Modeling Rheological Properties of Oil Well Cement

Slurries Using Artificial Neural Networks

Anjuman Shahriar1 and Moncef L. Nehdi2

Abstract: An artificial neural network (ANN) model was developed to predict the rheological properties of oil well cement slurries. The
slurries were prepared by using Class G oil well cement with a water-cement ratio (w=c) of 0.44 and incorporating three different chemical
Downloaded from ascelibrary.org by New York University on 05/15/15. Copyright ASCE. For personal use only; all rights reserved.

admixtures, including a new-generation polycarboxylate-based high-range water-reducing admixture (PCH), polycarboxylate-based mid-
range water-reducing admixture (PCM), and lingosulphonate-based midrange water-reducing admixture (LSM). The rheological properties
were investigated at different temperatures in the range of 23 to 60°C by using an advanced shear-stress/shear-strain controlled rheometer. A
back-propagation neural network was designed and trained by using the experimental flow curves. The shear rate, dosage of admixture, and
test temperature were considered as input parameters, and the measured shear stress was the output parameter. The trained ANN was not only
capable of accurately predicting the shear flow used for its training, but could also effectively predict the rheological properties of new slurries
designed within the range of input parameters of the experimental database with an absolute error of 3.43, 3.17, and 2.82% for slurries
incorporating PCH, PCM, and LSM, respectively. The flow curves developed by using the ANN model allowed the prediction of the
Bingham parameters (yield stress and plastic viscosity) of the slurries with adequate accuracy. DOI: 10.1061/(ASCE)MT.1943-5533
.0000340. © 2011 American Society of Civil Engineers.
CE Database subject headings: Slurries; Neural networks; Cement; Viscosity.
Author keywords: Oil well cement; Slurry; Yield stress; Plastic viscosity; Artificial neural network; Modeling.

Introduction
Oil well cement (OWC) slurries are subjected to a wide range of
temperature and pressure during well cementing, which makes the
rheology of such slurries more complicated than that of conven-
tional cement pastes. The rheological properties of OWC slurries
are important in assuring that the slurries can be mixed at the surface
and pumped into the well with minimum pressure drop and must
be optimized to achieve effective well cementing operation. Such
properties depend on various factors, including the water-cement
ratio (w=c), size and shape of cement grains, chemical composition
of the cement and relative distribution of its components at the
surface of grains, presence and type of additives, compatibility
between cement and chemical admixtures, and mixing and testing
procedures.
The interactions among the previously mentioned factors play a
vital role in altering the rheological properties of OWC slurries.
Therefore, a clear understanding of this complex behavior is impor-
tant to successfully predict the rheological properties of OWC slur-
ries. A number of shear-stress-strain rate relationships have been
developed for cement slurries, but there is a lack of understanding
of the relationships among the materials used for preparing such
1
Ph.D. candidate, Dept. of Civil and Environmental Engineering, Univ.
of Western Ontario, London, ON, N6A 5B9, Canada. E-mail: anjuman
.shahriar@gmail.com
2
Professor, Dept. of Civil and Environmental Engineering, Univ. of
Western Ontario, London, ON, N6A 5B9, Canada (corresponding author).
E-mail: mnehdi@uwo.ca
Note. This manuscript was submitted on November 1, 2010; approved
on May 26, 2011; published online on May 28, 2011. Discussion period
open until May 1, 2012; separate discussions must be submitted for indi-
vidual papers. This paper is part of the Journal of Materials in Civil En-
gineering, Vol. 23, No. 12, December 1, 2011. ©ASCE, ISSN 0899-1561/
2011/12-1703–1710/$25.00.
slurries and test conditions such as temperature and shear rate.
The power-law, Bingham, and Herschel-Bulkley models are the
most commonly used in the well cementing industry (Guillot
2006). Such models are composed of empirical expressions derived
from the analysis of limited experimental data and/or on the basis of
simplifying assumptions (El-Chabib and Nehdi 2005). Moreover,
they do not have true predictive capability outside the experimental
domain and/or when different materials are used (El-Chabib et al.
2003).
In this study, artificial neural networks (ANNs) were used to
predict the rheological properties of OWC slurries and to evaluate
the sensitivity of these properties to the variation of shear rate, ad-
mixture dosage, and test temperature. The ANN is a powerful tool
to overcome the difficulty of assessing the complex relationships
among parameters involved in the model. Unlike conventional
models, ANN does not need to assume a model structure between
input and output variables. Rather, it is generated by the data points
provided for the training of the network. An ANN solves problems
by creating parallel networks and the training/learning of those
networks rather than by a specific programming scheme on the
basis of well-defined rules or assumptions (Bruni et al. 2006).
Temperature has a drastic effect on the rheological behavior of
cement slurries. This effect also depends on the type of cement
and admixtures used. Thus, it has been argued that it would be very
difficult to find a general model to represent the temperature depen-
dency of cement slurry rheology (Guillot 2006). Ravi and Sutton
(1990) investigated the effect of temperature and pressure on the
plastic viscosity and yield stress of Class H cement slurries by us-
ing a high-pressure, high-temperature rheometer. They developed
a correlation to calculate the equilibrium temperature for plastic
viscosity and yield stress. Both plastic viscosity and yield stress
were found to increase with the increase in temperature. However,
plastic viscosity reached a constant value beyond the equilibrium
temperature, whereas there was no evidence for yield stress to attain

JOURNAL OF MATERIALS IN CIVIL ENGINEERING © ASCE / DECEMBER 2011 / 1703

J. Mater. Civ. Eng. 2011.23:1703-1710.

 a constant value beyond a certain temperature. By using the Bing-
ham plastic model, Ravi and Sutton (1990) developed equations to
represent the variation of rheological parameters with temperature
in which the yield stress and plastic viscosity values were measured
at 80°F (27°C) and are limited to a maximum temperature, T max .
The following equations were developed by using cement systems
containing specific additives and are thus dependent on the slurry
composition:
μp ðTÞ ¼ a þ ðb × TÞ þ ð0:00325 × T 2 Þ ð1Þ
in which μp is in mPa · s and T is in °F, and
a ¼ 65:0729 þ 1:3054 × μp at 80°F
Downloaded from ascelibrary.org by New York University on 05/15/15. Copyright ASCE. For personal use only; all rights reserved.
b ¼ 1:0734
0:00381 × μp at 80°F
Currently, there exists no reliable method to predict the rheo-
logical performance of OWC slurries or to relate its composition
(admixture type, dosage, and so on) and test conditions (e.g., shear
rate, temperature) to the expected rheological properties. Therefore,
the ANN technique has been used in the present study to create an
intelligent model to predict the shear stress at a given shear rate
as a function of the temperature and admixture dosage. Thus, a
shear stress-shear rate curve for OWC slurries can be predicted for
different temperatures before fitting the data to rheological models.
Experimental Setup
Materials
The OWC slurries used in this study were prepared by using a high
sulfate-resistant API Class G OWC with a specific gravity of 3.14.
Deionized distilled water was used for the mixing, and its temper-
ature was maintained at 23  1°C by using an isothermal container.
The chemical admixtures used to prepare cement slurries with a
w=c of 0.44 along with their respective dosages are presented in
Table 1.
Apparatus
The cement slurry preparation is very important because of the
influence of the shear history of the mixture on its rheological
properties (Orban et al. 1986). The cement slurries were prepared
by using a variable-speed high-shear blender-type mixer with
bottom drive blades as per the American National Standards Insti-
tute, New York/American Petroleum Institute (ANSI/API) Recom-
mended Practice 10B-2 (2005).
A high-accuracy advanced rheometer (TA instruments AR
2000) [Fig. 1(a)] was used to measure the rheological properties
of cement slurries. The rheometer is capable of continuous shear
rate sweep and stress sweep. The geometry of the test accessory
and the gap and friction capacity of its shearing surfaces have a
significant influence on the measured rheological properties of
cement pastes (Nehdi and Rahman 2004). The coaxial concentric
cylinder geometry was considered suitable for this study because of
the typically low viscosity of OWC slurries. The geometry consists
Table 1. Chemical Admixtures Used for Preparing Oil Well Cement Slurries
Type of admixture
A new-generation polycarboxylate-based high-range water-reducing admixture
Polycarboxylate-based midrange water-reducing admixture
Midrange lignosulphonate-based water-reducing admixture
1704 / JOURNAL OF MATERIALS IN CIVIL ENGINEERING © ASCE / DECEMBER 2011
J. Mater. Civ. Eng. 2011.23:1703-1710.
of a cylinder with a conical end that rotates inside a cylinder with a
central fixed hollow as shown in Fig. 1(b). A solvent trap was used
to prevent evaporation from the tested cement slurry sample by
covering the top of the hollow cylinder.
The rheometer was calibrated by using a certified standard
Newtonian oil with a known viscosity of 1:0 Pa · s and yield
stress ¼ 0 Pa at 20°C. The measured yield stress was 0 Pa and vis-
cosity was 1:009 Pa · s with an error of 0.9%, which is less than the
tolerated error of 4% specified by the manufacturer. This computer-
controlled rheometer is equipped with rheological data analysis
software, which can fit the shear-stress-strain rate data to several
rheological models. The Bingham model was used throughout this
study to calculate the rheological properties of cement slurries, that
is, yield stress, plastic viscosity, and thixotropy.
Experimental Procedure
Mixing and Preparing Cement Slurry
The cement slurries were prepared by using a high-shear blender-
type mixer with bottom-driven blades as per the ANSI/API
Recommended Practice 10B-2 (ANSI/API 2005) at a controlled
ambient room temperature of 23  1°C. The prepared slurry was
then placed into the bowl of a mixer for preconditioning over
20 min at the test temperature (23, 45, or 60°C) at a speed of
150 rpm. The total time between the beginning of mixing and
the start of the rheological tests was kept constant to avoid the
effects of exogenous variables on the results. The rheometer setup
was also maintained constant for all tested slurries. The concentric
cylinder test geometry was maintained at the test temperature to
avoid any thermal shock of the slurry.
Rheometric Tests
After mixing and preconditioning, the cement slurry sample was
placed in the coaxial cylinder of the rheometer. The temperature
was adjusted to the required level and the sample was then sub-
jected to a stepped ramp or steady-state flow at which rheological
measurements were assumed at 20 different shear rates, from 5.11
up to 511 s
1 after a continuous rotation of 10 s at each level. Sub-
sequently, the data were measured at a descending shear rate from
511 to 5:11 s
1 to obtain the down flow curve. The hysteresis loop
thus produced was used to characterize the thixotropy of cement
paste (Saak 2000). A schematic representation of the viscometric
testing scheme is illustrated in Fig. 2.
Experimental Results
Fig. 3 represents typical shear-stress-shear rate down curves for
OWC slurries prepared by using a new-generation polycarboxylate-
based high-range water-reducing admixture (PCH) at 60°C. The
data points in the curves can be fit to the Bingham plastic model
[Eq. (2)], and the rheological parameters (yield stress and plastic
viscosity) can be determined from the shear stress-shear rate down
curve. The rheological parameters of OWC slurries thus calculated
Abbreviation Solid content (%) Dosages by weight of cement (%)
PCH 44.5 0.25, 0.50, 0.75, and 1.0
PCM 23.5 0.25, 0.50, 0.75, and 1.0
LSM 23.5 0.5, 1.0, 1.5, and 2.0
 In this study, an ANN has been developed to determine the shear
stress as a function of test variables (temperature, admixture dos-
age, and shear rate), thus allowing prediction of the experimental
flow curve. Previous work by the writers (Shahriar and Nehdi
2010) showed that the superplasticizers considered had different
effects on OWC slurry rheology; therefore, a separate model was
developed for each superplasticizer.

Artificial Neural Network Approach

The ANNs are powerful computational tools capable of solving

complex and highly nonlinear functions through self-organization,
pattern recognition, and functional approximation. The ANNs sim-
Downloaded from ascelibrary.org by New York University on 05/15/15. Copyright ASCE. For personal use only; all rights reserved.

(a) ulate the structure and internal functions of the biological brain. An
Torque (M)
ANN is capable of learning the mapping between a set of input data
and its corresponding output. Through training, it creates memory
capable of predicting output when presented with a new set of data
within the practical range of the input used in the training process.
Rheometer Motor Among various kinds of ANNs, the feed-forward back-propagation
learning algorithm is the most commonly used in engineering
Angular velocity (rad/sec) applications, for instance, in modeling the behavior of cement-
based materials. A neural network consists of a number of layers
OWC slurry (an input layer, one or more hidden layers, and an output layer)
of several interconnected linear or nonlinear processing units
Inner cylinder (neurons). Each processing unit receives multiple inputs from the
neurons in the previous layer through the weighted connection and,
Outer cylinder after performing appropriate computation, transfers its output to
other processing units or as a network output by using an assigned
(b) transfer (activation) function. Details on the basics of ANNs can be
found in the literature (Rumelhart et al. 1986; Demuth et al. 2008,
Fig. 1. (a) Advanced rheometer with coaxial cylinder geometry (image
El-Chabib and Nehdi 2005).
by the writers); (b) coaxial concentric cylinder with cylindrical conical
Overfitting of the model leads to a precise prediction of the
end geometry
training patterns yet can also cause poor generalization of new
patterns. On the other hand, premature training may cause unsat-
isfactory performance because of not adequately learning the
are highly dependent on the temperature and admixture dosage, embedded relationships between inputs and outputs. The training
as described in previous work by the writers (Shahriar and Nehdi continues until it converges to a desired minimum error between its
2010).
predicted outputs and the desired targets provided in the training
τ ¼ τ 0 þ μP
γ ð2Þ process. The duration of training can be determined by (1) limiting
the number of iterations, called training epochs, (2) setting a desired
in which τ , τ 0 , μP , and
γ represent the shear stress, yield stress, minimum error, or (3) monitoring the trend of error improvement so
plastic viscosity, and shear rate, respectively. that training will be stopped when no or little improvement in the

600

500 Hysteresis loop

Shear Rate, (s-1)

Shear rate (s )
-1

400

300 Mixing and sample

preparation (1.83 min)
Denotes points of
200
measurement
100
Preconditioning (20 min)
Time 0
0 4 8 12 16 20 24 28 32
Shear Rate, (s-1)

Delay 10 sec Time (minutes)

(b)

Steady state flow

γ = constant

Time
(a)

Fig. 2. (a) Schematic representation of stepped ramp; (b) rheometer test sequence (shear rate history used in rheological tests)

JOURNAL OF MATERIALS IN CIVIL ENGINEERING © ASCE / DECEMBER 2011 / 1705

J. Mater. Civ. Eng. 2011.23:1703-1710.

 60
to the model, but within the range of training data, were used to test
50
0.25% 0.50% 0.75% 1.00% the performance of the network. Each flow curve consists of 20
data points at equal shear rate intervals from 5.11 to 511 s
1 . Each
Shear stress (Pa)

40 data pattern consists of an input vector of three elements, including

the admixture dosage, temperature, and shear rate, and a corre-
30 sponding output vector consisting of shear stress. Table 2 presents
the ranges, mean values, and SDs of all input and output variables
20 in the final database.
10 Network Architecture
0 In this study, a feed-forward back-propagation neural network was
0 100 200 300 400 500 600 developed to predict the rheological parameters of OWC slurries.
Shear rate (s-1) There are no generally accepted rules/guidelines to select the archi-
tecture or topography of a network. The topography and training
Downloaded from ascelibrary.org by New York University on 05/15/15. Copyright ASCE. For personal use only; all rights reserved.

Fig. 3. Shear-stress-shear rate down curve for OWC slurries prepared

by using PCH at 60°C
training error is reported over a given number or epochs. In each
iteration, the error [as shown in Eq. (3)] is compared with the
convergence tolerance; if it is not met, the iteration continues and
the calculated system error is back-propagated to the network to
adjust the weights and thresholds in a gradient search for the
desired minimum system error (El-Chabib and Nehdi 2005) as
follows:
1X P X K
Est ¼ ðt
opk Þ2
P p¼1 k¼1 pk
in which E st = system error; p = training pattern; P = number of
training patterns assigned to one epoch; and t pk and opk = predicted
output and provided target of pattern p at output unit k, respectively.
The construction of a successful neural network requires
considering three important steps: selection of database, network
architecture, and network training and testing.
Selection of Database
Although ANNs have been successfully used in predicting complex
nonlinear relationships and in modeling various aspects in cement
and concrete research, their efficiency depends on the quality of the
database used for training (El-Chabib et al. 2003). To account for
the primary aspects that influence the input-output relationship of
rheological properties of OWC slurries and capture the practical
range of key input parameters, the network should be trained by
using a large and comprehensive set of reliable experimental data.
To train the model, 190 data points were used for each of the three
admixtures tested (PCH, polycarboxlate-based midrange water-
reducing admixture [PCM], and lingosulphonate-based midrange
water-reducing admixture [LSM]). Fifty new data points unfamiliar
parameters obtained through trial and error for the ANN model thus
developed are presented in Fig. 4 and Table 3. The model param-
eters were selected on the basis of the lowest training and testing
error. For example, to determine the optimum number of hidden
nodes, the network performance was evaluated by changing the
number of hidden nodes alone and keeping all other parameters
unchanged. Fig. 5 shows that the addition of more hidden nodes
consistently improved the performance of the ANN in the training
process. However, in testing the model, its performance improved
up to Node 9, but the average absolute error (AAE) started to in-
crease thereafter. Therefore, nine hidden nodes were used in this
study to develop the ANN model. Different network architectures
can provide satisfactory performance for the same application.
ð3Þ
Training Process
Training a feed-forward back-propagation neural network to
predict the rheological properties of OWC involves teaching the
network the relationships between the input parameters (admixture
dosage, temperature, and shear rate) and the overall cement slurry
rheology. The success of the training process depends on (1) the
selection of network parameters, (2) the algorithm used for learn-
ing, and (3) the validation of the model by using experimental data
available for training along with new testing data unfamiliar to the
model. Specialized commercial computer software (Demuth et al.
2008) was used. Supervised training was implemented in this study
by providing the network with sets of data (inputs/targets) and the
network was instructed what to learn. Parameters such as the learn-
ing rate and convergence tolerance used for the ANN are presented
in Table 3. The unipolar log-sigmoid (logsig) function and linear
function were assigned as the transfer function for the processing
units in the input-hidden layers and the hidden-output layers,
respectively. A full connection for the processing units in adjacent
layers was adopted. No connection is permitted for neurons in the
same layer. The network receives the first input vector, performs the
appropriate computations and activation through the processing
units in the hidden layers, and produces an output value for shear

Table 2. Range, Average, and SD of Input and Output Variables

Training data Testing data
Range Average SD Range Average SD

1
Shear rate (s ), γ_ 5.11–511 258.1 153.6 5.11–511 257.9 157.52
Dosage of PCH (%) 0.25–1.0 0.625 0.281 0.25–1.0 0.631 0.320
Dosage of PCM (%) 0.25–1.0 0.624 0.287 0.25–1.0 0.619 0.332
Dosage of LSM (%) 0.5–2.0 1.267 0.563 0.5–2.0 1.301 0.654
Temperature (°C) 23–60 42.67 15.2 23–60 42.72 15.1
Shear stress (Pa), τ 1.57–144.9 35.1 29.1 1.97–140.6 36.0 29.3

1706 / JOURNAL OF MATERIALS IN CIVIL ENGINEERING © ASCE / DECEMBER 2011

J. Mater. Civ. Eng. 2011.23:1703-1710.

 1
Shear rate 2
t Results and Discussion
- 3
-
Admixture - 4 Validation of the Artificial Neural Network Model by
dosage - 5 Y Using Training and Testing Data
- 1 e = Y-t
6
-
-
7
The network model shown in Fig. 4 was trained by using 190
Temperature 8 training (input/target) patterns for each of the admixtures investi-
9 gated, and tested by using 50 patterns of data points unfamiliar
to the network. The ANN model predicted the shear stress of
the OWC slurries and the acceptability/rejection of the model
Input Hidden Output
Layer Layer Layer was evaluated by using the AAE given by Eq. (4) as follows:

Fig. 4. Architecture of developed ANN model 1X n

jY measured
Y predicted j
AAE ¼ ð4Þ
n i¼1 Y measured
Downloaded from ascelibrary.org by New York University on 05/15/15. Copyright ASCE. For personal use only; all rights reserved.

in which Y measured and Y predicted = shear stress value of OWC slurries

Table 3. Topography and Training Parameters for the Developed Artificial measured experimentally and the corresponding data predicted by
Neural Network Model the ANN, respectively; and n = total number of data points.
Fig. 6. shows the response of the ANN in predicting the shear
Name of the parameter Number/type used
stress of OWC slurries incorporating PCH. The figure shows that
Number of input nodes 3 all the measured and corresponding predicted data points are
Number of output nodes 1 located on or in the vicinity of the equity line with an AAE of
Number of hidden layers 1 3.43%. The acceptance/rejection of the ANN depends not only
Number of nodes in hidden layers 9 on its ability in predicting the training data, but more importantly
Activation function input-hidden layers Log-sigmoid on its performance to predict the shear stress of new sets of unfa-
Activation function hidden-output layers Linear
miliar data within the range of input variables of training patterns.
To validate the developed model, the network was presented with
Distribution of weights Gaussian
50 new sets of data that were not used in training the network. In
Momentum coefficient 0.03
this case, only an input vector of shear rate, dosage of admixture,
Learning rate 0.05 and temperature was presented to the network and no information
Convergence 5E
8 or knowledge about the shear stress was provided. The response of
the neural network is presented in Fig. 6(b) for OWC mixtures
made with PCH. The model predictions are accurate because the
testing points are located slightly over or under the equity line
8 but within the cluster of training data with an AAE of 2.76%.
Training data
Average absolute error (%)

Similar to PCH, the trained ANN models for PCM and LSM
Testing data
6 were able to accurately predict the shear stress corresponding to
each set of input data. The average, SD, and coefficient of variation
4
(COV) of the measured and predicted shear stress of OWC slurries
and the AAE for all the testing and training data are presented in
Table 4. The results reveal that the neural network has successfully
2 learned to map between input parameters and corresponding tar-
get used in the training process. The ANN model satisfactorily
0 predicted the shear stress with acceptable error. However, greater
0 5 10 15 20 25 AAE for OWC slurries incorporating PCH and PCM during train-
Number of hidden nodes
ing are probably attributable to the greater variability of shear stress
Fig. 5. Selection of number of hidden layer nodes for slurries prepared reading at lesser shear rates. Very small numerical values of shear
by using LSM stress can produce greater AAE even with very small errors of
prediction.

Performance of the Artificial Neural Network in

stress. The network compares its output with the corresponding
target provided in the output vector of the training pattern and
the difference is stored. The network is then provided with a second
training pattern, and so on, until the network goes through all data
available for training. This completes the first epoch. After com-
pletion of each epoch, calculations for the average sum-squared
of all errors were performed and back-propagated through the
Levenberg-Marquardt (Demuth et al. 2008) algorithm to adjust
the weights or connection strengths for the processing units (Nehdi
et al. 2001). This iterative process continues until the network
converges and a set of weights that minimizes the system error to
the desired level or the maximum number of iterations (epochs) has
been reached.
Predicting Rheological Properties of OWC Slurries
On the basis of the satisfactory performance of the developed ANN
model in predicting the shear stress of OWC slurries, it can be used
to develop the down flow curve by changing the shear rate and
keeping the admixture dosage and temperature unchanged. The
predicted curve can be used to predict rheological properties such
as the yield stress and plastic viscosity. Thus, one slurry mixture for
each of the admixtures was randomly selected from the testing data
and used to develop the down flow curve at different temperatures
(23, 45, and 60°C). These OWC mixtures were made with 0.5%
of each admixture. The yield stress and plastic viscosity were
determined by using the Bingham plastic model. The yield stress
was obtained by extrapolating the shear-stress-shear rate curve

JOURNAL OF MATERIALS IN CIVIL ENGINEERING © ASCE / DECEMBER 2011 / 1707

J. Mater. Civ. Eng. 2011.23:1703-1710.

 60 60
Training data Testing data

Predicted shear stress (Pa)

Predicted shear stress (Pa)
Equity line Equity line
45 45

30 30

15 15

AAE=3.43% (a) AAE=2.76% (b)

0 0
0 15 30 45 60 0 15 30 45 60
Measured shear stress (Pa) Measured shear stress (Pa)

Fig. 6. Measured versus predicted shear stress for OWC slurries incorporating PCH
Downloaded from ascelibrary.org by New York University on 05/15/15. Copyright ASCE. For personal use only; all rights reserved.

Table 4. Performance of the Artificial Neural Network in Predicting the Shear Stress of Cement Slurries Prepared with Different Chemical Admixtures
Y measured =Y predicted
Average absolute
error (%) Average SD Coefficient of variationa (%)
Type of admixture Training Testing Training Testing Training Testing Training Testing
New-generation polycarboxylate-based high-range 3.43 2.76 0.984 0.988 0.058 0.040 5.88 4.09
water-reducing admixture (PCH)
Polycarboxlate-based midrange water-reducing 3.17 2.77 0.998 1.001 0.062 0.040 6.18 4.01
admixture (PCM)
Lingosulphonate-based midrange water reducing 2.82 2.81 1.000 1.000 0.042 0.041 4.23 4.11
admixture (LSM)
a
Coefficient of variation ¼ SD=average × 100.

corresponding to a zero shear rate, and the plastic viscosity was the admixtures at 60°C is illustrated in Fig. 8. Admixture dosages not
slope of the curve. used in experiments were also included in ANN model predictions.
Figs. 7(a) and 7(b) represent the predicted yield stress and plas- Both experimental and predicted values of yield stress decreased
tic viscosity values, respectively, for OWC slurries incorporating with PCH and PCM admixture dosage. In the case of LSM, the
0.5% of PCH, PCM, and LSM at different temperatures, along with predicted yield stress values increased with the dosage up to 1.5%
the respective experimentally measured values. Both yield stress and then started to decrease, which is in good conformity with ex-
and plastic viscosity values predicted by the ANN followed the perimental results. At relatively low dosages, LSM did not improve
same trend as that of the experimental data. In addition to test the rheological properties of cement slurries, but rather acted as an
temperatures (23, 45, and 60°C), rheological parameters were also accelerator. It was documented that the combined effect of adsorp-
determined at 35 and 52°C to predict the model’s response within tion and nucleation is responsible for the hydration retardation
the range of input data. It can be observed that the yield stress for induced by admixtures (Ramachandran et al. 1997; Nelson et al.
OWC slurries incorporating PCH was less than that for slurries 2006). Admixtures inhibit the contact of cement grains with water
made with PCM and LSM. Both the yield stress and plastic viscos- by adsorbing onto the surface of cement grains and the hydration
ity were found to be sensitive to the change in temperature; the products throughout the hydration process and thereby delay the
greater the temperature the greater was the yield stress, which is in hydration process. Retarders may also adsorb onto the nuclei of
good agreement with experimental results. hydration products and can inhibit the further hydration. However,
The effect of the admixture dosage on the predicted rheological at greater temperature, this layer can break down and the rate of
parameters of cement slurries incorporating various dosages of hydration is accelerated. This may be one reason why LSM seemed

80 0.12
0.5% PCH Exp 0.5% PCH ANN 0.5% PCH Exp 0.5% PCH ANN
0.5% PCM Exp 0.5% PCM ANN 0.5% PCM Exp 0.5% PCM ANN
0.1
Plastic viscosity (Pa.s)

0.5% LSM EXP 0.5% LSM ANN 0.5% LSM EXP 0.5% LSM ANN
60
Yield stress (Pa)

0.08

40 0.06

0.04
20
0.02
(a) (b)
0 0
0 20 40 60 80 0 20 40 60 80
Temperature (°C) Temperature (°C)

Fig. 7. Variation of (a) yield stress and (b) plastic viscosity of OWC slurries at different temperatures

1708 / JOURNAL OF MATERIALS IN CIVIL ENGINEERING © ASCE / DECEMBER 2011

J. Mater. Civ. Eng. 2011.23:1703-1710.

 80
60
Yield stress (Pa)
40
20
0 0
0 0.5 1 1.5
Admixture dosage (% BWOC)
Fig. 8. Variation of (a) yield stress and (b) plastic viscosity of OWC slurries with admixture dosage
Downloaded from ascelibrary.org by New York University on 05/15/15. Copyright ASCE. For personal use only; all rights reserved.
to increase the yield stress and viscosity at high temperature.
Moreover, at a lesser dosage, the adsorbed layer of admixture might
not be sufficiently effective to act as a barrier to prevent the contact
of water and cement grains, which promotes the acceleration of
hydration reactions. This could be the reason why relatively greater
yield stress and viscosity values were observed at low dosages of
LSM. However, at 60°C and dosages greater than 1.5%, LSM
started to act as a dispersant (Shahriar and Nehdi 2010). The varia-
tion of yield stress with admixture dosage was reasonably estimated
for all the admixtures considered and the predicted values were
comparable to the corresponding measured data. Moreover, the
plastic viscosity of OWC slurries was found to be sensitive to the
change of temperature [Fig. 7(b)] and admixture dosage [Fig. 8(b)].
The plastic viscosity values predicted by the ANN model showed
irregular behavior, which may be associated with the error involved
in fitting the curve to the Bingham model (Nehdi and Al-Martini
2007; Saak 2000).
Figs. 7 and 8 reveal that the model was able to recognize and
evaluate the effects of the admixture dosage and temperature on
yield stress and plastic viscosity with an AAE in the range of 1.4
to 15.6% and 0.7 to 11.8% for yield stress and plastic viscosity,
respectively, depending on the admixture dosage and temperature
tested. The greater values of AAE are usually associated with the
lesser yield stress and plastic viscosity values because small pre-
diction errors may result in high AAE in such cases.
Predicting various properties of cementing systems such as
OWC slurries as a function of mixture ingredients is difficult to
achieve analytically, whereas a properly trained ANN model
can successfully predict such behavior with adequate accuracy
(El-Chabib et al. 2003). The present study investigated the feasibil-
ity of using ANNs to create an intelligent model that can be used to
predict the rheological properties of OWC slurries and the model
developed in this study has demonstrated the ability of ANNs to
recognize and evaluate the effect of individual input parameters
on the rheological behavior of OWC slurries. However, the predic-
tion capability of the ANN model thus developed is on the basis
of the mixture proportioning of OWC slurries with a single chemi-
cal admixture. Possible coupled effects or interactions with other
admixtures, mixing techniques, and handling or environmental
parameters have not been taken into consideration. To address these
limitations, the model can be extended by using a more compre-
hensive database encompassing a wider scope of OWC slurry mix-
ture compositions. The model can easily be retrained to incorporate
a wider range of input variables by providing additional training
sets that cover the new data range to capture other properties such
as thickening time, setting time, mechanical properties, and dura-
bility performance of OWC slurries. Therefore, the extended model
can be used on an industrial scale to limit the number of laboratory
J. Mater. Civ. Eng. 2011.23:1703-1710.
0.12
(a) (b)
0.1
Plastic viscosity (Pa.s)
0.08
PCH Exp PCH Exp
PCM Exp 0.06 PCM Exp
LSM Exp LSM Exp
0.04
PCH ANN PCH ANN
PCM ANN 0.02 PCM ANN
LSM ANN LSM ANN
2 2.5 3 0 0.5 1 1.5 2 2.5 3
Admixture dosage (% BWOC)
trial mixtures and to develop OWC slurry compositions with tailor-
made properties with less extensive laboratory work.
Concluding Remarks
In this study, the relationships among the shear stress, shear rate,
temperature, admixture type, and admixture dosage for OWC
slurries have been analyzed and the rheological properties were
modeled by using a feed-forward back-propagation ANN. The
model was then used to develop flow curves, which were used to
calculate the yield stress and plastic viscosity values for OWC
slurries with different admixtures and at different test tempera-
tures. On the basis of this study, the following conclusions can be
made:
1. The ANN developed in this study was able to learn the rela-
tionships among different shear flow parameters for various
OWC slurries and successfully predicted the rheological prop-
erties of OWC slurries used in the training process. It also de-
monstrated satisfactory performance when input parameters
(shear rate, temperature, and dosage of admixture) unfamiliar
to the neural network were used.
2. The flow curves developed by using the ANN model allowed
predicting the Bingham parameters (yield stress and plastic
viscosity) with an acceptable accuracy and were found to
be in good agreement with experimental results. The results
prove that the neural network is a powerful tool to quantita-
tively predict the rheological properties of OWC slurries within
the range of tested admixture dosages and temperatures.
3. The ANN model was found to be sensitive to the effect of
temperature increase and admixture dosage on the rheological
properties of OWC slurries.
4. The model was also found to generalize its predictions beyond
the training data to new slurries incorporating different dosages
of admixtures and considering different temperatures within
the practical range of training data.
References
American National Standards Institute, New York/American Petroleum
Institute (ANSI/API). (2005). “Recommended practice for testing well
cements.” RP 10B-2, Washington, DC, 171.
Bruni, C., Forcellese, A., Gabrielli, F., and Simoncini, M. (2006).
“Modelling of the rheological behaviour of aluminium alloys in
multistep hot deformation using the multiple regression analysis and
artificial neural network techniques.” J. Material Processing Technol-
ogy, 177(1-3), 323–326.
Demuth, H., Beale, M., and Hagan, M. (2008). Neural network tool box 6
for use with MATLAB R2008a, Math Works, Natick, MA, 5.2–5.72.
JOURNAL OF MATERIALS IN CIVIL ENGINEERING © ASCE / DECEMBER 2011 / 1709
 El-Chabib, H., and Nehdi, M. (2005). “Neural network modelling of prop-
erties of cement-based materials demystified.” Adv. Cem. Res., 17(3),
91–102.
El-Chabib, H., Nehdi, M., and Sonebi, M. (2003). “Artificial intelligence
model for flowable concrete mixtures used in underwater construction
and repair.” ACI Mater. J., 100(2), 165–173.
Guillot, D. (2006). “Rheology of well cement slurries.” Well cementing,
E. B. Nelson and D. Guillot, eds., Schlumberger, Sugar Land, TX,
93–142.
Nehdi, M., and Al-Martini, S. (2007). “Effect of chemical admixtures on
rheology of cement pastes at high temperature.” J. ASTM Int., 4(3), 17.
Nehdi, M., El-Chabib, H., and El-Naggar, M. (2001). “Predicting the
performance of self-compacting concrete mixtures using artificial
neural networks.” ACI Mater. J., 98(5), 394–401.
Nehdi, M., and Rahman, M. A. (2004). “Estimating rheological properties
Downloaded from ascelibrary.org by New York University on 05/15/15. Copyright ASCE. For personal use only; all rights reserved.
of cement pastes using various rheological models for different test
geometry, gap and surface friction.” Cem. Concr. Res., 34(11),
1993–2007.
Nelson, E. B., Michaux, M., and Drochon, B. (2006). “Chemistry and
characterization of portland cement.” Well cementing, E. B. Nelson
and D. Guillot, eds., Schlumberger, Sugar Land, TX, 23–48.
1710 / JOURNAL OF MATERIALS IN CIVIL ENGINEERING © ASCE / DECEMBER 2011
J. Mater. Civ. Eng. 2011.23:1703-1710.
Orban, J., Parcevaux, P., and Guillot, D. (1986). “Influence of shear
history on the rheological properties of oil well cement slurries.”
8th Int. Congress on the Chemistry of Cement, Vol. 6, Abla Gráfica
Editora, Rio de Janiero, Brazil, 243–247.
Ramachandran, V. S., Malhotra, V. M., Jolicoeur, C., and Spiratos, N.
(1997). Superplasticizers: Properties and applications in concrete,
Materials Technology Laboratory, CANMET, Natural Resources
Canada, Ottawa, 43–150.
Ravi, K. M., and Sutton, D. L. (1990). “New rheological correlation
for cement slurries as a function of temperature.” SPE 20499, 65th
Annual Technical Conf. and Exhibition of the Society of Petroleum
Engineers, Society of Petroleum Engineers, Richardson, TX,
455–462.
Rumelhart, D. E., Hinton, G. E., and Williams, R. J. (1986). “Learning
internal representation by error propagation.” Parallel distributed
processing, Vol. 1, MIT Press, Cambridge, MA, 318–362.
Saak, W. A. (2000). “Characterization and modeling of the rheology of
cement paste: With application toward self-flowing materials.” Ph.D.
thesis, Northwestern Univ., Evanston, IL, 283.
Shahriar, A., and Nehdi, M. (2010). “Effect of chemical admixture on
rheology of oil well cement slurries.” Const. Mater., in press.