Applications of Chemometrics to Building Mortar

Formulations, a Case Study: Cementitious Tile

Adhesives

Ezio Leone, Mapei S.p.A., Enrico Marcantoni.

Abstract
Chemometrics is the chemical discipline that applies mathematical and statistical methods in order to
obtain an optimization of experimental designs and to maximize the quality of information obtainable
from chemical data thorough multivariate statistical analysis.
The application of Chemometrics to cementitious tile adhesives formulations will be examined in this
paper. The definition of a proper design of experiments will be finalized to the implementation of a
mathematical model of physical-mechanical behaviors of cementitious tile adhesives, both from the point
of view of ISO 13007 testing and from the one of their rheological properties. Furthermore, multivariate
statistical analysis will be employed in order to identify possible correlations among mechanical and
rheological properties of cementitious tile adhesives. In addition, a Design of Experiment was also
designed in order to study the dustiness generation of dry blends.

Introduction
a. Cementitious Tile Adhesives
Cementitious tile adhesives formulations are ideally implemented on the basis of different objectives1:
1) Fulfillment of international norms requirements (in particular ISO 13007 and EN 12004).
2) Real world application and final users appreciation.
3) Maximization of performances and minimization of production costs.
It is self-evident how, depending upon these criteria, cementitious tile adhesives formulation represents
a textbook example of research for the perfect compromise. Historically, this research has always been
carried on following the linear approach of univariate analysis, optimizing independently each single
compositional variable.
 b. Chemometrics - Multivariate approach in Formulation Science
Choosing a multivariate approach in formulation development, both from the point of view of
Multivariate Design of Experiment and Multivariate Data Analysis, offers many non-negligible
advantages2:
1) It provides speed in obtaining real-time information from data.
2) It allows high quality information to be extracted from less resolved data.
3) It promises to improve measurements.
4) It improves knowledge of existing processes.
5) It has very low capital requirements.

c. Chemometrics – Experimental Design

Experimental Design allows to optimize the amount of available resources while carrying out an
experimental plan, having the possibility to predict the accuracy of the final model results before starting
any battery of tests, with a determination of the leverage level (model accuracy) in every single point of
the investigated space3.
All the experiments described in this paper are based on Full or Fractional – Factorial Designs (that allow
to obtain a linear responses model), developed with the addiction of Central Composite Designs,
indispensable in order to generate curve responses in mathematical modeling.
Practically, for each investigated variable three different composition levels have been set:
 -1 level = Minimum level
 0 level = Average level
 +1 level = Maximum level
The resulting experimental plan, is a rational combination of these levels for each variables, that provides
a full inspection of the multi-dimensional space determinate by the total number of components
investigated.
Fractioned-factorial designs have the same characteristics of Full-factorials but reducing the amount of
experiments realized in correspondence of the multidimensional space vertexes. This operation can be
conveniently used for designs with 5 or more variables as, in these cases, the use of this artifice does not
generate any confusion among the calculated coefficients (constant, linear, interaction and quadratic).
Using this approach with a reduced amount of variables can, instead, generate confounded effects among
interaction coefficients4. For this reasons, fractional factorial designs have been used in present work.
 a. Chemometrics – Principal Component Analysis
Principal Component Analysis (PCA) is a statistical procedure, invented in 1901 by Karl Pearson5, that
uses an orthogonal transformation to convert a set of observations of possibly correlated variables into a
set of values of linearly uncorrelated variables called principal components. This transformation is
defined in such a way that the first principal component has the largest possible variance and each
succeeding component in turn has the highest variance possible under the constraint that it is orthogonal
to the preceding components. The resulting vectors are an uncorrelated orthogonal basis set.
The results of a PCA are usually displayed in terms of component scores (the transformed variable values
corresponding to a particular data point), and loadings (the weight by which each standardized original
variable should be multiplied to get the component score).6

Materials

a. Tile Adhesives
In this work we analyzed several components of typical tile adhesives formulation:
a. Portland cement “Colacem 52.5R GREY”; batch: July 2017 (20% - 40%).
b. Cellulose ether (0.25% - 0.45%), modified or non-modified.
c. Redispersible polymeric powder (0% - 5%).
d. Fine calcium carbonate (<100 μm) (0% - 20%).
e. Coarse calcium carbonate (<600 μm) (0% - 20%).
f. Cellulose fibers (0% – 2%).
g. Silica sand (<450 μm).
h. Preconditioned water at 23 °C.

b. Concrete slabs
All the concrete slabs used for tile adhesive testing were in compliance with ISO 13007 / EN 1323 in
terms of water absorption, residual humidity and superficial resistance. They were all coming from the
same producer, with random production batches used during the experimental campaign.
Experimental methods

a. Evaluation of adhesion of cement based tile adhesives

All the investigated formulations were checked following the same preparation procedure and pulled off
with identical dynamometers (LBG M053, properly calibrated according to ISO 7500-1). The laboratory
was maintained by a proper system fully conditioned, keeping temperature, humidity and air speed within
the range dictated by the norm (i.e. 23±1°C, relative humidity of 50±5% and air speed lower than 0.2
m/s).
After a standardized mixing procedure, regulating the amount of water so that the Brookfield viscosity
of the wet mortar is around 500,000 mPa∙s, the samples are prepared spreading the adhesive on a concrete
slab in a single pass, with a 6mm-toothed square notched trowel and then applying 50 mm x 50 mm tiles
with a 20N weight following the ISO 13007 timing for each test.
The slabs prepared with each adhesive were submitted to all the curing procedures according to
ISO13007 and listed in table 2.

Days in air
Days in oven Freeze / Thaw
(23°C - 50% Days in water
(70°C) Cycles
R.H.)
24 hours 1 - - -
Open time (5’, 20’ and 7 - - -
30’)
Initial adhesion 28 - - -
Heat (5’, 10’ and 15’) 14 14 - -
Water (5’, 10’ and 15’) 7 - 21 -
Freeze / Thaw 7 - 21 25
Table 1: sample curing for pull-off test

Once the final curing time is reached, a metal plate is applied on the tiles with an epoxy adhesive and a
pull-off test is performed with a constant load rate of 250 N/s with pull-off dynamometers that satisfy
the minimum requirements for Class 1 according to ISO 7500-1, with a minimum sampling frequency
of 50Hz.
These tests allow to dictate the adhesive mortar classes according to ISO 13007.
 b. Evaluation of cement based tile adhesives fresh properties
Some fresh properties of adhesives were evaluated too.
Brookfield viscosity has been chosen as an indicator of adhesive consistence: in this way, the water ratio
used for mixing could be an output variable of this experimental campaign.
Specific gravity of the fresh material has been evaluated with internal methods, as well as adjustability
time of the adhesive, checked as the maximum time at which 100 mm x 100 mmm P1 tile can be
“adjusted” by the operator when applied on a non-absorbent substrate.
Slip tests were performed according to ISO 13007 / EN 1308, applying a layer of adhesive, leaving in on
a slab for two minutes, then applying a 100 mm x 100 mm V1 tile, measuring its distance from a reference
bar, then keeping it vertical for 20 minutes and finally measuring its slippage from the initial position.
Anyway, these measurements have a low reproducibility and are affected by the operator’s error and
thus, to fully understand the rheological properties of cement adhesives, these methods are not suitable.
Therefore, proper rheological evaluation is needed to provide a useful information for the optimization
of fresh cementitious adhesives performance. Rotational rheometry was used to measure the viscosity of
cement based adhesives by varying shear stress.
In this work, two geometries were used: the cross-hatched parallel-plate geometry and ball-measuring
system geometry in continuous, oscillatory tests and creep tests.
Rheological properties were evaluated using “AR-G2” of TA Instrument and “MCR 302” of Anton Paar.
For the TA rheometer the parallel-plate, cross-hatched geometry was used. For the Anton Paar rheometer
the ball-measuring system was used.

c. Evaluation of cement based tile adhesives deformability

All the investigated samples were tested for their deformability, according to ISO 13007 / EN 12002
norms. After a standardized mixing procedure, six samples were prepared for each formulation, cured
for 2 days under their molds, 12 days in plastic boxes and further 14 days in standard conditions.
Testing phase (3 points flexural test) has been performed using a Uniframe-Mini dynamometer from
Controls, with a sensitivity of 0.001N, a preload value of 0.1N and using deformability at maximum load
as breaking criterion.

d. Evaluation of cement based tile adhesives dustiness

Dustiness of cementitious tile adhesives is a real-world practical issue, generated every time a bag of
product is poured into a bucket. This property has been evaluated with an AnaTec DustMon L machine.
A fixed amount of cementitious product drops in an 80 cm tube and then generates a cloud of dust inside
a testing chamber: the measurement is taken with a laser light detector, that transforms the amount of
non-passing light in a dustiness index. These values (%) are recorded immediately after the dust cloud
generation and after 30 second from the impact.
The so-called “Dust Index” is the sum of these two numbers and can be considered as an indicator of air
quality during the use of cementitious products.

e. Data analysis software

An R-language based free Chemometrics software distributed from Italian Chemometrics Group7 has
been used for data analysis, in order to use the proper Design of Experiment and to obtain all the equations
coefficients necessary to describe the behavior of adhesives in the whole investigated domain.
An internal software based on Microsoft Excel© and Microsoft Visual Basic© has been developed in
order to obtain an intuitive graphical evaluation of the obtained results8.

Experimental designs

a. Tile adhesive model

This Full-Factorial with Central Composite Design model, ideated in order to understand the mechanical
and rheological behavior of tile adhesives includes the following formulation variables:
1) Ordinary Portland cement (20% ÷ 35%)
2) Cellulose ether (0.25% ÷ 0.45%)
3) Modification in cellulose ether (non-modified, highly modified and a 50/50 mixture of them)
4) Redispersible polymeric powder (0% ÷ 5%)
5) Fine calcium carbonate (0% ÷ 20%)

b. Tile adhesives dustiness model

This Fractional-Factorial with Central Composite Design model, ideated in order to investigate dust
generation in cementitious tile adhesives, includes the following formulation variables:
1) Ordinary Portland cement (20% ÷ 40%)
2) Fine calcium carbonate (0% ÷ 20%)
3) Coarse calcium carbonate (0% ÷ 20%)
4) Redispersible polymeric powder (0% ÷ 10%)
5) Cellulose fibers (0 ÷ 2%)
Experimental results

a. Correlations in tile adhesives testing

Using the Principal Component Analysis technique and the tile adhesives Experimental Design, it is
possible to evaluate the correlations among the whole performance profile of cementitious tile adhesives,
obtaining the following result:

Loading Plot (67% of total variance) Score Plot (67% of total variance)

Load
E'5
E'23
F/T Cycles Water 5'

6
E'50
0.3

12
Sp. grav. 24h 9
10

4
Water 10' 28 gg
Delta E' H.A. 5'
0.2
Component 2 (27.9% of variance)

Component 2 (27.9% of variance)

29
11
26 25 14
13

2
0.1

Water 15' 16 24 23
H.A.
O.T.10'
5' 21 17
18
15
+ Mapei
EN Deform
Slip
Slip .
H.A. 15' + 20
19
30 3231
0.0

Water ratio 4
33 27
Adjust. 1 22
2
-0.1

-2

28 6
O.T. 20'
O.T. 30'
-0.2

8
-4

3
7
5
-0.3

-6

-0.3 -0.2 -0.1 0.0 0.1 0.2 0.3 -5 0 5

Component 1 (39.1% of variance) Component 1 (39.1% of variance)

Figure 1: Cementitious tile adhesives loading and score plot.

1. All the elastic modulus from the DMA (E’5, E’23, E’50), 24 hours pull-off, water immersion,
freeze-thaw cycles and pull-off after heat aging are strictly correlated to each other and they are
all related to the cement amount and to the specific gravity of the adhesive. In this way, they all
seem inversely correlated to the amount of cellulose ether in formulation. The cement, among
other things, gives rigidity to the structure of the hardened adhesive, increasing its modulus. This
observation can be confirmed by observing the score plot in which we can identify some samples
that are positioned in the same area where, in the loading plot, the tests correlated to the cement
amount are present. In fact, if we take into account the samples that are present in that area (i.e.
9, 11 and 12), they all have a high content of cement and little amount of cellulose ether.
2. Open time at 20’, at 30’, adjustability and slip resistance are strictly correlated and they are all
related to the cellulose ether content. A high amount of cement and a high specific gravity
penalize these results. Water immersion and heat aging at 5’ are correlated to the cement amount
and at 10’ and especially at 15’ they move into the cellulose group.
3. Polymer generally improves all the performances, with a lower effect on water immersion and
freeze-thaw cycles and a very important contribution over heat, initial adhesion and deformability
results.
 4. The modified cellulose ether improves only the slip test (that, naturally, has better results when
the values are low) and strongly penalize all other results.

Response Surface Curves and Discussion

a. Slip and creep tests

The slip test is intended to measure the resistance of the adhesive to the sliding of the tile, which is carried
out by placing the tile vertically and measuring the sliding in millimeters after two minutes with a caliber.
In correspondence, to get a more accurate measurement, a creep test is performed with a rotational
rheometer. These two tests are therefore closely related to each other, in fact both demonstrate that: the
cellulose itself makes cement paste less thick and more fluent and for this reason, cellulose ether increases
the slippage of the adhesive, not desired in this test. Cellulose ether modification is a thickener ingredient
and possesses opposite properties to the unmodified cellulose. In fact, as can be seen from the graph, the
slippage decreases with the increase of the modification. The filler, which is the fine calcium carbonate,
dries the adhesives and decreases slip and creep. Increasing the amount of polymer, however, slippage
and creep increase.

Figure 2: RSC graphs of slip test and creep test.

b. Open time test at 5’ and 30’

While open time at 5 minutes is strongly dependent from cement an polymer amounts, open time after
20 and, even more, after 30 minutes, are strongly affected by other factors: cellulose ether amount
becomes fundamental to guarantee the overall performances: thanks to its property of incorporating water
into its molecular structure, it provides a good wettability and prevents drying and the formation of the
so called skin that generates detachments. On the other hand, filler content, cement content and cellulose
ether modification can significantly compromise the overall performances of tile adhesive open time.

Figure 3: RSC graphs of open time test at 5’ and 30’.

c. Dustiness evaluation
The emission of dust from cementitious tile adhesives can be described by two different phenomena: a
first release of dust immediately consequent to the act of poring powder in a bucket and the remaining
dust at the end of the precipitation process. These two mechanisms are described by the dustiness model
generated in present work.
The first mechanism seems to involve all the raw materials investigated: cement, calcium carbonates
(fine and coarse), polymers and fibers all contribute to the generation of powder, as shown in Figure 4.
Figure 4: RSC graphs of dustiness of tile adhesives (initial and permanent dust generation).

The second mechanism mostly involves light components of the formulation, like fibers and polymer
that, after 30 second from powder pouring are still fluctuating, generating the permanent cloud of dust,
as shown in Figure 4.

Conclusions

Chemometrics demonstrated to have the possibility to strongly improve the knowledge generated by
experimentations, as the definitions of rational Experimental Designs and the multivariate analysis
approach can give a wide-ranging amount of information that is usually lost when evaluating tests in
single points. This discipline, applied to cement based product science, also demonstrated its extreme
flexibility and adaptability to any kind of experimentation and its perfect applicability in formulation
science, clearly surpassing the classical univariate approach.
Many further studies can be carried out using these techniques; an example could be the realization of a
model able to foresee the coloring of the tile joints and therefore able to predict the right dosage of the
inorganic pigments.

1 Monaco M.; Carrà S. (2015) Statistical evaluation of pull-off adherence tests according to ISO 13007.
2 Workman J. Jr. (2002) Chemom. Intell. Lab. Syst., 60, 13-23.
3 Box G.E.; Hunter J.S.; Hunter W.G. (2005). Statistics for Experimenters: Design, Innovation, and

Discovery. 2nd edition. Wiley.

4 Oehlert G.W. (2010). A First Course in Design and Analysis of Experiment. Minneapolis, MN.

University of Minnesota.
5 Pearson K. (1901). On Lines and Planes of Closest Fit to Systems of Points in Space. Philosophical

Magazine. 2 (11): 559-572.

6 Joliffe I.T. (2010) Principal Component Analysis. 2nd edition. Springer.
7 Gruppo Italiano di Chemiometria, www.gruppochemiometria.it
8 Monaco M. (2018) C2 Tile Adhesive General Model v1.17. Mapei S.p.A.