2021-ALANEME-Mechanical Properties Optimization of Soil-Cement Kiln Dust Mixture Using Extreme Vertex Design

International Journal of Pavement Research and Technology
https://doi.org/10.1007/s42947-021-00048-8
ORIGINAL RESEARCH PAPER
Mechanical Properties Optimization of Soil—Cement Kiln Dust

Mixture Using Extreme Vertex Design
George Uwadiegwu Alaneme1 · Imoh Christopher Attah2 · Roland Kufre Etim2 ·
Mark Uzochukwu Dimonyeka1
Received: 23 October 2020 / Revised: 25 April 2021 / Accepted: 2 May 2021

© The Author(s), under exclusive licence to Chinese Society of Pavement Engineering 2021
Abstract
As a result of environmental issues, the use waste residue has gained much attention in the field of soil re-engineering, this is
because of the feasibility of using waste derivatives in soil amelioration protocols. This necessitated the current study to deal
with the utilization of an industrial residue termed as cement kiln dust (CKD) in enhancing the mechanical performance of
black expansive clayey material. The amelioration protocols were as a result of the poor engineering performance of black
cotton soil thereby becoming a road cancer material. The extreme vertex design (EVD) is a flexible approach and was adopted
for the mixture experimental design and modelling of the mechanical properties of problematic black cotton soil—cement
kiln dust blend. The statistical analyses and or approaches engaged in course of this study were carried out using Minitab 18
and Design Expert statistical software. In the current study, the responses considered include California bearing ratio (soaked
and unsoaked) and unconfined compressive strength test. The corresponding experimental responses were then achieved in
the laboratory and used for analysis and model development. Statistical diagnostics and influence tests carried out on the
developed model showed a good correlation with the actual results. However, using the EVD design of experiment approach,
the peak performance of soil-CKD was achieved at the mixture combination of 0.45, 0.443 and 0.107% for soil, CKD and
water, respectively. The strength outcomes indicate that cement kiln dust could be useful in ameliorating expansive soil for
sub-base material of low trafficked roads and as well reduce cost of cement kiln dust residue disposal.
Keywords Black cotton soil · Desirability function · Constrained simplex method · Design expert
1 Introduction soil may not be suitable for engineering works due to the
presence of soft sediments which makes them exhibit high
Black cotton soil is largely an inorganic clay which possesses volume change when there is significant change in moisture
higher fraction of fine grained particles, higher plasticity, content [4–15].
compressibility and expansive characteristics [1]. They are The production of Portland cement which is an essen-
confined to the temperate climatic and semi-arid zones and tial ingredient for construction works is accompanied by
very suitable for cotton cultivation [2]. Its mineralogical large quantities generation of by-product material in the
composition is dominated by montmorillonite clay mineral form of cement kiln dust (CKD) which is not suitable for
and is abundant in zones where the annual precipitation is re-use in the cement production process and is disposed
less than the evaporation [3]. Quite number of studies have in millions of tons annually [16]. There has been great
reported that this type of soil as well as other types deficient interest by researchers in providing applications for CKD
so as to curtail the high cost associated with the disposal
of these industrial waste and also related environmental
* Imoh Christopher Attah
[email protected] degradation challenges [17–19]. It has been applied in
various fields; first as a waste water streams stabilizer, as
1
Civil Engineering Department, Michael Okpara University asphalts anti-stripping agent, soil fertilizer and masonry
of Agriculture, P. M. B. 7267, UmudikeUmuahia 440109, products. In the cement manufacturing process, the kiln
Abia State, Nigeria
rotates the cement ingredients (raw materials) gradually
2
Department of Civil Engineering, Akwa Ibom State from the upper to the lower end and controlled by the
University, Ikot Akpaden, Nigeria
13
Vol.:(0123456789)
G. U. Alaneme et al.
rotational speed and slope of the kiln. In the hotter end of Table 1 Design constraints
the kiln, Chlorine, Potassium, Sodium and other elements Mixture coding: Actual
present in the raw materials are wholly or partially volatil- Low Constraint High
ize which are not allowed into the clinker [20, 21]. Rapid
flow of air is supplied for combustion of the fuel which 0.450 A: Soil 0.650
moves against the raw materials flow and the turbulent 0.250 B: CKD 0.450
nature of raw feed agitation and swift gas flow results in 0.100 C: Water 0.250
large volume of particulate matter to be entrained during A+B+C 1.000
gases combustion. The gas flow takes up the volatilized
and partially burned raw materials in the kiln while the
entrained precipitates consists of CKD which is removed Table 2 Design matrix *** Mixture component coding
and collected from the exhaust gases of the kiln by pollu- evaluation for mixture special is L-pseudo ***
cubic model 3 factors: A, B
tion control equipment [22]. To achieve low alkali clinker and C Degrees of freedom for evalu-
from high alkali raw materials and also to ensure consist- ation
ent compliant operation, there is larger quantity of CKD
Model 6
generated as industrial waste. Additionally, there are high
Residuals 5
concentration of developed volatiles deposited at the kiln’s
Lack of fit 4
walls which causes plants shut down frequently. Therefore,
Pure error 1
CKD is generated from the cement manufacturing plants
Corr. Total 11
so as to eject volatile alkalis, sulphates and chlorides
within the kiln system. Cement kiln dust (CKD) is a highly
alkaline waste and fine grained by-product obtained from
the cement exhaust gas control air pollution devices. The experimental runs to derive the optimum consistency-time
chemical and physical characteristics of CKD may vary for composite and consolidation properties with the substrate
from one cement manufacturing plant to another depend- OPC concrete at 34.10 and 34.90 sec. The apparent max
ing on the component ingredients (raw materials) utilized density was obtained at the range of 97.10 to 98.0% of the
and method of collection in the plant [23, 24]. free theoretical density.
There is need to improve the engineering properties of In line with the foregoing, expansive soils are blended
expansive soils to be used as pavement subgrade material with stabilizing agents to enhance its performance in con-
using industrial waste to encourage recycling and re-use of struction works using several mixture experimental design
industrial waste. To make clear the relationship between the techniques [29, 30]. Mixture experiment is a special case
factor levels under investigation and their corresponding of response surface method where the property understudy
responses (correlations), advanced mixture design methods compete with the existing ones. For the purpose of model-
have been employed by several researchers [25–27]. Sala- ling, all sort after responses are first experimentally meas-
hudeen et al. [21] performed soil stabilization assessment ured for each of the possible mixture combination in the
in CKD blended black soil using artificial neural network design space after which the generated responses were mod-
(ANN) multilayer with perception back propagation algo- elled as a function of the mixture components using polyno-
rithm. Ten input variables which were obtained experimen- mial fractions based on mathematical formulations [31]. As
tally and constitute the general engineering behaviour of the simplex design for mixture experiments places lower bounds
soil blended mixture in terms of effective grain size, grada- only on the factor levels, there are conditions where the use
tion coefficients, swell-shrinkage and specific gravity. The of complex constraints is appropriate or required. Extreme
output variables are two, namely max dry density (MDD) vertices design (EVD) is very important in this case as it is
and optimum mixture content (OMC). The developed model flexible enough to allow the imposition of additional con-
was validated using loss function parameters P value and straints factor levels by specification of both upper and lower
MSE. The simulated network performance was satisfac- bounds on the components through the definition of linear
tory with P value of 0.9884 and 0.983 for MDD and OMC, constraints for blends [32]. EVD is a mixture design which
respectively. Also, Olubanwo et al. [28] investigated the uti- occupies a smaller space or sub portion within the simplex.
lization of optimization techniques in the mixture of material The design is important when the design factor space cho-
experiments concept for designing and proportioning the sen is not L-simplex design. This condition is imposed by
cementitious portion of a bounded roller compacted fibre both upper and lower bound constraints in the factor levels
reinforcement polymer modified concrete (BRCFRPMC). when there are series of inter-dependencies between the
By constraining the variability range of the constituent mixture components which result to setting of lower and
paste, a feasible design space was generated having 13 upper bounds for the ingredients [33].
13
Mechanical Properties Optimization of Soil—Cement Kiln Dust Mixture Using Extreme Vertex Design
In design of experiments, optimal designs are experimen- to evaluate experimental designs. Design optimality depends
tal design types which are optimal compared to the required on statistical model evaluated with respect to prescribed sta-
statistical criterion. Optimal designs permit variables or fac- tistical criterion, which is associated with the estimator’s
tors to be predicted with minimum variance and without variance matrix. When several variables are assessed in the
bias. For non-optimal designs, more numbers of experimen- given statistical model, the inverse of the variance matrix is
tal runs would be needed using the same precision as opti- termed information matrix. The information matrix is com-
mal designs to predict the factor variables. This would result pressed or simplified with real valued summary statistics.
to more experimentation cost. Statistical criteria are utilized D-Optimality which is a popular criterion which maximizes
the information matrix( X ′ X or) minimizes the information
| |
determinant matrix | X � X −1 | of the design. However,
Table 3 Power calculations are performed using response type "Con- | |
tinuous" I-Optimality is a criterion associated with the variance of
predictions and minimizes the average prediction variance
Power at 5% alpha level to detect signal/noise ratio of
over the design factor space.
Term Std. err.1 VIF Ri-square 2 Std. dev. The objective of EVD is to select design points that
A 0.99 2.41 0.5851 21.7% appropriately cover the design space. This occurs due to
B 0.95 2.34 0.5734 21.8% imposition of additional constraints of lower and upper
C 1.94 7.01 0.8574 13.7% boundary conditions in the mixture components which result
AB 4.67 2.85 0.6496 28.7% in the design points occupying smaller portion of the sim-
AC 5.75 6.02 0.8338 20.6% plex termed the constrained region [34]. The constrained
BC 6.54 5.31 0.8118 17.1% mixture is of the general form for a single component con-
ABC 26.55 2.84 0.6474 37.8% straints (SCC) where q is the total number of ingredients
as presented in the Eqs. 1 and 2. When the component
1
Basis Std. dev. = 1.0
Fig. 1 Contour space and factor

space simplex
13
Fig. 2 Experimental factor space of the components in a three-component mixture space
13
proportions are imposed with SCC, the factor space will in Eq. 3. Pseudo-components are essentially utilized for mix-
now take the shape of irregular polyhedron within the sim- ture model fitting because there are relatively high levels of
plex [35]. Thus, the thrust of this study is that despite the multicollinearity among the factor levels at the constrained
numerous deployment of CKD in improving deficient soils, design space and computer aided designs like D-optimal,
to the best of the authors’ understanding they exist minute I-optimal designs in mixture experiments [36, 37].
attempts in optimizing additives proportions for soil treat-
∑
q
ment by means of EVD strategy.
Xj −Lj
Xj∗ = where L = Lj
1−L
j=1 (3)
X1 + X2 + ... + Xq − 1 = 0 (1) < 1 and is the lower boards total sum.
When constructing a design in pseudo components, we
Lj ≤ Xj ≤ Uj for j specify the design points in Xj∗ terms and then convert to the
= 1, 2 ... ... q. (2)
corresponding original component setting using the formula
where, Lj ≥ 0 is the lower region and Uj ≤ 1 is the upper in Eq. 4.
region. xj = Lj + (1 − L)Xj∗ (4)
It is possible to define new set of components consisting
of values ranging from 1 to 0 since the new constrained For a q-component experimental mixture with upper
region of the experimental design space is still a simplex. bound constraints with components where 0 ≤ × 1 ≤ Ui
This makes model fitting and design construction easier shown mathematically in Eq. 5.
through the constrained interest region. These newly gen-
erated components (× j) are termed as pseudo components U − pseudo components =
Uj −Xj ∑
q
U−1 , j = 1, 2..., q where U = Uj
which is defined using the mathematical formula expressed >1 j=1
(5)
The upper bound constraint causes the feasible experi-
Table 4 Measures derived from the information matrix mental portion to be situated entirely inside original simplex
Run Leverage Space type Build type in inverted form only if; as stated mathematically in Eq. 6.
1 0.3413 Interior Model ∑
q
2 0.7698 Edge Model Uj − Umin ≤ 1 (6)

j=1
3 0.3450 Center Center
4 0.3205 Interior Model The minimum of q-upper bound is represented by
5 0.8971 Vertex Model min and the experimental region in this case is termed
U
6 0.6700 Vertex Model U-simplex.
7 0.3194 Interior Lack of fit In this research study, a mixture experimental optimi-
8 0.4475 Edge Model zation and design were carried out using EVD method so
9 0.9747 Vertex Model as to optimize the utilization of cement kiln dust for the
10 0.7879 Edge Model stabilization of expansive clay (black cotton soil) in terms
11 0.6794 Vertex Model of its mechanical characteristics. The optimal ingredients’
12 0.4475 Edge Replicate content are then determined using multiple optimization
Average = 0.5833 criteria through the desirability function and the complex
Table 5 Design summary

File version 10.0.0.3
Study type Mixture Subtype Randomized

Design type I-optimal Coordinate exchange Runs 12
Design model Special cubic Blocks No blocks
Component Name Type Minimum Maximum Mean Std. dev.
A Soil Mixture 0.45 0.65 0.520484 0.0597638

B CKD Mixture 0.25 0.45 0.319125 0.0654976
C Water Mixture 0.1 0.25 0.160392 0.0528161
13
Table 6 Experimental mix proportions Table 8 Particle size fraction of D (mm) Passing (%)
unaltered soil
Actual components Pseudo-components
4.755 99.60
Z1 Z2 Z3 X1 X2 X3 2.000 99.09
1.700 96.915
0.520108 0.316261 0.163631 0.350539 0.331307 0.318155
0.850 95.095
0.548282 0.351718 0.1 0.491408 0.508592 0
0.425 89.15
0.5125 0.3125 0.175 0.3125 0.3125 0.375
0.300 87.025
0.52269 0.324926 0.152384 0.363452 0.374629 0.261918
0.212 84.75
0.45 0.45 0.1 0 1 0
0.075 70.978
0.5 0.25 0.25 0.25 0 0.75
0.039 45.000
0.494652 0.397233 0.108115 0.22326 0.736166 0.040574
0.029 30.555
0.573786 0.25 0.176214 0.618928 0 0.381072
0.020 27.917
0.65 0.25 0.1 1 0 0
0.016 22.566
0.45 0.376859 0.173141 0 0.634297 0.365703
0.011 18.333
0.45 0.3 0.25 0 0.25 0.75
0.008 15.575
0.573786 0.25 0.176214 0.618928 0 0.381072
0.007 13.083
0.005 9.980
nature of soil-additive blends has been simplified using EVD
method. This work provides an insight into the application
of constrained simplex method of experimental design for black expansive soil, it was broken into smaller fragments,
the evaluation of the soil-additive blend engineering proper- air dried, pulverised and as well sieved with BS sieve No.
ties [38]. 4 (4.75 mm aperture). The cement kiln dust used for the
laboratory exercise was collected from a “mole hill” of CKD
dumped at the discharge unit of Lafarge Cement Company in
2 Materials and Methods Calabar. Specific gravity test performed on the cement kiln
dust presented a value of 2.65.
2.1 Test Materials
2.2 Test Methods
The soil material was sourced via method of disturbed
sampling from a deposit in Deba, Gombe State, Nigeria The experimental examinations were investigated upon
(which lies within latitude 10° 12ʹ 42.73ʹʹ N, longitude 11° according to the guiding principles in BS 1377 [39] and
23ʹ 13.56ʹʹ E). The soil material has a greyish black colour BS 1924 [40] for both unaltered and treated soil mixtures,
based on eye inspection. Prior to the usage of the greyish respectively. The tests executed on the natural soil include
particle size analysis, Atterberg limits, specific gravity,
compaction test, California bearing ratio test and uncon-
fined compressive strength test. This study presents a three-
Table 7 Basic properties of the test soil
component mixture experiment consisting of soil, CKD
Description of soil Test results and water; due to the imposition of component constraint at
upper and lower boundaries, the simplex is constrained with
Natural moisture content (%) 20.10
the experimental points positioned at the edges and vertices
Specific gravity 2.40
of the constrained region. The mixture component ratios and
Percentage passing BS No. 200 sieve (75 μm aperture) 70.978
number of experimental runs were then determined using
Liquid limit (%) 56.30
I-optimal design through which the unconfined compres-
Plastic limit (%) 27.60
sive strength (UCS) and California bearing ratio (CBR) tests
Plasticity index (%) 28.70
were carried out in the laboratory. Data generated from the
Free swell (%) 53.50
experimental responses and their corresponding factor levels
Linear shrinkage (%) 18
were then analysed so as to model the soil-CKD mechani-
AASHTO A-7-6 (14)
cal blend behaviour. Statistical diagnostics and influence
USCS CH
tests were also carried out to validate the developed model;
Maximum dry density (Mg/m3) 1.64
numerical and graphical optimization is finally conducted
California bearing ratio (%) 3
using desirability function computation to maximize the
Optimum moisture content (%) 18
13
Table 9 Basic properties of the Oxides SiO2 CaO SO3 MgO TiO2 Fe2O3 Al2O3 Na2O K2O LOI
test soil
Mass fraction (%) Soil* 48.50 0.90 – 2.22 – 2.20 18.60 1.55 0.70 10.10
CKD* 18.82 66.82 2.01 0.01 0.40 2.05 6.34 0.20 1 1.03
*Attah et al. [9]
Table 10 Experimental results Runs Experimental ratios Mass conversion (Kg) Test results
Soil CKD Water Soil CKD Water CBR (S) CBR (US) UCS (7D)
1 0.52 0.32 0.164 2.158447 1.312485 0.679068 21 46 454

2 0.55 0.35 0.100 2.275368 1.459632 0.415 23 48 463
3 0.51 0.31 0.175 2.126875 1.296875 0.72625 20 47 448
4 0.52 0.32 0.152 2.169165 1.348442 0.632392 22 46 456
5 0.45 0.45 0.100 1.8675 1.8675 0.415 27 58 501
6 0.50 0.25 0.250 2.075 1.0375 1.0375 17 34 399
7 0.49 0.40 0.108 2.052806 1.648518 0.448677 26 53 478
8 0.57 0.25 0.176 2.38121 1.0375 0.73129 15 33 388
9 0.65 0.25 0.100 2.6975 1.0375 0.415 12 30 380
10 0.45 0.38 0.173 1.8675 1.563967 0.718533 24 54 470
11 0.45 0.30 0.250 1.8675 1.245 1.0375 19 45 442
12 0.57 0.25 0.176 2.38121 1.0375 0.73129 14 33.5 385
response criteria with respect to the factor levels to obtain were placed inside the loading frame of the UCS testing
the optimal mixture combination of problematic soil-CKD machine [43].
for the maximized mechanical strength response [41].
2.3 Components Mixture Design Formulation
2.2.1 California Bearing Ratio
The determination of the actual ratio of the ingredients to
California bearing ratio test is an indicator of soil strength be mixed for each particular experimental run and also the
parameter and it was executed according to the guidelines total number of experimental runs were carried out here. The
in BS 1377 [39] and BS 1924 [40] for both unaltered and effective ratios obtained here form fundamental bases for the
treated soil mixtures, respectively. The tests were carried model development so as to derive the optimal combination
out for soil materials compacted based on British Stand- ratio for the soil-additives blend and achieve improvement
ard Light (BSL) compaction energy for both soaked and in the problematic soil engineering properties [44].
unsoaked conditions. The soil specimens were compacted
in three layers with the aid of a 2.5 kg rammer and each of 2.3.1 Formulation of Constraints
the three layers received 62 nos. of blows. The soil speci-
mens compacted during the CBR tests were cured for 6 days The mixture components are imposed with upper and lower
and after the sixth day the soil specimens were immersed bounds established through the ingredient material charac-
in water for a period of 48 h. Thereafter, the cured speci- teristics which constitute the experimental blend. In most
mens were subjected to a static loading system by the CBR cases, economical, practical and environmental or physical
machine until failure took place [42]. considerations impose most of these boundary limits. For a
three-component mixture investigated in this state constitut-
2.2.2 Unconfined Compressive Strength ing of the problematic soil, CKD and water; here the soil
is treated with the CKD as admixture at varying values of
The method documented in BS 1377 [39] was used to deter- moisture content to enhance its mechanical properties. From
mine the unconfined compressive strength (UCS) of the soil relevant literatures [45, 46], the component constraints were
specimens. Both the natural soil and cement kiln dust treated formulated using single component constraints (SCC) are
soil mixtures were used for the UCS experimentation. The presented in Table 1.
soil specimens were compacted using BSL and cured for a
duration of 7 days. After the curing exercise, soil specimens
13
Table 11 Model summary Source Std. dev. R-squared Adjusted R-square Predicted R-square PRESS
statistics for UCS response
Linear 0.31 0.9193 0.9013 0.8570 1.50
Quadratic 0.17 0.9835 0.9698 0.8325 1.76
Special cubic 0.097 0.9955 0.9901 0.8312 1.77
Cubic 0.057 0.9994 0.9966 +
Special quartic 0.063 0.9988 0.9958 0.9237 0.80 Suggested
Quartic 0.054 0.9997 0.9969 + Aliased
+ Case(s) with leverage of 1.0000: PRESS statistic not defined

"Model Summary Statistics" Focus on the model maximizing the "Adjusted R-Squared" and the "Predicted
R-Squared"
Table 12 Lack of fit tests for Source Sum of squares df Mean square F value P value Prob > F
UCS response
Linear 0.84 8 0.11 36.22 0.1278
Quadratic 0.17 5 0.034 11.65 0.2187
Special cubic 0.044 4 0.011 3.78 0.3657
Cubic 3.508E-003 1 3.508E-003 1.21 0.4704
Special quartic 9.166E-003 2 4.583E-003 1.57 0.4909 Suggested
Quartic 0.000 0 Aliased
Pure error 2.911E-003 1 2.911E-003
"Lack of Fit Tests" Want the selected model to have insignificant lack of fit
Table 13 Sequential model sum Source Sum of squares df Mean Square F value P value Prob > F
of squares [Type I] for UCS
response Mean vs total 5253.52 1 5253.52
Linear vs mean 9.64 2 4.82 51.23 < 0.0001
Quadratic vs linear 0.67 3 0.22 7.82 0.0170
Sp. cubic vs quadratic 0.13 1 0.13 13.35 0.0147
Cubic vs Sp. cubic 0.041 3 0.014 4.21 0.1978
Quartic vs cubic 3.508E-003 1 3.508E-003 1.21 0.4704 Aliased
Residual 2.911E-003 1 2.911E-003
Sp. quartic vs quadratic 0.16 3 0.053 13.28 0.0308 Suggested
Quartic vs Sp. quartic 9.166E-003 2 4.583E-003 1.57 0.4909 Aliased
Residual 2.911E-003 1 2.911E-003
Total 5264.00 12 438.67
"Sequential Model Sum of Squares [Type I]": Select the highest order polynomial where the additional
terms are significant and the model is not aliased
Table 14 Model summary Source Std. dev. R-squared Adjusted Predicted PRESS
statistics for CBR response R-squared R-squared
Linear 0.20 0.9349 0.9205 0.8634 0.76

Quadratic 0.075 0.9939 0.9888 0.9606 0.22 Suggested
Special cubic 0.074 0.9951 0.9892 0.8825 0.65
Cubic 0.022 0.9998 0.9991 +
Special quartic 0.053 0.9985 0.9943 0.8668 0.74
Quartic 0.031 0.9998 0.9981 + Aliased
+ Case(s) with leverage of 1.0000: PRESS statistic not defined
13
Table 15 Lack of fit tests for Source Sum of squares df Mean square F value P value Prob > F
CBR response
Linear 0.36 8 0.045 47.90 0.1113
Quadratic 0.033 5 6.615E-003 7.04 0.2783 Suggested
Special cubic 0.026 4 6.547E-003 6.97 0.2760
Cubic 1.707E-005 1 1.707E-005 0.018 0.9147
Special quartic 7.646E-003 2 3.823E-003 4.07 0.3309
Quartic 0.000 0 Aliased
Pure error 9.399E-004 1 9.399E-004
"Lack of Fit Tests": Want the selected model to have insignificant lack of fit
Table 16 Sequential model sum Source Sum of squares df Mean square F value P value Prob > F
of squares [Type I] for CBR
response Mean vs total 521.95 1 521.95
Linear vs mean 5.19 2 2.59 64.65 < 0.0001
Quadratic vs linear 0.33 3 0.11 19.23 0.0018 Suggested
Sp. cubic vs quadratic 6.886E-003 1 6.886E-003 1.27 0.3111
Cubic vs Sp. cubic 0.026 3 8.724E-003 18.23 0.0524
Quartic vs cubic 1.707E-005 1 1.707E-005 0.018 0.9147 Aliased
Residual 9.399E-004 1 9.399E-004
Sp. quartic vs quadratic 0.025 3 8.477E-003 2.96 0.1982
Quartic vs Sp. quartic 7.646E-003 2 3.823E-003 4.07 0.3309 Aliased
Residual 9.399E-004 1 9.399E-004
Total 527.50 12 43.96
"Sequential Model Sum of Squares [Type I]": Select the highest order polynomial where the additional
terms are significant and the model is not aliased
2.3.2 Design of Factor Space and Simplex • Standard errors should be smaller and similar within
coefficient type.
The developed constraints which defined the upper and • The Best variance inflation factor (VIF) value is 1. VIFs
lower limits of the SCC imposed on the factor levels cause higher than 10 are cause for concern. VIFs higher than
the factor space to take the shape of a hyper-polyhedron sim- 100 are cause for alarm, signifying poor estimation of the
plex. The feasible experimental region within the simplex coefficients as a result of multicollinearity.
termed the constrained space is then obtained through the • Best Ri-squared is 0. Higher Ri-square indicates corre-
SCC evaluation [47]. The degree of freedom evaluation is lation of the terms which would possibly lead to poor
also carried out through the design matrix assessment com- models.
putation for the mixture design using special cubic model • For the experiment mixture designs the ratios of ingredi-
as shown in Table 2. A minimum of three lack of fit degrees ents must sum to one.
of freedom is recommended which ensures fit test validity. • This is a constraint on the system and causes multicol-
Fewer degrees of freedom will lead to a test that may not linearity to exist, thus increasing the VIFs and the Ri-
detect lack of fit [48]. squares, rendering these statistics useless.
Power computation tests were carried out on the gener- • Use precision-based metrics provided in this program via
ated mixture component constraints using Minitab 18 and fraction of design space (FDS) statistics.
design expert software to determine the standard error, vari-
ances and standard deviations on the mixture terms which
represent the model coefficients situated at the vertices, The design expert software also developed the three-
edges, design planes and centroid of the simplex at 5% level component simplex contour plot and conditions as shown
of alpha as shown in Table 3. in Figs. 1 and 2, showing the positioning of the actual
experimental points within the feasible design space. There
13
Table 17 ANOVA results for UCS response

Response 1 UCS
Transform: Square root Constant: 0

ANOVA for Special Quartic Mixture model
*** Mixture Component Coding is L-Pseudo. ***
Analysis of variance table [Partial sum of squares—Type III]
Source Sum of squares Df Mean square F value P value Prob > F
Model 10.47 8 1.31 325.08 0.0003 Significant

Linear mixture1 9.64 2 4.82 1196.70 < 0.0001
AB 0.21 1 0.21 51.38 0.0056
AC 0.027 1 0.027 6.80 0.0798
BC 1.444E-003 1 1.444E-003 0.36 0.5915
A2BC 0.032 1 0.032 7.97 0.0666
AB2C 0.026 1 0.026 6.35 0.0863
ABC2 2.914E-006 1 2.914E-006 7.239E-004 0.9802
Residual 0.012 3 4.026E-003
Lack of fit 9.166E-003 2 4.583E-003 1.57 0.4909 Not significant
Pure error 2.911E-003 1 2.911E-003
Cor total 10.48 11
1
Inference for linear mixtures uses Type I sums of squares
The Model F value of 325.08 implies the model is significant. There is only a 0.03% chance that an F value this large could occur due to noise
Values of "Prob > F" less than 0.0500 indicate model terms are significant
In this case A, B, C and AB are significant model terms. Values greater than 0.1000 indicate the model terms are not significant. If there are
many insignificant model terms (not counting those required to support hierarchy), model reduction may improve your model
The "Lack of Fit F value" of 1.57 implies the Lack of Fit is not significant relative to the pure error. There is a 49.09% chance that a "Lack of Fit
F value" this large could occur due to noise. Non-significant lack of fit is good the essence is for the model to fit
Table 18 Statistical summary for UCS provide minimum average estimation of the variance across
the experimental regions. It is desirable for (RSM) Response
Std. dev. 0.063 R-squared 0.9988
Surface Methods where estimation is very important and its
Mean 20.92 Adj R-squared 0.9958
algorithm selects the points which minimize the prediction
C.V. % 0.30 Pred R-squared 0.9237
variance across the design space [49, 50]. Condition Number
PRESS 0.80 Adeq precision 52.119
of Coefficient Matrix = 61.376.
− 2 log likelihood − 48.76 BIC − 28.88
AICc 15.24
• If this value is 100–1000, there is moderate to strong
The "Pred R-Squared" of 0.9237 is in reasonable agreement with the multicollinearity.
"Adj R-Squared" of 0.9958; i.e. the difference is less than 0.2 • Values above 1000 indicate severe multicollinearity.
are four space types through which these points where ran- Maximum Variance Mean = 0.975.
domly disposed which include, interior, edge, center and Average Variance Mean = 0.376.
vertex. Information matrix showing the leverages, build and Minimum Variance Mean = 0.205.
space type were calculated. Lack of fit were recorded on one G Efficiency = 59.8%
point at the interior space type while the replicates were also
observed at one point on the edge space type of the simplex • G Efficiency is inversely related to maximum variance.
as shown in Table 4. From the results, average leverage of • Lack of fit runs and replicates tend to reduce the G Effi-
0.5833 was calculated [49, 50]. ciency of a design.
The software determines data statistics for the experi- Scaled D-optimality Criterion = 82.331.
mental design and performs the multicollinearity design, • When comparing designs, a smaller value is better.
G-efficiency, scaled D-optimality and I-optimal design. Determinant of (X′X)−1 = 7.156E + 5.
I-optimal designs, also known as IV (integrated variance), Trace of (X′X)−1 = 808.230.
13
Table 19 ANOVA results for CBR response

Response 2 CBR
Transform: Square root Constant: 0

ANOVA for Quadratic Mixture model
*** Mixture Component Coding is L-Pseudo. ***
Analysis of variance table [Partial sum of squares—Type III]
Source Sum of squares df Mean square F value P value Prob > F
Model 5.51 5 1.10 194.55 < 0.0001 Significant

Linear mixture1 5.19 2 2.59 457.53 < 0.0001
AB 0.15 1 0.15 25.82 0.0023
AC 0.018 1 0.018 3.11 0.1281
BC 0.16 1 0.16 27.36 0.0020
Residual 0.034 6 5.669E-003
Lack of fit 0.033 5 6.615E-003 7.04 0.2783 Not significant
Pure error 9.399E-004 1 9.399E-004
Cor total 5.55 11
1
Inference for linear mixtures uses Type I sums of squares
The Model F value of 194.55 implies the model is significant. There is only a 0.01% chance that an F value this large could occur due to noise
Values of “Prob > F” less than 0.0500 indicate model terms are significant
In this case A, B, C, AB and BC are significant model terms. Values greater than 0.1000 indicate the model terms are not significant. If there are
many insignificant model terms (not counting those required to support hierarchy), model reduction may improve your model
The “Lack of Fit F value” of 7.04 implies the Lack of Fit is not significant relative to the pure error. There is a 27.83% chance that a “Lack of Fit
F value” this large could occur due to noise. Non-significant lack of fit is required
Table 20 Statistical summary for CBR I = 0.37922.

• These can be used in design comparisons with same
Std. dev. 0.075 R-squared 0.9939
number of runs, a smaller value is better. From the com-
Mean 6.60 Adj R-squared 0.9888
putation results the design summary is thus presented in
C.V. % 1.14 Pred R-squared 0.9606
Table 5.
PRESS 0.22 Adeq precision 39.180
− 2 Log Likelihood − 36.34 BIC − 23.91
AICc − 16.34
The “Pred R-Squared” of 0.9606 is in reasonable agreement with the

“Adj R-Squared” of 0.9888; i.e. the difference is less than 0.2 2.3.3 Experimental Mixture Proportion Design
The number of runs was assigned from the informa-

tion matrix where there are spaced at the corners of the
Table 21 Model coefficient Component Coefficient estimate Df Standard error 95% CI low 95% CI high VIF
estimates for UCS
A—Soil 19.51 1 0.063 19.31 19.71 2.42
B—CKD 22.37 1 0.063 22.17 22.57 2.54
C—Water 20.44 1 0.12 20.05 20.83 7.02
AB 2.23 1 0.31 1.24 3.21 3.14
AC − 0.95 1 0.37 − 2.11 0.21 6.02
BC 0.25 1 0.42 − 1.07 1.57 5.34
A2BC 82.36 1 29.18 − 10.50 175.23 99.19
AB2C − 57.03 1 22.64 − 129.08 15.02 61.55
ABC2 − 0.42 1 15.77 − 50.61 49.76 25.58
13
Table 22 Final equation in Sqrt(UCS) = Table 24 Final equation in Sqrt(CBR) =

terms of L-pseudo components terms of L-pseudo components
+ 19.51 *A + 5.48 *A
+ 22.37 *B + 7.56 *B
+ 20.44 *C + 5.85 *C
+ 2.23 *AB + 1.52 *AB
− 0.95 *AC + 0.65 *AC
+ 0.25 *BC + 2.10 *BC
+ 82.36 *A2BC
− 57.03 *AB2C
− 0.42 *ABC2 Table 25 Final equation in Sqrt(CBR) =
terms of real components
− 2.11110 *Soil
experimental region termed space type. Twelve (12) run of − 2.86936 *CKD
experiments were computed in the process and the number − 9.62606 *Water
of runs can be modified to reduce the lack of fit effects + 37.88364 *Soil *CKD
and screen for errors within the factor space. The mix- + 16.34046 *Soil *Water
ture components are bounded by developed constraints to + 52.45508 *CKD *Water
generate proportions for the run of experiments required.
The mix proportion matrix is presented in Table 6. Show-
ing the actual and pseudo components proportions for the 72% of the soil particles fall within the silt–clay fraction
mixture experiments. The results obtained for the run of (Table 8).
experiments which is the response would be utilized for The chemical composition of CKD and test soil is pre-
the model development process of the soil-additive blend sented in Table 9. The results showed that CKD has a high
overall behaviour [51]. content of lime at 66.82%, silica 18.82%, alumina 6.34%,
iron oxide 2.05% and a very low content of magnesium
oxide 0.01% while the black cotton soil has a high content
3 Results, Discussion and Model Validation of silica 48.5%, alumina 18.6%, iron oxide 2.2% and very
low lime content 0.9%. From the result obtained in terms
3.1 Characterization of Test Materials of elemental oxides present in the problematic soil, the
higher content of lime present in CKD would react when
The general classification and engineering behaviour of hydrated with the alumina and silica that is abundant in the
the test soil is presented in Table 7. The results indicate problematic soil to produce calcium silicate hydrate. The
that it possesses high plasticity and swelling potential; it is hydration products obtained from this reaction process are
also poorly graded and exhibit expansive properties with expected to enhance the mechanical behaviour that would
soft materials. Furthermore, the classification by AASHTO probably culminate to improvement of the problematic
approach [52] and USCS [53] produced A-7–6 (14) and CH, black cotton soil [54].
respectively, which shows an unsuitable soil for engineering
works with low CBR of 3%, MDD of 1.64 Mg/m3 and OMC
of 18%. The studied soil has a specific gravity of 2.40 and
from the grain size distribution test of the unaltered soil,
Table 23 Model coefficient Component Coefficient df Standard error 95% CI low 95% CI high VIF
estimates for UCS estimate
A—Soil 5.48 1 0.073 5.30 5.66 2.32

B—CKD 7.56 1 0.069 7.39 7.73 2.21
C—Water 5.85 1 0.13 5.52 6.17 5.79
AB 1.52 1 0.30 0.79 2.25 2.06
AC 0.65 1 0.37 − 0.25 1.56 4.40
BC 2.10 1 0.40 1.12 3.08 3.53
CI Confidence interval
13
Fig. 3 Normal plot of studentized residuals for UCS response
3.2 Mechanical Properties of the Treated Soil ranges from 30 to 58 with max to min ratio 1.9333 for
CBR responses. The fit summary, diagnostic tests, numeri-
From the formulated mixture component proportions cal and graphical optimization were carried out to deter-
through the twelve runs of experiment, the actual ratios mine the optimal mixture proportion of the problematic
were converted to the effective mass (Kg) and the mix- soil-CKD blend so as to maximize the mechanical strength
ture ingredients are then weighed according to the mass response. Post analysis, confirmation and coefficient tables
conversion values for each experimental run. The Califor- were then generated to validate the model results using
nia bearing ratio (CBR) and unconfined compressive test design expert and Minitab 18 software [56, 57].
(UCS) were carried out on the soil-additive blend with the
experimental results presented in Table 10. The results
indicate an improved mechanical strength performance 4.1 Fit Summary
due to the blending of CKD from 30–40%, soil at 40–50%
and water at 5–10% [55]. Fit summary is a collection of relevant statistical tools
which helps to choose the required final model initial
or starting point. The results presented include sum of
4 Model Development and Validation squares, lack of fit, R-squared and summary statistics.
Several relevant statistical computation table which
For the experimental response data processing, the would enable us to determine which model to select for
required transformation for the analysis quadratic (square in-depth study. The full-order model which meets the cri-
root) with the response ranges from 380 to 501 with a ratio teria specified is ‘suggested’. Aliased models are derived
of max to min 1.31842 for the UCS response and response through the software computation if there are not adequate
13
Fig. 4 Normal plot of studentized residuals for CBR response
unique experimental design points for the model coeffi- 4.1.1 Fit Summary for UCS Response
cients prediction. Through the sequential sum of squares
computation, the sum of squares and P value Prob > F is The fit summary for UCS response is presented in Tables 11,
expected to be minimum to indicate which polynomial 12, 13 and showed preference for the special quartic model
model improved the result the most. The total number of with R-squared, adjusted and predicted R-squared of 0.9988,
model coefficients added sequentially is equal to degree 0.9958 and 0.9237, respectively. The lack of fit test results
of freedom for each source. The lack of fit sum of squares showed sum of squares of 9.166 × 10–003, mean square of
utilizes the F value which is compared with the variations 4.583 × 10–003 and lack of fit p value (Prob > F) of 0.4909
in average response differences at the design points. The higher significant lack of fit p value is selected and used as
lack of fit tests compare the error residual with the error the response predictor [49, 50].
(pure) due to replicated design points. When lack of fit
error is bigger than error (pure), it indicates the residual 4.1.2 CBR Fit Summary
error contains some values that can be taken care of by
more appropriate model. The fit summary for CBR response is presented in
Tables 14, 15, 16 showed preference for the quadratic
model with R-squared, adjusted and predicted R-squared
of 0.9939, 0.988 and 0.9606, respectively. The lack of fit
test results showed sum of squares of 0.033, minimum
13
Fig. 5 Residuals vs. predicted plot for UCS response
Fig. 6 Residuals vs. predicted plot for CBR response
13
Design-Expert® Software Residuals vs. Run

Sqrt(UCS)
20.00
Color points by value of
Sqrt(UCS): 15.4435
22.383
Externally Studentized Residuals

19.494
10.00
0
0.00
-10.00
-15.4435
-20.00
1 3 5 7 9 11
Run Number
Fig. 7 Residuals vs. experimental run plot for UCS response
Design-Expert® Software
Residuals vs. Run
Sqrt(CBR)
6.00
4.9 825 3
Sqrt(CBR):
7.616
4.00
5.477
Externally Studentized esiduals
2.00
R
0
0.00
-2.00
-4.00
-4.9 82 53
-6.00
1 3 5 7 9 11
Run Number
Fig. 8 Residuals vs. experimental run plot for CBR response
mean square of 6.615 × 1 0 –003 and lack of fit p value Special quartic model was prescribed for UCS response
(Prob > F) of 0.2783 [49, 50]. while quadratic model was selected for CBR response to
determine the statistical significance for the mixture fac-
tor levels using pseudo coding. The ANOVA computation
4.2 Analysis of Variance (ANOVA) results using square root transformation is presented in
Tables 17 and 18 for UCS response and Tables 19 and 20
Analysis of variance was carried out with respect to the for CBR response [58].
model source selecting during fit summary computations.
13
Predicted vs. Actual
Sqrt(UCS)
23
Sqrt(UCS):
22.383
19.494
22
Predicted
21
20
19
19 20 21 22 23
Actual
Fig. 9 Predicted vs. actual plot for UCS response
Predicted vs. Actual
Sqrt(CBR)
8
Sqrt(CBR):
7.616
7.5
5.477
7
Predicted
6.5
5.5
5 5.5 6 6.5 7 7.5 8
Actual
Fig. 10 Predicted vs. actual plot for CBR response
"Adeq Precision" measures the signal to noise ratio. design points. If the factor is orthogonal with respect to
A ratio greater than 4 is desirable. The ratio of 52.119 all other factors in the model, the VIF = 1 [59] (Table 22).
indicates an adequate signal. This model can be used to The equation in terms of coded factors can be used to
navigate the design space [49, 50]. make predictions about the response for given levels of
each factor. By default, the high levels of the factors are
4.2.1 Coefficient Estimates and Model Equations for UCS coded as + 1 and the low levels of the factors are coded
as − 1. The coded equation is useful for identifying the
The components, coefficient estimate, degrees of freedom, relative impact of the factors by comparing the factor coef-
standard error, variance inflation factor (VIF) and final ficients [49].
equations terms of L-pseudo components computation “Adeq Precision” measures the signal to noise ratio.
results are presented in Tables 18 and 21. VIF measures A ratio greater than 4 is desirable. The ratio of 39.180
the extent to which the variance of the coefficient estimate indicates an adequate signal. This model can be used to
(predictor) is inflated by the lack of orthogonality in the navigate the design space [49, 50].
13
Design-Expert® Software Box-Cox Plot for Power Transforms

Sqrt(UCS)
4.4
Lambda
Current = 0.5
Best = 3
Low C.I. = 4.2
High C.I. =
Recommend transform: 4
None
Ln(ResidualSS)
(Lambda = 1)
3.8
3.6
3.4
3.2
2.8
-3 -2 -1 0 1 2 3
Lambda
Fig. 11 Power transforms Box-Cox plot for UCS response
Design-Expert® Software Box-Cox Plot for Power Transforms

Sqrt(CBR)
3.5
Lambda
Current = 0.5
Best = 2.79
Low C.I. = 0.31
High C.I. = 5.07
3
Recommend transform:
None
(Lambda = 1)
Ln(ResidualSS)
2.5
2
1 .84 537
1.5
-3 -2 -1 0 1 2 3
Lambda
Fig. 12 Power transforms Box-Cox plot for UCS response
4.2.2 Coefficient Estimates and Model Equations for CBR coded equation is useful for identifying the relative impact
of the factors by comparing the factor coefficients.
The components, coefficient estimate, degrees of freedom, The equation in terms of actual factors can be used to
standard error, variance inflation factor (VIF) and final equa- make predictions about the response for given levels of each
tions terms of L-pseudo components computation results for factor. Here, the levels should be specified in the original
the CBR response are presented in Tables 23, 24, 25. units for each factor. This equation should not be used to
The equation in terms of coded factors can be used to determine the relative impact of each factor because the
make predictions about the response for given levels of each coefficients are scaled to accommodate the units of each
factor. By default, the high levels of the factors are coded factor and the intercept is not at the center of the design
as + 1 and the low levels of the factors are coded as − 1. The space [60].
13
Cook's Distance
Sqrt(UCS)
12
Sqrt(UCS):
22.383
10
19.494
Cook's Distance
6
2
1
0
0
1 3 5 7 9 11
Run Number
Fig. 13 The Cook’s distance for UCS response
Cook's Distance
Sqrt(CBR)
3
Sqrt(CBR):
7.616
2.5
5.477
2
Cook's Distance
1.5
1
1
0.5
0
0
1 3 5 7 9 11
Run Number
Fig. 14 The Cook’s distance for UCS response
4.3 Diagnostics Plots studentized residual which is quotient of the residual and

its predicted standard deviation. Studentized residuals are
The regression diagnostics utilized for the verification of essentially used as outliers’ detector; Outlier is group of data
regression model assumptions and state if there are obser- which differs significantly from other observations due to
vations with huge or undue influence on the analysis using measurement variability or as an indication of experimental
13
Leverage vs. Run
Sqrt(UCS)
1.00
Sqrt(UCS):
22.383
19.494 0.80
0.7 5
0.60
Leverage
0.40
0.20
0
0.00
1 3 5 7 9 11
Run Number
Fig. 15 Leverage vs. Run plot for UCS response
Leverage vs. Run

Sqrt(CBR) 1.00
1

Sqrt(CBR):
7.616
0.80
5.477
0.60
Leverage
0 .5
0.40
0.20
0
0.00
1 3 5 7 9 11
Run Number
Fig. 16 Leverage vs. run plot for CBR response
error. Studentizing the residuals maps all the different nor- a better analysis [62]. Normal probability plot for UCS and
mal distributions to a single standard normal distribution. CBR responses are presented in Figs. 3 and 4.
Diagnostic statistical tests were carried out with respect to
UCS and CBR responses [61]. 4.3.2 Residual vs. Predicted Plot
4.3.1 Normal Probability Plot This statistical diagnostic test verifies the assumption of
constant variance with the externally studentized residuals
Normal probability plot checks that the errors are normally on the y-axis and the predicted values on the x-axis. The
roughly distributed which indicates that many of the residu- plot for UCS and CBR responses is presented in Figs. 5 and
als are positioned near the line of fit and not far away. It has 6, respectively. The result implies an expanding variance
very essential significance for the model estimation. Look which indicates the need for a transformation. The scattered
only for definite patterns, like an “S-shaped” curve, which plot were very close to the zero studentized residual points
indicates that a transformation of the response may provide with the maximum and minimum of 15.4435 and − 15.4435,
13
DFFITS vs. Run
Sqrt(UCS)
10
Sqrt(UCS):
22.383
19.494 5
2.5 980 8
0
0
DFFITS
-2.5 98 08
-5
-10
-15
1 3 5 7 9 11
Run Number
Fig. 17 DFFITS vs. runs plot for UCS response
DFFITS vs. Run
Sqrt(CBR)
6
Sqrt(CBR):
7.616
5.477 4
2 .12 132
2
DFFITS
0
0
-2 .12 132
-2
-4
1 3 5 7 9 11
Run Number
Fig. 18 DFFITS vs. runs plot for UCS response
respectively for UCS responses and 4.98253 and − 4.98253 experiments on the x-axis. Lurking variables are checked
for CBR responses [63]. which may have influenced the response during the experi-
ment in this statistical computation. The plot for the two
4.3.3 Residuals vs. Run Plot response cases are presented in Figs. 7 and 8. The plot shows
the studentized residuals are close to the line which indicates
This diagnostic statistic shows a plot of the externally stu- a time-related variable lurking in the background [64].
dentized residuals on the y-axis versus the run order of
13
Table 26 Report summary for the diagnostic plots and influence statistics for UCS
Response 1 UCS Transform: Square root Constant: 0.000
Diagnostics case statistics
Run order Actual value Predicted Residual Leverage Internally Externally Cook’s dis- Influence on Standard order
value studentized studentized tance fitted value
residual residual DFFITS
11 21.02 20.97 0.053 0.682 1.491 2.395 0.529 3.5041 12

6 19.97 20.03 − 0.054 0.671 − 1.489 − 2.382 0.503 − 3.4031 11
12 19.62 19.64 − 0.017 0.448 − 0.354 − 0.295 0.011 − 0.266 10
8 19.70 19.64 0.060 0.448 1.265 1.511 0.144 1.362 9
3 21.17 21.17 − 4.972E- 0.908 − 0.258 − 0.213 0.073 − 0.668 8
003
10 21.68 21.72 − 0.043 0.792 − 1.492 − 2.399 0.942 − 4.6821 7
1 21.31 21.29 0.012 0.422 0.258 0.213 0.005 0.182 6
4 21.35 21.36 − 8.457E- 0.733 − 0.258 − 0.213 0.020 − 0.353 5
003
7 21.86 21.86 3.571E-003 0.952 0.258 0.213 0.148 0.953 4
5 22.38 22.37 0.013 0.973 1.229 1.424 6.0451 8.5461 3
9 19.49 19.51 − 0.013 0.980 − 1.439 − 2.110 11.4971 − 14.9161 2
2 21.52 21.52 − 1.582E- 0.991 − 0.258 − 0.213 0.785 − 2.194 1
003
Constant 95% CI 95% CI Best Rec
K Low High Lambda Transform
0.000 3.00 None

1
Exceeds limits
Table 27 Report summary for the diagnostic plots and influence statistics for CBR
Response 2 CBR Transform: Square root Constant: 0.000

Diagnostics case statistics
Run order Actual value Predicted Residual Leverage Internally Externally Cook’s dis- Influence on Standard order
value studentized studentized tance fitted value
residual residual DFFITS
11 6.71 6.67 0.039 0.677 0.907 0.892 0.288 1.292 12

6 5.83 5.88 − 0.046 0.661 − 1.051 − 1.063 0.359 − 1.483 11
12 5.79 5.77 0.016 0.370 0.260 0.239 0.007 0.183 10
8 5.74 5.77 − 0.028 0.370 − 0.465 − 0.433 0.021 − 0.331 9
3 6.86 6.74 0.12 0.206 1.750 2.282 0.133 1.164 8
10 7.35 7.42 − 0.074 0.597 − 1.547 − 1.821 0.592 − 2.2181 7
1 6.78 6.76 0.026 0.210 0.392 0.362 0.007 0.187 6
4 6.78 6.83 − 0.048 0.217 − 0.717 − 0.684 0.024 − 0.360 5
7 7.28 7.35 − 0.066 0.304 − 1.044 − 1.054 0.079 − 0.696 4
5 7.62 7.56 0.052 0.844 1.762 2.314 2.8031 5.3871 3
9 5.48 5.48 − 2.113E- 0.939 − 0.011 − 0.010 0.000 − 0.041 2
004
2 6.93 6.92 0.011 0.604 0.235 0.215 0.014 0.266 1
Constant 95% CI 95% CI Best Rec
K Low High Lambda Transform
0.000 0.31 5.07 2.79 None

1
Exceeds limits
13
Table 28 Optimization criteria Name Goal Lower limit Upper limit Lower Upper weight Importance
definition weight
A: Soil is in range 0.45 0.65 1 1 3

B: CKD is in range 0.25 0.45 1 1 3
C: Water is in range 0.1 0.25 1 1 3
UCS Maximize 380 501 1 1 3
CBR Maximize 30 58 1 1 3
Table 29 Optimization Number Soil CKD Water UCS CBR Desirability

solutions
1 0.450 0.443 0.107 497.758 57.376 0.977 Selected
2 0.450 0.450 0.100 500.430 57.211 0.976
Fig. 19 Optimization ramps
4.3.4 Predicted vs. Actual 4.3.5 Box‑Cox Plot for Power Transforms
This diagnostic plot presents the estimated model response This diagnostic plot test provides guidelines for power law
values on the y-axis versus the actual values on the x-axis. transformation selection. Based on the derived best value for
This plot help to determine a value, or group of values, that lambda, a recommended transformation is then listed which
are not easily predicted by the model in terms of accuracy is situated at the lowest point of the curve generated by the
and is shown in Figs. 9 and 10 for the two response cases. natural log of the sum of squares of the residuals. Box-Cox
The result deduced from the plotted results indicates a strong Power Transforms plot for CBR and UCS responses are pre-
correlation between the experimental and the model presented in Figs. 11 and 12. The result showed best lambda at
dicted values [65]. 3 and 2.79 for UCS and CBR, respectively [66, 67].
13
Fig. 20 Optimization bar graph
4.4 Influence Plots 4.4.3 DFFITS vs. Runs
The statistical influence is evaluated through the cook’s DFFITS is a statistical diagnostic tool which shows how
distance, leverage vs. experimental run and difference in influential experimental points are in a regression analy-
fits (DFFITS) statistics vs. experimental run. The results sis computation. It is the change in the estimated value
are presented in graphical plots which provide a better for experimental point derived in regression when that
perspective on the data points. point is left out and also the product of the leverage fac-
tor and externally studentized residual shown in Figs. 17
and 18 for UCS and CBR responses. The plotted results
4.4.1 The Cook’s Distance indicate DFFITS points with respect to the experimental
runs lie very close to the zero points within the regions of
The cook’s distance is used commonly for the determina- ± 2.59808 and ± 2.12132 for UCS and CBR, respectively
tion of the data point influence when carrying out ordi- [70].
nary least square regression analysis. The influential points
which are particularly worth for validity checks and also
to show planes of the feasible experimental design space 4.4.4 Diagnostic Plots and Influence Statistics Summary
where better performance can be achieved. The cook’s Report
distance vs. experimental run plot for the two response
cases are shown in Figs. 13 and 14. The plotted result indi- The summary report for statistical diagnostic plots and
cates that the cook distance score for the UCS response are influences presenting the predicted and actual values,
mostly within 0 and 1 with run number 5 and 9 dispersed lambda values, the leverage, internally and externally stu-
while the same was observed for CBR response plot where dentized residuals with respect to the generated standard
only run number 5 was dispersed [68]. order for the two response cases as shown in Tables 26
and 27.
4.4.2 Leverage vs. Run
4.5 Numerical Optimization
Leverage measures how much each point influences the
model fit. If a point has a leverage of 1.0, then the model In a constrained design mixture both upper and lower
exactly fits the observation at that point. Experimental run bounds were at priori through a list of all combinations
with leverage greater than 2 times the average is generally expressed in the term [q(2)q−1 + 1]. For possible blends and
regarded as having high leverage, such runs have few other in addition to the model choice, desirability function using
runs near them in the factor space [69]. The average leverage multi-criteria optimization criteria were incorporated.
is the number of terms in the model divided by the number For each of the criteria, values ranging from 0 and 1 are
of experimental runs in the design and the plot for UCS and defined
( with ( the
) scale
) of desirability satisfying the condi-
CBR is presented in Figs. 15 and 16. tion 0 ≤ d yj ≤ 1 , in which 1 signifies corresponding
13
Fig. 21 Trace (piepel) plot
ideal response while 0 shows that one or some of the cri- ⎧ 0 yj < L ⎫
teria lie outside the acceptable values. The rejection or � � ⎪ � y−L �rj ⎪
acceptance condition depends generally on the set aim
d yj = ⎨
T−L
L ≤ yj ≤ T ⎬ (7)
⎪ yj > T ⎪
which is optimization direction to either minimization, ⎩ 1 ⎭
maximization or target [71, 72].
Where the maximization of the response shows that the The minimization of the response indicates that the
bigger value performs better and its desirability function lesser values performed better and its desirability func-
determined using Eq. 7. tion is determined using Eq. 8.
13
Fig. 22 Contour plot
13
Table 30 Confirmation report

Two-sided Confidence = name 95% level n = low level 1 high level Coding
component
A Soil 0.45 0.45 0.65 Actual

B CKD 0.43 0.25 0.45 Actual
C Water 0.12 0.1 0.25 Actual
Total = 1
Predicted Predicted
Response Mean Median Std. dev. N SE Pred. 95% PI low 95% PI high
UCS 493.536 493.532 2.8191 1 N/A 482.244 504.951

CBR 57.4739 57.4682 1.14161 1 N/A 54.036 61.006
⎧ 0 yj < T ⎫ The mixture optimization solution is presented in

� � ⎪ � U−yj �rj ⎪ Table 29. From the desirability function computation, the
d yj = ⎨
U−T
T ≤ yj ≤ U ⎬ (8) solution with the highest score is selected as the optimal
⎪ yj > U ⎪
⎩ 0 ⎭ solution based on the desirability criteria score of 0.977
with mix proportion 0.45:0.44259:0.1071 for the ratio of
While the target shows the best response where its soil, CKD and water, respectively.
desirability function is computed using Eq. 9.
4.5.1 Optimization Ramps and Bar Graph
⎧ 0 yj < L ⎫
⎪ 0 y ⎪
� � ⎪� � j <U ⎪ The numerical optimization ramps show the optimal solution
r
d yj = ⎨ U−yj T ≤ yj ≤ U ⎬ (9)
j
graphical view with the optimal predictor parameter factor
⎪� U−T �rj ⎪
⎪ y−L ⎪ settings in red and the optimal response parameter factor in
⎩ T−L L ≤ y ≤ T ⎭
j
blue. This factor tool helps to make required selection of the
optimal solution in a graphical view presentation as shown
where T is the target value, yj is the estimated result of
in Fig. 19.
jth response, L is the minimum acceptable result, U is the
The bar graph showing the desirability of the predictor
maximum acceptable result and rj is the weight parameter of
and response variables in blue and red colors, respec-
jth desirability function. Based on the above stated boundary
tively, is presented in Fig. 20. From the desirability bar
conditions, a numerical multi-response optimization were
graph, the predictor parameters produced a desirabil-
conducted where the optimum mixture ratio maximizes(the)
ity scores of 1 while the response variables produced
weight average of the individual desirability function d yj
0.973206 and 0.977709 for UCS and CBR, respectively.
within the feasible design space. In this process, an equally
Finally the overall durability score of 0.976957 was gen-
weighted model is selected utilization, the composite desir-
erated [76, 77].
ability of the form expressed in Eq. 10.
( ) ( ) ( ) 1 4.5.2 Optimization Trace Plot
D = [d y1 × d y2 × ... × d yn ] n (10)
where the total individual response number is denoted Trace plots are used to evaluate all the mixture compo-
by n. nents effects in the factor space. The essence of this plot
After analysis of variance (ANOVA) and diagnostic test is to find out sensitivity of the response function compared
statistics, numerical optimization is carried out through to the deviation from the formulation close to the reference
which appropriate ratio combination of factor levels which blend [78, 79]. The trace Piepel plot which has U-pseudo
simultaneously satisfy the criteria (maximize, minimize, coding units on the x-axis for the CBR, UCS and optimal
in range, equal to and target) for each of the predictors and desirability responses is shown in Fig. 20. The contour
response parameters with the imposition of the formulated plot is an important tool for the visualization of the fea-
single components constraints as shown in Table 28. In sible experimental region’s functional points in iteration
range and maximize, criteria were assigned to the predic- solution of mixture optimization. It is a graphical tool for
tors and response variables, respectively [73–75]. 3D-surface representation by contour plotting in terms of
constant slice in 2-D form [80]. The contour plots for the
13
Fig. 23. 3D-Surface plot
13
Table 31 Coefficient table

Response A B C AB AC BC A2BC AB2C ABC2
Sqrt(UCS) 19.5064 22.3702 20.4415 2.22512 − 0.952446 0.249054 82.3648 − 57.0304 − 0.424288
p= < 0.0001 < 0.0001 < 0.0001 0.0056 0.0798 0.5915 0.0666 0.0863 0.9802
Sqrt(CBR) 5.47744 7.56342 5.84685 1.51535 0.653619 2.0982
p= < 0.000 < 0.0001 < 0.0001 0.0023 0.1281 0.0020
P < 0.01; 0.01 <= p < 0.05; 0.05 <= p < 0.10; p >= 0.10
CBR, UCS and optimal desirability responses are shown mixture ingredients bounded by upper and lower lim-
in Fig. 21 its while the sum of the ingredients must be unity. The
Three dimension surface plots are diagrammatic pres- mix ratio designed in this process is utilized for labora-
entation of the three mixture component relationships with tory methodology to derive their respective responses in
respect to the response variables and also for the desir- terms of CBR and UCS characteristics. Data gotten from
ability function shown in Fig. 22 this process are utilized for the model development for
the soil-CKD behaviour. The results obtained showed an
improvement in the mechanical properties of the studied
4.6 Post Analysis soil makes it useful for pavement subgrade materials while
also encouraging the recycling and re-use of industrial
The post analysis computation results showing the confir- waste a very fundamental aspect of waste management
mation report at two-sided confidence of 95%, the descrip- for safe environment. Statistical fit test, ANOVA, diag-
tive statistics of the model predicted results and the com- nostic and influence tests were carried out after generat-
ponent constraints which must sum to one is presented in ing experimental response where model coefficients were
Table 30. The results indicate a component level of 0.45, derived with respect to UCS and CBR responses. UCS
0.43 and 0.12 for soil, CKD and water, respectively [81] response modelling was carried out using special quar-
(Fig. 23). tic model while quadratic model was utilized for CBR
response modelling. Numerical and graphical optimiza-
4.6.1 Coefficient Table tions were further conducted so as to derive the optimal
solution using the desirability scale ranging from 0 to 1.
The coefficient table showing the factor level combina- A desirability score of 0.977 was obtained at optimal mix
tion optimization coefficients of the black cotton soil- ratio of 0.45:0.44259:0.1071 for soil, CKD and water,
CKD blend is presented in Table 31. The special quartic respectively, to produce optimal response of 497.758 kN/
and quadratic models were simultaneously adapted for m2 and 57.3758% for UCS and CBR, respectively. Based
the complex mixture optimization computation where the on the upshot of this study, the incorporation of cement
former was used for UCS response modelling while the kiln dust in the amelioration of mechanical performance
later was for CBR response modelling [82, 83]. of an expansive clay has occasioned a considerable level
of enhancement. However, the optimization of other
inherent soil parameters like resilient modulus could as
5 Conclusion be investigated upon using EVD method and this is part
of the limitation of the study. Finally, from the economic
From the foregoing mixture optimization research using assessment point of view of the current study, it entails
EVD to evaluate the mechanical properties of problematic in numerous folds: (i) the reutilization of waste materials
black cotton soil-CKD blend the following conclusions in soil re-engineering as either cement or lime surrogate
can be drawn. materials facilitates reduction in cost of infrastructural
The treatment of the problematic soil with CKD leads constructions, (ii) it will reduce the amount of C02 emana-
to improvement in the soil’s mechanical property production during cement production process thereby promoting
ing a maximum response of 501 kN/m2 and 58% for UCS sustainable environment and (iii) it will eliminate the rate
and CBR, respectively, at 45% ratio of CKD and 10% of waste in the society and as well reduce the trouble of
ratio of water. Using I-optimal design for factor space, inadequate waste management practices.
the mixture ratios and run of experiments were derived
from the vertex, interior, center and edge of the simplex.
The single component constraints were imposed on the
13
References 14. Attah, I. C., Etim, R. K., & Usanga, I. N. (2021). Potentials of
cement kiln dust and rice husk ash blend on strength of tropi-
cal soil for sustainable road construction material. IOP Confer-
1. Mamatha, K. M., & Dinesh, S. V. (2017). Resilient modulus of
ence Series: Materials Science and Engineering, 1036, 012072.
black cotton soil. International Journal of Pavement Research,
https://doi.org/10.1088/1757-899X/1036/1/012072
10(2), 171–184. https://doi.org/10.1016/j.ijprt.2017.01.008
15. Ekpo, D. U., Fajobi, A. B., Ayodele, A. L., & Etim, R. K. (2021).
2. Sudharani, K., Abhishek, S. K., Adarsh, N., Harish, T., & Man-
Potentials of cement kiln dust-periwinkle shell ash blends on plas-
junath. (2017). Stabilization of black cotton soil using brick dust
ticity properties of two selected tropical soils for use as sustainable
and bagasse ash. International Journal for Scientific Research and
construction materials. IOP Conference Series: Material Science
Development, 5(5), 140–144.
and Engineering, 1036, 012033. https://doi.org/10.1088/1757-
3. Gidigasu, S. S. R., & Gawu, S. K. Y. (2013). The mode of forma-
899X/1036/1/012033
tion, nature and geotechnical characteristics of black cotton soils—a
16. Miller, G. A., & Zaman, M. (2000). Field and laboratory evaluation
review. Standard Scientific Research and Essays, 1(14), 377–390
of cement kiln dust as a soil stabilizer. Transportation Research
4. Etim, R. K., Eberemu, A. O., & Osinubi, K. J. (2017). Stabiliza-
Record, 1714, 25–32
tion of black cotton soil with lime and iron ore tailings admixture.
17. Salahudeen, A. B., Eberemu, O. A., & Osinubi, K. J. (2014).
Journal of Transportation Geotechnic, 10, 85–95. https://doi.org/
Assessment of cement kiln dust-treated expansive soil for the
10.1016/j.trgeo.2017.01.002
construction of flexible pavements. Geotechnical and Geological
5. Moses, G., Etim, R. K., Sani, J. E., & Nwude, M. (2019). Desicca-
Engineering, Springer, 32(4), 923–931
tion-induced volumetric shrinkage characteristics of highly expan-
18. Maslehuddin, M., Al-Amoudi, O. S. B., Shameema, M.,
sive tropical black clay treated with groundnut shell ash for barrier
Rehmana, M. K., & Ibrahim, M. (2008). Usage of cement kiln
consideration. Civil and Environmental Research, 11(8), 58–74.
dust in cement products—Research review and preliminary inves-
https://doi.org/10.7176/CER/11-8-06
tigations. Construction and Building Materials, 22, 2369–2375
6. Etim, R. K., Attah, I. C., Eberemu, A. O., & Yohanna, P. (2019).
19. Shoaib, M. M., Balaha, M. M., & Abdel-Rahman, A. G. (2000).
Compaction behaviour of periwinkle shell ash treated lateritic soil
Influence of cement kiln dust substitution on the mechanical proper-
for use as road sub-base construction material. Journal of Geo-
ties of concrete. Cement and Concrete Research, 30, 337–371
Engineering, 14(3), 179–190. https://doi.org/10.6310/jog.201909_
20. Sreekrishnavilasam, A., King, S., & Santagata, M. (2006). Char-
14(3).7
acterization of fresh and landfilled cement kiln dust for reuse in
7. Attah, I. C., Agunwamba, J. C., Etim, R. K., & Ogarekpe, N. M.
construction applications. Engineering Geology, 85, 165–173
(2019). Modelling and predicting of CBR values of lateritic soil
21. Salahudeen, A. B., Ijimdiya, T. S., Eberemu, A. O., & Osinubi,
treated with metakaolin for road material. ARPN Journal of Engi-
K. J. (2018). Artificial neural networks prediction of compaction
neering and Applied Sciences, 14(20), 3606–3618
characteristics of black cotton soil stabilized with cement kiln dust.
8. Etim, R. K., Attah, I. C., & Yohanna, P. (2020). Experimental study
Journal of Soft Computing in Civil Engineering, 2(3), 53–74
on potential of oyster shell ash in structural strength improvement
22. PCA: The Portland Cement Association. (1992). An analysis of
of lateritic soil for road construction. International Journal of Pave-
selected trace metals in cement kiln dust. PCA.
ment Research Technology, Chinese Society of Pavement Engineer-
23. El-Attar, M. M., Sadek, D. M., & Salah, A. M. (2017). Recycling
ing, 13(4), 341–351. https://doi.org/10.1007/s42947-020-0290-y
of high volumes of cement kiln dust in bricks industry. Journal
9. Attah, I. C., Okafor, F. O., & Ugwu, O. O. (2021). Optimization
of Cleaner Production, 143, 506–515
of California bearing ratio of tropical black clay soil treated with
24. Chaunsali, P., & Peethamparan, S. (2013). Influence of the com-
cement kiln dust and metakaolin blend. International Journal of
position of cement kiln dust on its interaction with fly ash and
Pavement Research and Technology, 14(6), 655–667. https://doi.
slag. Cement and Concrete Research, 54, 106–113. https://doi.
org/10.1007/s42947-020-0003-6
org/10.1016/j.cemconres.2013.09.001
10. Attah, I. C., Etim, R. K., Yohanna, P., & Usanga, I. N. (2021).
25. Alaneme, G. U., Onyelowe, K. C., Onyia, M. E., Van Bui, D.,
Understanding the effect of compaction energies on the strength
Mbadike, E. M., Dimonyeka, M. U., Attah, I. C., Ogbonna, C.,
indices and durability of oyster shell ash-lateritic soil mixtures for
Iro, U. I., Kumari, S., Firoozi, A. A., & Oyagbola, I. (2020).
use in road works. Engineering and Applied Science Research,
Modelling of the swelling potential of soil treated with quicklime-
48(2), 151–160
activated rice husk ash using fuzzy logic. Umudike Journal of
11. Etim, R. K., Attah, I. C., Ogarekpe, N. M., & Robert, E. E. (2018).
Engineering and Technology, 6(1), 1–22
Geotechnical behaviour of lateritic soil—oyster shell ash mixtures.
26. Fityus, S., & Buzzi, O. (2009). The place of expansive clays in the
Proceedings of 16th International Conference and Annual General
framework of unsaturated soil mechanics. Applied Clay Science,
Meeting 2018 of Nigerian Institution of Civil Engineers. Theme:
43, 150–155. https://doi.org/10.1016/j.clay.2008.08.005
Transforming National Economy through Sustainable Civil Engi-
27. Miao, L. C., & Liu, S. Y. (2001). Engineering characteristics of
neering Infrastructure, Paradise 2018. Calabar Intl. Convention
expansive soil and engineering measures. Advances in Science
Centre, Cal., Cross River State. 24–26 October, 2018, pp. 45–52
and Technology of Water Resource, 48(2), 37–40 in Chinese.
12. Etim, R. K., Attah, I. C., Yohanna, P., & Eshiet, S. J. (2018).
28. Olubanwo, A. O., & Karadelis, J. N. (2015). Applied mixture
Geotechnical properties of lateritic soil treated with periwinkle
optimization techniques for paste design of bonded roller-com-
shell ash. Proceedings of 16th International Conference and
pacted fiber reinforced polymer modified concrete (BRCFRPMC)
Annual General Meeting 2018 of Nigerian Institution of Civil
overlays. Materials and Structures, 48, 2023–2042. https://doi.
Engineers. Theme: Transforming National Economy through Sus-
org/10.1617/s11527-014-0291-x
tainable Civil Engineering Infrastructure, Paradise 2018. Calabar
29. Mishra, B. (2015). A study on engineering behavior of black cot-
Intl. Convention Centre, Cal., Cross River State. 24–26 October,
ton soil and its stabilization by use of lime. International Journal
2018, pp. 148–156
of Science and Research, 4(11), 290–294
13. Attah, I. C., Okafor, F. O., Ugwu, O. O. (2021). Experimental
30. Oja, J., & Gundaliya, P. (2012). Study of black cotton soil char-
and optimization study of unconfined compressive strength of
acteristics with cement waste dust and lime. In Proceedings of
ameliorated tropical black clay. Engineering and Applied Science
the Nirma University International Conference on Engineering
Research. 48(3), 238–248. https://doi.org/10.14456/easr.2021.26
(NUiCONE 2012), pp. 110–118, Gujarat, India, December 2013.
13
31. Scheffé, H. (1958). Experiments with mixtures. Journal of the of State Highway and Transportation Official (AASHTO), Wash-
Royal Statistical Society Series B., 20, 344–360 ington D.C.
32. Onyelowe, K. C., Alaneme, G. U., Van Bui, D., Van Nguyen, M., 53. American Standard for Testing Material. (1992). Annual book of
Ezugwu, C., Amhadi, T., Sosa, F., Orji, F., & Ugorji, B. (2019). standards Vol. 04.08. American Society for Testing and Materials.
Generalized review on EVD and constraints simplex method of 54. Alaneme, G. U., Onyelowe, K. C., Onyia, M. E., Van Bui, D.,
materials properties optimization for civil engineering. Civil Mbadike, E. M., Ezugwu, C. N., Dimonyeka, M. U., Attah, I. C.,
Engineering Journal, 5(3), 729–749 Ogbonna, C., Abel, C., Ikpa, C. C., & Udousoro, I. M. (2020).
33 Wangkananon, W., Phuaksaman, C., Koobkokkruad, T., & Modeling volume change properties of hydrated-lime activated rice
Natakankitkul, S. (2018). An extreme vertices mixture design husk ash (HARHA) modified soft soil for construction purposes by
approach to optimization of tyrosinase inhibition effects. Engi- artificial neural network (ANN). Umudike Journal of Engineering
neering Journal, 22(1), 175. https://d oi.o rg/1 0.4 186/e j.2 018.2 2.1. and Technology, 6(1), 88–110
175 55. Warren, K. W., & Kirby, T. M. (2004). Expansive clay soil: A wide-
34. Jian-Tong, D., Pei-Yu, Y., Shu-Lin, L., & Jin-Quan, Z. (1999). spread and costly geohazard. (pp. 24–28). Geostrata: Geo-Institute
Extreme vertices design of concrete with combined mineral of the American Society Civil Engineers.
admixtures. Cement and Concrete Research, 29(6), 957–960. 56. Salahudeen, A. B., & Sadeeq, J. A. (2019). California bearing ratio
https://doi.org/10.1016/S0008-8846(99)00069-1 prediction of modified black clay using artificial neural networks.
35. Cornell, J. A. (2011). Experiments with mixtures: Designs, models In S. Laryea, & E. Essah (Eds.), Proceedings West Africa Built
and the analysis of mixture data. (3rd ed.). John Wiley & Sons. Environment Research (WABER) Conference (pp. 268–281). 5–7
36. McLean, R. A., & Anderson, V. L. (1966). Extreme vertices August 2019 Accra: Ghana. https://doi.org/10.33796/waberconfe
design of mixture experiments. Technometrics, 8(3), 447–454. rence2019.19
https://doi.org/10.1080/00401706.1966.10490377 57. Brereton, R. G. (2003). Chemometrics: Data analysis for the labo-
37. Scheffé, H. (1963). The simplex-centroid design for experiments ratory and chemical plant. John Wiley & Sons.
with mixtures. Journal of the Royal Statistical Society: Series B, 58 Alaneme, G. U., & Mbadike, E. M. (2019). Modelling of the
25, 235–263 mechanical properties of concrete with cement ratio partially
38 Segad, M., Jönsson, B., Åkesson, T., & Cabane, B. (2010). Ca/Na replaced by aluminium waste and sawdust ash using artificial
montmorillonite: Structure, forces and swelling properties. Amer- neural network. SN Applied Sciences. https://doi.org/10.1007/
ican Chemical Society, Langmuir, 26(8), 5782–5790. https://doi. s42452-019-1504-2
org/10.1021/la9036293 59. Atkinson, A. C., Donev, A. N., & Tobias, R. D. (2007). Optimal
39. British Standard (BS) 1377. (1990). Method of testing soils for civil experimental design with SAS. Oxford University Press.
engineering purpose. British Standards Institution, London. 60. Smith, W. F. (2005). Experimental design for formulation. The
40. British Standard (BS) 1924. (1990). Method of testing for stabi- American Statistical Association and the Society for Industrial and
lized soils. British Standard Institution, London. Applied Mathematics.
41. Chen, F. H. (1988). Foundations on expansive soils. (2nd ed.). 61. Shobha, R., Hiremath, R., & Vanaja, K. (2016). Optimization tech-
Elsevier Services Publications. niques in pharmaceutical formulation and processing. Textbook of
42. Onyelowe, K. C. (2017). Mathematical advances in soil bear- Industrial Pharmacy. Drug Delivery Systems, Cosmetic and Herbal
ing capacity. Electronic Journal of Geotechnical Engineering, Drug Technology, pp. 158–168.
22(12), 4735–4743 62. Borkowski, J. J. (2003). Using genetic algorithm to generate small
43. Onyelowe, K. C., Alaneme, G. U., Igboayaka, C., Orji, F., exact response surface designs. Journal of Probability and Statisti-
Ugwuanyi, H., Van Bui, D., & Van Nguyen, M. (2019). Scheffe cal Science, 1, 65–88
optimization of swelling California bearing ratio, compressive 63. Stahle, L., & Wold, S. (1989). Analysis of variance (ANOVA).
strength, and durability potentials of quarry dust stabilized soft Chemometrics and Intelligent Laboratory Systems, 6, 259–272.
clay soil. Materials Science for Energy Technologies, 2(1), https://doi.org/10.1155/2018/9308580
67–77. https://doi.org/10.1016/j.mset.2018.10.005 64. Marquardt, D. W., & Snee, R. D. (1974). Test statistics for mixture
44. Alaneme, G. U., & Mbadike, E. M. (2019). Optimization of flex- models. Technometrics, 14, 533–537
ural strength of palm nut fibre concrete using Scheffe’s theory. 65. Bolton, S. (1997). Optimization techniques in pharmaceutical
Materials Science for Energy Technologies, 2, 272–287. https:// statistics. Practical and clinical applications. (3rd ed.). Marcel
doi.org/10.1016/j.mset.2019.01.006 Dekker.
45. Amhadi, T. S., & Assaf, G. J. (2018). Overview of soil stabilization 66. Cox, D. R. (1971). A note on polynomial response functions for
methods in road construction. GeoMEast, 21–33, 2019. https://doi. mixtures. Biometrika, 58, 155–159. https://doi.org/10.1093/biomet/
org/10.1007/978-3-030-01911-2_3 58.1.155
46. Eriksson, L. (2008). Design of experiments: Principles and applica- 67. Box, G. E. P., & Draper, N. R. (1959). A basis for the selection of a
tions. MKS Umetrics AB. response surface design. Journal of American Statistical Associa-
47. Kumar, R. G., & Sanghvi, I. (2015). Optimization techniques: An tion, 54, 622–654
overview for formulation development. Asian Journal of Pharma- 68. Damiri, S., Pouretedal, H. R., & Bakhshi, O. (2016). An extreme
ceutical Research, 5, 217–221 vertices mixture design approach to the optimization of methylal
48. Schwartz, J., Merck, S., & Dohme, R. L. (1981). Optimization tech- production process using p-toluenesulfonic acid as catalyst. Chemi-
niques in product formulation. Journal of the Society of Cosmetic cal Engineering Research and Design, 112, 155–162. https://doi.
Chemists, 32, 287–301 org/10.1016/j.cherd.2016.06.012
49. Design expert 11. (2018). Design of experiment software. Stat-Ease 69 Snee, R. D., & Marquardt, D. W. (1974). Extreme vertices designs
Inc. for linear mixture models. Technometrics, 16(3), 399–408
50. Minitab 18. (2018). Minitab statistical software. Minitab Inc. 70. Smith, W. F. (1931). Experimental design for formulation. Ameri-
51. Snee, R. D. (1979). Experiment design for mixture systems with can Statistical Association and Society for Industrial and Applied
multicomponent constraints. Communications in Statistics – Theory Mathematics.
and Methods, 17, 149–159 71. Simon, M. J., Lagergreen, E. S., & Synder, K. A. (1997). Concrete
52. Standard Specifications for Transportation, Material and Method of mixture optimization using statistical mixture design methods. In:
Sampling and Testing. (1986). 14th Edition, American Association
13
Proceedings of the PCI/FHWA International symposium on high Imoh Christopher Attah is a lec-
performance concrete, New Orleans, pp. 230–244. turer at the Department of Civil
72 Attah, I. C., Etim, R. K., Alaneme, G. U., & Bassey, O. B. (2020). Engineering, Akwa Ibom State
Optimization of mechanical properties of rice husk ash concrete University, Ikot Akpaden, Nige-
using Scheffe’s theory. SN Applied Science. https://doi.org/10. ria, and currently a Doctoral can-
1007/s42452-020-2727-y didate of the Department of Civil
73. Fedorov, V. V. (1972). Theory of optimal experiments. Academic Engineering, University of Nige-
press. ria, Nsukka, Nigeria. He has
74. Alaneme, G. U., Dimonyeka, M. U., Ezeokpube, G. C., Uzoma, published several scholarly arti-
I. I., & Udousoro, I. M. (2021). Failure assessment of dysfunc- cles in reputable journal houses
tional flexible pavement drainage facility using fuzzy analytical with research focus in: Optimi-
hierarchical process. Innovative Infrastructure Solutions. https:// zation of Civil Engineering
doi.org/10.1007/s41062-021-00487-z Materials, Soil Re-engineering,
75. Coetzer, R., & Haines, L. M. (2017). The construction of D- and Sustainable Civil Infrastructures
I-optimal designs for mixture experiments with linear constraints and Construction Engineering.
on the components. Chemometrics and Intelligent Laboratory He is a Registered Engineer with
Systems, 171, 112–124 the Council for the Regulation of Engineering in Nigeria (COREN) and
76. Esbensen, K. H., Guyot, D., Westad, F., & Houmoller, L. P. a Corporate Member of both Nigeria Society of Engineers (NSE) and
(2002). Multivariate data analysis, in practice: An introduction Nigerian Institution of Civil Engineers (NICE), respectively.
to multivariate data analysis and experimental design. Aalborg
University. Roland Kufre Etim is a lecturer in
77. Gorman, J. W. (1970). Fitting equations to mixture data with the Department of Civil Engi-
restraints on compositions. Journal of Quality Technology, 2, neering, Akwa Ibom State Uni-
186–194 versity, Ikot Akpaden, Nigeria,
78. Mitchell, T. J. (1974). An algorithm for the construction of D-opti- and he is currently on Ph.D pro-
mal experimental designs. Technometrics, 16, 203–210 gramme in the Department of
79. Alaneme, G. U., Mbadike, E. M., Iro, U. I., Udousoro, I. M., & Civil Engineering, Ahmadu
Ifejimalu, W. C. (2021). Adaptive neuro-fuzzy inference system Bello University, Zaria, Nigeria.
prediction model for the mechanical behaviour of rice husk ash He has published several schol-
and periwinkle shell concrete blend for sustainable construction. arly articles in the field of Geo-
Asian Journal of Civil Engineering. https://doi.org/10.1007/ technical and Geoenvironmental
s42107-021-00357-0 Engineering, Sustainable Civil
80. Ozol-Godfrey, A., Anderson-Cook, C. M., & Montgomery, D. Engineering Materials and
C. (2005). Fraction of design space plots for examining model MICP in soil Engineering and
robustness. Journal of Quality Technology, 37, 223–235 Environmental applications. He is a Registered Engineer with the
81. Ding, J. T., Yan, P. Y., Liu, S. L., & Zhu, J. Q. (1999). Extreme Council for the Regulation of Engineering in Nigeria (COREN) and a
vertices design of concrete with combined mineral admixtures. Corporate Member of Nigeria Society of Engineers (NSE).
Cement and Concrete Research, 29(6), 957–960. https://doi.org/
10.1016/S0008-8846(99)00069-1 Mark Uzochukwu Dimonyeka is
82. Hardin, R. H., & Sloane, N. J. A. (1993). A new approach to con- a postgraduate student of
struction of optimal designs. Journal of Statistical and Planning Michael Okpara University of
Inference, 37, 339–369 Agriculture, Umudike, Nigeria.
83. Syafitri, U., Sartono, B., & Goos, P. (2015). I-optimal design of He obtained his B.Eng. (2008) in
mixture experiments in the presence of ingredient availability Civil Engineering Department at
constraints. Journal of Quality Technology, 47, 220–234 Anambra State University, Uli,
and M.Eng. (2021) in Structural
Engineering at Michael Okpara
George Uwadiegwu Alaneme is University of Agriculture,
a postgraduate student of Umudike, Nigeria. He is a mem-
Michael Okpara University of ber of Council of the Regulation
Agriculture, Umudike, Nigeria. of Engineering in Nigeria
He obtained his B.Eng. (2014) in (COREN) and Nigeria Society of
Civil Engineering Department Engineers (NSE). He has pub-
and M.Eng. (2021) in Structural lished several research papers in
Engineering in the same school. reputable journal house with research interest in application of comput-
He has published several ing techniques on sustainable structural and geotechnical materials
research papers in reputable modelling and optimization.
journals with research interests
in applications of soft computing
techniques on sustainable struc-
tural and geotechnical materials,
modelling and optimization.
13

2021-ALANEME-Mechanical Properties Optimization of Soil-Cement Kiln Dust Mixture Using Extreme Vertex Design

Uploaded by

Copyright:

Available Formats

2021-ALANEME-Mechanical Properties Optimization of Soil-Cement Kiln Dust Mixture Using Extreme Vertex Design

Uploaded by

Document Information

Original Title

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

2021-ALANEME-Mechanical Properties Optimization of Soil-Cement Kiln Dust Mixture Using Extreme Vertex Design

Uploaded by

Copyright:

Available Formats

International Journal of Pavement Research and Technology

ORIGINAL RESEARCH PAPER

Mechanical Properties Optimization of Soil—Cement Kiln Dust

Received: 23 October 2020 / Revised: 25 April 2021 / Accepted: 2 May 2021

Fig. 1 Contour space and factor

Fig. 2 Experimental factor space of the components in a three-component mixture space

2 0.7698 Edge Model Uj − Umin ≤ 1 (6)

Table 5 Design summary

Study type Mixture Subtype Randomized

A Soil Mixture 0.45 0.65 0.520484 0.0597638

*Attah et al. [9]

1 0.52 0.32 0.164 2.158447 1.312485 0.679068 21 46 454

+ Case(s) with leverage of 1.0000: PRESS statistic not defined

Linear 0.20 0.9349 0.9205 0.8634 0.76

+ Case(s) with leverage of 1.0000: PRESS statistic not defined

Table 17 ANOVA results for UCS response

Transform: Square root Constant: 0

Model 10.47 8 1.31 325.08 0.0003 Significant

Table 19 ANOVA results for CBR response

Transform: Square root Constant: 0

Model 5.51 5 1.10 194.55 < 0.0001 Significant

Table 20 Statistical summary for CBR I = 0.37922.

The “Pred R-Squared” of 0.9606 is in reasonable agreement with the

The number of runs was assigned from the informa-

Table 22 Final equation in Sqrt(UCS) = Table 24 Final equation in Sqrt(CBR) =

A—Soil 5.48 1 0.073 5.30 5.66 2.32

Fig. 3 Normal plot of studentized residuals for UCS response

Fig. 4 Normal plot of studentized residuals for CBR response

Fig. 5 Residuals vs. predicted plot for UCS response

Fig. 6 Residuals vs. predicted plot for CBR response

Design-Expert® Software Residuals vs. Run

Externally Studentized Residuals

Fig. 7 Residuals vs. experimental run plot for UCS response

Fig. 8 Residuals vs. experimental run plot for CBR response

Fig. 9 Predicted vs. actual plot for UCS response

5 5.5 6 6.5 7 7.5 8

Fig. 10 Predicted vs. actual plot for CBR response

Design-Expert® Software Box-Cox Plot for Power Transforms

Fig. 11 Power transforms Box-Cox plot for UCS response

Design-Expert® Software Box-Cox Plot for Power Transforms

Fig. 12 Power transforms Box-Cox plot for UCS response

Fig. 13 The Cook’s distance for UCS response

Fig. 14 The Cook’s distance for UCS response

4.3 Diagnostics Plots studentized residual which is quotient of the residual and

Fig. 15 Leverage vs. Run plot for UCS response

Leverage vs. Run

Color points by value of

Fig. 16 Leverage vs. run plot for CBR response

Fig. 17 DFFITS vs. runs plot for UCS response

Fig. 18 DFFITS vs. runs plot for UCS response

11 21.02 20.97 0.053 0.682 1.491 2.395 0.529 3.5041 12

0.000 3.00 None

Response 2 CBR Transform: Square root Constant: 0.000

11 6.71 6.67 0.039 0.677 0.907 0.892 0.288 1.292 12

0.000 0.31 5.07 2.79 None

A: Soil is in range 0.45 0.65 1 1 3

Table 29 Optimization Number Soil CKD Water UCS CBR Desirability

Fig. 19 Optimization ramps

4.3.4 Predicted vs. Actual 4.3.5 Box‑Cox Plot for Power Transforms

4.3 Diagnostics Plots studentized residual which is quotient of the residual and

4.3.4 Predicted vs. Actual 4.3.5 Box‑Cox Plot for Power Transforms

4.4 Influence Plots 4.4.3 DFFITS vs. Runs