Professional Documents
Culture Documents
Measurement Quality and Uncertainty Evaluation
Measurement Quality and Uncertainty Evaluation
4, 127–134 (2013)
c EDP Sciences 2013
DOI: 10.1051/ijmqe/2013043
Abstract. Civil engineering is a branch of science that covers a broad range of areas where experimental
procedures often plays an important role. The research in this field is usually supported by experimental
structures able to test physical and mathematical models and to provide measurement results with accept-
able accuracy. To assure measurement quality, a metrology probabilistic approach can provide valuable
mathematical and computational tools especially suited to the study, evaluation and improvement of mea-
surement processes in its different components (modeling, instrumentation performance, data processing,
data validation and traceability), emphasizing measurement uncertainty evaluation as a tool to the analysis
of results and to promote the quality and capacity associated with decision-making. This paper presents
some of the research held by the metrology division of the Portuguese civil engineering research institutes,
focused on the contribution of measurement uncertainty studies to a variety of frameworks, such as testing
for metrological characterization and physical and mathematical modeling. Experimental data will be used
to illustrate practical cases.
testing specimens with assured quality in order to support measurement uncertainties. This will help users to select
research studies. In this topic, the contribution is given by sensors and to establish the required accuracy for the mea-
the characterization of geometrical quantities of metal and surement systems.
steel moulds [3, 4] used to manufacture these specimens.
600 500
400
300 P2
300
200
200 P3 P1
100 P2
100
c' P1
0 a
0
0 200 400 600 800 1000 1200
0 100 200 300 400 500 600 700 800 900 1000
Effective normal stress, (kPa)
Effective normal stress, p ' (kPa)
Fig. 1. Evaluation of the resistance parameters using Mohr
circles at failure. Fig. 2. Straight-line regression for the evaluation of the friction
angle φ and the cohesion c .
Table 1. Results of tests for different consolidation stresses.
Table 2. Results from MCM.
Specimen P1 P2 P3
Consolidation stress/kPa 100 200 300 Specimen q /kPa uq / kPa p /kPa up /kPa
q at failure/kPa 132 252 355 1 132 1.63 228 0.47
p at failure/kPa 228 452 655 2 252 1.63 452 0.47
3 355 1.62 655 0.47
500
Table 3. Results of straight-line regression for specimens 1
to 3.
400 y = 0,5091x + 19,0377 α
φ /◦ uφ /◦ c /kPa uc /kPa 2
R = 1,0000
P3
500 200
P1
y = 0,5235x + 14,2101
400 2 α 100
R = 0,9997 P3
Shear stress, q ' ( kPa)
a
300 P2 0
0 100 200 300 400 500 600 700 800 900 1000
a
0
0 100 200 300 400 500 600 700 800 900 1000 implicit scheme is implemented over φ = arcsin(b) and
Effective normal stress, p ' (kPa)
c = a/cos(φ ) yielding two Jacobian matrices with sensi-
tivity coefficients with respect to (c , φ ) and (a, b):
Fig. 3. Straight-line regression with uncertainties evaluated
using MCM.
cos(φ) −c sin(φ)
Jy =
0 cos(φ)
Table 4. Results of straight-line regression for specimens 1 to
5 with experimental dispersion. −1 0
Jx =
0 −1
φ /◦ uφ /◦ c /kPa uc /kPa
Validation accepted 30.6 0.36 22.12 2.55 which in conjunction with the covariance matrix for (a,b)
2
u (a) u(a, b)
uncertainties, produced the values of a, b, φ (friction an- Vx =
u(a, b) u2 (b)
gle) and c (cohesion) shown in Table 3 and in Figure 3.
The validation referred to in the previous table is es- permits the propagation of uncertainty by solving Jy ·
tablished by comparing the observed R value with the V y · Jy T = Jx · V x · JxT and extracting the standard
chi-squared distribution with v = m − 2 degrees of free- uncertainty and covariance.
dom and probability 95%. As the observed R value was
outside the corresponding interval, a value σ̂ 2 = R/(m−2) u(c) = [Vy (1, 1)]
1/2
of strain gauges connected to bridges to measure the dis- The analysis of the experimental results shown in Figure 6
placement variations observed at the specimen faces. may point out some conclusions:
The quality of the measurement is strongly related i. The nonlinearity found avoids the need to consider a
with the fact that both specimens and strain gauges have single estimator of the thermal expansion coefficient
similar coefficients of thermal expansion, being these usu- to be given to the specimens.
ally known with some accuracy for the strain gauges but ii. Heterogeneity of material gives similar behaviour but
not well known for the specimens worsened by the fact different estimates for each temperature interval. It
that the pavement material can be quite heterogeneous. explains the memory effect observed on the decreas-
In order to obtain estimates of the specimens coeffi- ing cycle relative to the high temperature result and
cient of thermal expansion (αs ), LNEC metrology division also the gap between the first and last points, both
provided a setup to estimate that parameter for a set of obtained at the same reference temperature.
pavement specimens. The setup had a 300 mm vertical iii. A reversibility effect was detected, with the displace-
column with horizontal arm and a granite base plate, a ments higher during the increase of temperature (top
displacement transducer (50 mm range and 0.1 μm of res- curves) than during the decrease of temperature.
olution) with a digital device indicator and a chamber able iv. The analysis of temperature cycles appear to indicate
to control temperature (stability and uniformity) within that a drift exists, with the shift calculated as the dif-
0.1 ◦ C and humidity within 0.5%rh. ference between the final and the initial displacement
The procedure developed was two stage, being the first of each cycle.
one the evaluation of the setup column coefficient of ther-
mal expansion (αc ), and the second one, the evaluation of Regarding the measurement uncertainty, estimates were
the specimens coefficient of thermal expansion (αs ). obtained again using first order GUM approach applied
For the first step, three 100 mm pure metallic bars to equation (9), using as estimate the average of the setup
(invar, steel and aluminium) were used as reference mate- column thermal coefficient given on Table 6 (u (αc ) =
rials with known αs and the testing was carried doing the 9×10−7 ◦ C−1 ). The result obtained, expressed in Figure 6,
increment of temperature from 5 ◦ C to 30 ◦ C with 5 ◦ C was of (u (αs ) = 2.0 × 10−6 ◦ C−1 ).
steps. Taking into consideration the studies carried out, it
Considering a change in temperature of Δθ, the dis- was possible to conclude that the use of extensometers to
placement d measured by the transducer, for each temper- measure displacements in specimens used on road pave-
ature step and reference material, is approx. given by (7) ment testing, was within the measurement range of ex-
and the estimates of the setup column αc by (8). tensometers and their own thermal expansion coefficient.
However, an accurate measurement of temperature during
testing is strongly advisable.
di = (ls αs Δθi ) − (lc αc Δθi ) (7)
In this process, the thermal coefficient of the glue (for
[(ls αs Δθi ) − di ] several compositions) used to stick the extensometers to
αc = . (8)
(lc Δθi ) the specimens was also evaluated previously at labora-
tory, being obtained estimates between 1 × 10−5 ◦ C−1
Calibration results are given in Table 5, being the mea- and 2 × 10−5 ◦ C−1 . Further studies are intended to be
surement uncertainty obtained using first order GUM made in order to evaluate if the overall accuracy obtained
approach. The estimate obtained is according with the is acceptable to measure the phenomena to be studied.
132 International Journal of Metrology and Quality Engineering
0.0000150
0.0000100
0.0000050
Δ l l mm
0.0000000
8 10 12 14 16 18 20 22 24 26 28 30
-0.0000050
-0.0000100
-0.0000150
Temperature / ºC
Table 6. Experimental results obtained for a triple mold tested Table 7. MCM results for ∠AB orthogonality deviation.
at LNEC’s metrology laboratory.
mm
Measure Nominal Min. Max. d −9, 1 × 10−3
length and experimental experimental U95% (d) 0.6 × 10−3
tolerance value/mm value/mm Skewness (d) <3 × 10−2
AC (40.0 ± 0.2) mm 39.956 39.973 Kurtosis (d) <2 × 10−2
BD (160 ± 1) mm 160.051 160.011
ET (40.1 ± 0.1) mm 40.167 40.142
Measure Nominal Experimental Deviation/
angle and meas. mm
tolerance angle/◦
∠ AB 90 ◦ / 0.2 mm 89.987 –0.009
∠ BC 90 ◦ / 0.2 mm 89.986 −0.010
∠ CD 90 ◦ / 0.2 mm 90.001 0.001
∠ DA 90 ◦ / 0.2 mm 90.002 0.015
∠ AE 90 ◦ / 0.2 mm 89.994 −0.004
∠ BE 90 ◦ / 0.2 mm 90.022 0.015
∠ CE 90 ◦ / 0.2 mm 89.992 −0.006
∠ DE 90 ◦ / 0.2 mm 89.915 −0.059
/ / CA 0 ◦ / 0.2 mm 0.016 0.011
/ / BD 0 ◦ / 0.2 mm 0.039 0.027 Fig. 9. PDF for the ∠AB orthogonality deviation.
/ / ET 0 ◦ / 0.2 mm 0.025 0.017
Measure Nominal length Deviation/mm
(flatness) and tolerance
Plan A ±0.03 mm 0.009 calibration uncertainty (U95% = 30 ) with the measure-
Plan B ±0.03 mm 0.006 ment repeatability. The results obtained are found in
Plan C ±0.03 mm 0.008 Table 7.
Plan D ±0.03 mm 0.004
Plan E ±0.03 mm 0.010 The values obtained, combining the estimate and the
uncertainty interval have an order of magnitude less than
1/10th of the tolerance, allowing to assure that the con-
In the first step of the procedure the reference CMM formity assessment can be properly done by this way.
provides an angle, θ, between every pair of planes, P1 and Based on the output deviation probability density
P2 , constructed based on two sets of points, followed by function (Fig. 9) obtained using MCM it was possible to
the numerical evaluation of their normal vectors, n1 and evaluate if the nature of the mathematical model, includ-
n2 , being: ing a trigonometric function, would introduce any asym-
|n1 · n2 | metry or change of the Gaussian shape. The evaluation of
cos θ = . (10) skewness and kurtosis confirm the expected normality of
n1 n2
the output distribution.
The second step of the procedure is to use the estimate
of each measured angle and of the length related to the
maximum angular deviation, l, to evaluate the deviation
related to orthogonality (d⊥ ) and to parallelism (d ), re- 5 Conclusions
spectively given by:
π Metrology probabilistic approach has mathematical and
d⊥ = l tan −θ (11a) computational tools particularly suited to improve the
2
d// = l tan (θ) . (11b) quality of measurement thus fulfilling the growing tech-
nological demands of the modern society.
Considering the nonlinearity of the mathematical model, In the context of civil engineering, where diverse frame-
the use of Monte Carlo Method (MCM) was found suit- works can be found, measurement is often a major support
able for the purpose of uncertainty evaluation. Simula- for the analysis and decision-making, whether it concerns
tions were performed for each of the experimental val- simply the relative quality of products or processes up to
ues presented in Table 6, using MatLab validated routines extreme cases where integrity, safety and risk assessment
being performed numerical simulations with sequences of determines the establishment of special requirements for
106 draws. conformity assessment.
In the case of the measurement of the angle ∠ AB, Thus, the metrological analysis and, in particular, the
the model applied was (11a), being the input variables in- evaluation of measurement uncertainty, can have a role in
troduced considering the length of (40 ± 1) mm, and a the process of Quality assurance, which should be one of
measured average angle of (89.987 ± 0.017)◦, being the the aims of the activity related to Research, Development
angle uncertainty obtained combining the CMM angular and Innovation.
134 International Journal of Metrology and Quality Engineering