Construction and Building Materials 31 (2012) 86–93

Contents lists available at SciVerse ScienceDirect

Construction and Building Materials

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

Reliability of the pull-off test for in situ evaluation of adhesion strength

N.M.M. Ramos ⇑, M.L. Simões, J.M.P.Q. Delgado, V.P. de Freitas
LFC  Laboratório de Física das Construções, Departamento de Engenharia Civil, Faculdade de Engenharia, Universidade do Porto, Rua Dr. Roberto Frias, s/n, 4200-465 Porto, Portugal

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

Article history: The pull-off test is a very popular way of evaluating the adhesion strength of renderings or ceramic tiles.
Received 7 October 2011 Several standards frame the application of this test and interpretation of its results. In our laboratory, the
Received in revised form 9 December 2011 in situ pull-off test is frequently applied to support the diagnosis of causes for building anomalies. The
Accepted 23 December 2011
statistical meaning of the results, however, is clearly different from the one found in laboratory tests
Available online 21 January 2012
and hence reliability issues arise.
In this article, 55 case studies based on in situ tests are analysed. The sample was retrieved from 15 test
Keywords:
reports conducted by LFC-FEUP. In each of these reports different situations are included in terms of wall
Pull-off test
Adhesion strength
components, expected hygrothermal loads and ageing history. This large number of tests supports a dis-
In situ tests cussion on the applicability of the pull-off test regarding its reliability as a tool for two different subjects:
Reliability decisions on repair strategies and building forensics.
Typical variability found in field tests is described and analysed. The reliability of the test is analysed in
detail with the help of one specific case where 40 individual pull-offs were conducted in the same
building.
Ó 2011 Elsevier Ltd. All rights reserved.

1. Introduction and so as the converse of the term probability of failure [1]. But
in the case of ceramic tiles detachment from a façade, for instance,
The adhesive bonding of exterior revetments applied on façades the situation is more complicated. The behaviour of the system it-
is an important factor to ensure the safety and durability of the self is highly uncertain due to several factors that make it hetero-
building. The failure of adhesive bonding has immediate conse- geneous by nature, ranging from substrate to workmanship. If a
quences and therefore is a common concern for the building indus- façade is under study, it is impossible to define its safety level, un-
try and building owners. This paper focusses on the behaviour of der the same standards that are usually applied in structural
renderings and ceramic tiles applied on exterior surfaces where behaviour, using a semi-destructive technique such as the pull-
substrates are typically concrete or brick walls. This type of enve- off test. However, variability associated with the results of the
lope frequently develops pathologies that can have different in situ pull-off test should be addressed. One way of doing so is
causes. The pull-off test is frequently used as a tool to help on to investigate statistical parameters such as mean and variance
the correct diagnosis of causes for degradation of the façades exte- of available results and apply interval estimation techniques to
rior layers. The authors have applied its results in several cases and those parameters.
therefore use that hands-on knowledge as a starting point to ad- In this article, 55 case studies are analysed. The sample was re-
dress the subject of the test reliability. trieved from 15 reports of in situ inspections conducted by LFC-
The pull-off test results have been used by the authors in other FEUP. This large number of tests supports a discussion on the
contexts with two main purposes. The first one is to support the reliability of the pull-off test for in situ inspections where variability
decision on how to repair building anomalies and the second one of results is an important part of the decision process. By synthesiz-
is defining who is responsible for the defects. In both cases, reli- ing the information that resulted from the large number of pull-off
ability is a key aspect. test results available at LFC behaviour patterns can be detected and
Reliability of an engineering system can be defined with differ- a statistical analysis can provide benchmarks for similar cases.
ent terms. A common approach in structural engineering is to
define reliability as the probability of successful performance,
2. Literature review

⇑ Corresponding author. Fax: +351 225081440. 2.1. Variability assessment

E-mail addresses: [email protected] (N.M.M. Ramos), [email protected]
(M.L. Simões), [email protected] (J.M.P.Q. Delgado), [email protected] Ideally, when performing pull-off tests on a façade, the proba-
(V.P. de Freitas). bility distribution of adhesive strength should be determined. This

0950-0618/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.conbuildmat.2011.12.097
 N.M.M. Ramos et al. / Construction and Building Materials 31 (2012) 86–93 87

would allow for a supported risk assessment based on the estima- the candidate areas to test for searching the lower adhesion values.
tion of the probability of failure. However, performing a large But the number of causes can actually be very wide as demon-
number of pull-offs would be economically unfeasible and could strated by literature.
ultimately lead to the necessity of replacing the façade revetment. The causes of external wall tile failures have been widely re-
If the type of probability distribution was known, for instance nor- viewed by Ho et al. [6]. The author considered that the major causes
mal or lognormal, than knowing mean and variance would be en- are: the environmental effects (weathering, attacks and vibration
ough to define the distribution [1]. Mean and variance, however, and loading), the design and material factors (improper design
cannot be known precisely as it would imply testing the whole faç- and selection of materials specifications, joints and grouting and de-
ade surface. Nevertheless, the samples where the test is performed sign and maintenance), the construction factors (improper surface
supply an estimation of the parameters. The quality of that estima- preparation such as inadequate cleaning, no provision of proper
tion can be defined by the associated confidence intervals for the keys, improper sequence of work, application/workmanship and
mean. The interval estimation method can also be of interest in an- adhesives/rendering) and the structural and substrate factors (sub-
other perspective. strate defects, substrate movement and substrate properties).
As in our case the population variance of adhesion strength is Other important common causes are: the differential move-
not known in advance, it needs to be estimated from collected ments between the tile, adhesive and the immediate substrates
samples of size n. Thus, for a sample with mean  x and standard due to thermal, moisture or other effects; the failure of the cement
deviation s, the (1  a) confidence interval for the mean can be de- rendering behind the adhesive; the structural movements, shrink-
fined by expression (1), as the population has a Gaussian type dis- age and creep, vibrations and settlement problems and deforma-
tribution, where ta=2;n1 are the values of Student’s T distribution tion of adhesive (or mortar) onto which the tiles have been laid
with (n  1) degrees of freedom evaluated at probabilities of due to shrinkage [7,8].
(1  a/2) and a/2, respectively [1]: Related to the environmental factors, some authors showed the
influence of pollutant weather, temperature, relative humidity, wind,
s s
 
x  t a=2;n1 pffiffiffi ; x þ ta=2;n1 pffiffiffi ð1Þ sunshine and air pressure, in the performance of tiles [9–12]. For exam-
n n ple, Guan et al. [13] demonstrated experimentally that high tempera-
The confidence interval for the variance can be estimated with tures (an increase of 10 °C) accelerate the thermal degradation of
expression (2), where c1a=2;n1 and ca=2;n1 are the values of polymeric materials in the adhesives. Briffett [9] and Davies [14]
Chi-square distribution with (n  1) degrees of freedom evaluated stressed the consequence of moisture penetration behind the tile sur-
at probabilities of (1  a/2) and a/2, respectively: face through cracks. Experimental laboratory test carried out by Chew
[7] showed the influence of thermal induced and moisture-induced
ðn  1Þs2 ðn  1Þs2 movements on adhesive strength. The author obtained an approxi-
 
; ð2Þ
c1a=2;n1 ca=2;n1 mately 20% reduction in pull-off strength after one 24-h thermal cycle
in adhesives with polymer as additives, and a 6–37% reduction in bond
In this type of analysis, the a value is usually taken as 0.05, strength after wet and dry treatment.
defining a 95% confidence interval for the estimated mean and var- The construction factors, such as workmanship, the tile setting
iance. The estimation of the confidence interval may be used to de- pressure, the adhesive inner cavity and the adhesive open time
fine how close a sample of size n is to the population it intends to were widely studied by Zhao and Zhang [15]. The authors showed
represent. If the confidence interval is small, the sample mean is a that the pull-off strength decreases significantly when the setting
good predictor of the population mean. pressure is below 50% standard setting pressure; decreases propor-
The variability of pull-off test results for in situ conditions was tionally with the increase of the adhesive inner cavity ratio; and
addressed by Bungey and Soutsos [2] in a wider study on the reli- decreases significantly when open time exceeds by 50% the speci-
ability of partially destructive tests on concrete on site. The conclu- fied open time limit.
sions of that study pointed to the interest of using pull-off test on Related to the influence of the exposure temperature on the
site since it is quick and straightforward to use and its results external wall tile failures, Chew [7] demonstrated experimentally
showed good correlation with cube compressive strength, but the that the effect of temperature during application on the develop-
lack of statistical confidence due to few in situ available tests ment of bond strength is significant. Application at 40 °C and
was defined as a problem. 10 °C was found to have reduced the bond strength from 10% to
A study by Flores-Colen et al. [3,4] found, in a set of in situ tests, 47% compared to an identical application at 25 °C.
variation coefficients from 32% to 104%. Several causes for the var- Finally, Maranhão et al. [16] showed that moisture content
iability found included: using manual equipment with difficult above 6% is sufficient to reduce 50% of mortar deformability; the
control on load rate application, the aggressive conditions during drying process allows the mortar to recover flexibility to a value
test procedure and the difficulty to guarantee equal depth of the similar to that prior to saturation; a logarithmic function gives
surface cores, equal glue thickness or nil load eccentricity in all the best fit for correlations between moisture content and flexibil-
cases. A correlation between causes and variability could not be ity and the water saturation increases mortar rigidity, as character-
precisely determined. ised by Young’s Modulus.
The study of pull-off test intrinsic aspects influence on variability
of results was addressed by Costa and Carasek [5]. Those aspects 2.3. Pull-off test standardization
included not only parameters already mentioned above but also
geometry and dimension of samples. It was concluded that the ten- The adhesive strength is determined as the maximum tensile
sile stress concentration that will develop in the borders of square strength applied by a direct load perpendicular to the surface being
samples by comparison with a more uniform distribution in circular tested. The pull-off test is classified as a near-to-surface, partially
samples results in higher adhesion values for the latter ones. destructive method which is able to measure tensile strength of
various materials. The tensile load is applied by means of a defined
2.2. Adhesion failure pull-head plate glued to the test area. The adhesive strength is the
ratio between the failure load and the test area. On all the tests
A good knowledge of the causes for adhesion failure is impor- described in this paper, the test area was pre-cut in situ, fitting
tant to define the most sensitive areas of a façade and hence reduce the size of the pull-head plates. The pre-cut was usually performed
88 N.M.M. Ramos et al. / Construction and Building Materials 31 (2012) 86–93

5 mm deep inside the substrate or 5 mm beyond the interface un- esting approach is introduced as expected coefficients of variation are
der analysis. These could be square metallic plates of 50 mm  presented as reference for the application of the tests. The results were
50 mm with a thickness of 15 mm or circular metallic plates with based in round-robin tests and pointed to intra-laboratory coefficients
diameter of 50 mm and 18 mm thick. In each case study, only one of 12% and 20% for inter-laboratory comparisons.
type of plate was used which means that results were not affected
by plate geometry variation inside each case study. 3. Experimental study
The test machine for direct pull tensile force test is in accor-
dance with standard requirements, with the additional feature of This experimental study is based on data retrieved from inspec-
automatic control of the applied force provided by an attached tions of buildings on operation phase. This implies that several fac-
electrical engine, as presented in Fig. 1. tors conditioned the behaviour of the tested components such as
Several standards are available to frame the evaluation of adhe- the initial quality of the materials, their actual application condi-
sive strength by means of the pull-off test. The evaluation of tensile tions, substrate, façade exposure, age of the buildings and mainte-
adhesion strength for cementitious adhesives, applied with cera- nance operations. The results of these tests cannot be directly
mic tiles, can be found in EN 1348 [17]. The application of the compared since the conditions that lead to each set of results are
pull-off test in cement based renderings is described in EN 1015- unique.
12 [18]. Both standards refer to laboratory tests and, therefore, The available data, however, presented an interesting opportu-
do not focus on two subjects that are important for a successful nity to develop a statistical portrait of the adhesive strength that
implementation of the procedure in situ. can actually be measured in revetments of building façades. The fo-
The first subject that can be decisive for the implementation of cus of the study is not only on the average values found but, more
the tests in situ is to perform a low disturbance pre-cut of each importantly, the spread of those values and the factors that influ-
specimen. Usually, square specimens are preferred, as they are eas- ence it. The sample originated by the available data is random as
ier to produce. Another aspect is where to take the pre-cut. If it is it resulted from the study of buildings whose owners asked for
performed deep in a render that is acting as substrate for the adhe- inspections due to façade anomalies.
sive mortar, the failure will probably occur in the substrate, result- The results from 15 different in situ inspections were analysed.
ing in a false low result. It can however be interpreted as a In each inspection, different zones in the building were tested,
performance indicator of the substrate and not of the adhesive. characterised by different components and/or different exposure.
The pre-cut on the two different systems application of the pull- The data was therefore grouped by zones of identical conditions,
off test is presented in Fig. 2. In Fig. 2b a template is used to im- allowing isolating more accurately the variables that contributed
prove the quality of the pre-cut by reducing an eventual eccentric- to the measured adhesion strength. The selected studies focused
ity of the core drill. on ceramic tiles applied with cement based adhesives and cement
The second aspect, crucial to the reliability of the test, is the based renderings as they were the ones with more data available.
amount of specimens needed and the admissible variability of the Hence, the studied sample includes 232 measurements grouped
results. The EN 1015-12 [18] defines a minimum of five specimens in 55 case studies. All the invalid measurements were discarded
for the test. The EN 1348 [17] also imposes a minimum of five spec- from the statistical analysis. Table 1 presents the relation between
imens but the results must fall inside the range of ±20% from the the different tests and the original inspection report.
mean value. The MDT.D.3 [19] defines six as the minimum number The grouping of case studies attended to the material that was
of test specimens to perform in situ tests on rendering but states that actually tested. When testing in situ ceramic tiles adhesion the
the final number should be higher if considerable heterogeneity is pre-cut is taken inside the mortar substrate. If the failure mode
found. is cohesive inside the mortar instead of the adhesive the strength
The standard D 4541-02 [20] is focused on pull-off test application value must be associated with the mortar. Regarding substrates,
to coatings. Although focused on paints and varnishes testing, an inter- concrete (7% of the cases) and brick masonry (75% of the cases)
are typical solutions for buildings of the last decades while stone
masonry (16% of the cases) is common for older buildings. The only
example of a non traditional solution included in this study was
the application of ceramic tiles on wood–cement panels. Regarding
substrate, there is a clear predominance in these tests of brick ma-
sonry, which corresponds to the Portuguese reality.
The inspection reports 16 and 17 correspond to laboratory tests
that were added to this study for reference comparison of variabil-
ity. Report 16 refers to a straightforward test on rendering samples.
Report 17, however, is a bit more complex test of a cement based
adhesive. Case 17A is the reference test without ageing while cases
17B–17F are pull-offs after ageing using different processes.

4. Results and discussion

An overview of the results retrieved from the 55 case studies is

presented in Figs. 3 and 4, divided between rendering and cement
based adhesives. Each case study is represented by the mean value
and standard deviation. The graphs also include a reference value
for each material, namely 0.3 MPa for renderings (NF P 15-201
[21]) and 0.5 MPa for cement based adhesives (EN 12004 [22]).
It can be observed that the mean values for adhesive strength of
renderings were typically low and frequently below the recom-
Fig. 1. Pull-off test machine. mended value of 0.3 MPa. The tested cement based adhesives
 N.M.M. Ramos et al. / Construction and Building Materials 31 (2012) 86–93 89

Fig. 2. Pre-cut of (a)-ceramic tiles and (b)-rendering.

Table 1
Inspection reports and connected case studies.

Inspection report Case studies Measurements Material tested

A B C D E F G H
1 5 4 4 13 Rendering
2 2 3 3 3 11 Rendering
3 5 3 8 Rendering
4 4 4 5 13 Rendering
5 2 5 2 7 16 Rendering
6 2 3 3 8 Rendering
7 3 3 3 3 12 Rendering
8 3 3 3 6 3 3 21 Rendering
9 5 5 4 14 Rendering
10 10 10 20 Cement based adhesive
11 3 3 6 Cement based adhesive
12 5 5 10 Cement based adhesive
13 4 5 5 5 5 5 29 Cement based adhesive
14 3 6 2 11 Cement based adhesive
15 5 5 5 5 5 5 5 5 40 Cement based adhesive
16a 4 4 Rendering
17a 3 3 3 3 3 2 17 Cement based adhesive
a
The inspection reports refer to laboratory tests.

2.0
Average values and standard deviation of adhesive strength
1.8
Adhesive strength (MPa)

1.6
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
01A
01B
01C
02A
02B
02C
02D
03A
03B
04A
04B
04C
05A
05B
05C
05D
06A
06B
06C
07A
07B
07C
07D
08A
08B
08C
08D
08E
08F
09A
09B
09C

Case studies

Fig. 3. Average values and standard deviation of adhesive strength for rendering case studies.

exhibited higher strength and were frequently above the reference A deeper analysis of the adhesive strength found for renderings
value of 0.5 MPa and even above 1.0 MPa. On both types of tested and cement based adhesives is presented in Figs. 5 and 6. The dis-
materials, significant variations of the standard deviations could be tributions found for each tested material were significantly differ-
found. The low values observed, especially for renderings, could be ent. While renderings presented a log-type distribution with
related to the fact that these inspections were done in façades predominant results below 0.3 MPa, for cement based adhesives
where anomalies could be observed. no theoretical model could be easily assigned. The mean was found
90 N.M.M. Ramos et al. / Construction and Building Materials 31 (2012) 86–93

2.0
Average values and standard deviation of adhesive strength
1.8

Adhesive strength (MPa)

1.6
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
10A
10B
11A
11B
12A
12B
13A
13B
13C
13D
13E
13F
14A
14B
14C
15A
15B
15C
15D
15E
15F
15G
15H
Case studies

Fig. 4. Average values and standard deviation of adhesive strength for cement based adhesives case studies.

2.0

Adhesive strength (MPa)

80
Rendering
1.5
60
Frequency

1.0

40
0.5

20 0.0
Cement based adhesive Rendering
Tested materials
0
0.0 0.5 1.0 1.5 2.0
Fig. 6. Adhesive strength box-plots for the tested materials.
Adhesive strength (MPa)

80
2.0
Cement based adhesives Average values and Average values and standard
1.8 standard deviation deviation for cement based
60 adhesives
1.6 for rendering
Adhesive strength (MPa)
Frequency

1.4
40
1.2
1.0
20
0.8
0.6
0
0.0 0.5 1.0 1.5 2.0 0.4
Adhesive strength (MPa) 0.2
0.0
Fig. 5. Adhesion strength distribution for the tested materials.
16A

17A

17B

17C

17D

17E

Case studies

above the reference value of 0.5 MPa. The box-plots for each mate- Fig. 7. Average values and standard deviation for laboratory tests.
rial confirm this observation since, for cement based adhesives, no
outliers were defined while for renderings, several values were de-
fined as outliers, all corresponding to high resistance values. standards, even for the test where ageing processes were applied
The results from the laboratory experiments are presented in to samples.
Fig. 7. They show not only higher average values but also, and more Looking at the variability observed in situ, the conclusion is to-
important, much lower standard deviations. This is important since tally different. In the tests on rendering, cases from reports 1 to 9,
the tests were performed by the same technicians and using the the average CV was 67% ranging to a maximum value of 173% in
same equipment as the tests performed in situ. case study 8F. In the tests on cement based adhesives, cases from
A careful analysis of the data revealed that the laboratory test reports 10 to 15, the average CV was 57% ranging to a maximum
on rendering, case 16A, resulted in a coefficient of variation (CV) value of 99% in case study 15E. These values are quite revealing
of 11.7%. The highest CV in the laboratory tests on cement based of the huge difference between laboratory and in situ conditions.
adhesives, cases 17A–17E, was 10.9% in case 17A with a maximum The analysis of the number of measurements influence on the
difference to mean of 8.7%, clearly below the 20% limit defined in variability of results is presented in Fig. 8, for the available results.
EN 1348. This means that the tests conducted in laboratory, pre- The number of cases with 6 and 10 pull-offs was small, with only
sented as example, had low variability, within limits defined by two for each, and so they are not comparable with the results with
 N.M.M. Ramos et al. / Construction and Building Materials 31 (2012) 86–93 91

100%
Average values and standard deviation of the coefficient of variation

80%
Coefficient of Variation

60%

40%

20%

0%
2 3 4 5 6 10
Number of measurements

Fig. 8. Coefficient of variation correlation with the number of measurements per

case study.

the remaining cases. Looking at the results for cases with 2, 3, 4

and 5 valid pull-offs, the standard deviation of the coefficients of
variation is smaller for 5 measurements. The correlation between
the coefficient of variation and the number of valid measurements
seems to indicate that the greater the number of valid pull-offs the
greater the probability of not missing the actual variability of adhe-
sive strength under evaluation. The results indicate that the in situ
coefficient of variation for a specific set of pull-offs can range from
35% to 80%. The eventual lack of trust in a method that results in
such high variability is controlled by standards as minimum values
are imposed for single valid tests and not only to the mean value.
The problem of in situ tests is that often the adequate number of Fig. 9. Pull-off tests in the East facing façade.
valid tests is not achieved due to the obvious difficulties in repeat-
ing the tests.

5. Improving reliability

To propose strategies for improving the reliability of the pull-off

tests the discussion will now focus on report 15 since it has the
largest number of tests on the same building. The corresponding
study was requested by the building owner, a public authority,
with the objective of investigating on the need to totally replace
the existing ceramic tiles revetment. The East facing façade is pre-
sented in Fig. 9 and West facing façade is presented in Fig. 10. The
pull-offs performed in both façades, A–G, are defined in Figs. 9 and
10. Pull-off H was performed in the North facing façade, at ground
level.
This case study presented a good opportunity for deepening the
variability analysis. The 40 pull-offs performed in the façades cor-
respond to a sample large enough to explore the application of sta-
tistical tools. The sample distribution of these data is illustrated by
the boxplot in Fig. 11. Applying the Kolmogorov–Smirnov test to
evaluate the goodness of fit of Gaussian distribution to the data,
we obtained a p-value equal to 0.57 > 0.05. Thus we can conclude
that the adhesive strength, in this case, has a Gaussian type distri-
bution with mean 0.68 MPa and standard deviation 0.49 MPa. The
results from the tests were highly scattered as it can be seen in
Fig. 4. The minimum individual value found was 0.04 MPa, in case
study A, and the maximum individual value found was 1.86 MPa,
in case study C. The coefficients of variation ranged from 33% in
case study D to 103% in area E.
If one was to decide on whether to keep the existing tiles or
replacing them, clearly the second option should prevail. Although
the mean is above 0.5 MPa, an important part of the tiles may pres-
ent low adhesion, as it is clear from Fig. 11. Fig. 10. Pull-off tests in the West facing façade.
92 N.M.M. Ramos et al. / Construction and Building Materials 31 (2012) 86–93

2.0
studies, an additional feature was added, as the confidence inter-
vals were estimated using 3, 4 or the total 5 of the pull-offs per-
formed in each case study. It was assumed that, when the
Adhesive strength (MPa)

number of pull-offs was 3 or 4, the extreme values of each case

1.5
study were included. As a result of this exercise, Figs. 12 and 13
were obtained, allowing for some conclusions:
1.0
– The confidence interval for the mean and the upper confidence
limit for the variance showed a strong decrease especially when
stepping from 3 to 4 pull-offs.
0.5
– Both the variance upper limit and the size of the confidence
interval for the mean corresponded to high values, clearly indi-
cating that some results were not reliable enough to conclude
0.0
on the expected adhesion values.
– The interpretation of the results must always be careful. Case
Cement based adhesive
study A indicates that, that specific area has surely low adhe-
Fig. 11. Box-plot of the complete set of pull-offs in case study 15. sion values. Case study C indicates that the adhesion in that area
may be acceptable but one cannot be sure unless additional
pull-offs were performed.
3.5
6. Conclusions
95% confidence interval for mean of

n= 5 n= 4 n= 3
3.0
The analysis of case studies of in situ pull-off test applications to
2.5 adhesion strength evaluation retrieved from inspections to build-
adhesive strength

ings resulted in the following conclusions:

2.0
– Pull-off test is used frequently as a decision support tool for
1.5
building pathology studies.
1.0 – Adhesion strength can be measured for different systems and
components according to international standards. These stan-
0.5 dards are however focused on laboratory tests. Documents for
in situ measurements are scarce and not adapted to the entire
0.0 range of components applied on façades.
A B C D E F G H
Case studies – The adhesion strength distribution found for the overall tests on
renderings was logarithmic, presenting many values below rec-
Fig. 12. The 95% confidence interval for the mean of adhesive strength. ommended resistance while for cement based adhesives no the-
oretical distribution could be assigned and the mean value
corresponded to an acceptable resistance.
– The variability found in these tests was usually very high, corre-
sponding to variation coefficients ranging from 40% to 100%.
25
– There is no sufficient information to decide on the acceptable
Upper confidence limit for variance of

n= 5 n= 4 n= 3
coefficient of variation for the adherence observed in a specific
20 area. However, the studied sample revealed that 40% is a typical
low value and therefore acceptable in situ, providing at least
adhesive strength

five valid pull-offs are done. The definition of an upper limit

15
should be linked to each specific type of technical solution
and could be a parameter indicated by manufacturers.
10 – The application of confidence interval estimation for mean and
variance can be a useful tool to decide on the reliability of
the results. Criteria for the acceptable values of those estima-
5
tions could be useful for a practitioner to decide on whether
to trust the test results or not and should therefore be object
0 of standardization.
A B C D E F G H – A sample increase from 3 to 5 showed a strong decrease in the
Case studies size of confidence intervals highlighting the importance of the
number of tests to achieve reliable results for each tested area.
Fig. 13. The upper confidence limit for the variance of adhesive strength.
– A large number of pull-offs can lead to the definition of the
adhesion strength distribution in a whole building with a good
degree of confidence, leading to more robust decisions.

This case study also presents an interesting opportunity of ana- Although a large statistical sample was studied, the number of
lysing how the reliability of the results can be influenced by the cases correlated to certain variables were not sufficiently high to
number of samples. Hence, expressions (1) and (2), defined in Sec- support definitive conclusions. Nevertheless, certain patterns were
tion 2.1, were applied in the estimation of the confidence intervals identified and the variability found for adhesive strength in situ
for the mean and variance of the overall population of tiles adhe- tests inspires a cautious interpretation of results and a need for
sion strength and for each of the case studies, A–H. For the case standards adapted to this specific subject.
 N.M.M. Ramos et al. / Construction and Building Materials 31 (2012) 86–93 93

References [12] Yiu CY, Ho DCW, Lo SM. Weathering effects on external wall tiling systems.
Constr Build Mater 2007;21(3):594–600.
[13] Guan WL, Alum J, Liu ZJ, Yang T. Performance of external tiled-wall systems
[1] Haldar A, Mahadevan S. Probability reliability and statistical methods in
under tropical weathering. J Perform Constr Facil 1997;11(1):24–34.
engineering design. Wiley; 2000.
[14] Davies H. Repair methods for tile–clad buildings in Hong Kong. Struct Surv
[2] Bungey JH, Soutsos MN. Reliability of partially-destructive tests to assess the
1998;16(1):34–8.
strength of concrete on site. Constr Build Mater 2001;15(2–3):81–92.
[15] Zhao ZY, Zhang WL. Influence of workmanship on the bonding strength of tiles
[3] Flores-Colen I, Brito J, Freitas VP. Expected render performance assessment
to external walls. Int J Adhes Adhes 1997;17(1):47–53.
based on impact resistance in situ determination. Constr Build Mater
[16] Maranhão FL, Loh K, John VM. The influence of moisture on the deforma-
2009;23(9):2997–3004.
bility of cement–polymer adhesive mortar. Constr Build Mater 2011;25(6):
[4] Flores-Colen I, Brito J, Branco F. In situ adherence evaluation of coating
2948–54.
materials. Exp Tech 2009;23(3):51–60.
[17] CEN. EN 1348: Adhesives for tiles – determination of tensile adhesion strength
[5] Costa E, Carasek H. Recommendations for the execution of the pull-off test in
for cementitious adhesives. Brussels: European Committee for Standardi-
mortar renderings. VII Simpósio Brasileiro de Tecnologia das Argamassas,
zation; 2007.
Curitiba, Brasil; 19–22, 2009 [in Portuguese].
[18] CEN. EN 1015-12: Methods of test for mortar for masonry. Determination of
[6] Ho DCW, Lo SM, Yiu CY. A study on the causes of external finishes defects in
adhesive strength of hardened rendering and plastering mortars on substrates.
Hong Kong. Struct Surv 2005;23(5):386–402.
Brussels: European Committee for Standardization; 2000.
[7] Chew MYL. Adhesion of tiles for external cladding. Struct Surv 1999;17(1):12–7.
[19] RILEM. Recommendation MDT D. 3. Determination in situ of the adhesive
[8] Mahaboonpachai T, Matsumoto T, Inaba Y. Investigation of interfacial fracture
strength of rendering and plastering mortars to their substrate. Mater Struct
toughness between concrete and adhesive mortar in an external wall tile
2004;37:488–490.
structure. Int J Adhes Adhes 2010;30(1):1–9.
[20] ASTM. D4541-02. Standard test method for pull-off strength of coatings using
[9] Briffett C. The performance of external wall systems in tropical climates.
portable adhesion testers. West Conshohocken, PA: ASTM International; 2002.
Energy Build 1991;16(3–4):917–24.
[21] AFNOR. NFP 15 201-1 building works – plasterworks made from cement
[10] Simpson JW, Horrobin PJ. The weathering and performance of building
mortars, lime mortars, mixed gypsum and slaked lime mortars; 1993 [in French].
materials. 1st ed. United Kingdom: Wiley-Interscience; 1970.
[22] CEN. EN 12004. Adhesives for tiles – definitions and specifications. Brussels:
[11] Wallis K. Buildings crumbling at rate of three each day. Hong Kong: South
European Committee for Standardization; 2001.
China Morning Post; February 20, 1995.