Construction and Building Materials 163 (2018) 149–159

Contents lists available at ScienceDirect

Construction and Building Materials

journal homepage: www.elsevier.com/locate/conbuildmat

Formulating and optimizing the compressive strength of a raw earth

concrete by mixture design
S. Imanzadeh a,b,⇑, A. Hibouche b, A. Jarno b, S. Taibi b
a
Normandie Univ., INSA Rouen Normandie, LMN, 76000 Rouen, France
b
Normandie Univ., UNIHAVRE, CNRS, L0MC, 76600 Le Havre, France

h i g h l i g h t s

Economical, ecological and workability constraints used to build the mixture design.

Statistical validation: quadratic model for short curing times, linear one for 90-day.

Drying process study from response trace plots: negative role of silt pointed out.

Optimization from surface response plots: minimizing cement and water proportions.

Validation of model with a raw earth concrete implemented on a construction site.

a r t i c l e i n f o a b s t r a c t

Article history: Earthen construction is one of the most common construction technique used since the earliest times.
Received 14 April 2017 The raw material is abundant, requires very low energy to manufacture and does not generate waste.
Received in revised form 3 December 2017 Building today with raw materials, requires noticeable mechanical performance. For this, a raw earth
Accepted 9 December 2017
treatment using the binders is one of the methods used to improve its durability and strength. This paper
presents the use of mixture design as a tool to optimize a raw earth concrete formulation to reach a desir-
able compressive strength. The results show that the mixture design approach can be an important tool
Keywords:
to help develop and optimize a raw earth concrete formulation.
Raw earth concrete
Formulation
Ó 2017 Elsevier Ltd. All rights reserved.
Design of experiments
Mixture design
Response surface methods
Unconfined compressive strength

1. Introduction
From the first centuries of human civilization, the raw earth, by
its properties and performances, was used as a building material at
different levels of complexity [1,2]. The prodigious constructions
made of raw earth by our ancestors have proved a resistance and
a durability that have crossed through the centuries. Nowadays,
contemporary building materials (concrete, steel, etc.) are very
energy-intensive in terms of grey energy, responsible for the emis-
sion of large quantities of greenhouse gases. Hence the need to
develop new ecological and economical materials, known as eco-
geomaterials of construction, mainly raw earth-based materials
with low energy consumption [3,4]. These raw earth materials as
a natural building material have received increasing attention.
⇑ Corresponding author at: Normandie Univ., INSA Rouen Normandie, LMN,
76000 Rouen, France.
E-mail address: saber.imanzadeh@insa-rouen.fr (S. Imanzadeh).
Some of the main advantages of these materials, made from the
earth, are: i) Natural materials are everywhere in the world, avail-
able in large quantities with a low and affordable cost. ii) In com-
parison with industrial building materials such as concrete, earth
material needs approximately 99% less energy during the produc-
tion process [5]. iii) They are recyclable, then they prevent or
reduce the amount of waste. iv) There is no need to use a very
advanced technology for their in-situ implementation, when it is
not the case for materials such as concrete or steel. v) They contain
a fraction of clay which provides the natural cohesion of the mate-
rial and contribute to the strength. In contrast, for a concrete, cohe-
sion requires an amount of cement which is very energy-intensive.
Some important properties of earth material are the mechanical
strength, the shrinkage and swelling, the cracking and the
hygrothermal properties [6].
It is not always adequate to meet the performance required
from a building material and stabilizers are used to enhance the
properties [7]. Among them, when the mechanical strength is con-

https://doi.org/10.1016/j.conbuildmat.2017.12.088
0950-0618/Ó 2017 Elsevier Ltd. All rights reserved.
150 S. Imanzadeh et al. / Construction and Building Materials 163 (2018) 149–159
cerned, stabilizers such as lime, cement and gypsum can be added.
Reinforcement can also be assured by an addition of strong fibers.
Some studies explore the influence of these binders on the raw
earth properties [6,8–10]. As was reported by Delgado and Guer-
rero in a review paper on earth construction in Spain [7], for unsta-
bilized soils, the compressive crushing strength varies in the range
0.60–1.80 MPa for rammed earth and between 0.75 and 2.25 MPa
for Adobe. It can be enhanced by a factor 3.6–5 when cement is
used as a stabilizer, 2.6–3.6 when a mixing of lime and cement is
used and a factor 2–3 when lime is used alone [7]. The given ranges
correspond to different dosages of the stabilizer and different ini-
tial compressive strengths of the unstabilized tested earth mate-
rial. Zak et al. [5] studied the influence of reinforcement by
natural fibers, gypsum and cement on compressive strength of
unfired earth bricks materials. They pointed out that the mixing
of earth with gypsum has no favorable influence on the compres-
sive strength. Pakbaz and Farzi (2015) [11] studied the effect of
mixing methods (dry vs. wet) on the mechanical and hydraulic
properties of treated soil with cement or lime. They showed that
the strength of wet cement treated samples was higher than dry
cement treated samples and this was opposite for lime treated
samples. Minke [12] proposed that for dry building elements made
of raw earth materials, a compressive strength value between 2
and 5 MPa should be considered.
A new concrete based on raw earth material was developed by a
firm from Normandy called Cematerre, in collaboration with the
University of Le Havre, Normandie. Its originality is its ability to
be cast in place like a traditional concrete. For the tested raw earth
concrete, preliminary studies according to an experimental proto-
col under CSTB (Building scientific and technical center (FR)) spec-
ifications have led to the so-called ‘‘ATEx A” official qualification.
Traditionally, in the great majority of studies dealing with the
characterization of materials resulting from mixing, the chosen
experimental design is built on varying successively the different
components of the mixture. However this method requires a great
number of tests to finally obtain the desired product characteris-
tics. Design of Experiments (DoE) method is an alternative
methodology [13]. Based on statistics analysis, it provides the max-
imum amount of relevant information with a selected number of
experimentations. Different families of DoE exist: full factorial,
fractional, composite and D-optimal designs. The application of
these methods in the field of civil engineering is recent. Full facto-
rial, fractional and composite designs are specifically used for reg-
ular experimental regions where every corner of the region of
experiments is accessible [14]. For example they make it possible
to study mortar composition [15] or the influence of Nano-Silica
on the compressive strength and water absorption of mortar mixes
[16]. A factorial design was also built for assessment of properties
of recycled concrete for optimization of the compressive strength
of rubberized concrete [17] and for development of self-
compacting concretes [18]. Concerning the composite design, it
was used for statistical research on phase formation and modifica-
tion of alite polymorphs in cement clinker [19]. D-optimal design is
a computer-aided design, more specifically used for irregular
experimental regions [20,21]. This irregularity is due to restrictions
of the experimental region imposed by specific constraints [14].
The mixture design is a particular design, adapted to study
responses depending on the proportion of mixture components.
Such a design originally developed in the fields of chemistry and
agronomy [22,23] was also recently used in civil engineering. Some
authors used mixture design to develop formulations of low-
strength materials containing polymer concrete [24] or mine tail-
ings [25]. Enhancing the properties of ceramic and pozzolanic
products through mixture design was studied by Nardi et al.
[26]. More application of mixture design in civil engineering can
be found in Yeh et al. [27], Chen et al. [28], Senff et al. [29]. In
the search for economic and ecological materials, Kupaei et al.,
[30] used the mixture design to produce geopolymer lightweight
concrete using locally available waste materials.
The main objective of this paper was to optimize the formula-
tion of raw earth concrete with the aim of improving the mechan-
ical strength. To fulfill this objective, the statistical combinations of
four-constituent mixtures composed of Portland cement, silt, lime
and water were formulated by a D-optimal mixture design to study
unconfined compressive strength. The experimental domain was
defined based on different constraints and choices that are pre-
sented. The mixture design of experiments used to establish model
formulations after 7, 28 and 90-day of curing times. The derived
models were validated. The influence of each mixture component
on the unconfined compressive strength of raw earth concrete
was then studied with the contour plots. Finally, results were ana-
lyzed to improve and optimize formulations of raw earth concrete
materials with two binders (cement and lime) using silt and water.
2. Materials and experimental methods
2.1. Materials
2.1.1. Soil material
The used material is natural silt, chosen because it is a local material, available
in abundance at the site of the planned construction.
Concerning particle size, the analysis is carried out by different methods,
depending on the particle diameters:
– by sieving (dry after washing according to NF P 94-056 standard [31], or wet
according to XP P 94-041 standard [32]) for particles with diameters between
80 lm and 100 mm.
– by sedimentation for particles with diameters less than 80 lm according to NF P
94-057 standard [33].
Table 1 and Fig. 1 show the grading size curve, the effective diameter (D10), the
size of the 60% passing granulates (D60), the Hazen uniformity factor (Cu = D60/D10)
and the curvature factor (Cc = D230/(D60  D10)). The results highlight a sandy loami-
ness material. The sands (0.06 mm–2 mm) are largely predominant with a content
of around 67%. There is also a silt fraction of less than 25% and a small gravel frac-
tion of less than 8%. Moreover, the material has a spreading grading size curve.
Concerning the liquid limit, and given the presence of a large silty fraction, the
test required the use of the Casagrande box with a rough cup.
Concerning the plastic limit, it is almost impossible to measure it according to
the NF P 94-051 standard [34]. To estimate the plastic limit parameter, one of the
methods consists in using the correlation of Biarez and Favre (1975) [35] which
gives the plasticity index value according to the following expression:
IP ¼ 0:73ðW L  13Þ ð1Þ
where Ip is the plasticity index and WL is liquid limit.
Based on grading size curve and Atterberg limits, and according to LPC-USCS
(ASTM D2487-11) standard [36], this material is classified as silty sand (SM).
2.1.2. Binders
Two binders are used, lime and cement. The lime used comes from the Provia-
calÒ DD range, produced by Lhoist group, according to the European standard EN
459-1. It is a calcic quicklime CL 90-Q (R5, P3), containing 90.9% available CaO
and reactivity t60 = 3.3 min [37]. The cement is CEM I 52.5 N, according to the NF
EN197-1 [38], NF P15-318 [39] and NF EN196-10 [40] standards. It is a Portland
Table 1
Silt properties.
Grain distribution Fines content (<80 lm) 35%
Clay particles (<2 lm) 0
Effective size D10 (lm) 32
Uniformity coefficient Cu = D60/D10 4.37
Gradation coefficient Cc 0.94
Atterberg limits Liquid limit wL 20%
Plasticity index Ip 6%
Clay Methylene blue test value VBS 0.5
D60: Diameter corresponding to 60% finer in the particle-size distribution.
 S. Imanzadeh et al. / Construction and Building Materials 163 (2018) 149–159 151

100
90
80
Passing material (%)

70
Force sensor Displacement sensor
60
50
40 Sample
30
Mobile base
20
10
0
1 10 100 1000 10000 Fig. 3. Loading device for UCS test.
Particle size (µm)

Fig. 1. Silt grading size distribution.

3. Statistical analysis of regression models

3.1. Model chosen

cement, composed of 95 to 100% of clinker, with an UCS strength obtained at 28
days of 52.5 MPa (minimal value), with an ordinary short-term strength class (2 Quadratic polynomial model was used for four-components to
or 7 days). find the model that fulfilled the best fit to the experimental mea-
surements so as to make predictions of the unconfined compres-
sive strength (UCS) of a raw earth concrete for any mixture of
2.1.2. Water
The water used for preparing the raw earth concrete was potable tap water
constituents. The quadratic model takes into account binary inter-
from the pipeline in the lab. action effects for all possible pairs of constituents (Eq. (2)).

X
n X
i;j¼n

2.2. Experimental methods UCS ¼ a0 þ ai xi þ aij xi xj ðquadratic modelÞ ð2Þ

2.2.1. Sample preparation
The preparation procedure was established with care in order to obtain homo-
geneous samples. Specimens were prepared in a laboratory mixer with a capacity of
4 L. After a dry mixing for two minutes, the water was added for wet mixing for
three minutes. Thereafter, the molds of 50 mm in diameter and 100 mm in height
are filled by vibration for 2 min using a vibrating table (Fig. 2). Finally, the speci-
mens were stored in controlled laboratory conditions (temperature: T  22 °C and
relative humidity RH  50%) till tests were performed.
2.2.2. Unconfined compressive strength test (UCS)
Three curing times were studied: 7; 28 and 90-day. After a target curing time,
the specimens were sheared on unconfined compression strength path according to
NF P94-420 [41], NF P94-425 [42] French standards.
The experimental device consists of a press with a maximum capacity of 100
kN, connected to an acquisition center (HP model Agilent 34970A) controlled by
a computer. The press is equipped with a force and a displacement sensor which
measure these parameters in the axial direction of the specimen (Fig. 3). Several
strain rates were tested (0.1, 0.5, 1 and 1.5 mm/min). The results showed that there
is no influence of the loading rate on the behavior of the material, i.e. the used raw
earth in the mixture a silty sand, it does not exhibit a viscous behavior as in case of
clays. Consequently, the rate of 0.1 mm/min was chosen for all the tests. All speci-
mens are sheared at water content of about 10% whatever the curing time.
i¼1 i;j¼1
where UCS is the response, i.e. the unconfined compressive strength
of a raw earth concrete (MPa). a0 is a constant. n is the number of
constituents. xi and xj are the amount of constituents in weight per-
cent (wt%). ai the regression coefficients for the linear terms and aij
are the regression coefficients for the binary interaction terms.
If the statistic indicators remain high and quasi-constant when
ignoring some terms in the model, the insignificant model terms
are removed and the model is simplified. As a special case, if the
second order is neglected, the quadratic model (Eq. (2)) converges
to the linear model. The linear model is only acceptable in the
absence of interaction effects between components.
3.2. Diagnostic tools
The Analysis of variance (ANOVA) includes several diagnostic
tools that are used to validate models. The first diagnostic tools
are the regression coefficients R2 and R2adj (see Anex), informing
on the ability of the model to fit the data [43]. They vary between

Fig. 2. Sample preparation a) mixing, b) vibration.

152 S. Imanzadeh et al. / Construction and Building Materials 163 (2018) 149–159

0 and 1. For model lack-of-fit estimation, R2 must be adjusted with Table 2

the number of degrees of freedom. R2adj coefficient is called ANOVA table for the fitting of a model.

explained variance. Moreover, Q2 coefficient (see Annex) estimates Source of variation Degrees of Sum of Mean squares (MS)
the model validity i.e. its capacity to predict new data. A first indi- freedom (DF) squares (SS)
cation of the pertinence of models is given if a high R2 (or R2ajd) Regression p-1 SSR MSR = SSR/DFR
value is detected (R2 or R2adj > 0.9) ensuring a high degree of Residual n-p SSr MSr = SSr/DFr
correlation between the experiments and predicted responses Lack of fit m-p SSLOF MSLOF = SSLOF/DFLOF
Pure error (PE) n-m SSPE MSPE = SSPE/DFPE
and furthermore if models exhibit a very good predictive relevance Total (T) n-1 SST
(Q2 > 0.9). Model validity might be low in very good models due to
extremely good replicates. Significance tests based on variance m = number of distinct levels of the independent variables, n = total number of
observations, p = number of coefficients.
analysis are used to evaluate statistically the quality of the regres-
sion models. The method consists in comparing the ratio of two
variances and calculating the probability that these two variances
this proportion, Ca2+ ions necessary for chemical reactions
belongs to the same distribution [44]. Table 2 shows the ANOVA
(cation exchanges and pozzolanic reactions) saturate the silt
table for regression model evaluation and validation.
material, and the excess lime does not plays its binder role any-
The first F-Test is the regression model significance test. It com-
more. For these reasons, the maximum quantity of lime is lim-
pares variance of the regression (SSR) to residual variance (SSr). SSR
ited to 12%. Then, the following condition must be verified: %
is a sum of squares of differences between predicted values by the
Cement + % Lime < 16%.
regression and the grand average of all the response values. SSr is a
– Workability constraints: The use of the material imposes fluid-
sum of squares of differences or residuals between the experimen-
ity conditions for a suitable workability. This condition was ful-
tal values and the predicted values from the model. It must be ver-
filled by a S3 consistency condition in concrete slump test i.e.,
ified that explained variance of the model is greatly higher than
very plastic concrete according to the standard NF EN 206-1
unexplained variance. If the critical Fisher value for a probability
[46]. Then, preliminary formulation tests have resulted in the
at a fixed 95% confidence interval found in the Fisher-Snedecor
following condition [4]:
table adjusted for degrees of freedom (DF model, DF residual) is
2.5 < – 9 Silt – 22 lime – 9 cement + 42 water < 3.
lower than estimated F-value for the model (F-value is the ratio
– According to the considered constraints above, mixing range
between explained variance and unexplained variance; F-value =
chosen for each of the components in this study is given in
MSR/MSr), the criterion is considered as validated and one can con-
Table 3.
clude on the statistical ability of prediction of the model. In prac-
tise, the P-value is associated with testing the null hypothesis
4.2. Generation of the experimental design
that the model does not explain any of the variability in the
response [45].
In the present study, experimental region is constrained by
For the second F-Test also called the lack of fit test, residual
three conditions (see Section 4.1) to an irregular 3D-polyhedron.
error is splitted in two parts: lack of fit (LOF) due to imperfection
Then, D-optimal computer-aided design is adapted to generate
of the model and pure error (PE) estimated from replicates data
the set of experiments. The selection process of the best candidate
error (SSr = SSLOF + SSPE, SS: Sum of squares, Table 2). SSLOF data is
set of experiments is fixed by a D-optimal criterion, applied to
a sum of squares of differences between the predicted values at
maximize the information contained in the different possible data
each level and the average experimental value at that level and
sets, and the quadratic model that will be fitted. Two additional
SSPE is a sum of squares of differences between all the individual
evaluation criteria: G-efficiency criterion and leverage value are
measured values and the average of the measured values at the
then calculated to select the best D-optimal data set. A G-
same level. The test compares variances due to the model and
efficiency criterion compares the efficiency of a D-optimal design
due to replications. The two errors should be low and of the same
to a fractional factorial design. This G-efficiency (formula given
order. This condition is fulfilled if the calculated F-value (F-value =
in Annex) considers not only the number of coefficients of the
MSLOF/MSPE) is lower than the critical acceptable value at 95% con-
model but also the number of experiments and the variance func-
fidence interval. A small P-value, i.e., <.05 at a 95% confidence
tion [47]. A G-efficiency value equal to 100% corresponds to the
interval, indicates an inadequate model. In this case, model error
optimum case of a fractional factorial design returned by the D-
is higher than replicate error and a lack of fit test is detected. Some
optimal search. A design can be considered as acceptable when
equations concerning ANOVA were presented in the Annex.
the G-efficiency value is at the least 50% [48]. Leverage value is
used to verify that no point of a design has a statistical weight
4. Experimental design greater than 3 times the average statistical weight calculated from
one observation [49].
4.1. Constraints Leverage values vary from zero to one. Low leverage values
indicate that there is a good agreement between predicted and
Mixture constraints are: experimental data. On the other hand, high leverage points have
a significant influence on the estimation of regression coefficients.
– Fundamental constraint where the sum of the ingredients of the Hence, candidate sets with leverage points must be excluded.
mixture is 100% for all mix of the design: % Silt + % Lime + %
Cement + % water = 100%.
– Economical and ecological constraints on constituents: The raw Table 3
earth concrete has to be designed to provide suitable mechani- Mixing range.
cal and durability properties in order to be used as a building xi Lower limit (%) Upper limit (%)
material. Furthermore, it must be non-energy-intensive in x1: Silt 42 76
terms of gray energy. Therefore the percentage of added binders x2: Lime 0 12
(cement and lime) must be limited. Concerning lime, a thresh- x3: Cement 4 16
old of a few percent should not be exceeded. Indeed, beyond x4: Water 20 30
 S. Imanzadeh et al. / Construction and Building Materials 163 (2018) 149–159 153

In this research study, the mixture design was developed using Table 5
a D-optimal design with a good G-efficiency value (86.19%) and Values of R2, R2adj and Q2 for linear model.

insignificant leverage points (less than 0.085) indicating the high Linear model Curing time
probability of predictability of the selected set of experiments. 7-day 28-day 90-day
The 21 mixture experiments (Table 4) were carried in the ran-
R2 0.860 0.932 0.981
dom run order proposed by MODDEÓ software [50] with three rep- R2adj 0.853 0.928 0.980
etitions including three center points. Q2 0.847 0.924 0.979

4.3. Resolution of design

Table 6
A multivariate regression method namely PLS analysis (Partial Values of R2, R2adj and Q2 for quadratic model.
Least Square projections to latent structures) was used to analyze Quadratic model Curing time
the design. PLS technique improves prediction accuracy for com-
7-day 28-day 90-day
plex experimental design data through different constraints on
R2 0.975 0.987 0.979
model. PLS was initially proposed in economics and chemometrics
R2adj 0.971 0.985 0.976
[51]. It has been successfully extended to other scientific areas, Q2 0.972 0.983 0.978
including bioinformatics, economics, computer vision, medicine
and civil engineering [52]. The objective of PLS is to predict a set
of response variables from a set of predictor variables by means
of latent variables. It is generally achieved via an iterative process. (i) Consumption of part of the water of mixing by the lime:
It has been extensively described in the literature [53]. In this
study, the Partial Least Square projections to latent structures were CaO þ H2 O ! CaðOHÞ2 þ quantity of heatð1155 kJ=kg of limeÞ
performed to estimate coefficients of the model for complex exper-
imental design data. (ii) Cation exchange reactions within the micro-structure
responsible for agglomeration of particles.
4.4. Model validation
In the soil, clays as well as organic materials have the property
Linear (first-order) and quadratic (second-order) polynomial of fixing the calcium ion. For clays, the cation exchange capacity
models defined in Section 3.1 were first fitted to the experimental corresponds to exchanges:
data following an analysis of R2, R2adj, Q2 in order to determine the
most adequate models that represent the experimental data (i) Interfoliar cations: these exchanges concern more particu-
(Tables 5 and 6). There is a notable increase in all three coefficients larly the TOT clays (smectite, vermiculite etc.). The substitu-
for the first two curing times (7-day and 28-day) when the quad- tion of the interfoliar cations is all easier than the cations
ratic model is chosen. The percentage of improvement is nearly have a lower charge and are of smaller size. Thus, cations
12% after 7-day and 5.9% after 28-day. On the other hand, for 90- can be classified in increasing order of their ability to be
day, the differences between the three indicators calculated with exchanged: Li+ < Na+ < H+ < K+ < Mg2+ < Ca2+.
quadratic and linear models are negligible. This suggests that in (ii) Border cations: on the edges of a crystal, the electric charges
the short term the material is more influenced by binary interac- are no longer balanced. SiOH and AlOH groups appear. These
tions of components than in the long term. It is consistent with groups are destabilized in a basic medium and then consti-
what one would expect. Indeed, it is possible to associate the tute an additional source of binding sites for the calcium ion.
model evolution from quadratic to linear to processes taking place
at different time scales in the material. In the short term, treatment On the other hand, in the long term, only pozzolanic reactions
with binders results in two types of reactions: remain. These slow reactions can continue for several months.

Table 4
Experimental design of the D-optimal design for a raw earth concrete.

Combination reference Silt (x1) Lime (x2) Cement (x3) Water (x4)
1 0.7283 0.0000 0.0400 0.2317
2 0.5784 0.1200 0.0400 0.2616
3 0.5784 0.1200 0.0400 0.2616
4 0.6088 0.0000 0.1600 0.2312
5 0.6088 0.0000 0.1600 0.2312
6 0.7381 0.0000 0.0400 0.2219
7 0.7381 0.0000 0.0400 0.2219
8 0.5882 0.1200 0.0400 0.2518
9 0.6185 0.0000 0.1600 0.2215
10 0.6185 0.0000 0.1600 0.2215
11 0.6784 0.0400 0.0400 0.2416
12 0.6284 0.0800 0.0400 0.2516
13 0.6382 0.0800 0.0400 0.2418
14 0.5886 0.0800 0.0800 0.2514
15 0.5987 0.0400 0.1200 0.2413
16 0.5983 0.0800 0.0800 0.2417
17 0.6084 0.0400 0.1200 0.2316
18 0.6483 0.0400 0.0800 0.2317
19 0.6434 0.0400 0.0800 0.2366
20 0.6434 0.0400 0.0800 0.2366
21 0.6434 0.0400 0.0800 0.2366
154 S. Imanzadeh et al. / Construction and Building Materials 163 (2018) 149–159

Therefore, adopting short-term quadratic models and a long-term Table 8

linear model is physically validated. First F-Test (model significance test: (F-value)critical < (F-value)model).

From the first F-Test (model significance test), the model F- Models F-value (critical) F-value (model)
values are significantly higher than the critical Fisher values for a Quadratic model (curing time: 7-day) 1.97 339.50
probability at a fixed 95% confidence interval (Tables 7 and 8). Quadratic model (curing time: 28-day) 1.97 555.10
The P-value is smaller than the significance level and it was found Linear model (curing time: 90-day) 2.49 1018.22
that chosen models passed the first F-Test. The second F-Test, com-
pares the residual error to the pure error from replication and cal-
culates F-values for all the models. According to the criterion, the Table 9
F-value should be lower if a particular model is to be significant. Model coefficients for the unconfined compressive strength (UCS) of raw earth
concrete for 7, 28 and 90-day of curing time.
From the second F-Test the model F-values are higher than the crit-
ical Fisher values for the given degrees of freedom at the 95% con- Coefficients Quadratic model Quadratic model Linear model
fidence level. In this particular case where zero P-value for the lack (curing time: (curing time: (curing time:
7-day) 28-day) 90-day)
of fit test means that both model error and replicate errors are low.
The lack of fit is due to a very small replication error. Thus, a0 7.733 1.990 45.979
a1 14.256 6.755 40.056
although the model validity error on the highlighted model is
a2 79.177 165.150 26.451
low, it will always be greater than the pure error whatever the a3 152.923 318.324 72.799
attempt to improve the model is performed. The model validity a4 40.157 3.138 82.965
error can be low even though the model is good and complete. a12 32.920 69.285 –
Thus, if one relies on the analysis of the whole computed statistical a13 63.541 199.153 –
a14 69.815 27.258 –
data set and preceding remark on the low replication error, one can
a23 119.302 236.282 –
consider that models found are valid. Once the best-fitting model a24 264.655 535.266 –
was determined for each curing time, an equation describing the a34 420.748 642.008 –
prediction of UCS was provided for raw earth concrete formula-
tions. The regression coefficients of the model for each curing time
were presented in Table 9.
The steep slope for cement indicates the major influence of this
component. Augmentation of the proportion of cement in the mix-
5. Results ing increases the unconfined compressive strength. This positive
effect is reinforced with longer curing time. It can be mainly attrib-
Response trace plots were processed and show the influence of uted to the necessary time for pozzolanic to efficiently take place. A
changing each mixture component while holding all the compo- similar result was reported by Alqadi et al. [18] and Eid [54]. Water
nents at a constant ratio (fundamental constraint, see Section 4.1). exhibits an important negative influence that increases with curing
The response (unconfined compressive strength) was plotted time. This result is in agreement with Poon et al. [55]. They showed
around a reference mixture. In this study, the considered reference that the higher compressive strength of concrete was obtained
mixture is the centroid of the constrained experimental domain with the lower amount of water [55]. Furthermore, the straight line
with the amounts of silt, lime, cement and water respectively equal for water for every curing time indicates a minor role of interac-
to 64%, 4%, 8% and 24%. tions with regard to the important main effect of water compo-
Response trace plots of UCS with deviation from reference mix- nent. Comparisons of slopes of both cement and water shows
ture in proportion for each curing time were presented in Fig. 4. that as curing time increases, the positive cement influence
They show how each component affects UCS relative to the refer- increases faster than the negative influence of water. Indeed, in
ence mixture. For the reference mixture, the UCS values are 0.50, the short term, water is necessary to trigger the chemical reactions
1.52 and 5.42 MPa respectively for 7, 28 and 90-day of curing time within the binders. Then, once the amount of water required for
(Fig. 4). The values of UCS are very sensitive to changes in cement the chemical reactions is consumed, the excess water – necessary
and water proportions for the three curing times. for workability requirement – evaporates slowly. Water influence,

Table 7
ANOVA table for 7, 28 and 90-day of curing time for raw earth concrete.

Source of variation Degrees of freedom (DF) Sum of squares (SS) Mean square (MS) F-value P-value
Quadratic model (curing time: 7-day)
Regression 10 67.95 6.79 339.50 <.001
Residual error 73 1.78 0.02
Lack-of-fit 4 1.33 0.33 55 <.001
Pure error 69 0.44 0.006
Total 83 69.73 0.84
Quadratic model (curing time: 28-day)
Regression 10 555.12 55.51 555.10 <.001
Residual error 73 7.42 0.10
Lack-of-fit 4 2.60 0.65 9.29 <.001
Pure error 69 4.82 0.07
Total 83 562.54 6.78
Linear model (curing time: 90-day)
Regression 4 2158.63 539.66 1018.22 <.001
Residual error 79 41.89 0.53
Lack-of-fit 10 33.20 3.32 25.54 <.001
Pure error 69 8.69 0.13
Total 83 2200.51 26.51
 S. Imanzadeh et al. / Construction and Building Materials 163 (2018) 149–159 155

2.0 ponent has little effect on the UCS resistance. Indeed, the maxi-
Silt Lime Curing time : 7-day
mum percentage of lime necessary for the chemical reactions is
Cement Water about of 3–4% as reported by Basma et al. [57], Delfaut [58]
1.5 Hibouche [4] and Al-Mukhtar et al. [9]. Beyond this percentage,
Predicted UCS (MPa)

lime no longer reacts, and the excess lime behaves like a very fine
granular additive in the mixture, which may explain the slight
1.0 decrease of the UCS measured values in the short curing time.
Lasledj [10] and Al-Mukhtar et al. [9] confirmed these results by
their work on lime consumption of different clayey soils.
0.5
6. Discussion

0.0 6.1. Response surface plot of UCS-optimization

-0.15 -0.10 -0.05 0.00 0.05 0.10 0.15 0.20
Deviation from reference mixture in proportion
a) In this section, the aim is to find the best formulation of the
material satisfying a threshold of acceptable UCS around 5 MPa
for 90-day curing time better addressing economic and ecologic
5
Silt Lime Curing time: 28-day constraints but also taking account of the negative roles of water
Cement Water and silt highlighted earlier in Section 5. For this phase of optimiza-
4 tion, response surface plots and projections of response surface
Predicted UCS (MPa)

plots are used.

First, it was chosen to reinforce economic constraint by mini-
3
mizing the cement proportion till condition on UCS was satisfied.
The ranges of UCS which can be reached are shown in Fig. 5 as a
2 function of cement fraction in mixing. It was found that cement
fraction in raw earth concrete must be fixed to a minimum of
nearly 8%. Otherwise, if UCS constraint is analyzed through water
1
parameter, a range of 22.5–24.5% is acceptable (Fig. 6). If amount
of water is smaller than 22.5%, workability condition is not ful-
0 filled. If amount of water is more than 25.5%, UCS value greatly
-0.15 -0.10 -0.05 0.00 0.05 0.10 0.15 0.20 reduces and the threshold condition cannot be verified anymore
Deviation from reference mixture in proportion for any mixing.
b)
Response surface plot for 8% of cement is presented in Fig. 7.
The response domain is limited because of the applied constraints
10
Silt Lime Curing time: 90-day (see Section 4.1). Iso-values of unconfined compressive strength
Cement Water are between 5.10 and 5.60 MPa, the plotted black lines intersect
8 at the reference mixture (64% silt, 4% lime, 8% cement and 24%
Predicted UCS (MPa)

water). UCS value is equal to 5.42 MPa (Figs. 4c and 7). Reference
mixture meets UCS threshold constraint. As mentioned earlier, a
6
UCS value above 5 MPa is achieved for all the mixtures if the per-
centage of cement is set to 8% with a variation range of UCS of
4 about 10%. Once the 8% proportion of cement is selected, the other
mix components can be chosen by economic and ecological
criteria.
2
For example, in formulation Mix 1 (dotted black lines in Fig. 7,
Table 10), lime fraction was decreased from 4% to 2% to minimize
0 binder proportions. Then, the percentages of binders for reference
-0.15 -0.10 -0.05 0.00 0.05 0.10 0.15 0.20
mixture and for Mix 1 are respectively equal to 12% and 10%
Deviation from reference mixture in proportion
c) (Table 10). The proportion of water was also slightly decreased to
reduce shrinkage risks [55].
Fig. 4. Response trace plots of UCS for three different curing times: a) 7-day, b) 28-
day and c) 90-day.
6.2. Experimental validation

especially in the long term, is weaker in comparison with that of
cement. Concerning silt influence, Fig. 4c shows that an increase
of silt proportion in the mixing tends to decrease significantly
UCS at 90-day. Indeed, in the long term, concrete hardens but at
the same time it dries. The drying process results in shrinkage
induced micro-cracking of raw-earth concrete which reduces the
mechanical strength. Kanema et al. [8] studied cracks and shrink-
age evolution of a natural silt intended for a new earth building
material. They found a similar behavior. Shrinkage increases as
the proportion of silt increases [56]. The response trace for lime
is nearly a horizontal line (Fig. 4c) except for short curing time
(Fig. 4a, b). This indicates that beyond 4% of added lime, this com-
In order to complete validation of the models, additional exper-
imental tests were conducted on a formulation implemented on a
construction site and compared to the models prediction. The
tested raw earth concrete was composed of 67.6% of silt, 2.4% of
lime, 7.5% of cement and 22.5% of water. Specimens of 16 cm in
diameter and 32 cm in height were core-drilled at 90-day curing
time, and stored under different humidity conditions in order to
study the effect of the water content (w) (varying from 4% to
22%) on the unconfined compressive strength. Samples were
sheared at the age of 90-day. Fig. 8 shows the variation of axial
compressive stress as a function of axial deformation (Fig. 8a)
and the variation of unconfined compressive strength, correspond-
156 S. Imanzadeh et al. / Construction and Building Materials 163 (2018) 149–159

7
UCS (MPa)

3
6.5 7 7.5 8 8.5 9 9.5 10
Fraction of cement (%)

Fig. 5. Ranges of UCS as a function of cement fraction in mixing.

18

16

14

12
UCS (MPa)

10

0
22 22.5 23 23.5 24 24.5 25 25.5
Fraction of water (%)

Fig. 6. Ranges of UCS as a function of water fraction in mixing.

Fig. 7. Response surface contour plot of UCS for 90-day curing time (cement = 8%).
 S. Imanzadeh et al. / Construction and Building Materials 163 (2018) 149–159
Table 10
Two formulations of raw earth concrete for 8% of cement.
Water (%)
Reference mixture 24
Mix1 23
Fig. 8. For the curing time of 90-day a) Variation of the unconfined compressive stress as a function of axial strain, b) Variation of the unconfined compressive strength as a
function of the water content (predicted values in red dotted lines). (For interpretation of the references to colour in this figure legend, the reader is referred to the web
version of this article.)
ing to maximum axial stress, as a function of water content
(Fig. 8b).
Fig. 8a shows that for weak axial strains (0.5% for 90-day of cur-
ing time), there is no influence of the water content on the axial
stress, and hence the stiffness of the material. This can be
explained by the closing phase of micro-pores and micro-cracks
under the effect of mechanical stress. Beyond these values of axial
strains, it is clear that the water content of the material plays a
major role on the evolution of the stiffness. The lower the water
content, the higher the stiffness and maximum axial stress.
Fig. 8b shows that the values of UCS decrease linearly with the
increasing of water content for considered curing time. It could be
due to the softening of binders by water and development of pore
water pressures as proposed by Morel et al. [59] in their work on
the compressive strength testing of compressed earth blocks.
Experimental linear regression laws can be derived for the 90-
day curing time as follows:
UCS ¼ 0:1318w þ 6:1134 with R2 ¼ 0:99
where water content (w) is in percentage.
Note that, the water content of the specimens used to elaborate
models of DoE is estimated to 10%. Thus, experimental value of UCS
to choose for the comparison from Fig. 8b must be for the same
water content of 10% (dashed lines).
Predicted values are indicated by red dotted lines on Fig. 8b
(UCS = 5 MPa corresponding to w = 8.5%). Experimental law (Eq.
(3)) gives UCS = 4.8 MPa corresponding to w = 10%. This shows that
there is a good agreement between experimental data and pre-
dicted ones from mix design.
7. Conclusions
In the present work, it was shown that Design of Experiments
method is an optimization tool adapted to analyze unconfined
compressive strength of a raw earth concrete. The tests of the
157
Silt (%) Lime (%) Binders (%)
(Lime + Cement)
64 4 12
67 2 10
design were defined by economical, ecological and fluidity criteria.
The curing time was varied from 7 to 90-day to study the mechan-
ical response of the geo-material during the drying process.
Models derived from a mixing design were discussed and vali-
dated by statistical tools. It has been found that a quadratic model
is necessary for a short curing time whereas for 90-day curing time
a linear model is sufficient. The amount of water and binders
(cement and lime) were optimized to design an economic and eco-
logical workable raw earth concrete with a desired unconfined
compressive strength of around 5 MPa for 90-day curing time. In
the long time, the negative role of the increase of the proportion
of water and silt in the mixture was highlighted thanks to the trace
plots around a reference mixture. It was attributed to shrinkage in
the material responsible for the development of micro-cracks. The
discussion, built on response surface plots has allowed to propose
an optimized formulation of the material, taking account technical,
economic and ecological constraints.
Additional experimental tests focusing on effect of water con-
ð3Þ tent on unconfined compressive stress were presented on a raw
earth concrete implemented on a construction site. A good agree-
ment between experimental results and model prediction was
found.
Future work is necessary to take into account directly shrinkage
influence to optimize raw earth concrete formulation. Water con-
tent or suction in the material should be useful additional param-
eters in the design to investigate the hydro-mechanical behavior of
stabilized raw earth and perform a more relevant optimization in
terms of economical, mechanical and durability.
Acknowledgements
The authors would like to thank the company Cematerre, the
CNRS-France (National Center for Scientific Research) and the
FEDER-Normansdie-France for their support.
158 S. Imanzadeh et al. / Construction and Building Materials 163 (2018) 149–159

Annex [3] P. Prikryl, Á. Török, M. Theodoridou, M. Gómez-Heras, K. Miskovsky,

Geomaterials in construction and their sustainability, Sustain. Use Tradit,
Geomater. Constr. Pract. (2016) 1–22, https://doi.org/10.1144/SP416.21.
[4] A. Hibouche, Sols traités aux liants, performances hydro-mécaniques et hygro-
Expressions and their definitions. thermiques, applications en BTP (dissertation), Normandie Univ., Unihavre,
2013.
[5] P. Zak, T. Ashour, A. Korjenic, S. Korjenic, W. Wu, The influence of natural
reinforcement fibers, gypsum and cement on compressive strength of earth
Expression Definition bricks materials, Constr. Build. Mater. 106 (2016) 179–188, https://doi.org/
10.1016/j.conbuildmat.2015.12.031.
Cc ¼
D230 Cc: Gradation coefficient
D60 D10 [6] T. Ashour, A. Korjenic, S. Korjenic, W. Wu, Thermal conductivity of unfired
D10: Effective sizePlease check earth bricks reinforced by agricultural wastes with cement and gypsum,
the typeset of Annex section, Energy Build. 104 (2015) 139–146, https://doi.org/10.1016/j.
and correct if necessary. enbuild.2015.07.016.
[7] M. Carmen Jimenez Delgado, I. Canas Guerrero, Earth building in Spain, Constr.
D30: Diameter corresponding Build. Mater. 20 (2006) 679–690, https://doi.org/10.1016/
to 30% finer in the particle-size j.conbuildmat.2005.02.006.
distribution [8] J.M. Kanema, J. Eid, S. Taibi, Shrinkage of earth concrete amended with recycled
aggregates and superplasticizer: impact on mechanical properties and cracks,
D60: Diameter corresponding Mater. Des. 109 (2016) 378–389, https://doi.org/10.1016/
to 60% finer in the particle-size j.matdes.2016.07.025.
distribution [9] M. Al-Mukhtar, A. Lasledj, J.F. Alcover, Lime consumption of different clayey
Pm Pni 2 SSR: Sum of squares of the
soils, Appl. Clay Sci. 95 (2014) 133–145, https://doi.org/10.1016/
SSR ¼ i j
^y
ðy Þ j.clay.2014.03.024.
regression [10] A. Lasledj, Traitement des sols argileux à la chaux: processus physico-
ni: Number of replicates at the chimique et propriétés géotechniques (Ph.D. thesis), University of Orleans,
France, 2009, p. 361.
ith level
[11] M.S. Pakbaz, M. Farzi, Comparison of the effect of mixing methods (dry vs. wet)
m: Number of planning levels on mechanical and hydraulic properties of treated soil with cement or lime,
^: Predicted value
y Appl. Clay Sci. 105–106 (2015) 156–169.
: Mean of measured value [12] G. Minke, Building with Earth, Design and Technology of Sustainable
y
Pm Pni SSr: Sum of squares of the
Architecture (Text Book), Birkhaeser, Publishers for Architecture, Basel,
SSr ¼ i j
^ i Þ2
ðyij  y Berlin, Boston, 2006.
residualy: Measured value [13] R.H. Myers, D.C. Montgomery, C.M. Anderson-Cook, Response Surface
Pm Pni 2 SSLOF: Sum of squares of lack of Methodology: Process and Product Optimization Using Designed
SSLOF ¼ i j
^i  y
ðy i Þ Experiments, fourth ed., Wiley, New York, 2016, p. 856.
fit [14] L. Eriksson, E. Johansson, N. Kettaneh-Wold, C. Wikström and S. Wold, Design
Pm Pni
SSPE ¼ i j
i Þ2
ðyij  y SSPE: Sum of squares of pure of Experiments: Principles and Applications, Learnways AB, Ume°a, 2000, pp.
error 425, ISBN 91-973730-4-4.
Pm Pni [15] L. Senff, P.A. Barbetta, W.L. Repette, D. Hotza, H. Paiva, V.M. Ferreira, J.A.
SST ¼ i j
 Þ2
ðyij  y SST: Total sum of squares of Labrincha, Mortar composition defined according to rheometer and flow table
measured value tests using factorial designed experiments, Constr. Build. Mater. 23 (2009)
Pm Pni _ 2 PRESSPLS: Predictive residual 3107–3111, https://doi.org/10.1016/j.conbuildmat.2009.06.028.
PRESSPLS ¼ i j ðyij  y icv Þ [16] B.B. Mukharjee, S.V. Barai, Assessment of the influence of Nano-Silica on the
sum of squares behavior of mortar using factorial design of experiments, Constr. Build. Mater.
_
y icv :
Predicted left-out values 68 (2014) 416–425, https://doi.org/10.1016/j.conbuildmat.2014.06.074.
[17] R.K. Vieira, R.C. Soares, S.C. Pinheiro, O.A. Paiva, J.O. Eleutério, R.P. Vasconcelos,
from the cross-validation Completely random experimental design with mixture and process variables
procedure. for optimization of rubberized concrete, Constr. Build. Mater. 24 (2010) 1754–
R2 ¼ 1  SS
SSr R2 : Coefficient of 1760, https://doi.org/10.1016/j.conbuildmat.2010.02.013.
T [18] A.N. Alqadi, K.N. Bin Mustapha, S. Naganathan, Q.N. Al-Kadi, Development of
determination self-compacting concrete using contrast constant factorial design, J. King Saud
SSr =DF r
R2adj ¼ 1  SS T =DF T
R2adj : Fraction of variation of Univ. Eng. Sci. 25 (2) (2013) 105–112, https://doi.org/10.1016/j.
jksues.2012.06.002.
the response explained by the [19] X. Li, H. Haung, J. Xu, S. Ma, X. Shen, Statistical research on phase formation
modelDFr: Degrees of freedom and modification of alite polymorphs in cement clinker with SO3 and MgO,
of the estimate of the Constr. Build. Mater. 37 (2012) 548–555, https://doi.org/10.1016/
j.conbuildmat.2012.07.099.
population error variance [20] R. Phan-Tan-Luu, M. Sergent, Nonclassical experimental designs, in: S.D.
DFT: Degrees of freedom of the Brown, R. Tauler, B. Walczak (Eds.), Comprehensive Chemometrics: Chemical
population variance of the Biochemical Data Analysis, 2009, pp. 453–499.
[21] A. Herrero, M.C. Ortiz, L.A. Sarabia, D-optimal experimental design coupled
dependent variable
with parallel factor analysis 2 decomposition a tool in determination of
Q 2 ¼ 1  PRESS PLS Q2: Estimates the predictive triazines in oranges by programmed temperature vaporization-gas
SST
ability of the model chromatography-mass spectrometry when using dispersive-solid phase
extraction, J. Chromatogr. A. 1288 (2013) 111–126, https://doi.org/10.1016/j.
G  eff ¼ nd
P G-eff: G-efficiency, p: number
chroma.2013.02.088.
of terms in the model, n: [22] J.L. Glajch, J.J. Kirkland, K.M. Squire, J.M. Minor, Optimization of solvent
number of runs in the design, strength and selectivity for reversed-phase liquid chromatography using an
d: maximum relative interactive mixture-design statistical technique, J. Chromatogr. A. 199 (1980)
57–79, https://doi.org/10.1016/S0021-9673(01)91361-5.
prediction variance over the [23] J.A. Cornell, Mixture experiments, in: S. Kotz, N. Johnson (Eds.), Encyclopedia
candidate set of Statistical Sciences, Wiley, New York, 1985, pp. 569–579.
[24] M. Muthukumar, D. Mohan, Optimization of mechanical properties of polymer
concrete and mix design recommendation based on design of experiments, J.
Appl. Polym. Sci. 94 (2004) 1107–1116, https://doi.org/10.1002/app.21008.
[25] S. Bouzalakos, A.W. Dudeney, B.K. Chan, Formulating and optimising the
compressive strength of controlled low-strength materials containing mine
References tailings by mixture design and response surface methods, Miner. Eng. 53
(2013) 48–56, https://doi.org/10.1016/j.mineng.2013.07.007.
[1] L. Rainer, A.B. Rivera, D. Gandreau, Terra 2008: The 10th International [26] J.V. Nardi, D. Hotza, Mixture design and response surface analysis of
Conference on the Study and Conservation of Earthen Architectural Heritage, pozzolanic products, Mater. Sci. Forum (2003).
Getty Publications, 2011. [27] I.C. Yeh, I.C. Chen, T.Z. Ko, C.C. Peng, C.C. Gan, J.W. Chen, Optimum mixture
[2] L. Fontaine, R. Anger, Bâtir en terre – Du grain de sable à l’architecture, Editions design of high-performance concrete using artifcial neural networks, J.
Belin, 2009. Technol. 17 (4) (2002) 583–591.
 S. Imanzadeh et al. / Construction and Building Materials 163 (2018) 149–159
[28] S.H. Chen, C.S. Chang, H.Y. Wang, W.L. Huang, Mixture design of high
performance recycled liquid crystal glasses concrete (HPGC), Constr. Build.
Mater. 25 (2011) 3886–3892, https://doi.org/10.1016/j.conbuildmat.
2011.04.012.
[29] L. Senff, A. Castela, W. Hajjaji, D. Hotza, J.A. Labrincha, Formulations of
sulfobelite cement through design of experiments, Constr. Build. Mater. 25
(2011) 3410–3416, https://doi.org/10.1016/j.conbuildmat.2011.03.032.
[30] R.H. Kupaei, U.J. Alengaram, M.Z. Bin Jumaat, H. Nikraz Mix, Design for fly ash
based oil palm shell geopolymer lightweight concrete, Constr. Build. Mater. 43
(2013) 490–496, https://doi.org/10.1016/j.conbuildmat.2013.02.071.
[31] Standard AFNOR: NF P94-056, Sols: reconnaissance et essais-Analyse
granulométrique-Méthode par tamisage à sec après lavage, 1996.
[32] Standard AFNOR: XP P94-041, Sols: reconnaissance et essais-Identification
granulométrique-Méthode de tamisage par voie humide, 1995.
[33] Standard AFNOR: NF P94-057, Sols: reconnaissance et essais-Analyse
granulométrique des sols-Méthode par sédimentation, 1992.
[34] Standard AFNOR: NF P94-051,, Sols: reconnaissance et essais-Détermination
des limites d’Atterberg-Limite de liquidité à la coupelle-Limite de plasticité au
rouleau, 1993.
[35] J. Biare, J.L. Favre, Parameters filing and statistical analysis of data in soil
mechanics, in: Proceedings of the 2nd Int. Conf. on Application of Statistics and
Probabilities in Soil Mechanics, Aachen, Germany, 1975.
[36] ASTM D2487-11, Standard Classification of Soils for Engineering Purposes
(Unified Soil Classification System), 2012.
[37] Standard AFNOR: NF EN 459-1, Chaux de construction - Partie 1: définitions,
spécifications et critères de conformité, 2012.
[38] Standard AFNOR: NF EN197-1, Cement Part 1: Composition, Specifications,
and Conformity Criteria for Common Cements, 2001.
[39] Standard AFNOR: NF P 15-318, liants hydrauliques ciment à faible chaleur
d’hydratation initiale et à teneur en sulfures limitée, 1991.
[40] Standard AFNOR: NF EN196-10, Méthodes d’essais des ciments-Partie 10:
Détermination de la teneur en chrome (VI) soluble dans l’eau des ciments,
2015.
[41] Standard AFNOR: NF P94-420, Détermination de la résistance à la
compression uniaxiale, 2000.
[42] Standard AFNOR: NF P94-425, Méthodes d’essai pour roches - Détermination
du module d’Young et du coefficient de Poisson, 2002.
[43] G.E. Box, J. Stuart Hunter, W.G. Hunter, in: Statistics for Experimenters: Design,
Innovation, and Discovery, John Wiley and Sons, 2005, p. 629.
159
[44] R.A. Fisher, The Design of Experiments, Hafner Libraries Australia, New York,
1971, p. 248.
[45] J. Goupy, Plans d’expériences: les mélanges, DUNOD, Paris, 2001, p. 285.
[46] Standard AFNOR: NF EN 206-1, CN, Béton–Partie 1: Spécification,
performances, production et conformité (in French), 2012.
[47] P.F. De Aguiar, B. Bourguignon, M.S. Khots, D.L. Massart, R. Phan-Than-Luu, D-
optimal designs, Chemom. Intell. Lab. Syst. 30 (1995) 199–210, https://doi.org/
10.1016/0169-7439(94)00076-X.
[48] P. Krajnik, J. Kopac, A. Sluga, Design of grinding factors based on response
surface methodology, J. Mater. Process. Technol. 162–163 (2005) 629–636,
https://doi.org/10.1016/j.jmatprotec.2005.02.187.
[49] J. Johnston, Econometric Methods, McGraw-Hill, New-York, 1972, p. 447.
[50] Software for design of experiments and optimization, Umetrics: http://www.
umetrics.com, 2016.
[51] H. Wold, Path Models with Latent Variables: The NIPALS Approach, Academic
Press, 1975.
[52] A. Sharma, M. Al Haj, J. Choi, L.S. Davis, D.W. Jacobs, Robust pose invariant face
recognition using coupled latent space discriminant analysis, Comput. Vis.
Image Underst. 116 (2012) 1095–1110, https://doi.org/10.1016/j.
cviu.2012.08.001.
[53] B. Li, A.J. Morris, E.B. Martin, Generalized partial least squares regression based
on the penalized minimum norm projection, Chemom. Intell. Lab. Syst. 72
(2004) 21–26, https://doi.org/10.1016/j.chemolab.2004.01.026.
[54] J. Eid, Elaboration d’un éco-géo-matériau à base de terre crue (Ph.D. thesis),
University of Le Havre Normandy, France, 2016.
[55] C.S. Poon, Z.H. Shui, L. Lam, Effect of microstructure of ITZ on compressive
strength of concrete prepared with recycled aggregates, Constr. Build. Mater.
18 (2004) 461–468, https://doi.org/10.1016/j.conbuildmat.2004.03.005.
[56] J.M. Fleureau, S. Kheirbek-Saoud, R. Soemitr, S. Taibi, Behavior of clayey soils
on drying-wetting paths, Can. Geotech. J. 30 (2) (1993) 287–296, https://doi.
org/10.1139/t93-024. NRC.
[57] A.A. Basma, E.R. Tuncer, Effect of lime on volume change and compressibility
of expansive clays, Transp. Res. Rec. 1295 (1990) 52–61.
[58] A. Delfaut, Traitement à la chaux vive et au ciment des argiles plastiques du
Sparnacien et de la région parisienne: étude de laboratoire, LCPC 169 (1990)
16–20.
[59] J.C. Morel, A. Pkla, P. Walker, Compressive strength testing of compressed
earth blocks, Constr. Build. Mater. 21 (2007) 303–309, https://doi.org/10.1016/
j.conbuildmat.2005.08.021.