60.82 2.68
8 47.50 2.71 16 46.88 2.61 24 68.78 2.71
Table 1: Natural moisture content of soil samples
3.3. Atterberg limits
Atterberg limits were determined for air-dried samples. It was carried out based on the
standard according to ASTM (2000). Air-dried soil samples were prepared by spreading the
material out in trays and leaving them open in the air for at least 5-7 days until completely
dry. The room temperature was approximately 25º C. Atterberg limit values for soils in the
study area are summarized in Table 2.
The result of the Atterberg limit of soil sample used presented in Table 2 shows which were
determined by using Casagrande’s and plastic limit method is performed. From this, it is
observed that the liquid limit is between 58% and 106%; the plastic limit ranges from 24.8-
52.3%, and the plastic index is between 31% and 61%. The test outcome displays that the soils
in the study area are highly plastic with high plasticity index values.

Sample Liquid Plastic Plasticity Sample Liquid Plastic Plasticity

No Limit, LL Limit, PL Index, PI No Limit, LL Limit, PL Index, PI
(%) (%) (%) (%) (%) (%)
1 81.30 36.60 44.70 13 65.00 25.00 40.00
2 77.30 37.30 40.00 14 90.00 40.00 50.00
3 59.50 28.50 31.00 15 77.00 37.00 40.00
4 59.80 24.80 35.00 16 75.60 33.60 42.00
5 61.00 30.00 31.00 17 65.70 30.70 35.00
6 60.00 29.00 31.00 18 83.50 38.50 45.00
7 79.70 44.70 35.00 19 67.40 31.40 36.00
8 87.00 46.00 41.00 20 79.20 32.20 47.00
9 67.00 32.50 34.50 21 70.20 33.20 37.00
10 72.30 34.80 37.50 22 71.00 32.00 39.00
11 63.00 32.00 31.00 23 101.00 52.30 48.70
12 58.00 27.00 31.00 24 106.00 45.00 61.00
Table 2: Atterberg limits of soils of the study area
3.4. Particle size analysis
The test was conducted according to ASTM D422-63(2007) (ASTM 2007) particle distribution
finer than 75μm can be done by hydrometer test and courser than 75μm can be done by
mechanical sieve. Therefore, the samples collected from the site were first air-dried and a
representative sample was taken by quartering. The weight of the sample was measured and
then washed in sieve No.200. The mechanical sieve was performed on soil samples retained
by No. 200 sieve after oven drying for 24 hours. In the hydrometer test, 50 g of soil has taken
and soaked for 24 hours by adding a sodium hexametaphosphate that is used as a dispersing
agent. Then, soaking the sample was further dispersed using a stirring apparatus. Then it was
poured into a 1000 ml cylinder and mixed again for one minute by covering it with the palm
(glove). The hydrometer reading and the test temperature were taken for 1, 2, 4, 8, 15, 30, 60,
120, 240, 480 and 1440 minutes. The summarized combined grain size and hydrometer
analysis curves are shown in Figure 1. From the particle size results, it was observed that there
are several particle size variations. Grain size analysis yielded a clay content ranging from 6.38-
22.97%, silt fraction 74.98-90.86%, sand fraction 0.67-12.89% and gravel content from 0.0 –
0.41%.
Sample 1
Grain size distribution curve Sample 2
Sample 3
100 Sample 4
Percentage Passing (%)
Sample 5
80 Sample 6
Sample 7
Sample 8
60 Sample 9
Sample 10
40 Sample 11
Sample 12
20 Sample 13
Sample 14
0 Sample 15
Sample 16
10 1 0,1 0,01 0,001 Sample 17
Sample 18
Particle size (mm) Sample 19
Sample 20
Figure 1: Summary of combined grain size distribution curves from sieve and
hydrometer analysis

3.5. Soil classification

The soils under investigation have been classified according to USCS (Das and Sobhan 2012)
and AASHTO (2000). For both, the classification test results are summarized in the following
figures. Most of the soil of the study area falls in Highly plastic clay (CH) and three soil samples
fall in the MH region as USCS classification scheme (Das and Sobhan 2012). From visual
observations and field tests, the soils of the study area are classified as clay with high plasticity.
The soils are classified as CH or MH (clay with high plasticity, clay with high elasticity)
conducted as USCS. According to AASHTO classification, the soil is classified as A-7-5 and A-7-
6, which are clayey soils.
Sample 1
Plasticity Chart Sample 2
70 Sample 3
Sample 4
Sample 5
60 Sample 6
Sample 7
Plasticity Indecx, PI (%)
50 Sample 8
Sample 9
Sample 10
40 Sample 11
Sample 12
30 Sample 13
Sample 14
Sample 15
20 Sample 16
Sample 17
10 Sample 18
Sample 19
Sample 20
0 Sample 21
0 10 20 30 40 50 60 70 80 90 100 110 120 Sample 22
Sample 23
Liquid Limit, LL (%) Sample 24
Figure 2: Unified soil classification systems of the soil in the study area using a
plasticity chart
Plasticity Chart Sample 1

70 Sample 2
Sample 3
Sample 4
60 Sample 5
Sample 6
50 Sample 7
Plasticity Indecx, PI (%)
Sample 8
Sample 9
40 Sample 10
Sample 11
Sample 12
30 Sample 13
Sample 14
20 Sample 15
Sample 16
Sample 17
10 Sample 18
Sample 19
Sample 20
0 Sample 21
0 10 20 30 40 50 60 70 80 90 100 110 120 Sample 22
Sample 23
Liquid Limit, LL (%) Sample 24
Figure 3: Plasticity chart of soil in the study area according to AASHTO system of
classification
From the plot of the plasticity chart in Figure 2, the soils found in Jimma town are highly plastic
clay except for samples 7, 8, and 23, which are highly elastic. Figure 3 also shows that most

soil samples of the study area fall in the A-7-5 subgroup, in which the plasticity index is equal
to or less than the liquid limit minus 30 (LL-30) and below the A-line. In addition, some soil
samples fall in the A-7-6 subgroup, in which the plasticity index is greater than the liquid limit
minus 30 and above or equal to the A-line.
3.6. Consolidation test
Consolidation properties of soils test were done on the ASTM standard, Designation (ASTM
2011). A small representative undisturbed soil sample is carefully trimmed and fitted into the
rigid metal ring. The specimen was mounted on a bottom porous stone base, and a similar
porous stone was placed on top to permit water, which has squeezed out of the sample, to
escape freely at the top and bottom and a sitting load of 7 kPa was applied. A consolidation
test was carried out to study the compressibility of the soil using the apparatus called
oedometer. Diameters of 50 mm soil samples having a height of 20 mm were loaded from 50
kPa to 800 kPa by doubling the loading. The loads were doubled every 24 hours, starting from
50 kPa to 800 kPa. Similarly, for all samples, the procedure is followed.
The undisturbed soil samples are performed for all the twelve test pits; twenty-four
oedometer tests (i.e., 1.5 m and 3.0 m depths) are conducted. Then, the compression index
(Cc), which is used in establishing correlations with index properties and reconsolidation
pressure (Pc), is determined from the consolidation test. Results of compressibility parameters
for the twenty-four undisturbed soil samples collected from test pits around Jimma town are
summarized in Table 3. The results revealed that clays in the study area are highly plastic with
a marginal degree of expansion. They also have relatively moderate to high free swell values.
Correlation between the results of the twelve stations and collected data shows that the Cc
values, in general, increase with increasing liquid limit and plastic index of soil. This serves to
suggest that soil compressibility generally increases with plasticity and vice versa. Generally,
void ratios for all samples were reduced to lower values since the increasing intensity of
loadings at each step of loading brought soil grains closely to each other. The test results of
the collected data test in the study area are conducted for the oedometer test. The
compression index (Cc) range is 0.227 to 0.33 and the swelling index (Cs) lies in a range of 0.02
to 0.11.
Sample Compression Swelling index Sample Compression Swelling index

No index (Cc) (Cs) No index (Cc) (Cs)
1 0.287 0.02 13 0.256 0.11
2 0.274 0.06 14 0.305 0.05
3 0.232 0.05 15 0.296 0.04
4 0.227 0.07 16 0.294 0.04
5 0.248 0.03 17 0.26 0.04
6 0.244 0.03 18 0.29 0.04
7 0.286 0.03 19 0.264 0.07
8 0.31 0.03 20 0.324 0.07
9 0.264 0.06 21 0.267 0.02
10 0.243 0.05 22 0.272 0.02
11 0.252 0.04 23 0.321 0.06
12 0.241 0.04 24 0.33 0.07
Table 3: Results of Consolidation Test for twenty-four soil samples for the study
area

4. Numerical Modelling
Artificial neural networks and regression analysis are the two approaches to get the
correlation between compression index (Cc) and soil index properties (Nesamatha and
Arumairaj 2015). Determination of the parameter model needs various premises, which the
residuals (observed value small estimated values) corresponding to different observations are
uncorrelated random variables with zero mean and constant variance (Danial
Mohammadzadeh et al. 2019). In addition, one assumes that the order of the model is correct;
that is, if one fits a simple linear regression model, one is assuming that the phenomenon
behaves in a linear or first-order manner (Chen 1975). During regression analysis, a regression
model with a higher value of R2 (coefficient of determination), which quantifies the proportion
of the variance of one variable by the other, is usually accepted. In this study, the compression
index (Cc) is the dependent variable, whereas the LL, PL, and PI are regressor variables. To
carry out statistical analysis, Microsoft® Excel and SPSS-20 (Gabrosek 2013) were used.
Twenty-four numbers of samples are used in correlating Cc with LL, PL, and PI, and seven
numbers of secondary samples are used in correlating Cc with LL, PL, and PI. While carrying
out the statistical analysis, different regression models are used and those models with a
higher value of the coefficient of determination (R2) are accepted.
4.1. Scatter plot
In developing correlations, the first step is creating a scatter plot of the data to visually assess
the strength and form of the relationship. In Figure 4, the scatter plot of Cc with LL, PL, and PI
is presented.
Cc vs LL
0,36
0,34 y = 0,0021x + 0,1196
0,32 R² = 0,82
0,3
0,28
Cc
0,26
0,24
0,22
0,2
50 60 70 80 90 100 110
LL
(a)
Cc vs PL Cc vs PI
y = 0,0036x + 0,1342
0,36 y = 0,0034x + 0,1573 0,32 R² = 0,6059
R² = 0,6135
0,34 0,3
0,32
0,28
0,3
Cc
0,26
Cc
0,28
0,26 0,24
0,24 0,22
0,22 0,2
0,2 30 35 40 45 50 55
20 40 60 PI
PL
(b) (c)
Figure 4: (a), (b), (c): Scatter plot of Cc versus LL, PL and PI

From Figure 4, the available test points are not sufficient to give reliable relationships between
the independent and dependent variables. Nevertheless, different models (linear and non-
linear) have been employed to examine the trend of the scatter. Figure 4(a) shows that the
compression index is increased with the increment of liquid limit and has a good relationship
between Cc and LL as R2 = 0.84. Figure 4(c) indicates the compression index is increased with
the plasticity index increased and has a positive relationship between Cc and PI as the value
of R2 = 0.6059. Figure 4(b) shows that the compression index is increased with the increment
of plastic limit, but it is not an intrinsic variable on Cc value as LL and PI. Three different values
of compression index predicted when liquid limit, plastic limit, and plasticity index are given
as input are shown in Table 4.
Sample Cc Pred. Cc Pred. Cc Pred. Cc Sample Cc Pred. Cc Pred. Cc Pred. Cc
No actual LL as PL as PI as No actual LL as PL as PI as
input input input input input input
1 0.287 0.290 0.281 0.293 13 0.241 0.241 0.248 0.246
2 0.274 0.281 0.283 0.277 14 0.256 0.256 0.242 0.277
3 0.232 0.244 0.253 0.246 15 0.305 0.308 0.292 0.311
4 0.227 0.245 0.241 0.260 16 0.296 0.281 0.282 0.277
5 0.248 0.247 0.258 0.246 17 0.294 0.278 0.271 0.284
6 0.244 0.245 0.255 0.246 18 0.26 0.257 0.261 0.260
7 0.286 0.286 0.308 0.260 19 0.29 0.294 0.287 0.294
8 0.31 0.302 0.312 0.280 20 0.264 0.261 0.263 0.263
9 0.264 0.260 0.267 0.258 21 0.324 0.285 0.266 0.301
10 0.243 0.271 0.275 0.268 22 0.267 0.266 0.269 0.267
11 0.252 0.251 0.265 0.246 23 0.272 0.268 0.265 0.273
12 0.287 0.290 0.281 0.293 24 0.321 0.331 0.334 0.307
Table 4: Predicted values of compression index (Pred. Cc) using regression model
4.2. Summary of correlation between Cc and LL, PL and PI for soils of Jimma
A linear model is used. The correlation coefficient with least standard error is considered, and
the following relationships are found in Table 5.
Ser. Model equation R2 (coefficient of No. of
No. determination) samples, n
1 Cc = 0.164ln(LL-0.4296) 0.8418 24
2 Cc = 0.1233ln (PL) -0.161 0.6282 24
3 Cc = 0.145ln (PI) – 0.2556 0.7395 24
Table 5: Summary of the regression analysis by Microsoft excel and SPSS-20
4.3. Artificial Neural Network (ANN)
Atterberg limits that are detected (i.e., LL, PL, and PI) are given as input and the observed
compression index is given as a target for the ANN modeling. After the values are given as
input, the training process will take place in the developed hidden layer to predict the R-value.
From the trained results, the output values are obtained. By changing the number of the
hidden layer, the R-value changes. In this training, the maximum R-value is obtained by giving
20 hidden layers for the ANN modeling. The correlation coefficient value (R 2) obtained from
the output is 0.961.
Three different types of predicted values of compression index (Cc) using regression model
when liquid limit, plastic limit as input, and plasticity index as input are shown in Table 6. The
output of the software after the training is shown in Figure 5.

Figure 5: Output after training
Sample Cc Pred. Cc Pred. Cc Pred. Cc Sample Cc Pred. Cc Pred. Cc Pred. Cc

No actual LL as PL as PI as No actual LL as PL as PI as
input input input input input input
1 0.287 0.289 0.293 0.285 13 0.241 0.258 0.247 0.278
2 0.274 0.294 0.273 0.278 14 0.256 0.306 0.294 0.305
3 0.232 0.236 0.234 0.243 15 0.305 0.294 0.329 0.278
4 0.227 0.237 0.248 0.252 16 0.296 0.289 0.275 0.295
5 0.248 0.244 0.246 0.243 17 0.294 0.260 0.263 0.252
6 0.244 0.237 0.232 0.243 18 0.26 0.292 0.295 0.289
7 0.286 0.290 0.288 0.252 19 0.29 0.269 0.253 0.263
8 0.31 0.310 0.330 0.308 20 0.264 0.291 0.297 0.322
9 0.264 0.267 0.293 0.252 21 0.324 0.267 0.281 0.260
10 0.243 0.254 0.261 0.245 22 0.267 0.261 0.262 0.272
11 0.252 0.254 0.262 0.243 23 0.272 0.324 0.323 0.321
12 0.287 0.234 0.240 0.243 24 0.321 0.329 0.330 0.330
Table 6: Predicted values of compression index using ANN Model

4.4. Correlations from the Artificial Neural Network

Figure 6 shows the relation between the observed and predicted compression index with
liquid limit as input.
Figure 6: Predicted and observed compression index from ANN model using liquid
limit (LL). A relationship between compression index and liquid limit was arrived
based on the above plot

Figure 7 shows the relation between the observed and predicted compression index with
plastic limit as input.
Figure 7: Predicted and observed compression index from ANN model using plastic
limit (PL)

A relationship between the compression index and liquid limit was arrived based on the above
plot. Figure 8 shows the relation between the observed and predicted compression index with
the plasticity index as input.
Figure 8: Predicted and observed compression index from ANN model using
plasticity index (PI)

A relationship between the compression index and plasticity index was arrived based on the
above plot. The predicted Cc vs. actual Cc that was obtained through ANN trained model is
shown in Figure 9.
Figure 9: The predicted compression index versus the observed compression index
by ANN trained model

4.5. Comparison of predicted Cc by Microsoft Excel, SPSS-20 and MATLAB-ANN

In this study the value of Cc found from Oedometer tests and the predicted Cc are almost
similar as it is shown in Table 7 below.
Ser. Laboratory RA MLP ANN Ser. Laboratory RA MLP ANN
No test No test
1 0.287 0.291 0.290 0.289 13 0.256 0.259 0.256 0.249
2 0.274 0.281 0.274 0.295 14 0.305 0.309 0.330 0.305
3 0.232 0.244 0.248 0.241 15 0.296 0.280 0.274 0.293
4 0.227 0.247 0.256 0.230 16 0.294 0.279 0.294 0.295
5 0.248 0.246 0.248 0.247 17 0.26 0.257 0.264 0.258
6 0.244 0.244 0.248 0.243 18 0.29 0.295 0.290 0.291
7 0.286 0.282 0.274 0.288 19 0.264 0.261 0.264 0.267
8 0.31 0.298 0.274 0.329 20 0.324 0.288 0.330 0.323
9 0.264 0.259 0.264 0.267 21 0.267 0.266 0.264 0.267
10 0.243 0.270 0.264 0.243 22 0.272 0.269 0.272 0.257
11 0.252 0.250 0.264 0.255 23 0.321 0.328 0.290 0.321
12 0.241 0.241 0.244 0.236 24 0.33 0.344 0.330 0.325
Table 7: Comparison of the value of Cc found from the oedometer test and
predicted Cc by LR (Microsoft Excel), NN- MLP (SPSS-20) and MATLAB-ANN analysis
From Table 7 and shown in Figure 10, the accuracy of the present ANN model was checked by
comparing the laboratory values of Cc with predicted values of Cc. It was found that the mean
target value for input data was 0.274, whereas the mean target value for ANN output was
0.276. But the mean target value using regression analysis and Multilinear perception is
similar, which is 0.275. The accuracy of the proposed model was also checked by calculating
the correlation coefficient (R-value) and it was found that the ANN model was 0.944, whereas
for models proposed using regression analysis and Multilinear perception were 0.822 and
0.742. It was found that compression index values predicted using the ANN model have better
distribution around the trend line in comparison to another model (Nesamatha and Arumairaj
2015). It can be calculated that the proposed models are much more accurate and are in good
agreement with laboratory values.
ANN Model
0,34
0,32
Predicted Cc value
y = 1,006x - 0,0005
0,3
R² = 0,9434
0,28
0,26
0,24
0,22
0,2
0,2 0,22 0,24 0,26 0,28 0,3 0,32 0,34
Actal Cc value
(a)

RA Analysis
0,36 y = 0,8203x + 0,0494
0,34 R² = 0,8221
Predicted Cc value
0,32
0,3
0,28
0,26
0,24
0,22
0,2
0,2 0,22 0,24 0,26 0,28 0,3 0,32 0,34
Actal Cc value
(b)
MLP Analysis
0,34 y = 0,7315x + 0,0745
R² = 0,7418
0,32
Predicted Cc value
0,3
0,28
0,26
0,24
0,22
0,2
0,2 0,22 0,24 0,26 0,28 0,3 0,32 0,34
Actal Cc value
(c)
Figure 10: (a), (b) and (c): The plot of different mathematical model for relationship
between predicted and observed target values of the study
Ser. Model equation R2 (coefficient of No. of Mathematical model used

No. determination) samples, n
1 Cc = 0.164ln (LL) - 0.4296 0.8418 24 Regression analysis (LR)
2 Cc = 0.1233ln (PL) -0.161 0.6282 24 Regression analysis (LR)
3 Cc = 0.145ln (PI) – 0.2556 0.7395 24 Regression analysis (LR)
4 Cc = 0.1637ln (LL) - 0.4283 0.939 24 Artificial neural network (ANN)
5 Cc = 0.1637ln (LL) - 0.4283 0.696 24 Artificial neural network (ANN)
6 Cc = 0.149ln (PI) - 0.2719 0.869 24 Artificial neural network (ANN)
Table 8: R2-Value for Regression and ANN Model
Table 8 shows the variation of R2 for both Regression analysis (LR) and Artificial neural network
(ANN). ANN model shows more goodness of fit and it has higher reliability than the regression
analysis from these two correlations. Figure 11 shows a comparison of different mathematical
models in one plot.

Figure 11: Comparison of different mathematical models in one plot
5. Conclusions
The main objective of this paper was to obtain valid relationships between index properties
and compression index of Jimma clay soil. However, from regression analysis, Cc has a very
strong correlation with Atterberg limits by achieving a coefficient of determination of 84%,
63%, and 74% using ANN model, Regression and analysis and MLP, respectively, while
parameters like void ratio, dry unity weight, and natural moisture content have little influence
on the compression index in this study. The R2 (correlation coefficient) of the ANN model for
LL, PL, and PI is 93%, 69%, and 86%, respectively, indicating that the Cc has a very strong
correlation with Atterberg limits. In this study, the ANN model is the best fit to achieve a
greater R-value than the regression analysis model. Therefore, the compression index can be
computed from the known value of LL, PL, and PI by the correlation equations.
From the statistical analysis and ANN analysis, one observes a relatively good indicative
correlation between Cc and liquid limit (LL), Cc and plastic limit (PL), and Cc and plasticity index
(PI), while parameters like void ratio, dry unity weight, and natural moisture content have little
influence on the compression index in this study. From the developed correlations, one would
be in a position to determine the compression index from the index properties for undisturbed
soil samples of Jimma town. The compression index is influenced by LL and PI for this category;
Cc can be estimated from the oedometer by:
𝐶𝑐 = 0.1641𝑛 (𝐿𝐿) − 0.4296, 𝑅2 = 0.84, 𝑛 = 24, 𝑎𝑛𝑑 𝐶𝑐 = 0.145𝑙𝑛 (𝑃𝐼) – 0.2556, 𝑅2
(1)
= 0.74, 𝑛 = 24
The corresponding settlement equation of:
𝐶𝑐 𝑙𝑜𝑔10 ((𝜎′𝑣+∆𝜎′𝑣)/𝜎′𝑣)
𝑆𝑐 = 𝐻𝑜 + (2)
1+𝑒𝑜
In this study, ANN’s (curve fitting) practice has been made to predict the compression index
based on the geotechnical characteristics of different test pits data collected from Jimma
town. ANN is a powerful tool in predicting the consolidation parameters and the best fit model
than conventional methods are obtained. In the proposed ANN model, the soil properties,
such as the liquid limit, plastic limit, and plasticity index, are input parameters. The proposed
model of the ANN results compared with the experimental values and the predicted
compression index values have been found close to the experimental values. In this research,
the observed compression index is performed by ANN proposed model to obtain the predicted
Cc.
𝐶𝑐 = 0.1637𝑙𝑛 (𝐿𝐿) − 0.4283, 𝑅2 = 0.939 𝑎𝑛𝑑 𝐶𝑐 = 0.149𝑙𝑛 (𝑃𝐼) − 0.2719, 𝑅2 = 0.869. (3)

By engineer judgment, one of the formulae given above may be used for computing Cc due to
the absence of consolidation test data.
References
AASHTO. 2000. Classification of soil and soil-aggregate mixtures for highway construction
purposes. AASHTO M-145-91. American Association of State Highway and Transportation
Officials.
Abbasi, N., A. A. Javadi, and R. Bahramloo. 2012. "Prediction of compression behaviour of
normally consolidated fine-grained soils". World Applied Sciences Journal 18, no. 1: 6-14.
https://doi.org/10.5829/idosi.wasj.2012.l8.01.2675.
Akayuli, C. F. A., and B. Ofosu. 2013. "Empirical model for estimating compression index from
physical properties of weathered Birimian phyllites". Electronic Journal of Geotechnical
Engineering 18 Z: 6135-44.
Alptekin, A., and H. Taga. 2019. "Prediction of compression and swelling index parameters of
quaternary sediments from index tests at mersin district". Open Geosciences 11, no. 1: 482-
91. https://doi.org/10.1515/geo-2019-0038.
ASTM International. 1998. Standard test method for laboratory determination of water
(moisture) content of soil and rock by mass. ASTM D2216-98. ASTM International.
https://doi.org/10.1520/D2216-98.
———. 2000. Standard test methods for liquid limit, plastic limit, and plasticity index of soils.
ASTM D4318-00. ASTM International. https://doi.org/10.1520/D4318-00.
———. 2006. Standard test methods for specific gravity of soil solids by water pycnometer.
ASTM D854-06e1. ASTM International. https://doi.org/0.1520/D0854-06E01.
———. 2007. Standard test method for particle-size analysis of soils. ASTM D422-63(2007).
ASTM International. https://doi.org/10.1520/D0422-63R07.
———. 2011. Standard test method for one-dimensional consolidation properties of soils.
ASTM D2435-96. ASTM International. https://doi.org/10.1520/D2435_D2435M-11.
Chen, F. H. 1975. Foundations on Expansive Soils. Developments in Geotechnical Engineering.
https://doi.org/10.1016/B978-0-444-41393-2.50002-8.
Danial Mohammadzadeh, S., S. F. Kazemi, A. Mosavi, E. Nasseralshariati, and J. H. M. Tah.
2019. "Prediction of compression index of fine-grained soils using a gene expression
programming model". Infrastructures 4, no. 2. Article number infrastructures4020026.
https://doi.org/10.3390/infrastructures4020026.
Das, B. M., and K. Sobhan. 2012. Principles of Geotechnical Engineering. 8th ed. Cengage
Learning.
Dwivedi, P., R. Kumar, and P. K. Jain. 2016. "Prediction of compression index (Cc) of fine grained
remolded soils from basic soil properties". International Journal of Applied Engineering
Research 11, no. 1: 592-98.
Gabrosek, J. 2013. SPSS manual for introductory applied statistics: A variable approach. John
Gabrosek.
Giasi, C. I., C. Cherubini, and F. Paccapelo. 2003. "Evaluation of compression index of
remoulded clays by means of Atterberg limits". Bulletin of Engineering Geology and the
Environment 62, no. 4: 333-40. https://doi.org/10.1007/s10064-003-0196-3.
Giri, S. 2019. "Correlation of compression index (Cc) with liquid limit (LL) and plasticity index
(Ip) for white soil deposit (CL, low plastic inorganic clays)". Journal of Advances in
Geotechnical Engineering 2, no. 1: 1-8. https://doi.org/10.13140/RG.2.2.33725.69605.

Hermans, T., and J. Irving. 2017. "Facies discrimination with electrical resistivity tomography
using a probabilistic methodology: Effect of sensitivity and regularization". Near Surface
Geophysics 15: 13-25. https://doi.org/10.3997/1873-0604.2016047.
Ibrahim, N. M., N. L. Rahim, R. C. Amat, S. Salehuddin, and N. A. Ariffin. 2012. "Determination
of plasticity index and compression index of soil at perlis". APCBEE Procedia 4: 94-98.
https://doi.org/10.1016/j.apcbee.2012.11.016.
Işik, N. S. 2009. "Estimation of swell index of fine grained soils using regression equations and
artificial neural networks". Scientific Research and Essays 4, no. 10: 1047-56.
Jain, V. K., M. Dixit, and R. Chitra. 2015. "Correlation of plasticity index and compression index
of soil". International Journal of Innovations in Engineering and Technology 5, no. 3: 263-
70.
Jayalekshmi, S., and V. Elamathi. 2020. "A review on correlations for consolidation
characteristics of various soils". IOP Conference Series: Materials Science and Engineering
1006: Article number 012007. https://doi.org/10.1088/1757-899X/1006/1/012007.
Legget, R. F. 1964. "American society for testing and materials". Nature 203, no. 4945: 565-
68. https://doi.org/10.1038/203565a0.
Namdarvand, F., A. Jafarnejadi, and G. Sayyad. 2013. "Estimation of soil compression
coefficient using artificial neural network and multiple regressions ". International Research
Journal of Applied and Basic Sciences 4, no. 10: 3232-36.
Nesamatha, R., and P. D. Arumairaj. 2015. "Numerical modeling for prediction of compression
index from soil index properties". IOSR Journal of Mechanical and Civil Engineering 12, no.
3: 68-76. https://doi.org/10.9790/1684-12316876.
Ng, K. S., Y. M. Chew, and N. I. A. Lazim. 2018. "Prediction of consolidation characteristics from
index properties". In E3S Web of Conferences, Article Number 06004.
https://doi.org/10.1051/e3sconf/20186506004.
Onyejekwe, S., X. Kang, and L. Ge. 2015. "Assessment of empirical equations for the
compression index of fine-grained soils in Missouri". Bulletin of Engineering Geology and
the Environment 74, no. 3: 705-16. https://doi.org/10.1007/s10064-014-0659-8.
Rashed, K. A., N. B. Salih, and T. A. Abdalla. 2017. "Correlation of consistency and
compressibility properties of soils in Sulaimani city". Sulaimani Journal for Engineering
Sciences 4, no. 5: 86-94. https://doi.org/10.17656/sjes.10061.
Sarmadian, F., R. Taghizadeh Mehrjardi, and A. Akbarzadeh. 2009. "Modeling of some soil
properties using artificial neural network and multivariate regression in Gorgan Province,
North of Iran". Australian Journal of Basic and Applied Sciences 3, no. 1: 323-29.
Tiwari, B., and B. Ajmera. 2012. "New correlation equations for compression index of
remolded clays". Journal of Geotechnical and Geoenvironmental Engineering 138, no. 6:
757-62. https://doi.org/10.1061/(ASCE)GT.1943-5606.0000639.
Vinod, P., and J. Bindu. 2010. "Compression index of highly plastic clays–an empirical
correlation". Indian Geotechnical Journal 40, no. 3: 174-80.
Yoon, G. L., B. T. Kim, and S. S. Jeon. 2004. "Empirical correlations of compression index for
marine clay from regression analysis". Canadian Geotechnical Journal 41, no. 6: 1213-21.
https://doi.org/10.1139/T04-057.
Zumrawi, M. M. E. 2012. "Prediction of swelling characteristics of expansive soils". Sudan
Engineering Society Journal 58, no. 2: 55-62.

Numerical Stress Strain Curve Oedometer

Uploaded by

Copyright:

Available Formats

Numerical Stress Strain Curve Oedometer

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Numerical Stress Strain Curve Oedometer

Uploaded by

Copyright:

Available Formats

U.

Porto Journal of Engineering, 8:6 (2022) 102-120 Received: 29 March, 2022

Numerical Modelling for Prediction of Compression Index

U.Porto Journal of Engineering, 8:6 (2022) 102-120 103

U.Porto Journal of Engineering, 8:6 (2022) 102-120 104

3. Index Properties Test Results

U.Porto Journal of Engineering, 8:6 (2022) 102-120 105

Sample Liquid Plastic Plasticity Sample Liquid Plastic Plasticity

U.Porto Journal of Engineering, 8:6 (2022) 102-120 106

3.5. Soil classification

Plasticity Chart Sample 1

U.Porto Journal of Engineering, 8:6 (2022) 102-120 107

Sample Compression Swelling index Sample Compression Swelling index

U.Porto Journal of Engineering, 8:6 (2022) 102-120 108

U.Porto Journal of Engineering, 8:6 (2022) 102-120 109

U.Porto Journal of Engineering, 8:6 (2022) 102-120 110

Figure 5: Output after training

Sample Cc Pred. Cc Pred. Cc Pred. Cc Sample Cc Pred. Cc Pred. Cc Pred. Cc

U.Porto Journal of Engineering, 8:6 (2022) 102-120 111

4.4. Correlations from the Artificial Neural Network

U.Porto Journal of Engineering, 8:6 (2022) 102-120 112

U.Porto Journal of Engineering, 8:6 (2022) 102-120 113

U.Porto Journal of Engineering, 8:6 (2022) 102-120 114

U.Porto Journal of Engineering, 8:6 (2022) 102-120 115

4.5. Comparison of predicted Cc by Microsoft Excel, SPSS-20 and MATLAB-ANN

U.Porto Journal of Engineering, 8:6 (2022) 102-120 116

Ser. Model equation R2 (coefficient of No. of Mathematical model used

U.Porto Journal of Engineering, 8:6 (2022) 102-120 117

Figure 11: Comparison of different mathematical models in one plot

U.Porto Journal of Engineering, 8:6 (2022) 102-120 118

U.Porto Journal of Engineering, 8:6 (2022) 102-120 119

U.Porto Journal of Engineering, 8:6 (2022) 102-120 120

You might also like