The Journal of Strain Analysis for Engineering

Design
[Link]

Uncertainties Associated with Strain-Measuring Systems Using Resistance Strain Gauges

W Montero, R Farag, V Díaz, M Ramirez and B L Boada
The Journal of Strain Analysis for Engineering Design 2011 46: 1
DOI: 10.1243/03093247JSA661

The online version of this article can be found at:

[Link]

Published by:

[Link]

On behalf of:

Institution of Mechanical Engineers

Additional services and information for The Journal of Strain Analysis for Engineering Design can be found at:

Email Alerts: [Link]

Subscriptions: [Link]

Reprints: [Link]

Permissions: [Link]

Citations: [Link]

>> Version of Record - Jan 1, 2011

What is This?

Downloaded from [Link] at University of Birmingham on October 9, 2014

 1

Uncertainties associated with strain-measuring systems

using resistance strain gauges
W Montero, R Farag, V Dı́az, M Ramirez, and B L Boada
Mechanical Engineering Department, Carlos III University, Leganés, Madrid, Spain

The manuscript was received on 19 April 2010 and was accepted after revision for publication on 16 July 2010.

DOI: 10.1243/03093247JSA661

Abstract: Resistance strain gauges have been used for the measurement of strain for more
than 50 years; however, research to quantify the inherent uncertainty in a strain-measuring
system has been scarce hitherto. Nevertheless, resistive strain gauges are the most widely used
tool to measure strain owing to their simplicity, apparent accuracy, low cost, and ease of use. In
spite of this, at times they are used improperly, and the sources of error are neglected. Every
type of measurement has an uncertainty associated with it. As it is impossible to eliminate error
completely, the goal must be to quantify it and to reduce it to a value that is acceptable for the
purposes of the measurement being taken. The novelty of the present research is to put forward
a new methodology for determining the uncertainty in a strain gauge measuring system. To
achieve this, the principal sources of error that influence the measuring system are formulated
in order to develop an error model. Subsequently, the law of propagation of uncertainty is
applied, together with a type A and B evaluation approach to determine the combined
uncertainty of the entire measuring system, taking into consideration the correlation between
variables, when applicable. The new methodology is then applied to a series of strain
measurements taken on an aluminium flat bar subject to a bending load, and the results are
discussed.

Keywords: instrumentation, strain gauge, uncertainty

1 INTRODUCTION The expression of the final result of a measure-

ment being taken with a specific instrument de-
Strain gauges are one of the tools most often used in pends mainly on the purpose of the measurement:
strain measurement owing to their apparent accu- industrial, quality control, research, etc. Quality
racy, low cost, and ease of use; however, frequently standards have determined the need to express the
they are misused, and the causes of their measure- final measurement not as ‘raw data’, and therefore
ment errors are badly estimated [1]. Their working it involves the use of a model of measurement
principle is simple: the electrical resistance of a to determine the uncertainty associated with the
conducting wire is a function of the material’s elec- best estimate of the value of the quantity to be
trical resistivity, cross-sectional area, and length. A measured [2].
strain gauge is basically a thin wire or a thin metallic Every measurement, no matter what type it is, is
foil formed in a grid shape. The grid is then glued to affected by an uncertainty. This uncertainty reflects
the specimen through a thin backing or carrier for the lack of exact knowledge of the value of the
support and insulation. This way, the strain on the measured value or measurand [3]. Therefore, the
specimen is directly transferred to the gauge, gen- uncertainty of a measured value is of high impor-
erating a change in its electrical resistance. This tance. There are two reasons for measurement
phenomenon is known as the piezoresistive effect. uncertainty: the first is due to the measurand, and
the second is due to the error introduced by the
*Corresponding author: Universidad Carlos III de Madrid, Avd. measuring system. It is also important to note that
de la Universidad nu 30, Leganés, Madrid 28911, Spain. systematic errors have an effect on the global
email: mrami@[Link] accuracy of the measuring system, while random

J. Strain Analysis Vol. 46

Downloaded from [Link] at University of Birmingham on October 9, 2014

2 W Montero, R Farag, V Dı́az, M Ramirez, and B L Boada

errors affect the system’s precision and conse- for transverse sensitivity [7]. In general, the gauge
quently its accuracy [4]. factor takes into account the effect caused by the
Poisson strain; nevertheless, if a strain gauge is used
under different conditions from those set by the
2 SOURCES OF ERRORS IN STRAIN manufacturer to obtain the gauge factor, there is
MEASUREMENTS always some degree of error due to transverse
sensitivity. This transverse sensitivity error is a
Strain gauges are not measuring instruments by function of the corrected values of the strains
themselves – before they can be of any use, the user parallel and perpendicular to the primary sensing
must install them [5]. Therefore, their precision is axis of the gauge, ea and et respectively, as well as the
associated with various factors. Error sources in transverse sensitivity coefficient Kt and Poisson’s
resistance strain gauge measuring systems are ratio nF used by the manufacturer to determine Kt
numerous; for example, errors due to problems in (usually nF 5 0.285), as indicated in the following
the installation of the gauge, measuring procedure, equation [7]
signal processing, data acquisition, and manage-
ment. In this sense, the errors that can be quantified Kt ðet zea nF Þ
EST ~ ð1Þ
are of greater importance, as they establish the 1{nF Kt
relationships that connect the input or influence
quantities to the measurand, and error analysis can For the particular case of a two-gauge 90u rosette,
be regarded as a consistent methodology for devel- the corrected values of the strains parallel to the
oping models that can be used as a basis for sensing axis, ea, i , of each gauge are defined by
evaluating measurement uncertainty [2]; and this equations (2) and (3) [7]. Note that the strain parallel
even though it is impossible to determine the true to the maximum sensitivity axis of one of the gauges
value of these errors, for the same reason that it is is equal to the strain in the transverse direction of
impossible to determine the true value of a mea- the other gauge
surement [6]. In the following sections, the most
1{nF Kt
important sources of errors for two-gauge 90u ea, 1 ~et, 2 ~e1 ~ ð^e1 {Kt^e2 Þ ð2Þ
rosettes (Fig. 1) under biaxial strain states with 1{Kt2
known principal axis directions are analysed. Note
that, in the following equations, the subscript i takes 1{nF Kt
the value 1 or 2, in reference to the gauges in the ea, 2 ~et, 1 ~e2 ~ ð^e2 {Kt^e1 Þ ð3Þ
1{Kt2
rosette, as can be seen in Fig. 1.
where ^e1 and ^e2 are the measured (indicated and
uncorrected) strains in gauges 1 and 2 of the rosette
2.1 Errors due to the transverse sensitivity of the
respectively. Finally, the errors due to transverse
strain gauges
sensitivity for both gauges can be obtained by
This source of error refers to the response of the substituting equations (2) and (3) into equation (1)
gauge to strains that are perpendicular to the
primary sensing axis of the gauge. In general, it is Kt
EST, 1 ~ ½^e1 ðnF {Kt Þz^e2 ð1{nF Kt Þ ð4Þ
relatively small in comparison with the response of 1{Kt2
the gauge in its primary sensing axis. However, in
biaxial strain fields characterized by extreme ratios
Kt
between principal strains, the percentage error in the EST, 2 ~ ½^e2 ðnF {Kt Þz^e1 ð1{nF Kt Þ ð5Þ
smaller strain can be very significant if not corrected 1{Kt2

2.2 Errors due to temperature

The change in resistance of the gauge can also be
produced by a change in temperature of the material
whose strain is being measured or by a change in the
temperature of the environment surrounding the
Fig. 1 Tee strain gauge rosette gauge. This apparent temperature-induced strain

J. Strain Analysis Vol. 46

Downloaded from [Link] at University of Birmingham on October 9, 2014

 Uncertainties associated with strain-measuring systems 3

can become the main source of uncertainty of the exception of the temperature and eap , are given by
measuring system [8]. Once the gauge is installed, the manufacturer in the technical specifications
connected to a strain indicator instrument, and sheet of the strain gauge.
balanced to zero strain, a subsequent change in There are two possible ways to determine the
temperature will produce a change in the electrical value of the temperature-induced strain for any
resistance of the gauge independently from the given temperature. The first approach involves the
mechanical strain produced by the applied forces use of information regarding thermal output that
on the object. This change has two causes [8]: is provided by the manufacturer. This information
generally comes in the form of a graph and an
(a) the electrical resistance of the conductive grid
equation by which the thermal output can be
is, to a certain degree, temperature dependent;
determined for any given temperature. Before using
(b) The difference in thermal expansion between
this information, however, four points must be noted
the gauge’s conductive grid and the material of
[8].
the specimen being measured.
1. The material of the part whose strain is to be
In these cases, the measurement can be corrected or measured must be the same as that used by the
compensated for by using a compensating (dummy) manufacturer to determine the temperature-
strain gauge, a self-temperature-compensated strain induced strain curve and equation. It is also
gauge, or algebraically by means of the information important to note that even materials with the
provided by the manufacturer. It is also important to
same designation as the material used by the
note that, during static strain measurements, if the
manufacturer may have different thermal proper-
measuring system has been balanced for zero strain at
ties, as these vary from one lot to another and
the same temperature as the test temperature, the
with the shape of the part.
temperature-induced strain will be zero [8].
2. The information provided by the manufacturer
In addition to this temperature-induced strain, the
has been determined using a gauge factor F’
gauge factor also varies with temperature. As a
(usually F’ 5 2). Thus, in order to use it, the
consequence, the gauge factor used to calculate the
measuring instrument must also be set at this
strain has to be corrected for. In many cases, the
gauge factor.
error is small and can be neglected, especially if the
3. As the manufacturer has calculated the thermal
temperature change is not great. However, there are
output e’ap after zero balance at a temperature TF ,
cases in which a correction for the gauge factor is
the Wheatstone bridge must be balanced for zero
needed, depending on the alloy of the grid, the
strain at this same temperature in order to obtain
temperature during the test, and the accuracy
accurate results from equation (6).
required [8]. The gauge factor can be corrected for
4. Finally, the equation provided represents the
the effect of temperature by means of the informa-
average of a lot, and there will always be small
tion provided by the manufacturer about the gauge
differences in the behaviour between one gauge
factor variation with temperature.
and another, even if they come from the same lot.
Finally, it is possible to determine the error
induced simultaneously by both of these sources The second approach to obtaining the tempera-
and for each one of the gauges in the rosette through ture-induced strain is to use a much more general
the following equation [8] equation that can be applied to any type of material
  and under any circumstance, thus allowing for the
FI ^ei {eap evaluation of the thermal output when the condi-
ET , i ~^ei { ð6Þ
Fi ½1zCT ðT {TF Þ tions for the use of the graph and equation provided
by the manufacturer cannot be met. The equation is
where FI is the gauge factor at which the measuring as follows [8]
instrument has been set up, ^ei is the measured
   
(incorrect) strain, eap is the temperature-induced 1zKt
bG zFi ðaS {aG Þ ðT {T0 Þ
strain (thermal output), Fi is the gauge factor at 1{nF Kt
eap ~ ð7Þ
room temperature, CT is the variation coefficient of FI
the gauge factor with temperature, T is the tempera-
ture during the test, and TF is the temperature at where bG is the temperature coefficient of resistance
which the manufacturer determined CT (usually of the grid conductor, aS and aG are the thermal
room temperature). All of these values, with the expansion coefficients of the substrate and grid

J. Strain Analysis Vol. 46

Downloaded from [Link] at University of Birmingham on October 9, 2014

4 W Montero, R Farag, V Dı́az, M Ramirez, and B L Boada

respectively, T0 is the temperature at which the (c) the angular mounting error b between the axis
Wheatstone bridge has been balanced to zero strain, of the gauge and the strain axis that is intended
and FI is the gauge factor setting of the measuring to be measured.
instrument. It is important to note that this equation
takes into account the transverse sensitivity of the The Mohr’s circle illustrated in Fig. 2 shows a
gauge in the term ð1zKt Þ=ð1{nF Kt Þ. The reason for biaxial strain field and a two-gauge 90u rosette with
this is that in isotropic materials the thermal an angle h~w+b with respect to the principal strain
expansion/contraction will be uniform in all direc- axis. As can be seen in the figure, in any uniform
tions, while the gauge factor refers only to the strain biaxial strain field the error due to the misalignment
gauge’s sensitivity in a uniaxial stress state with a of the gauge is maximum while measuring the strain
principal strain ratio of 1=nF . In addition, from the at a 45u angle from the principal strain axis, and
shape of this equation it should not be assumed that minimum while measuring in the direction of the
the temperature-induced strain changes linearly principal strain axis. If ^e1 and ^e2 are the strains
with temperature, because all of the terms inside measured by gauges 1 and 2 respectively, with an
the brackets are themselves functions of tempera- incorrect angle h between the longitudinal axis of the
ture [8]. This equation clearly shows that the thermal gauge and the maximum principal strain axis p, the
output depends not only on the strain gauge’s equation for the error is as follows [9]
characteristics but also on the thermal and mechan-
ical properties of the material to which the gauge is EA ~ew+b {ew ð8Þ
bonded.
where ew is the strain at an angle w with respect to the
maximum principal strain axis, and ew+b is the
2.3 Errors due to misalignment measured strain (incorrect) at an angle w+b with
respect to the maximum principal strain axis. Next, it
When a gauge is bonded to the surface of a test
is possible to write the previous expression for the
surface at a small angular error with respect to the
error in terms of the centre C and the radius R of
intended axis of strain measurement (for example, p
Mohr’s circle (Fig. 2) as follows
and q in Fig. 2), the measured strain will also be in
error owing to the gauge’s misalignment. In general,
EA ~ew+b {½CzR cos ð2wÞ ð9Þ
for a gauge over a uniform biaxial strain field, the
magnitude of the error caused by misalignment
depends upon three factors [9]: According to Fig. 2, it is geometrically possible to
determine equations for the centre C and the radius
(a) the ratio of the maximum and minimum  R of Mohr’s circle in terms of the measured strains ^e1
algebraic strain in the two principal axes, ep eq ; and ^e2 and the mounting angle h. Substituting C and
(b) the angle w between the maximum principal R into equation (9) results in equations (10) and (11),
strain axis and the strain axis that is intended to which correspond to the error due to the misalign-
be measured; ment of gauges 1 and 2 of the rosette

Fig. 2 Mohr’s circle of strain, representing the difference between the intended strain
measurement and the actual strain measured owing to a mounting error b

J. Strain Analysis Vol. 46

Downloaded from [Link] at University of Birmingham on October 9, 2014

 Uncertainties associated with strain-measuring systems 5

  achieve this, it is necessary to use the law of

^e1 z^e2 ^e1 {^e2 propagation of uncertainty, by which the combined
EA, 1 ~^e1 { z cos ð2wÞ ð10Þ
2 2 cos ð2hÞ standard uncertainty uc ð y Þ can be found by calculat-
ing the square root of the sum of variances u2c ð y Þ
  given by [6]
^e1 z^e2 ^e1 {^e2
EA, 2 ~^e2 { { cos ð2wÞ ð11Þ
2 2 cos ð2hÞ N 
X 
Lf 2  
u2c ð y Þ~ u2 x j ð13Þ
j~1
Lxj

 
2.4 Error due to the Wheatstone bridge’s where each u xj is a standard uncertainty deter-
non-linearity mined through a type A or type B evaluation of the

variables xj , and the partial derivatives Lf Lxj are
Errors due to the non-linearity of the Wheatstone
bridge take place when the measurement is carried called the sensitivity coefficients (cj ). When the input
out with an unbalanced Wheatstone bridge. This is variables are correlated, it is necessary to take into
because the output voltage of the circuit is a non- consideration these correlations. In this case, the
linear function of the change in resistance. The error appropriate expression for the combined variance
is, in general, small, and negligible in many cases [10]. u2c ð y Þ associated with the result of a measurement is
The output voltage obtained from an unbalanced [6]
Wheatstone bridge is a function of the amount of
N 
X 
unbalance and is therefore directly related to the Lf 2  
u2c ð y Þ~ u2 x j
strain applied to the strain gauge. For the particular j~1
Lxj
case of a two-gauge 90u rosette in which each gauge
X
N N 
{1 X  
 
is wired in a quarter-bridge configuration, the non- Lf Lf
z2 u xj , xr ð14Þ
linearity error is [10] j~1 r~1
Lxj Lxr

2^ei
ENL, i ~^ei { ð12Þ A significant correlation between two input quan-
2{Fi^ei
tities can be present if the measurement conditions
are the same when they are determined. This situ-
ation occurs when the magnitudes have been ob-
tained from simultaneous observations and also
3 APPROACHES TO FINDING THE
UNCERTAINTY OF A MEASURING SYSTEM when the following are the same for both of them [6]:

(a) the measuring instrument;

Regardless of the source of the uncertainty, there are (b) the physical measurement standard;
two approaches to its estimation: the type A evalu- (c) the measurement methodology;
ation and the type B evaluation. Both methods are (d) the reference datum (having a significant
based on probability distributions, and the uncer- standard uncertainty).
tainty resulting from each type of evaluation can be
 
quantified through a standard deviation [6]. More- Finally, the term u xj , xr in equation (14) is called
over, the transformation of an error model into an the covariance of xj and xr , and it is given by the
uncertainty model is made by applying the law of following equation [6]
propagation of uncertainty [2], which is explained in
the following section.   1 X
n {1 Xn  
u xj , xr ~ xj {xj ðxr {x
r Þ ð15Þ
nðn{1Þ j~1 r~jz1

3.1 Propagation of the uncertainty

Very often the measurand is not measured directly;
instead, it is determined on the basis of other 4 UNCERTAINTY OF A STRAIN GAUGE
magnitudes for which a direct measurement can be MEASURING SYSTEM
made. For this reason, the uncertainties caused by
different sources of error are combined to obtain the As stated previously, to find the uncertainty asso-
uncertainty of the whole measuring system. To ciated with a measuring system, the law of propaga-

J. Strain Analysis Vol. 46

Downloaded from [Link] at University of Birmingham on October 9, 2014

6 W Montero, R Farag, V Dı́az, M Ramirez, and B L Boada

 2  2
tion of uncertainty has to be applied to the error LEs LEs
u2 ðEs Þ~ uðEST Þ z uðET Þ
model. However, first it is necessary to define the LEST LET
error of a measurement, E, as the deviation of the  2  2
LEs LEs
measured value (in this case the strain ei, k ) with z uð E A Þ z uðENL Þ ð21Þ
LEA LENL
respect to the true value of the measurand (in this
case the strain ei ). More specifically, the error is the
result of a measurement minus the true value of the
u2 ðEs Þ~u2 ðEST Þzu2 ðET Þzu2 ðEA Þzu2 ðENL Þ ð22Þ
measurand [5]. It is impossible to know the true
value of the measurand; however, it is possible to
determine the upper and lower limits where the true As a result, the variance of all the systematic errors
value lies. Therefore, according to the definition affecting the measurement is equal to the sum of the
variances of each systematic error. Finally, substitut-
ei ~ei, k {E ð16Þ ing equation (22) into equation (19) and applying the
square root on both sides of the equation, an
expression for the combined standard uncertainty
On the other hand, the measurement error E is
of the strain ei can be obtained
also defined as the sum of the systematic error Es
and the random error Er . The random error can also qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 
be defined as the difference between the measured uc ðei Þ~ u2 ð^ei Þzu2 EST;i zu2 ðET;i Þzu2 ðEA ;i Þzu2 ðENLi Þ
value in the kth observation (ei, k ) and the arithmetic
ð23Þ
mean of a series of measurements (in this case the
strain ^ei ) [6]; thus
Finally, the true (corrected) strain can be expressed
E~Er zEs ~ei, k {^ei zEs ð17Þ as

ei ~ð^ei {Es Þ+uc ðei Þ ð24Þ

Substituting equation (17) into equation (16) yields

ei ~^ei {Es ð18Þ

4.1 Uncertainty of the measuring device
Next, the law of propagation of uncertainty [6] is The uncertainty of the measuring instrument is of
applied through equation (13) in order to find the great importance, as it has consequences in all the
combined variance u2c ðei Þ, as shown below other uncertainties. The signal amplifier and condi-
 2  2 tioner equipment used during this research was
Lei Lei
u2c ðei Þ~ uð^ei Þ z uðEs Þ ~u2 ð^ei Þzu2 ðEs Þ a NEC San-ei AS2101 DC strain amplifier, with
L^ei LEs external 120 V Wheatstone bridges for each of the
ð19Þ strain gauges. This device gives an output signal in
volts, and thus it has to be converted into mm/m
where uðEs Þ is the uncertainty of the systematic error units using the following formula
and uð^ei Þ is the uncertainty of the arithmetic mean ^ei .
 
On the other hand, the systematic error Es is the V i 2
error associated with the measuring system and is ^ei ~ ec, i ð25Þ
Vc, i Fi
therefore the summation of all the errors that have
affected it and that have been described in previous where Vi is the mean output voltage, Vc, i is the
sections. Thus output voltage when a calibration strain ec, i is
applied, and Fi is the gauge factor. Next, the
Es ~EST zET zEA zENL ð20Þ
uncertainty of the measuring instrument is obtained
by applying equation (13) of the law of propagation
where EST is the transverse sensitivity error, ET is of uncertainty to equation (25)
the error due to the test temperature, EA is the
error due to the misalignment of the gauge, and ENL    2
2 L^ei    2 L^ei
is the error due to the non-linearity of the Wheat- u ð^ei Þ~  u V i z uðec, i Þ
LV i Lec, i
stone bridge. By applying equation (13), the variance  2  2
of the systematic error u2 ðEs Þ can be calculated as L^ei L^ei
z uðVc, i Þ z uð F i Þ ð26Þ
follows LVc, i LFi

J. Strain Analysis Vol. 46

Downloaded from [Link] at University of Birmingham on October 9, 2014

 Uncertainties associated with strain-measuring systems 7

 
where the standard uncertainty u V i is obtained In a similar way to the case of transverse sensitivity,
by applying the procedure for type A or type B the magnitudes of ^e1 , ^e2 , and T have been obtained by
evaluations, depending on the case [6] simultaneous observations of V1 , V2 , and T, and thus a
    correlation could exist between them. For this reason,
u V i ~max uðVi Þ, s V i ð27Þ equation (14) has to be applied in order to find the
variance of the error due to temperature u2 ðET ;i Þ
 2  2
2 LET , i LET , i
u ðET , i Þ~ uð^ei Þ z uð F i Þ
4.2 Uncertainty due to transverse sensitivity L^ei LFi
 2  2
As the magnitudes ^e1 and ^e2 have been obtained by LET , i LET , i
simultaneous observations of V1 and V2 , a correla- z uðFI Þ z uðbG Þ
LFI LbG
tion between them could exist. Because of this, it is  2  2
necessary to apply equation (14) to equations (4) and LET , i LET , i
z u ð aS Þ z uð a G Þ
(5) in order to find the variance of the transverse LaS LaG
sensitivity error u2 ðEST ;i Þ as seen in equation (28).  2  2
LET , i LET , i
Note that the subscript i takes the value 1 or 2, in z uðKt Þ z uðT0 Þ
LKt LT0
reference to the gauges in the rosette, as can be seen  2  2
in Fig. 1 LET , i LET , i
z u ð nF Þ z uðTF Þ
LnF LTF
 2  2
LEST, i LEST, i  2  2
u2 ðEST, i Þ~ uð^e1 Þ z uð^e2 Þ LET , i LET , i
L^e1 L^e2 z uðCT Þ z uð T Þ
LCT LT
 2  2
LEST, i LEST, i LET , i LET , i
z uð n F Þ z uðKt Þ z2 uð^ei , T Þ ð31Þ
LnF LKt L^ei LT
LEST, i LEST, i
z2 uð^e1 , ^e2 Þ ð28Þ where uð^ei , T Þ is the estimated covariance obtained
L^e1 L^e2
through equation (15), and given by
where uð^e1 , ^e2 Þ is the estimated covariance given by
1 Xn  
equation (15), which for this case is uð^ei , T Þ~ ei, k {^ei ðTk {T Þ ð32Þ
nðn{1Þ k~1
1 Xn   
uð^e1 , ^e2 Þ~ e1, k {^e1 e2, k {^e2 ð29Þ
nðn{1Þ k~1 where ei, k is the measured strain value of the kth
observation of gauge i that has been obtained using
where e1, k and e2, k are the measured strain values of equation (25), T is the arithmetic mean of the
the kth observation in gauges 1 and 2 that have been temperature, Tk is the measured temperature value
calculated through equation (25), and n is the total during the kth observation, and n is the total number
number of observations. of observations.

4.3 Uncertainty due to thermal output and the 4.4 Uncertainty due to misalignment of the strain
variation of the gauge factor with gauge
temperature
Again, it is necessary to consider for this case the
In order to use the general approach discussed correlation between the magnitudes of the strains ^e1
previously to find the temperature-induced strain and ^e2 , and it is necessary to apply equation (14) to
eap , equation (7) has to be substituted into equation find the variance of the error due to misalignment
(6) to obtain the final expression of the error due to u2 ð E A ; i Þ
temperature
 2  2
    2 LEA, i LEA, i
1zKt u ðEA, i Þ~ uð^e1 Þ z uð^e2 Þ
FI^ei { bG zFi ðaS {aG Þ ðT {T0 Þ L^e1 L^e2
1{nF Kt  2
ET , i ~^ei { LEA, i LEA, i LEA, i
Fi ½1zCT ðT {TF Þ z uðhÞ z2 uð^e1 , ^e2 Þ ð33Þ
ð30Þ Lh L^e1 L^e2

J. Strain Analysis Vol. 46

Downloaded from [Link] at University of Birmingham on October 9, 2014

8 W Montero, R Farag, V Dı́az, M Ramirez, and B L Boada

where uð^e1 , ^e2 Þ is the estimated covariance given by screw at the other (Fig. 3). The rosette was bonded
equation (29). 44.6 mm away from the fixed end. The strain field
was known to be biaxial and with the maximum
principal strain axis parallel to the long side of the
4.5 Uncertainty due to the non-linearity of the flat bar; thus, the rosette was bonded so that the
Wheatstone bridge maximum sensitivity axis of gauge No. 1 would
For this final case, equation (13) of the law of match the maximum principal strain axis (w 5 0).
propagation of uncertainty is applied to equation Ten measurements have been made at three differ-
(12) to find the variance of the error due to the non- ent deflections of the beam, giving a total of 30
linearity of the Wheatstone bridge u2 ðENL ;i Þ for each observations. The flat bar was bent by adjusting the
gauge screw to deflections of 3.82, 5.82, and 7.82 mm at a
distance of 254 mm measured from the fixed end of
 2  2 the bar.
LENL, i LENL, i
u2 ðENL: i Þ~ uð F i Þ z uð^ei Þ ð34Þ The mean output voltages of the strain gauges
LFi L^ei
through the tests can be seen in Table 1. For each
series of observations, the standard deviations has
been obtained. Subsequently, the standard uncer-
5 ESTIMATION OF THE UNCERTAINTY OF A tainty has been calculated according to the proce-
MEASURING SYSTEM dure for type A evaluations [6]. However, it must  be
noted that the type A standard uncertainties s V i
Once the uncertainties associated with the error that have been obtained are smaller than the
factors considered before have been identified and standard uncertainty u(V) obtained according to a
determined, it is possible to quantify the value of the type B evaluation of the resolution of the measuring
uncertainty of the entire measuring system. This is instrument, which is 0.01 V. For this reason, the
done by solving equation (23) for each strain gauge. uncertainty obtained through the type B evaluation
The procedure described throughout the previous is the one used in all calculations (see Table 2).
sections has been applied in a test carried out by Table 2 shows the values of all the variables used
bending an aluminium flat bar (25.4 mm66.35 mm) in the preceding equations and the associated
while measuring the strain using a two-gauge 90u standard uncertainty of most of them. These were
rosette. The flat bar was fixed by a clamp on one end, calculated by means of type B evaluation [6], using
and a bending force was applied by an adjustable the data supplied by the manufacturer of the rosette,

Fig. 3 Bending device used to test a strain gauge bonded to an aluminium flat bar

Table 1 Temperature and output voltage of the strain gauges measured at different deflections of the beam
Output voltage V at a Output voltage V at a Output voltage V at a
deflection of 3.82 mm deflection of 5.82 mm deflection of 7.82 mm
Gauge No. 1 Gauge No. 2 Gauge No. 1 Gauge No. 2 Gauge No. 1 Gauge No. 2 Temperature (uC)
Mean V̄i 2.370 20.768 3.611 21.168 4.857 21.572 33
Standard deviation 0.047 0.0079 0.0032 0.0042 0.0082 0.0042 0.00
Standard uncertainty s(V̄i) 0.0015 0.0025 0.0010 0.0013 0.0026 0.0013 0.00

J. Strain Analysis Vol. 46

Downloaded from [Link] at University of Birmingham on October 9, 2014

 Uncertainties associated with strain-measuring systems 9

Table 2 Values and estimated uncertainties of all the variables

Variable Value Units Interval Semi-range a Variance u2 5 a2/3 Uncertainty u
22 22 25
Vi Varies V ¡1.00610 1.00610 3.33610 5.7761023
Fi 2.12 Dimensionless ¡1.0061022 1.0061022 3.3361025 5.7761023
Kt 0.00 Dimensionless ¡3.0061023 3.0061023 3.0061026 1.7361023
CT 1.0061024 uC21 ¡5.0061025 5.0061025 8.33610210 2.8961025
FI 2.12 Dimensionless ¡1.0061022 1.0061022 3.3361025 5.7761023
ec, i 1.561023 m/m ¡1.0061026 1.0061026 3.33610213 5.7761027
Vc, i 8.07 V ¡1.0061022 1.0061022 3.3361025 5.7761023
T0 33 uC ¡5.0061021 5.0061021 8.3361022 2.8961021
T 33 uC ¡5.0061021 5.0061021 8.3361022 2.8961021
b 8.7361023 Radians ¡5.8261024 5.8261024 1.1361027 3.3661024
wi 0.00 Radians — — — —
bG 7.24361026 uC21 — — — —
aS 2.461025 uC21 — — — —
aG 1.561025 uC21 — — — —
aF 1.561025 uC21 — — — —
nF 0.285 Dimensionless — — — —
TF 23 uC — — — —
FF 2 Dimensionless — — — —

 
the thermometer’s resolution, and the user’s manual e’ap FF 1zKt
of the strain amplifier and the digital goniometer. A bG ~ {Fi ðaF {aG Þ ð37Þ
T {TF 1{nF Kt
rectangular probability distribution has been con-
sidered. As for the last eight variables, their asso- This equation, however, cannot be evaluated
ciated uncertainty is not taken into account as they directly for temperatures around T 5 23 uC because
are considered to have constant values. it is not a continuous function, as at this temperature
At this point it must be said that the value of the the term (T {TF ) will be zero and bG will tend to
temperature coefficient of resistance of the grid infinity. Nevertheless, it is highly unlikely that bG will
conductor bG had to be calculated, as this informa- behave in this way. To overcome this mathematical
tion was not specified by the manufacturer of the problem, equation (37) has been plotted for a range
strain gauge. This coefficient is a property of the of temperatures T matching the range of tempera-
strain gauge that specifies the change in its resis- tures used by the manufacturer to obtain e’ap . Next,
tance with temperature. To determine bG , equation the points near T 5 23 uC, corresponding to tem-
(7) has been evaluated under the same conditions as peratures between 0 and 45 uC, were eliminated from
those used by the manufacturer of the strain gauges. the plot, and the remaining data were fitted using
Solving for bG in equation (7), and with T0 ~TF , non-linear regression. The result is a sixth-degree
aS ~aF , and FI ~FF , the following equation is polynomial equation that relates the temperature
obtained coefficient of resistance of the grid conductor bG
  with the temperature T. The plot and the sixth-
eap FF 1zKt degree polynomial equation can be seen in Fig. 4.
bG ~ {Fi ðaF {aG Þ ð35Þ
T {TF 1{nF Kt

The only unknown variable from the previous 6 RESULTS

equation is the temperature-induced strain eap .
However, as a result of evaluating equation (7) When the law of propagation of uncertainty is
under the same circumstances as those used by the applied to the error model through equation (14), it
manufacturer to obtain the equation for thermal is convenient that it is organized
  in such a way that
output e’ap , it can be used to determine bG . the contributions of u2 xj are emphasized. The
According to the specifications sheet of the strain reasons for this are to allow for a global vision of the
gauges, the thermal output is given by balance of the uncertainty, to be aware of the weight
of the different contributions, and to develop ways to
e’ap ~{29:4z2:32T {4:6|10{2 T 2 improve the measurement process [11].
An example of this can be seen in Table 3 for the
z1:67|10{4 T 3 z5|10{7 T 4 ð36Þ uncertainty of the error due to thermal output and
the variation in the gauge factor with temperature
and with eap ~e’ap in equation (35) that has been discussed previously. In the first three

J. Strain Analysis Vol. 46

Downloaded from [Link] at University of Birmingham on October 9, 2014

10 W Montero, R Farag, V Dı́az, M Ramirez, and B L Boada

Fig. 4 Plot of the variation in bG with temperature

Table 3 Balance of the contributions of the uncertainties to the uncertainty of the error due to temperature
xj  
Lf    
2 cj ~
Symbol Units Value uj aj u (xj) Lxj u2j ET , 1 ~cj2 u2 xj
Fi Dimensionless 2.12 — 1.0061022 3.3361025 1.9661024 1.28610212
Kt Dimensionless 0.000 — 3.0061023 3.0061026 0.00 0.00
CT uC21 1.0061024 — 5.0061025 8.33610210 4.1561023 1.43610214
bG uC21 7.2461026 — 0 0 0 0
aS uC21 2.4061025 — 0 0 0 0
aG uC21 1.5061025 — 0 0 0 0
nF Dimensionless 0.285 — 0 0 0 0
TF uC 23 — 0 0 24.1561028 0
FI Dimensionless 2.12 — 1.0061022 3.3361025 21.9661024 1.28610212
T0 uC 33.0 — 5.0061021 8.3361022 21.2461025 1.28610211
T uC 33.0 — 5.0061021 8.3361022 1.2461025 1.29610211
ê1 m/m 4.1661024 1.56610 26
— 2.42610212 9.9961024 2.42610218
u(ê1, T) 0.00
  
Lf Lf 0.00
2 uð^e1 , T Þ
L^e1 LT
  P   PP  
u2 ET , 1 ~ u2j ET , 1 z2 cj cr u xj , xr 2.83610211
  q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 
u ET , 1 ~ u2 ET , 1 5.3261026

columns of the table, the symbols, units, and values the values of the sensitivity coefficients calculated
of the independent variables are given. The fourth for this specific case. It is also important to note that
column contains the contributions of statistically the covariance between e1 and T is zero, even though
evaluated uncertainties, meaning type A evaluations both variables have been measured simultaneously
as well as uncertainties obtained from previous during the tests. Finally, the uncertainty of the error
calculations or calibration certificates. In the next due to temperature is obtained. This result can also
column, the values of the half-width, aj , of the be seen in Table 4 (for gauge No. 1 and a deflection
symmetric rectangular distribution assumed for the of 3.82 mm), where all of the respective errors and
variables with type B evaluations
  can be seen. The their uncertainties are quantified following the same
2 procedure used in Table 3.
column with header u xj contains the variances of
the variables calculated according to the type A or B As can be seen, the error source that clearly has
evaluations, and following is a column that contains more weight as an error itself and in the uncertainty,

J. Strain Analysis Vol. 46

Downloaded from [Link] at University of Birmingham on October 9, 2014

 Uncertainties associated with strain-measuring systems 11

Table 4 Magnitudes of the errors that affected the measurement and their uncertainty
Deflection 5 3.82 mm Deflection 5 5.82 mm Deflection 5 7.82 mm
Gauge No. 1 Gauge No. 2 Gauge No. 1 Gauge No. 2 Gauge No. 1 Gauge No. 2
EST, i (mm/m) 0.000 0.000 0.000 0.000 0.000 0.000
u(EST, i) (mm/m) 0.028 0.653 0.042 0.996 0.057 1.339
ET, i (mm/m) 0.415 20.135 0.633 20.205 0.851 20.275
u(ET, i) (mm/m) 5.320 5.088 5.634 5.121 6.051 5.169
EA, i (mm/m) 20.042 0.042 20.064 0.064 20.086 0.086
u(EA, i) (mm/m) 0.003 0.003 0.005 0.005 0.007 0.007
ENL, i (mm/m) 20.183 20.019 20.425 20.044 20.770 20.081
u(ENL, i) (mm/m) 0.001 0.000 0.003 0.001 0.005 0.001

Table 5 Final results for gauge No. 1 and gauge No. 2 (the level of confidence of Uc is 95 per cent)
Deflection 5 3.82 mm Deflection 5 5.82 mm Deflection 5 7.82 mm
Gauge No. 1 Gauge No. 2 Gauge No. 1 Gauge No. 2 Gauge No. 1 Gauge No. 2
êi (mm/m) 415.585 2134.671 633.198 2204.812 851.687 2275.654
u(êi) (mm/m) 1.556 1.082 2.065 1.168 2.624 1.280
Es, i (mm/m) 0.190 20.112 0.143 20.185 20.005 20.270
Es, i (%) 0.05 0.08 0.02 0.09 20.001 0.10
u(Es, i) (mm/m) 5.320 5.130 5.634 5.217 6.052 5.340
ei (mm/m) 415.395 2134.559 633.054 2204.626 851.692 2275.384
Uc(ei) (mm/m) 11.085 10.485 12.001 10.692 13.192 10.982
Uc(ei) (%) 2.67 27.79 1.90 25.23 1.55 23.99

both for gauge No. 1 and for gauge No. 2, is the it represents less than 0.1 per cent of the measured
temperature during the test. For gauge No. 2, the value and could be added directly to the final
transverse sensitivity error has a significant weight uncertainty.
on the combined uncertainty as well. The reason for
this is that the ratio between the principal strains is
large, and gauge No. 2 has been bonded perpendi- 7 CONCLUSIONS
cularly to the maximum principal strain axis. Finally,
it can be observed that, for this particular case, the This paper gives a general model, based on error
errors due to misalignment and to the non-linearity modelling, to calculate the uncertainty of a strain
of the Wheatstone bridge can be neglected for both gauge measuring system as it offers a qualitative
gauges. analysis of the principal sources of error. Because
Table 5 shows the final results for the true strain strain gauges are not measuring devices by them-
(corrected) ei and the uncertainty Uc ðei Þ with a selves, it is essential to use a model that considers all
coverage factor of k 5 2 for the three deflections the error sources, and their correlation, that are
considered. The relative error Es, i (%) and the possibly present during a particular test where the
relative uncertainty Uc ðei Þ (%) are calculated in order uncertainty is to be known.
to analyse their contribution to the quality of the The general expression developed to determine
measurement. the uncertainty is then applied in a bending test
As can be seen, the relative error is not clearly under controlled conditions of load, temperature,
dependent on the deflection magnitude, as its values and gauge bonding, but even then the calculated
remain mainly constant, except for gauge No. 1, at a uncertainty can be appreciable and is inversely
deflection of 7.82 mm. This is not the case for the proportional to the magnitude of the measured
relative uncertainty, as an increase in the measured strain. These results demonstrate the impossibility
strain implies a decrease in the relative uncertainty. of giving a unique uncertainty of a strain gauge
For instance, if the strain is doubled, the relative measurement device, as it is strongly dependent on
uncertainty is 58 per cent lower, as the magnitude of the range of the strain measured, the control of the
uncertainty due to temperature remains practically temperature conditions, the ratio between the max-
constant for the three deflections considered. imum and minimum principal strains under the
Moreover, the incidence of the error in the gauge and its alignment with respect to them, and
expression of the final result could be neglected, as the installation procedure.

J. Strain Analysis Vol. 46

Downloaded from [Link] at University of Birmingham on October 9, 2014

12 W Montero, R Farag, V Dı́az, M Ramirez, and B L Boada

F Authors 2011 FF gauge factor used by the manufac-

turer during testing (dimensionless)
REFERENCES Fi gauge factor of strain gauge i
(dimensionless)
1 Chalmers, G. F. The dangers of the 5% mentality. FI gauge factor setting of the measuring
In Proceedings of the International Conference of instrument (dimensionless)
the BSSM and The Institution of Production i subscript that takes the value 1 or 2
Engineers, 1980. in reference to the gauges in the
2 Veen, A. M. H. and Cox, M. G. Error analysis in the
rosette
evaluation of measurement uncertainty. Metrolo-
gia, 2003, 40, 42–50. Kt transverse sensitivity (dimensionless)
3 Bell, S. A beginner’s guide to the uncertainty of n number of observations
measurement, 2001 (National Physical Laboratory, (dimensionless)
Teddington, UK. T mean temperature during the test
4 Window, A. L. and Holister, G. S. Strain gauge (uC)
technology, 2nd edition, 1992 (Applied Science TF temperature at which the manu-
Publishers, London, UK).
facturer determined the gauge
5 Hoffmann, K. An introduction to measurements
using strain gauges, 1989 (Hottinger Baldwin factor (uC)
Messtechnik GmbH, Darmstadt, Germany). Tk temperature measured in the kth
6 Joint Committee for Guides in Metrology: Evalua- observation (uC)
tion of measurement data – guide to the expression T0 temperature at zero calibration (uC)
of the uncertainty in measurement, JCGM, 100:2008, Vc, i calibration voltage in gauge i (V)
2008. Vi output voltage in gauge i (V)
7 Vishay Micro-Measurements, Technical note TN-
509, Errors due to transverse sensitivity in strain
V̄i mean output voltage in gauge i (V)
gauges, 2007, pp. 91–99.
8 Vishay Micro-Measurements, Technical note TN-
504-1, Strain gauge thermal output and gauge aF thermal expansion coefficient of the
factor variation with temperature, 2007, pp. 35–47. material used by the manufacturer
9 Vishay Micro-Measurements, Technical note TN- during testing (uC21)
511, Errors due to misalignment of strain gauges, aG thermal expansion coefficient of the
2007, pp. 107–111.
gauge’s grid (uC21)
10 Vishay Micro-Measurements, Technical note TN-
507-1, Errors due to Wheatstone bridge nonlinear- aS thermal expansion coefficient of the
ity, 2007, pp. 77–81. substrate (uC21)
11 Barbato, G. Misurare per decidere: misure e b angular mounting error (rad)
statistica di base, 2nd edition, 2005, pp. 218–222 bG temperature coefficient of resistance
(Progetto Leonardo, Bologna, Italy). of the grid conductor (uC21)
ea strain along the gauge’s main axis
(mm/m)
APPENDIX
eap temperature-induced strain (mm/m)
ec, i calibration strain in gauge i (mm/m)
Notation
ei true (corrected) strain in gauge i
CT variation coefficient of Fi with tem- (mm/m)
perature (uC21) êi mean strain measured in gauge i
E measurement error (mm/m) (mm/m)
EA error due to misalignment (mm/m) ei, k strain measured in the kth observa-
ENL error due to non-linearity of the tion in gauge i (mm/m)
Wheatstone bridge (mm/m) ep maximum principal strain (mm/m)
Er random error (mm/m) eq minimum principal strain (mm/m)
Es systematic error (mm/m) et strain perpendicular to the gauge’s
EST error due to transverse sensitivity main axis or transverse strain
(mm/m) (mm/m)
ET error due to temperature (mm/m) ew strain measured at an angle w with
F9 gauge factor used by the manufacturer respect to the maximum principal
to determine e9ap (dimensionless) strain (mm/m)

J. Strain Analysis Vol. 46

Downloaded from [Link] at University of Birmingham on October 9, 2014

 Uncertainties associated with strain-measuring systems 13

ew ¡ b strain measured at an angle w ¡ b nF Poisson’s ratio of the material used

with respect to the maximum princi- by the manufacturer to obtain Kt
pal strain (mm/m) (dimensionless)
h true angle between the maximum w angle between the maximum princi-
principal strain axis and the measured pal strain axis and the strain axis that
strain axis (rad) is intended to be measured (rad).

J. Strain Analysis Vol. 46

Downloaded from [Link] at University of Birmingham on October 9, 2014