Documente Academic
Documente Profesional
Documente Cultură
DIMITRIS TSAMATSOULIS
Halyps Building Materials S.A., Italcementi Group
17th Klm Nat. Rd. Athens – Korinth
GREECE
d.tsamatsoulis@halyps.gr http://www.halyps.gr
Abstract: - This study is aiming to search the optimum settings of two kinds of dynamic linear models
predicting cement typical strength and afterwards to couple them to achieve optimality. The modeling is
based on physical and chemical characteristics and on the early strength of the Portland cement types studied.
More than 3000 sets of industrial results were used for this purpose. Models parameters are calculated using a
moving past period of length TD and are tested in a future period of length TF. The moving period
characterizing the dynamical models attempts to assure that changes in the process are taken into account
during the parameters calculation. Among the models the coupled one is noticeably superior as regards the
mean square residual error. The implementation of these methods in the daily quality control is an
essential factor of quality improvement by maintaining a low variance of typical strength.
ISBN: 978-1-61804-279-8 98
Advances in Electrical and Computer Engineering
order reactions. Tsamatsoulis [14] presented a dynamic linear model as concerns the prediction of
similar kinetic model of strength development using future 28 days strength, by taking into account the
different cement and aggregate types. De Siqueira information becoming from the second one. This
Tango [15] presented an extrapolation method for latter information is filtered and added to the result
compressive strength prediction of cement products of the first model. The filtering parameters are also
considering that the typical 28 days strength is a optimized. The above constitutes the coupled linear
function only of the two earlier ages’ of 2 and 7 days model of 28 days strength prediction. The study is
strengths. Tsamatsoulis [16] elaborated a restricted to Portland cement.
multivariable polynomial model by utilizing the 1
and 7 days early strength, physical and chemical 2 Experimental
data to predict the typical strength. The main feature Two Portland cement types produced according to
of the two last contributions [15,16] is the EN 197-1:2011 were studied: CEM II A-L 42.5 N
introduction of the early strength as an independent and CEM II B-M (P-L) 32.5 N. The first type,
variable, despite that this strength is a function of except clinker and gypsum, contains also limestone,
the physical and chemical structure of cement. while the second type pozzolane and limestone, as
The main feature of the described models is that main components. The modeling is based on the
they are static: (a) A set of data is used to estimate results of the daily average samples of cement
the model parameters; (b) the future strength is produced in two cement mills (CM) of Halyps plant.
computed based on these set of parameters. Relis et The analyses made on these samples were the
al. [17] developed a linear regression model to following:
predict the strength of Portland cement by (i) Residue at 40 μm sieve, R40 (%), measured
introducing a time-sequence dynamic correction with air sieving .
procedure to enhance the model accuracy. Their (ii) Specific surface, Sb (cm2/gr), measured
final model includes constant coefficients of the according to EN 196-6.
linear model and the addition of the dynamical (iii) Loss on ignition, LOI (%), and insoluble
correction. Tsamatsoulis [18] performed an initial residue, Ins_Res (%) of the cement
comparison of the static polynomial equations measured according to EN 196-2.
referred in [16] and movable time horizon models (iv) SO3 (%) measured with X-ray fluorescence.
based on linear regression methods. The latter (v) Compressive strength at 1, 7 and 28 days
models incorporate the uncertainty due to the time (MPa). The preparation, curing and
variability of non involved factors during the measurement of the specimens were made
modelling procedure and they can be characterized according to the standard EN 196-1.
as dynamic. In [19] the particularities of these two The modeling predicting the 28-day strength was
classes of models have been investigated in detail. based on more than 3400 data sets of cement
This study is initially based on the analysis of fineness, chemical analyses, 1, 7 and 28 days
Tsamatsoulis in [18, 19] where dynamic linear strength.
models have been utilized to predict the typical
cement strength. The models are characterized as
dynamic because the parameters were estimated
3 Mathematical Models Predicting
from a moving set of data belonging to a predefined Strength
past time interval. Two classes of equations The common independent variables in all models
developed in these earlier studies: Except physical are: LOI, SO3, Ins_Res, R40, Sb. The reason to
and chemical data, the first one utilized the one day utilize the chemical analysis of cement instead of the
strength results while the second one the results of cement composition used in the earlier modeling
seven days strength too. The first model is broadly presented in [18], is to generalize as much as
utilized in Halyps cement plant to regulate the possible the derived equations: The direct usage of
cement composition according to the 28 days chemical analysis do not need prior knowledge of
strength estimation. The second model is much more raw materials and clinker composition. Two basic
accurate from the first one but due to its much and independent equations are initially implemented
bigger delay time compared with the first one, to predict the 28 days strength. (i) The one where the
generally it cannot be used for direct control one day strength - Str_1- constitutes model variable,
purposes. However, it constitutes additional except the set of physical, chemical data. This model
information for the cement composition adjustment. consists of 6 independent variables and is named
These two models operate independently. The main Str_28_1. (ii) The second one where the seven days
objective of this study is to optimize the first class of strength variable –Str_7- is also included and the
ISBN: 978-1-61804-279-8 99
Advances in Electrical and Computer Engineering
model, named Str_28_7, is a 7 variables linear without new parameters estimation. Consequently
equation. (iii) The third model contains all the TF is ≥ 1.
independent variables of Str_28_1 and an additional (vii) According to step (vi) for the dates belonging
correction computed from the moving difference of to the interval [t, t+TF-1], the future strength of the
Str_28_7 and the actual 28 days strength measured cement produced in the time intervals [t-2, t+TF-3],
in the date the model is to be applied. The model is [t-8, t+TF-9] is computed according to the equation
named Str_28_EW. The linear function describing of step
the three models is given by equation (1). (v). Otherwise if the date is greater than t+TF-1, new
parameters estimation is performed starting from
step (i).
(viii) As the time span remains TD, when the results
For all the models the parameters AI, I=(0..N) of TF days are completed, then, the time interval is
were estimated from a moving set of data belonging moved on by TF days. Thus the future 28 days
to a past time interval of predefined size TD in days strengths are calculated using models applied to data
and the model is applied for a determined time sets of movable time span TD and in steps of length
interval TF before the parameters to be recalculated. TF.
The latest date of the time interval TD has two (ix) Parameters TD, TF shall be optimized
characteristics: (i) the 28-day strength has been considering the following two criteria: (a) minimum
measured; (ii) its distance from the first date of sPast during modeling and (b) minimum error sFutur
prediction is minimal. The parameters AI, I=(0..N) during the future application of the models.
are computed via regression by minimizing the (x) For each TD and TF and for each past and future
objective function given by equation 2. time interval, a set (AI, sPast, sFutur) is computed from
the samples belonging to this interval. Depending on
TD and TF values, the number of the consecutive sets
(AI, sPast, sFutur) is KTD, the number of data sets in
each past interval I is NTD(I) and in each future
where Yact = actual 28 days strength, Ycalc = the interval J is NTF(J). The mean square residual
calculated one from the model, M = number of data errors, sPast, during modeling and sFutur during future
sets, p = number of independent variables. prediction are calculated by equations (3) and (4)
respectively.
3.1 Parameters Estimation Algorithm
The parameters of models Str_28_1, Str_28_7 were
computed by the following algorithm:
(i) At date t a new 28-day strength result appears.
The specimen was prepared 28 days ago. The
production date is in distance t-29 days from the
current date t.
(ii) A time interval of TD days and the samples
belonging to the period [t-29-TD, t-29] are The described procedure is a generalization of the
presumed. The dynamic data set consists of this respecting procedure presented in [18] and
population of samples. constitutes a step forward as to the method
(iii) Using multiple regression the model parameters optimization.
AI (I=0 .. N) and sres are computed.
(iv) At day t, the chemical and physical results of the 3.2 Combination of Linear Models and
cement produced in the previous day, the 1 day Dynamical Correction
strength of the cement produced 2 days ago and the The third model – Str_28_EW – constitutes an
7 days strength of cement produced 8 days ago have extension of Str_28_1 and first of all needs the
been measured. definition and implementation of the moving
(v) With the set of parameters computed in step (iii) average filter. The exponentially weighted moving
the 28 days strength of cement produced at t-2 and t- average (EWMA) technique was used for this
8 days are estimated, by applying the models purpose. As analyzed in [16], for a variable X and
Str28_1 and Str28_7 respectively. discrete time I, the EWMA variable Y is defined by
(vi) The steps (iv), (v) are repeated up to the date the procedure:
t+TF-1, where TF is a predetermined time interval (i) For time I=0 the initial moving average
Y(0) is expressed by the relation (5):
of these models. For this reason dynamic models [5] Osbaeck, B. and Johansen, V., Particle Size
have been elaborated, correlating the 28 days Distribution and Rate of Strength Development
strength with physical and chemical features and of Portland Cement, Journal of the American
early strength: The first one named Str_28_1 which Ceramic Society, Vol. 72, 2005, pp. 197-201.
takes into account the strength measured at one day, [6] Tsivilis, S. and Parissakis, G., A Mathematical
and the second one, the Str_28_7, including also the Model for the Prediction of Cement Strength,
seven days strength. A third improved model takes Cement and Concrete Research, Vol. 25, 1995,
into account the moving average difference between pp. 9-14.
the Str_28_7 values and the actual ones to correct [7] García-Casillas, P. E., Martinez, C. A., Montes,
the Str_8_1 model. In this way a coupling of the two H. and García-Luna, A., Prediction of Portland
independent models is succeeded resulting in a Cement Strength Using Statistical Methods,
combined linear model. The software developed Materials and Manufacturing Processes, Vol.
optimizes two parameters per cement type, to 22, 2007, pp. 333-336.
determine the minimum MSQE of the test sets. The [8] Kheder, G. F., Gabban, A.M. and Abid, S.M.,
models are applied to a moving data set belonging to Mathematical model for the prediction of cement
a past time interval of predefined size TD in days, to compressive strength at the ages of 7 and 28
calculate the parameters and then to a future time days within 24 hours, Materials and Structures,
interval TF. The first set is the training set, while the Vol. 36, 2003, pp. 693-701.
second one is the test set. [9] Abd SM, Zain MFM, Abdul Hamid R. Modeling
The main conclusions of the presented analysis the Prediction of Compressive Strength for
are the following: (a) for each TD value, as TF Cement and Foam Concrete, Proc. of
decreases the MSRE of the test set decrease too; (b) International Conference on Construction and
the Str_28_EW model results in a severe reduction Building Technology, Kuala Lumpur, 2008, pp.
of the error compared with the St_28_1. The MSRE 343 – 354.
of Str_28_EW remains continuously lower than the [10] Tepecik, A., Altin, Z. and Erturan, S.,
respecting of Str_28_1. The average reduction is Modeling, Compressive Strength of Standard
6%, reaching up to 10%. (c) The minimum MSRE is CEM I 42.5 Cement Produced in Turkey with
achieved with Str_28_EW and TD=240, TF=1, e.g. Stepwise Regression Method, Journal of
these are the optimum time parameters. The further Chemical Society of Pakistan, Vol. 31, 2009, pp.
improvement of these techniques can follow the 213-220.
next directions. [11] Tsamatsoulis, D. and Nikolakakos, N.,
- Combination of the dynamical models with Optimizing the Sulphates Content of Cement
neural network techniques to investigate Using Multivariable Modeling and Uncertainty
possible synergies. Analysis, Chemical and Biochemical
- Exploitation of the dynamic models to develop Engineering Quarterly, Vol. 27, 2013, pp. 133-
robust controllers based on Model Predictive 144.
Control (MPD) techniques or other advanced [12] Mechling, J. M., Lecomte, A. and Diliberto, C.,
methods. Relation between Cement Composition and
Compressive Strength of Pure Pastes, Cement
and Concrete Composites, Vol. 31, 2009, pp.
References: 255-262.
[1] Lee, F.M., The Chemistry of Cement and [13] Popovics, S., Model for the Quantitative
Concrete, 3rd ed. Chemical Publishing Company, Description of the Kinetics of Hardening of
New York, 1971. Portland Cements, Cement and Concrete
[2] Odler, I., Cement Strength, Materials and Research, Vol. 17, 1987, pp. 821-838.
Structures, Vol. 24, 1991, pp. 143-157. [14] Tsamatsoulis, D., Kinetics of Cement Strength
[3] Zhang, Y.M. and Napier-Munn, T.J., Effects of Development Using Different Types of Cement
Particle Size Distribution, Surface Area and and Aggregates, WSEAS Transactions on
Chemical Composition on Portland Cement Systems, 2009, Vol. 8, pp. 1166 - 1176.
Strength, Powder Technology, Vol. 83, 1995, pp. [15] De Siqueira Tango C.E, An extrapolation
245-252. method for compressive strength prediction of
[4] Celik, I.B., The Effects of Particle Size hydraulic cement products, Cement and
Distribution and Surface Area upon Cement Concrete Research, Vol. 28, 1998, pp. 969-983.
Strength Development, Powder Technology, [16] Tsamatsoulis, D., Control Charts and Models
Vol. 188, 2009, pp. 272-276. Predicting Cement Strength: A Strong Tool