Download as pdf or txt
Download as pdf or txt
You are on page 1of 27

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

net/publication/322499893

Evaluation of Shear Strength of Cohesionless Soil from Maximum, Minimum


Dry Density and Fines Content using Polynomial Surface Fitting Method

Article in Electronic Journal of Geotechnical Engineering · January 2018

CITATIONS READS

14 2,146

4 authors:

Md. Enayet Chowdhury Md Azijul Islam


Bangladesh University of Engineering and Technology University of Texas at Arlington
16 PUBLICATIONS 83 CITATIONS 61 PUBLICATIONS 380 CITATIONS

SEE PROFILE SEE PROFILE

Tahsina Islam Nafeesa Khan

12 PUBLICATIONS 79 CITATIONS
Southern Illinois University Carbondale
3 PUBLICATIONS 23 CITATIONS
SEE PROFILE
SEE PROFILE

All content following this page was uploaded by Md Azijul Islam on 30 January 2018.

The user has requested enhancement of the downloaded file.


Evaluation of Shear Strength of
Cohesionless Soil from Maximum,
Minimum Dry Density and Fines
Content using Polynomial Surface
Fitting Method
Md. Enayet Chowdhury1, Md. Azijul Islam2*, Tahsina Islam3,
Nafeesa Khan4
1Undergraduate Research Assistant, Department of Civil Engineering,

Bangladesh University of Engineering and Technology (BUET), Dhaka-1000, Bangladesh;


email: enayet108@gmail.com
2*Lecturer, Department of Civil Engineering,
Bangladesh University of Engineering and Technology (BUET), Dhaka-1000, Bangladesh;
*Corresponding author; email: azijul@ce.buet.ac.bd
3Graduate Research Assistant, Department of Civil Engineering,

Bangladesh University of Engineering and Technology (BUET), Dhaka-1000, Bangladesh;


email: tahsinabini@yahoo.com
4Lecturer, Department of Civil Engineering,
University of Asia Pacific (UAP), Dhaka-1215, Bangladesh;
email: nafeesa14.khan@gmail.com

ABSTRACT
This study establishes a mathematical relationship between angle of internal friction, fines content and
maximum or minimum dry density for non-plastic fines. Sand samples of three different fineness
moduli (1.3, 1.5 and 1.7) were used and six different percentages (0, 3, 5, 8, 10, 15% by weight) of
fines content were simulated for each fineness modulus. Direct Shear Test, Maximum Index Density
Test and Minimum Index Density Test were performed on the corresponding samples. Results from
laboratory tests were used as numerical inputs in the software, assigned in three distinct variables.
Among 120 completely different combinations with respect to degrees of independent variables and
variables allocation in X, Y and Z data, some combinations were chosen to validate with data available
in the existing literature on the basis of goodness of fit parameters. Among them, two equations
(generated from polynomial surface fitting) of third-order polynomial mostly converged with the
validation data. It seems that the proposed equations can perform satisfactorily to estimate any of the
variables considered when the other two are known.
KEYWORDS: Shear strength, fine content, polynomial surface fitting, Levenberg-
Marquardt Algorithm.

INTRODUCTION
Shear strength of a soil mass is the internal resistance per unit area that the soil mass can
offer to resist failure along any plane inside it (Das, 1990). The soil shear strength

- 31 -
Vol. 23 [2018], Bund. 1 32

is to some extent surface dependent, any action that will hinder or promote the interlocking
of soil particles will invariably affect soil shear strength (Ayininuola, 2009).
A review of the studies performed by Das (1990), Murthy et al. (2007), Shahriar (2016) reveals
that fines content affects the engineering behavior of soil. Therefore, research interest to understand
the impact of fines content on the physical properties of soil seems to increase subsequently. Murthy
et al. (2007) indicated that the Ottawa sand with non-plastic fines (5, 10, and 15%) showed a more
contractive and collapse tendency. Nevada sand with non-plastic fines content (10, 20, and 30%) also
showed a more contractive behavior than the clean sand in both drained and undrained monotonic
compression test (Lade and Yamamuro 1998). Thevanayagam and Mohan (2000) reported that the
shear stress at undrained peak state was smaller for the soil with plastic fines than that for the clean
sand at the same void ratio when the fines content was smaller than 30%. A reduction of shear stress
at the quasi-steady state caused by an addition of plastic fines (10 and 20%) was reported by Ni et al.
(2004). The instability zone became larger with the increase in plastic fines content up to 10%,
whereas it decreases when the fines content is greater than 20% (Abedi and Yasrobi, 2010). Here the
instability zone is defined as the zone between the line connecting the origin to the undrained peak
state and the line connecting the origin to the phase transformation state in the effective stress plane
for cohesionless soil. With the increase in fines content; angle of internal friction generally decreases
(Islam T. et al., 2017).
In a different study, Gupta and Trivedi (2009) showed that for non-plastic fines, maximum and
minimum void ratios of clean sand decreases as fines content increases from 0 to 20% and increases if
fines content exceeds 20%. Results also indicate that angle of internal friction and bearing capacity
decreases on the addition of fines due to compressibility of fines. Townsend (1973) presented
comparisons of vibrated density and standard compaction tests on sands with varying amounts of
fines. Similar densities were produced by impact and vibratory compaction at higher percentage of
fines added to the well graded sand compared to the percent fines added to the uniform sand.
Ojha and Trivedi (2013) showed that for silty sand, maximum and minimum void ratio decreases with
increase in silt content and critical angle of internal friction increases with increase in percent of silt.
Prasad and Pandey (2013) studied the effects of fines on void ratio of sands and their investigation
showed that both maximum dry density and void ratio increased up to 30 % with the increase of fines,
however, further increment of fines led to decrease of the parameters. The decrease of void ratio was
from 0.7 to 0.24 up to 30% fines content. Ayodele (2008) studied the effect of fines content on the
performance of soil as sub-base material for road construction.
Moreover, Meyerhof (1956) suggested that the relative density (Dr) and angle of internal friction
(𝜑) of granular soils or sands can be estimated as:
𝜑 = 30 + 0.15𝐷𝑟 When, silt > 5% (1)
𝜑 = 25 + 0.15𝐷𝑟 When, silt < 5% (2)
The relative density estimated by Meyerhof′s method (1957) was in the range
of +15% to -45% of the experimental results. Meyerhof′s method (1957) was modified by
Tokimatsu and Yoshimi (1983) by considering the effect of fines content on the SPT N-value. The
relative density estimated by Tokimatsu and Yoshimi′s method (1983) was in the
range of +25%to-20% of the experimental results. Thus the underestimation of relative density
of Meyerhof′s method (1957) was modified. Hatanaka et. al. (2006) and Mujtaba et. al. (2006) also
estimated relative density of sandy soils and friction angle based on SPT-N value.
Sinha and Wang (2008) developed Artificial Neural Network (ANN) prediction models for soil
compaction and permeability. The test soils were prepared from bentonite, limestone dust, sand and
gravel and these were blended in different proportions to form 55 different mixes. The standard
Vol. 23 [2018], Bund. 1 33

Proctor compaction tests were adopted, and both the falling and constant head test methods were used
in the permeability tests. Three sets of ANN prediction models were developed, one each for the
maximum dry density (MDD), optimum moisture content (OMC) and permeability (PMC). A
comparison with the test data indicated that predictions within 95% confidence interval can be
obtained from the ANN models developed. Abdel-Rahman (2008) predicted the compaction of
cohesionless soils using ANN. The model was developed to predict compaction parameters: the
maximum dry unit weight and the optimum moisture content.
Tatlisoz et al. (1997) studied the effect of fines on mechanical properties of soil tyre chip
mixtures and found out that the fines have significant effect on the mechanical properties of the soil
tyre mixture. According to Wang et al. (2009) fines content could affect the dynamic response of soils
significantly. During dynamic compact loading, dynamic forces disrupt the soil skeleton and force the
particles to compact into a denser arrangement. Fines have also been found to affect the liquefaction
potential, slope stability, compression characteristics, strength recovery during aging, pullout force
generated from retaining structure and stress-strain behavior of soil (Cabalar, 2008; Naeini and
Bazier, 2004, Islam et al. 2016, Shahriar 2015, Shahriar et al. 2016, Islam et al. 2017, Shahriar et al.
2018). Belkhatir et al. (2011) shows that the increase of the inter-granular void ratio and the fine
content accelerate the liquefaction phenomenon for the stress ratio under study and the liquefaction
resistance decreases with the increase of the inter-granular void ratio and the loading amplitude. From
practical perspective, Chowdhury et al. (2017) designed an artificial steel tray rainfall simulator
which can be used to analyze the effect of rainfall on loss of fines content in a slope.
Research efforts have been observed to establish a mathematical relationship between at best two
variables related to geotechnical properties of soil. In this study, a mathematical modeling is
conducted to correlate three variables (angle of internal friction, fines content and maximum or
minimum dry density) of cohesionless soil. Simulation is mainly based on the Curve-Fitting tool of
MATLAB software. Levenberg-Marquardt algorithm is used for the fitting procedure. Among 120
simulations, two equations (generated from polynomial surface fitting) of third-order polynomial are
suggested (between a certain range of fines content) at the end of the study.

DATA ACQUISITION
Sample Preparation and Test Procedure
Test Sample
Sand samples, collected from the bank of Turag River were at first oven dried. Then samples
were sieved through a stack of sieves where ASTM No. 4 sieve (4.75 mm) was the top one to
remove the coarse aggregates from the sample. The stack contained sieves of ASTM No.4
(4.75 mm), No. 8 (2.38 mm), No. 16 (1.19 mm), No. 30 (0.60 mm), No. 50 (0.30 mm), No.100
(0.15 mm), No. 200 (0.075 mm) in sequence to collect the sand particles retained on each of
the sieves. The fines content that passed through No. 200 were separated (Figure 1). Obtained fines
were added to each of three base sand samples, S-1, S-2 and S-3 in varying amount (0, 3, 5, 8, 10 and
15% by weight). The gradation curve (Figure 2) shows a median diameter, D50 of 0.25 mm, which is
close to the median diameter of the soil sample collected from Padma river (D50 = 0.24 mm) used
inside the geotextile bags (Chowdhury, Hossain and Muktadir, 2017). Pertinent physical properties of
the three base samples are presented in Table 1.
Vol. 23 [2018], Bund. 1 34

Experimental Program
A direct shear device was used to estimate the angle of internal friction with varying fines content
and the test was performed as per ASTM D 3080 standard test procedure. Maximum and Minimum
dry density of the sampled soil was determined following the test procedures of ASTM D 4253 and
ASTM D 4254 respectively.
Table 1: Direct Shear and Dry Density Test Result

Physical Property S-1 S-2 S-3

Fineness Modulus 1.3 1.5 1.7

Specific Gravity 2.75 2.75 2.75

D10 (mm) 0.15 0.15 0.15

D30 (mm) 0.19 0.23 0.27

D60 (mm) 0.28 0.35 0.40

Uniformity
1.86 2.30 2.63
Coefficient, Cu

Coefficient of
0.86 0.99 1.20
Curvature, Cz

100

80

60
% Finer

40

20

0
0.01 0.1 1 10
Particle size (mm)

Figure 1: Fines Passing through Figure 2: Particle Size Distribution of Sample-1


No. 200 sieve
Vol. 23 [2018], Bund. 1 35

Test Results
Direct shear tests were done on sand samples having varying fineness modulus (1.30, 1.50
and 1.70) and varying fines content (0%, 3%, 5%, 8%, 10% and 15%). Also, minimum dry density
(𝛾𝑑𝑚𝑖𝑛 ) and maximum dry density (𝛾𝑑𝑚𝑎𝑥 ) were determined for six specimens of each sample for
corresponding fines content (0, 3, 5, 8, 10, 15%). Experimental results are shown in Table 2.
Table 2: Direct Shear and Dry Density Test Result

Angle of
Percent Fines 𝛾𝑑𝑚𝑖𝑛 𝛾𝑑𝑚𝑎𝑥
Fineness Modulus Internal Friction
content (kN/m3) (kN/m3)
(𝜑 ° )
0 37 12.91 16.67
3 35.9 13.04 16.92
5 35.7 13.17 17.21
1.3
8 34.2 13.19 17.38
10 34.5 13.23 17.36
15 34 13.24 17.92

0 38 12.94 17.14
3 35.8 13.06 17.30
5 36.4 13.15 17.25
1.5
8 34.7 13.20 17.74
10 35.3 13.24 17.84
15 33.6 13.31 18.17

0 36.6 13.29 17.46


3 35.3 13.31 17.48
5 35.6 13.36 17.74
1.7
8 36.1 13.46 17.91
10 35.9 13.53 18.16
15 34.8 13.52 18.46

SURFACE FITTING
Surface fitting is a process where a surface is fitted in three dimensions instead of a straight or curved
line, enabling it to correlate three different variables where a straight or curved line is capable of
correlating only two variables. In this study, a surface fitting is performed to correlate three different
characteristics of sandy soil: Percentage of fines content, maximum or minimum dry density and
angle of internal friction. Fitting was performed following “Levenberg-Marquardt algorithm”.
Levenberg-Marquardt Algorithm
For solving non-linear least squares problems, Levenberg-Marquardt method is widely
recognized. Any non-linear least squares method reduces the Sum of the Squares of the Errors (SSE)
iteratively between the desired function to be fitted and the experimentally measured data points
through a series of updates to parameter values. The Levenberg-Marquardt fitting method is a
combination of two different minimization methods: the gradient descent method and the Gauss-
Newton method (Gavin H.P., 2017)
Vol. 23 [2018], Bund. 1 36

While fitting a function 𝑦̂(𝑥; 𝒒)of an independent variable x and a vector of e parameters q to a
set of m data points (xi , yi); it is convenient to minimize the sum of weighted residuals between the
measured data y(xi) and the fitting function 𝑦̂(𝑥𝑖 ; 𝒒). So the chi-squared error criterion:
𝑦(𝑥𝑖 )−𝑦̂(𝑥𝑖 ;𝒒) 2
𝜒 2 (𝒒) = ∑𝑚
𝑖=1 [ ]
𝜎𝑦𝑖
(3)
𝑇
̂(𝒒)) 𝑾(𝒚 − 𝒚
= (𝒚 − 𝒚 ̂(𝒒))
(4)
= 𝒚𝑻 𝑾𝒚 − 𝟐𝒚𝑻 𝑾𝒚
̂+𝒚̂𝑻 𝑾𝒚̂
(5)
Where, 𝜎𝑦 is the measurement error for measurement y(xi) [Here any expression containing T as a
1
superscript indicates transpose of a matrix]. The weighting matrix W is diagonal with 𝑾𝑖𝑖 = 𝜎2 . In
𝑦𝑖
the gradient descent method, the gradient (first derivative) of the chi-squared objective function with
respect to the parameters is used.
𝜕 2 𝑇 𝜕
̂(𝒒)) 𝑾 (𝒚 − 𝒚
𝜒 = 2(𝒚 − 𝒚 ̂(𝒒))
𝜕𝒒 𝜕𝒒 (6)
𝑇 ̂(𝒒)
𝜕𝒚
= −2(𝒚 − 𝒚̂(𝒒)) 𝑾 [ ] (7)
𝜕𝒒
= −2(𝒚 − 𝒚̂)𝑇 𝑾𝑱 (8)
𝜕𝒚
̂ (𝒒)
̂ to the
Where, the m×e Jacobian matrix 𝑱 = [ 𝜕𝒒 ] represents the local sensitivity of the function 𝒚
variation in the parameters q. The parameter update h (𝒉′for gradient descent method, 𝒉′′for Gauss-
Newton method and 𝒉′′′ for Levenberg-Marquardt method) that moves the parameters in the
direction of steepest descent is given by,
(9)
𝒉′ = 𝜶𝑱𝑻 𝑾(𝒚 − 𝒚
̂)
Where, the length of the steps in the steepest-descent direction is determined by the positive scalar 𝜶.
In case of Gauss-Newton method, through a first order Taylor series expansion, the function
evaluated with perturbed model parameters may be locally approximated.
̂
𝜕𝒚
̂(𝒒 + 𝒉′′) ≈ 𝒚
𝒚 ̂(𝒒) + [ ] 𝒉′′ = 𝒚
̂ + 𝑱𝒉′′ (10)
𝜕𝒒
̂ + 𝑱𝒉′′for 𝒚
Substituting 𝒚 ̂ in the equation (3), the approximation for the perturbed function is,

𝑻
𝝌𝟐 (𝒒 + 𝒉′′) ≈ 𝒚𝑻 𝑾𝒚 − 𝟐𝒚𝑻 𝑾𝒚 ̂𝑻 𝑾𝒚
̂+𝒚 ̂)𝑻 𝑾𝑱𝒉′′ + 𝒉′′ 𝑱𝑻 𝑾𝑱𝒉′′ (11)
̂ − 𝟐(𝒚 − 𝒚
𝝏𝜒2
The parameter update 𝒉′′ that minimizes chi-square is found from 𝝏𝒉′′ = 0; and the resulting normal
equation for the Gauss-Newton update are;

[𝑱𝑻 𝑾𝑱]𝒉′′ = 𝑱𝑻 𝑾(𝒚 − 𝒚


̂) (12a)

The Levenberg-Marquardt method is an adaptive algorithm which varies the parameter updates
between the gradient descent update and the Gauss-Newton update,
Vol. 23 [2018], Bund. 1 37

[𝑱𝑻 𝑾𝑱 + 𝝀𝑰]𝒉′′′ = 𝑱𝑻 𝑾(𝒚 − 𝒚


̂) (12b)

Where, 𝝀 is an algorithmic parameter and 𝑰 is an identity matrix of the same order (m×e) of the
Jacobian matrix. Here, small values of 𝝀 results in a Gauss-Newton update and large values of 𝝀
result in a gradient descent update. If any iteration happens in an approximation such that 𝝌𝟐 (𝒒 +
𝒉′′′) > 𝝌𝟐 (𝒒); then 𝝀 is increased. Otherwise, as the solution forwards, 𝝀 is decreased. In
Marquardt’s (Marquardt D.W., 1963) updated relationship,

[𝑱𝑻 𝑾𝑱 + 𝝀 𝒅𝒊𝒂𝒈 (𝑱𝑻 𝑾𝑱)]𝒉′′′ = 𝑱𝑻 𝑾(𝒚 − 𝒚


̂) (13)

In this study, the Levenberg-Marquardt algorithm is used to carry out fitting in three dimensions (for
three different variables). Two independent variables x and y are expressed as a function of z such
that,
𝒛𝒊 = 𝒇(𝒙𝒊 , 𝒚𝒊 ) (14)

Where, i = 1,2,3,……,m and x and y both are column vectors of independent variables. Finally, fitting
is carried out containing x, y and z variables in different three dimensions.

Goodness of Fit
Different statistical parameters were calculated to determine the suitability of the proposed equation.
These parameters are briefly discussed below:
(i) Sum of Squares due to Error (SSE)

𝑺𝑺𝑬 = ∑𝒏𝒊=𝟏 𝒘𝒊 (𝒚𝒊 − 𝒇𝒊 )𝟐 (15)


where, 𝑦𝑖 is the observed data, 𝑓𝑖 is the predicted value from the fit and 𝑤𝑖 is the weighing
applied to each point, here 𝑤𝑖 =1.
(ii) R-Square
∑𝒏
𝒊=𝟏 𝒘𝒊 (𝒚𝒊 −𝒇𝒊 )
𝟐
𝑹 − 𝑺𝒒𝒖𝒂𝒓𝒆 = 𝟏 − ∑𝒏 𝟐
(16)
𝒊=𝟏 𝒘𝒊 (𝒚𝒊 −𝒚𝒂𝒗 )

where, 𝑦𝑎𝑣 is the mean of the observed data.


(iii) Adjusted R-Square
(𝟏−𝑹𝟐 )(𝑵−𝟏)
𝑨𝒅𝒋𝒖𝒔𝒕𝒆𝒅 𝑹 − 𝒔𝒒𝒖𝒂𝒓𝒆 = 𝟏 − (17)
𝑵−𝒑′−𝟏

where, 𝑅 2= Sample R-square, N= Total sample size, 𝒑′= Number of predictors


(iv) Root Mean Squared Error (RMSE)
𝟏
𝑹𝑴𝑺𝑬 = √𝒏 ∑𝒏𝒊=𝟏 𝒘𝒊 (𝒚𝒊 − 𝒇𝒊 )𝟐 (18)

where, n is the number of total sample size.


Vol. 23 [2018], Bund. 1 38

Surface Fitting Session


The surface fitting session was performed by creating a function named “createFit1”, writing
an m-file script. This function works as the following steps:
1. Step 1: Taking the input data and assigning it in three variables (X data, Y data and
Z data)
2. Step 2: Transforming row (Y input) and column (X input) headers into different two
arrays (Y out) and (X out) that are the same size as Z input using
“prepareSurfaceData” function which returns data as column shapes regardless of the
input data. This function also converts complex numbers to real numbers by removing
imaginary parts, non-double to double numbers [all numerical values are stored as double
precision floating-point values that are 8 bytes (64 bits)] and removes all Nan (not a number)
and Inf (infinity) elements from the data. All the time any changes (conversion or removal)
occur, a warning is provided.
3. Step 3: Choosing a fittype for fitting the variables in a surface like Interpolant, Lowess,
Polynomial etc.
4. Step 4: Using the fit function to fit the data by using Levenberg-Marquardt algorithm.
5. Step 5: Evaluating the performance of fit by using gof (goodness of fit) function which
returns four parameters: Sum of Squares due to Error (SSE), R-square, Adjusted R-square
and Root Mean Square Error (RMSE)

Figure 3: A flow chart depicting the steps of Surface Fitting function createFit1
In this study, the polyfit fittype is used for fitting the surface in case of polynomial equations
which solves equations in a form of Vandermonde matrix.
Vol. 23 [2018], Bund. 1 39

SIMULATION
The MATLAB software uses four different fitting methods for any kind of surface or curve
fitting: (1) MATLAB custom equation, (2) Interpolant method, (3) Locally Weighted Scatterplot
Smoothing (Lowess) method and (4) Polynomial method. For comparative analysis between these
methods for the purpose of selecting a suitable equation which can correlate percentage of fines
content, maximum or minimum dry density and angle of internal friction, Interpolant and Lowess
methods are excluded in this study because in spite of showing satisfactory goodness of fit, these two
methods do not generate any equation from which other values can be extrapolated.
Surface fitting with polynomial equations and MATLAB custom equation offer an equation for
extrapolating other values using the equation. The generalized equation for the polynomial method is:

𝒑(𝒙) = 𝒑𝟏 𝒙𝒕 + 𝒑𝟐 𝒙𝒕−𝟏 + ⋯ + 𝒑𝒕 𝒙 + 𝒑𝒕+𝟏 (19)

Where, the coefficients are represented as p and the length of p is (t+1).


The generalized equation for the MATLAB custom equation is:

𝟐
𝒛(𝒙, 𝒚) = 𝒂′ + 𝒃′𝒔𝒊𝒏(𝒎′𝝅𝒙𝒚) + 𝒄′𝒆−(𝒘′𝒚) (20)

Here, 𝒂′, 𝒃′, 𝒎′, 𝒄′, 𝒘′are the coefficients which are determined after the fitting; z is the function of x
and y.

On the basis of goodness of fit (minimum SSE and RMSE value; maximum R-square and
Adjusted R-square value), some polynomial combinations (Table 3) were chosen among 120 different
degree of polynomials combinations (Appendix Table 5~Table 12) to validate with the data of
Alshameri B. et al (2017). As the values of maximum dry density follow the same trend as minimum
dry density (decreases with the increase in fines content; Table 2), the performance of goodness of fit
parameters is expected to follow the same chronology as the minimum dry density with different
combinations of degrees of independent variables. The best performance according to goodness of fit
is shown by degrees of independent variables combination x(4) and y(4) (Appendix Table 5); but any
equation with fourth order polynomial in any one variable (x or y) or both variables, was in an ill
condition for the repeated points of fines content. So combinations with fourth order polynomial in
any one independent variable or both variables, were avoided for the comparison. MATLAB can
perform surface fitting up to fifth order polynomial, but because of insufficient number of data set,
combination with fifth order polynomial, in any one variable (x or y) or both variables, is not
considered. As combinations with fourth order polynomial, set up equations which are badly
conditioned for the repeated data points of fines content, it can be easily assumed that combination
with fifth order polynomial would produce ill conditioned equations also, therefore, disregarding
combination with fifth order polynomial will not affect the study. In Table 3, validation was
performed with maximum dry density data (as there was no data available for minimum dry density
validation); for minimum dry density the polynomial equations were reproduced.
Vol. 23 [2018], Bund. 1 40

Table 3 (a): Comparison of Validation Data with Values of Angle of Internal Friction from
Polynomial Equations [a: Percentage of Fines Content; d: Maximum Dry Density; c: Tangent
of Angle of Internal Friction (in radian); (a-d-c): X data (a), Y data (d), Z data (c); (d-a-c): X
data (d), Y data (a), Z data (c); all the angles are in degree unit]

Validation Data (Alshameri B. et al, 2017) Values of Angle of Internal Friction from Polynomial Equations

Maximum Angle of
Fines x(3), y(3) x(3), y(3) x(3), y(2) x(3), y(2) x(2), y(3)
Dry Density Internal
Content (a-d-c) (d-a-c) (a-d-c) (d-a-c) (a-d-c)
(MDD) (kN/m3) Friction
0.2 18.93 45.8 -56.50 -56.50 32.58 -46.99 -46.99
0.3 18.54 52.6 -61.53 -61.53 14.97 -52.75 -52.75
0.4 17.75 46.7 -70.50 -70.50 69.30 -36.01 -36.01
0.5 16.77 62.1 -84.78 -84.78 85.44 -68.69 -68.69
0.6 16.08 53.5 -87.96 -87.96 88.07 -83.18 -83.18
0.7 15.49 68.3 -88.97 -88.97 88.97 -86.90 -86.90

Table 3 (b): Comparison of Validation Data with Values of Angle of Internal Friction from
Polynomial Equations [a: Percentage of Fines Content; d: Maximum Dry Density; c: Tangent
of Angle of Internal Friction (in radian); (a-d-c): X data (a), Y data (d), Z data (c); (d-a-c): X
data (d), Y data (a), Z data (c); all the angles are in degree unit]

Values of Angle of Internal Friction from Polynomial


Validation Data (Alshameri B. et al, 2017)
Equations

Maximum Angle of x(3),


Fines x(2), y(3) x(1), y(3) x(1), y(3) x(3), y(1)
Dry Density Internal y(1)
Content (d-a-c) (a-d-c) (d-a-c) (d-a-c)
(MDD) (kN/m3) Friction (a-d-c)
0.2 18.93 45.8 32.58 -20.60 81.67 31.14 -20.60
0.3 18.54 52.6 14.97 -35.66 83.87 -1.53 -35.66
0.4 17.75 46.7 69.30 -14.60 84.85 -53.62 -14.60
0.5 16.77 62.1 85.44 -58.09 85.25 -74.18 -58.09
0.6 16.08 53.5 88.07 -80.71 85.19 -81.64 -80.71
0.7 15.49 68.3 88.97 -85.92 84.48 -85.01 -85.92

From Table 3, it is observed that considering the absolute values of the angles of internal friction,
polynomial equations with combination x(3), y(2) (d-a-c) and x(2), y(3) (a-d-c) matches mostly with
the validation data, showing considerably good performance with respect to goodness of fit (SSE:
0.001942, R-square: 0.8686, Adjusted R-square: 0.7519, RMSE: 0.01469) for both combinations. As
the deviation with the actual data was increasing with the lowering of degrees of polynomial, further
computation with lower degrees of polynomial than 3, was stopped there.
Vol. 23 [2018], Bund. 1 41

(a) (b)

(c) (d)

(e) (f)

Figure 4: Polynomial Surface Fit [(a) and (b)], Residuals plot [(c) and (d)] and Contour plot [(e)
and (f)] for x(2), y(3) (a-d-c) combination and x(3), y(2) (d-a-c) combination respectively for
maximum dry density
Vol. 23 [2018], Bund. 1 42

(a) (b)

(c)
(d)

(e) (f)

Figure 5: Polynomial Surface Fit [(a) and (b)], Residuals plot [(c) and (d)] and Contour plot [(e)
and (f)] for x(2), y(3) (a-b-c) combination and x(3), y(2) (b-a-c) combination respectively for
minimum dry density
Vol. 23 [2018], Bund. 1 43

The equations associated with above polynomial surface fitting are represented here:

for maximum dry density:

𝒕𝒂𝒏(𝝋) = 𝒑𝟎 + 𝒑𝟏 𝜸𝒅𝒎𝒂𝒙 + 𝒑𝟐 𝒇 + 𝒑𝟑 𝜸𝒅𝒎𝒂𝒙 𝟐 + 𝒑𝟒 𝜸𝒅𝒎𝒂𝒙 𝒇 + 𝒑𝟓 𝒇𝟐


+ 𝒑𝟔 𝜸𝒅𝒎𝒂𝒙 𝟑 + 𝒑𝟕 𝜸𝒅𝒎𝒂𝒙 𝟐 𝒇 + 𝒑𝟖 𝜸𝒅𝒎𝒂𝒙 𝒇𝟐
(21)
for minimum dry density:

𝒕𝒂𝒏(𝝋) = 𝒑𝟎 + 𝒑𝟏 𝜸𝒅𝒎𝒊𝒏 + 𝒑𝟐 𝒇 + 𝒑𝟑 𝜸𝒅𝒎𝒊𝒏 𝟐 + 𝒑𝟒 𝜸𝒅𝒎𝒊𝒏 𝒇 + 𝒑𝟓 𝒇𝟐


+ 𝒑𝟔 𝜸𝒅𝒎𝒊𝒏 𝟑 + 𝒑𝟕 𝜸𝒅𝒎𝒊𝒏 𝟐 𝒇 + 𝒑𝟖 𝜸𝒅𝒎𝒊𝒏 𝒇𝟐
(22)

After the surface fitting session is performed, the numerical value of the coefficients 𝒑𝟎 , … … , 𝒑𝟖
are estimated. Figure 4 shows the corresponding polynomial surface fit for Eq. (21) and figure 5
shows for Eq. (22). The proposed equations with the numerical values of coefficients are given
below:

for maximum dry density:

𝒕𝒂𝒏(𝝋) = | − 𝟏𝟕𝟒𝟑 + 𝟑𝟎𝟑. 𝟑𝜸𝒅𝒎𝒂𝒙 − 𝟏𝟑𝟓𝟐𝒇 − 𝟏𝟕. 𝟓𝟖𝜸𝒅𝒎𝒂𝒙 𝟐


+ 𝟏𝟓𝟑. 𝟖𝜸𝒅𝒎𝒂𝒙 𝒇 − 𝟓𝟒. 𝟗𝟏𝒇𝟐 + 𝟎. 𝟑𝟑𝟗𝟑𝜸𝒅𝒎𝒂𝒙 𝟑
− 𝟒. 𝟑𝟕𝟓𝜸𝒅𝒎𝒂𝒙 𝟐 𝒇 + 𝟑. 𝟏𝟔𝜸𝒅𝒎𝒂𝒙 𝒇𝟐 |
(23)
for minimum dry density:

𝒕𝒂𝒏(𝝋) = | − 𝟓𝟓𝟖. 𝟏 + 𝟗𝟒. 𝟑𝟖𝜸𝒅𝒎𝒊𝒏 + 𝟑𝟒𝟗𝟎𝒇 − 𝟒. 𝟔𝟑𝟓𝜸𝒅𝒎𝒊𝒏 𝟐


− 𝟓𝟒𝟖. 𝟖𝜸𝒅𝒎𝒊𝒏 𝒇 + 𝟏𝟓𝟏𝟗𝒇𝟐 + 𝟎. 𝟎𝟓𝟐𝟒𝟓𝜸𝒅𝒎𝒊𝒏 𝟑
+ 𝟐𝟏. 𝟓𝟒𝜸𝒅𝒎𝒊𝒏 𝟐 𝒇 − 𝟏𝟏𝟒. 𝟓𝜸𝒅𝒎𝒊𝒏 𝒇𝟐 | (24)

Eq. (23) and Eq. (24) both estimate negative values of 𝒕𝒂𝒏(𝝋), which indicates that angle of
internal friction will also be a negative value. However, there is no meaning of a negative value of
angle of internal friction. So, absolute values of 𝒕𝒂𝒏(𝝋) are taken here to estimate the values of angle
of internal friction as positive. The equations (23) and (24) are returning negative values because
many higher order polynomial (2 or 3) terms hold a negative sign in the equations as well as the first
term of the equations. Eq. (23) is applicable for both x(2), y(3) (a-d-c) and x(3), y(2) (d-a-c)
combinations for maximum dry density; same goes for Eq. (24) in case of minimum dry density.
Normalizing the data by centering it to zero mean and scaling it to unit standard deviation was not
done here because it was observed that in which simulations this was done by “center and scale”
option of MATLAB, it returned some localized angle of internal friction value varying very little for
Vol. 23 [2018], Bund. 1 44

different fines content and dry density. The ‘robust’ parameter was selected as ‘off’ in this study,
resulting in better goodness of fit performance for polynomial equations. Moreover, surface fitting is
avoided here for MATLAB custom equation (Eq. 20) as it was returning different values of
coefficients a′, b′, m′, c′, w′ (Table 4) in spite of allocating the same variables in X, Y and Z data
every time.
Table 4: Different coefficients for same variable allocation in X, Y and Z data
of MATLAB Custom Equation (20)
[a: Percentage of Fines Content; b: Minimum Dry Density; c: Tangent of Angle of Internal
Friction (in radian)]

Simulation Values of Coefficients X Y Z


No. Data Data Data
a′ b′ m′ c′ w′
1 0.7509 -0.07183 0.2093 0.8143 0.9293
2 0.7508 0.07223 -0.207 10.7 0.4364 a b c
3 0.7509 -0.07207 0.208 0.7513 0.506

On the other hand, polynomial equations returned the same values of coefficients in every single
run for same variable allocation in X, Y and Z data; depicting a stable nature. It is observed that
MATLAB custom equation holds both a trigonometric function and a natural logarithmic function.
Due to its fluctuating nature in coefficients values, other trigonometric and logarithmic fits are also
avoided here, by far getting a stable and better (goodness of fit) result from polynomial surface fit.

100 Actual values from Alshameri B. et


al (2017)
Angle of Internal Friction (in degree)

80 Values estimated from equation


(23) for the same maximum dry
density
60

40

20

0
15 15.5 16 16.5 17 17.5 18 18.5 19 19.5
Maximum Dry Density (in kN/cubic meter)

Figure 6: A comparative analysis between the actual data values and the estimated ones from
the equation
Figure 6 shows a comparison between the values of angle of internal friction (validation data) and
the estimated values of angle of internal friction from x(3), y(2) (a-d-c) or x(2), y(3) (d-a-c)
polynomial combinations [corresponding equation: Eq. (23)], Table 3. The graph clearly shows that
Vol. 23 [2018], Bund. 1 45

both values converge very much at higher maximum dry density (lower fines content percentage). As
the maximum dry density approaches lower values (higher fines content percentage), both the values
start to diverge slightly from each other. The angle between the best fit lines in the graph is less than
45 degrees which indicates a good convergence. The equation performs well despite the fluctuating
nature of angle of internal friction. The slight divergence may occur because of the different fines
content used in Alshameri B. et al (2017), where the equation overestimates due to the higher amount
of fines content.

CONCLUSIONS
This study represents a simplified mathematical model to estimate any one of the variables among
the fines content, angle of internal friction and maximum or minimum dry density of sandy soil if the
other two is known. Usually, the fines content is easier to determine than the other two variables.
Therefore, if any one of the Direct Shear Test or Dry Density Test is performed, using the polynomial
equations (23) and (24), the other one parameter can be estimated. Some points should be addressed
from this study:
a) The polynomial equations (Eq. 23 and Eq. 24) can be used effectively in an adverse
situation where either Direct Shear Test or Dry Density Test apparatus are unavailable.
b) The polynomial surface fitting method gives better performance than other three
methods (Interpolant, Lowess and MATLAB Custom Equation) available in
MATLAB for surface fitting. Performance was evaluated in terms of convergence with
validation data and goodness of fit parameters.
c) The proposed polynomial equation best performs at fines content range of 0%~30%
where it gives almost the exact values from actual data. The fines content percentage is
mostly responsible here for the divergence of the estimated values from Eq. (23) from
the actual data values, as it ranges with a high variation (20%~70%); but maximum dry
density ranges in a very low variation (15kN/m3~20kN/m3) according to the validation
data. So, it is recommended to use these equations at this range (0%~30%) of fines
content. Otherwise, at higher fines content, the equations overestimate the angle of
internal friction value.
d) The absolute values of tangent of angle of internal friction from Eq. (23) and Eq. (24)
are considered here as there is no meaning of negative values of angle of internal
friction.
e) For minimum dry density, it is recommended to use Eq. (24) instead of Eq. (23) to
estimate the desired parameter because Eq. (24) is separately fitted with minimum dry
density data from Table 2.
f) Every possible combination (Appendix Table 5~Table 12) with respect to degree of
independent variables and X, Y and Z data has been attempted here, so it can be
concluded that equations (23) and (24) represent the possible best combination here
with respect to goodness of fit performance (that mostly converges with the actual
data).
Vol. 23 [2018], Bund. 1 46

ACKNOWLEDGEMENTS
This research work is conducted in the Geotechnical Engineering Laboratory of Bangladesh
University of Engineering and Technology (BUET) under the supervision of Dr. Md. Zoynul Abedin,
Professor, Military Institute of Science and Technology (MIST). The authors would like to express
profound gratitude to Mr. Abedin for his sincere supervision, guidance, lively encouragement and
useful suggestions throughout the research work. Also Mr. Azmayeen Rafat Shahriar, Lecturer,
Department of Civil Engineering, BUET has mentored the full study with constructive comments and
valuable recommendations.

NOTATIONS
𝒑𝟎 , … 𝒑𝒕 , … = Coefficients of polynomial equations
𝜎𝑦 = Measurement error
𝜒2 = Chi-squared criterion
𝜸𝒅𝒎𝒂𝒙 = Maximum dry density
𝜸𝒅𝒎𝒊𝒏 = Minimum dry density
a = Assigned variable for percentage of fines content in the software
b = Assigned variable for minimum dry density in the software
Assigned variable for tangent of angle of internal friction (in radian) in
c =
the software
d = Assigned variable for maximum dry density in the software
Dr = Relative density
e = Number of parameters
f = Ratio of fines content to the total mass of the system
fi = Predicted values from the fit in SSE (i=1,………,n)
I = Identity matrix
J = Jacobian matrix
JT = Transpose of Jacobian matrix
m = Number of data points in Levenberg-Marquardt algorithm
N = Total sample size in Adjusted R-square values
n = Total sample size in SSE, R-square and RMSE values
Q = A vector of e parameters to a set of m
W = Weighting matrix
wi = Weighing applied in the goodness of fit parameters (i=1,………,n)
yav = Mean of observed data in R-square values
𝒂′, 𝒃′, 𝒎′, 𝒄′, 𝒘′ = Coefficients of MATLAB Custom Equation [Eq. 20]
𝒉′ = Parameter update for gradient-descent method
𝒉′′ = Parameter update for Gauss-Newton method
𝒉′′′ = Parameter update for Levenberg-Marquardt method
𝒑′ = Number of predictors in Adjusted R-square values
𝜶 = A positive scalar that determines the length of the steps in the steepest
descent direction
𝝀 = An algorithmic parameter in Levenberg-Marquardt method
𝝋 = Angle of internal friction
MDD = Maximum dry density
RMSE = Root mean square error
SSE = Sum of the squares due to error
Vol. 23 [2018], Bund. 1 47

REFERENCES
[1] Abdel-Rahman, A. H. (2008). “Predicting Compaction of Cohesion less Soils using ANN”,
Ground Improvement, Issue 161, pp. 3-8

[2] Abedi, M., and Yasrobi, S.S. (2010), “Effects of Plastic Fines on the Instability of Sand”, Soil
Dynamics and Earthquake Engineering, 30(3), pp. 61–67

[3] Alshameri B., Madun A. and Bakar I. (2017), “Comparison of the Effect of Fine Content and
Density towards the Shear Strength Parameters”, Geotechnical Engineering Journal of the
SEAGS & AGSSEA, Vol. 48 No. 2 June 2017 ISSN 0046-5828

[4] ASTM (2000) D 4253-00: Standard Test Methods for Maximum Index Density and Unit
Weight of Soils Using a Vibratory Table.

[5] ASTM (2000c), Standard Test Method for Direct Shear Test of Soils Under Consolidated
Drained Conditions, Annual Book of ASTM Standards, D 3080, Vol. 04.08, ASTM,
Philadelphia, USA.

[6] ASTM (2002) D 422-63: Standard Test Method for Particle-Size Analysis of Soils.

[7] ASTM (2002) D 854-02: Standard Test Methods for Specific Gravity of Soil Solids by Water
Pycnometer.

[8] ASTM (2006) C 136-06: Standard Test Method for Sieve Analysis of Fine and Coarse
Aggregates.

[9] ASTM (2006) D 4254-00 (Reapproved 2006): Standard Tests Methods for Minimum Index
Density and Unit Weight of Soils and Calculation of Relative Density

[10] Ayininuola, G.M., Agbede, O.A. and Franklin, S.O. (2009), “Influence of Calcium Sulphate
on Subsoil Cohesion and Angle of Friction”, Journal of Applied Sciences Research, 5(3), pp.
297-304

[11] Ayodele, A.L. (2008), “A Study of the Effect of Fines Content on the Performance of Soil as
Sub-Base Material for Road Construction”, Unpublished M. Sc. Thesis, Obafemi Awolowo
University, Ile-Ife, Nigeria

[12] Belkhatir, M., Missoum, H., Arab, A., Della, N., and Schanz, T. (2011), “Undrained Shear
Strength of Sand-silt Mixture: Effect of Inter-Granular Void Ratio and Other Parameters”,
KSCE Journal of Civil Engineering (2011) 15(8):1335-1342, DOI 10.1007/s12205-011-1051-
x

[13] Cabalar, A. F. (2008), “Effects of Fines Content on the Behavior of Mixed Samples of a Sand”
Electronic Journal of Geotechnical Engineering, 13 (D), 2008, pp. 1-13

[14] Chowdhury, M.E., Hossain, A. and Muktadir, H.M. (2017), “A Case Study of the River
Training Work of Padma River: Assessment of Local Slope Protection Measures”,
Vol. 23 [2018], Bund. 1 48

Proceedings of Civil and Water Resource Engineering Conference, 3-4 November 2017,
BIAM Foundation, 63 Eskaton, Dhaka, Bangladesh.

[15] Chowdhury, M.E., Islam, M.A., Islam, M.S., Shahriar, M.S. and Alam, T. (2017), “Design,
Operation and Performance Evaluation of Portable Perforated Steel Tray Rainfall Simulator”,
Proceedings of 14th Global Engineering and Technology Conference, 29-30 December, 2017,
BIAM Foundation, 63 Eskaton, Dhaka, Bangladesh.

[16] Das, M.B. (1990), Soil mechanics (California: Brooks/Cole publishers)

[17] Gupta, R. and Trivedi, A. (2009). “Bearing Capacity and Settlement of Footing Resting on
Confined Loose Silty Sands”, Electronic Journal of Geotechnical Engineering, Volume 14A,
2009

[18] Gupta, R. and Trivedi, A. (2009). “Effects of Non-Plastic Fines on the Behavior of Loose
Sand-An Experimental Study”, Electronic Journal of Geotechnical Engineering, Volume 14B,
2009

[19] Hatanaka, M and Feng, L. (2006). “Estimating Relative Density of Sandy soils”, Soils and
Foundations, JGS, Vol. 46, No. 3, pp. 299-313

[20] Islam, M. A., Islam, M. S., Islam, T., “Landslides in Chittagong Hill Tracts and Possible
Measures”. Proceedings of International Conference on Disaster Risk Mitigation, Dhaka,
Bangladesh, September 23 - 24, 2017

[21] Islam, M. S., Sarker, L., Islam, M. A., Islam, M. A., Karim, R., “Consideration of soil
properties for stability analyses of Padma and Jamuna river bank”. Proceedings of 3rd
International Conference on Advances in Civil Engineering, CUET, Chittagong, Bangladesh,
21-23 December 2016

[22] Islam, T., Islam, M.A., Islam, M.S. and Abedin, M.Z. (2017), “Effect of Fine Content on
Shear Strength Behavior of Sandy Soil”, Proceedings of 14th Global Engineering and
Technology Conference, 29-30 December, 2017, BIAM Foundation, 63 Eskaton, Dhaka,
Bangladesh.

[23] Lade, P.V., Liggio, Jr. C.D. and Yamamuro, J.A. (1998), “Effects of Non-Plastic Fines on
Minimum and Maximum Void Ratios of Sand”, Geotechnical Testing Journal, 21(4), pp. 336–
347

[24] Marquardt D. (1963), “An Algorithm for Least Squares Estimation of Non-linear Parameters”,
SIAM Journal of Applied Mathematics, 11(2): 431-441

[25] Meyerhof, G. G. (1956). “Penetration Tests and Bearing Capacity of Cohesion less Soils”,
Proceedings of ASCE, Vol. 82, pp 1-19.

[26] Meyerhof, G.G. (1957). “Discussion on Research on Determining the Density of Sands by
Penetration Testing”. Proc. 4th International Conference on Soil Mechanics and Foundation
Engineering., Vol. 1, pp. 110.
Vol. 23 [2018], Bund. 1 49

[27] Mujtaba, H., Farooq. K., Sivakugan. N., and Das. B. M. (2017) “Evaluation of Relative
Density and Friction Angle Based on SPT Values”, KSCE Journal of Civil Engineering, pp. 1-
10. https://doi.org/10.1007/s12205-017-1899-5

[28] Murthy, T.G., Loukidis, D., Carraro, J.A.H., Prezzi, M., and Salgado, R. (2007), “Undrained
Monotonic Response of Clean and Silty Sands”, Géotechnique, 57(3), pp. 273–288.

[29] Naeini, S.A. and Baziar, M.H. (2004), “Effect of Fines Content on Steady State Strength of
Mixed and Layered Samples of a Sand”, Soil Dynamics and Earthquake Engineering, 24 (3),
pp. 181 – 187

[30] Ni, Q., Tan, T.S., Dasari, G.R., and Hight, D.W. (2004), “Contribution of Fines to the
Compressive Strength of Mixed Soils”, Géotechnique, 54(9), pp. 561–569.

[31] Ojha, S. and Trivedi, A. (2013). “Shear Strength Parameters for Silty Sand using Relative
Compaction”, Electronic Journal of Geotechnical Engineering, Volume 18/A, 2013

[32] Prasad A. and Pandey B. (2013), “Effect of Fines on the Mechanical Behavior of Sand”
International Journal of Structural and Civil Engineering Research, ISSN 2319 – 6009, Vol.
2, pp. 41-45

[33] Shahriar, A.R. (2015). “Effect of Moisture Content on the Thixotropic Strength Recovery of
Dhaka Soil” Proceedings of the 1st International Conference on Advances in Civil
Infrastructure. and Construction Materials, Dhaka.

[34] Shahriar, A.R. (2016). “Thixotropic Hardening of Clay-Water System” B.Sc. Engineering
Thesis, Department of Civil Engineering, Bangladesh University of Engineering and
Technology, Dhaka.

[35] Shahriar, A.R., Abedin, M.Z., Islam, M.A. (2016). “Factors Affecting Thixotropic Hardening
of Dhaka Clays” Proceedings of the 1st Bangladesh Civil Engineering Summit, Dhaka.

[36] Shahriar, A.R., Abedin, M.Z., Jadid, R. (2018). “Thixotropic Aging and its Effect on 1-D
Compression Behavior of Soft Reconstituted Clays” Applied Clay Science, Vol. 153, pp. 217-
227.

[37] Sinha, S.K. and Wang, M.C. (2008). “Artificial Neural Network Prediction Models for Soil
Compaction and Permeability”, iie, Vol. 26, pp. 47–64.

[38] Tatlisoz, N., Benson, C. and Edil, T. (1997), “Effect of Fines on Mechanical Properties of
Soil-Tyre Chip Mixtures, Testing Soil Mixed with Waste or Recycled Materials”, ASTM STP
1275, Mark A. Wasemiller, Keith B. Hoddinott, Eds., (American Society for Testing and
Materials)

[39] Tokimatsu, K. and Yoshimi, Y. (1983). “Empirical Correlation of Soil Liquefaction based on
SPT-N value and Fines Content”, Soils and Foundations, JGS, Vol. 23, No. 4, pp. 56-74

[40] Townsend, F.C. (1973). “Comparisons of Vibrated Density and Standard Compaction Tests on
Sands with Varying Amounts of Fines”, ASTM, Digital Library / STP / STP523-EB /
STP37882S.
Vol. 23 [2018], Bund. 1 50

[41] Wang, S., Chan, D. and Lam, K.C. (2009), “Experimental Study of the Effect of Fines Content
on Dynamic Compaction Grouting in Completely Decomposed Granite of Hong Kong”,
Construction and Building Materials, 23, pp. 1249–1264

APPENDIX
Table 5: Performance of Polynomial Equation (Degree: x=4; y= 1,2,3,4)
[a: Percentage of Fines Content; b: Minimum Dry Density; c: Tangent of Angle of Internal
Friction (in radian)]

Degree of
Independent Goodness of Fit Parameters
Simulation Variables X Y Z
No. Data Data Data
Adjusted
x y SSE R-squared RMSE
R-square
1 0.00053 0.9645 0.7989 0.01322 a b c
2 0.02506 0.9522 0.7293 0.09139 a c b
3 4 4 0.00053 0.9645 0.7989 0.01322 b a c
4 0.00206 0.9532 0.7347 0.02586 b c a
5 0.02506 0.9522 0.7293 0.09139 c a b
6
7 0.00206
0.005582 0.9532
0.9622 0.7347
0.8395 0.02586
0.01181 ca b
b ac
8 0.0479 0.9223 0.6699 0.1009 a c b
9 0.000616 0.9584 0.823 0.01241 b a c
4 3
10 0.00246 0.9428 0.7559 0.0248 b c a
11 0.06233 0.8812 0.4949 0.1248 c a b
12 0.004361 0.8982 0.5678 0.03302 c b a
13 0.00129 0.9128 0.7529 0.01466 a b c
14 0.04921 0.9062 0.7342 0.09056 a c b
15 0.001279 0.9135 0.755 0.0146 b a c
4 2
16 0.002685 0.9373 0.8225 0.02115 b c a
17 0.1073 0.7953 0.4201 0.1338 c a b
18 0.005624 0.8688 0.6282 0.03062 c b a
19 0.00129 0.9128 0.7529 0.01466 a b c
20 0.04921 0.9062 0.7342 0.09056 a c b
21 0.001279 0.9135 0.755 0.0146 b a c
4 1
22 0.002685 0.9373 0.8225 0.02115 b c a
23 0.1073 0.7953 0.4201 0.1338 c a b
24 0.005624 0.8688 0.6282 0.03062 c b a
Vol. 23 [2018], Bund. 1 51

Table 6: Performance of Polynomial Equation (Degree: x=1,2,3,4; y=4)


[a: Percentage of Fines Content; b: Minimum Dry Density; c: Tangent of Angle of Internal
Friction (in radian)]
Degree of
Independent Goodness of Fit Parameters
Simulation Variables X Y Z
No. Data Data Data
Adjusted
x y SSE R-squared RMSE
R-square
25 0.00053 0.9645 0.7989 0.01322 a b c
26 0.02506 0.9522 0.7293 0.09139 a c b
27 0.00053 0.9645 0.7989 0.01322 b a c
4 4
28 0.00206 0.9532 0.7347 0.02586 b c a
29 0.02506 0.9522 0.7293 0.09139 c a b
30 0.00206 0.9532 0.7347 0.02586 c b a
31 0.000616 0.9584 0.823 0.01241 a b c
32 0.06233 0.8812 0.4949 0.1248 a c b
33 0.000558 0.9622 0.8395 0.01181 b a c
3 4
34 0.004361 0.8982 0.5675 0.03302 b c a
35 0.04074 0.9223 0.6699 0.1009 c a b
36 0.002461 0.9426 0.7559 0.0248 c b a
37 0.001279 0.9135 0.755 0.0146 a b c
38 0.1073 0.7953 0.4201 0.1338 a c b
39 0.00129 0.9128 0.7529 0.01466 b a c
2 4
40 0.005624 0.8688 0.6282 0.03062 b c a
41 0.04921 0.9062 0.7342 0.09056 c a b
42 0.002685 0.9373 0.8225 0.02115 c b a
43 0.001602 0.8916 0.7953 0.01334 a b c
44 0.2021 0.6148 0.2723 0.1498 a c b
45 0.00206 0.8606 0.7367 0.01513 b a c
1 4
46 0.007217 0.8316 0.6819 0.02832 b c a
47 0.1187 0.7738 0.5727 0.1148 c a b
48 0.005366 0.8748 0.7635 0.02442 c b a
Vol. 23 [2018], Bund. 1 52

Table 7: Performance of Polynomial Equation (Degree: x=3; y=1,2


[a: Percentage of Fines Content; b: Minimum Dry Density; c: Tangent of Angle Internal
Friction (in radian)]

Degree of
Independent Goodness of Fit Parameters
Simulation Variables X Y Z
No. Data Data Data
Adjusted
x y SSE R-squared RMSE
R-square
49 0.001491 0.8992 0.7857 0.01365 a b c
50 0.1145 0.7818 0.5363 0.1196 a c b
51 0.001491 0.8992 0.7857 0.01365 b a c
3 3
52 0.004795 0.8881 0.7622 0.02448 b c a
53 0.1145 0.7818 0.5363 0.1196 c a b
54 0.004795 0.8881 0.7622 0.02448 c b a
55 0.001491 0.8991 0.8095 0.01287 a b c
56 0.1286 0.7548 0.5368 0.1195 a c b
57 0.001492 0.8991 0.8094 0.01288 b a c
3 2
58 0.004811 0.8877 0.7879 0.02312 b c a
59 0.1149 0.781 0.5864 0.113 c a b
60 0.006939 0.8381 0.6941 0.02777 c b a
61 0.002298 0.8445 0.7597 0.01445 a b c
62 0.1699 0.6761 0.4995 0.1243 a c b
63 0.002006 0.8643 0.7903 0.01351 b a c
3 1
64 0.005614 0.869 0.7975 0.02259 b c a
65 0.2144 0.5912 0.3682 0.1396 c a b
66 0.007565 0.8235 0.7272 0.02622 c b a
Vol. 23 [2018], Bund. 1 53

Table 8: Performance of Polynomial Equation (Degree: x=1,2,3; y=3)


[a: Percentage of Fines Content; b: Minimum Dry Density; c: Tangent of Angle of Internal
Friction (in radian)]

Degree of
Independent Goodness of Fit Parameters
Simulation Variables X Y Z
No. Data Data Data
Adjusted
x y SSE R-squared RMSE
R-square

67 0.001491 0.8992 0.7857 0.01365 a b c


68 0.1145 0.7818 0.5363 0.1196 a c b
69 0.001491 0.8992 0.7857 0.01365 b a c
3 3
70 0.004795 0.8881 0.7622 0.02448 b c a
71 0.1145 0.7818 0.5363 0.1196 c a b
72 0.004795 0.8881 0.7622 0.02448 c b a
73 0.001492 0.8991 0.8094 0.01288 a b c
74 0.1149 0.781 0.5864 0.113 a c b
75 0.001491 0.8991 0.8095 0.01287 b a c
2 3
76 0.006939 0.8381 0.6941 0.02777 b c a
77 0.1286 0.7548 0.5368 0.1195 c a b
78 0.004811 0.8877 0.7879 0.02312 c b a
79 0.002006 0.8643 0.7903 0.01351 a b c
80 0.2144 0.5912 0.3682 0.1396 a c b
81 0.002298 0.8445 0.7597 0.01445 b a c
1 3
82 0.007565 0.8235 0.7272 0.02622 b c a
83 0.1699 0.6761 0.4995 0.1243 c a b
84 0.005614 0.869 0.7975 0.02259 c b a
Vol. 23 [2018], Bund. 1 54

Table 9: Performance of Polynomial Equation (Degree: x=2; y=1,2,3)


[a: Percentage of Fines Content; b: Minimum Dry Density; c: Tangent of Angle of Internal
Friction (in radian)]
Degree of
Independent Goodness of Fit Parameters
Simulation Variables X Y Z
No. Data Data Data
Adjusted
x y SSE R-squared RMSE
R-square
85 0.002543 0.828 0.7563 0.01456 a b c
86 0.1668 0.682 0.5494 0.1179 a c b
87 0.002543 0.828 0.7563 0.01456 b a c
2 2
88 0.007041 0.8357 0.7672 0.02422 b c a
89 0.1668 0.682 0.5494 0.1179 c a b
90 0.007041 0.8357 0.7672 0.02422 c b a
91 0.002819 0.8093 0.7506 0.01473 a b c
92 0.1834 0.6503 0.5427 0.1188 a c b
93 0.002548 0.8276 0.7746 0.014 b a c
2 1
94 0.007329 0.829 0.7763 0.02374 b c a
95 0.2178 0.5848 0.4571 0.1294 c a b
96 0.007636 0.8218 0.767 0.02424 c b a

Table 10: Performance of Polynomial Equation (Degree: x=1,2,3; y=2)


[a: Percentage of Fines Content; b: Minimum Dry Density; c: Tangent of Angle of Internal
Friction (in radian)]
Degree of
Independent Goodness of Fit Parameters
Simulation Variables X Y Z
No. Data Data Data
Adjusted
x y SSE R-squared RMSE
R-square
97 0.002543 0.828 0.7563 0.01456 a b c
98 0.1668 0.682 0.5494 0.1179 a c b
99 0.002543 0.828 0.7563 0.01456 b a c
2 2
100 0.007041 0.8357 0.7672 0.02422 b c a
101 0.1668 0.682 0.5494 0.1179 c a b
102 0.007041 0.8357 0.7672 0.02422 c b a
103 0.002548 0.8276 0.7746 0.014 a b c
104 0.2178 0.5848 0.4571 0.1294 a c b
105 0.002819 0.8093 0.7506 0.01473 b a c
1 2
106 0.007636 0.8218 0.767 0.02424 b c a
107 0.1834 0.6503 0.5427 0.1188 c a b
108 0.007329 0.829 0.7763 0.02374 c b a
Vol. 23 [2018], Bund. 1 55

Table 11: Performance of Polynomial Equation (Degree: x=1; y=1)


[a: Percentage of Fines Content; b: Minimum Dry Density; c: Tangent of Angle of Internal
Friction (in radian)]

Degree of
Independent Goodness of Fit Parameters
Simulation Variables X Y Z
No. Data Data Data
Adjusted
x y SSE R-squared RMSE
R-square

109 0.005074 0.6568 0.611 0.01839 a b c


110 0.3239 0.3825 0.3001 0.1469 a c b
111 0.005074 0.6568 0.611 0.01839 b a c
1 1
112 0.0113 0.7362 0.701 0.02745 b c a
113 0.3239 0.3825 0.3001 0.1469 c a b
114 0.0113 0.7362 0.701 0.02745 c b a

Table 12: Performance of MATLAB Custom Equation


[a: Percentage of Fines Content; b: Minimum Dry Density; c: Tangent of Angle of Internal
Friction (in radian)]

Goodness of Fit Parameters


Simulation
X Data Y Data Z Data
No.
Adjusted
SSE R-squared RMSE
R-square
115 0.005068 0.6572 0.5517 0.01975 a b c
116 6.364 -11.13 -14.87 0.6997 a c b
117 0.01473 0.003585 -0.303 0.03366 b a c
118 0.0106 0.7525 0.6764 0.02856 b c a
119 0.2925 0.4424 0.2708 0.15 c a b
120 0.02349 0.4517 0.283 0.04251 c b a

© 2018 ejge
Vol. 23 [2018], Bund. 1 56

Editor’s note.
This paper may be referred to, in other articles, as:
Md. Enayet Chowdhury, Md. Azijul Islam, Tahsina Islam, Nafeesa Khan:
“Evaluation of Shear Strength of Cohesionless Soil from Maximum,
Minimum Dry Density and Fines Content using Polynomial Surface
Fitting Method” Electronic Journal of Geotechnical Engineering, 2018
(23.01), pp 31-56. Available at ejge.com.

View publication stats

You might also like