ACTA IMEKO

ISSN: 2221‐870X
November 2016, Volume 5, Number 3, 9‐15

Application of optimization methods for improved electrical

metrology
Marija Cundeva‐Blajer
Ss. Cyril and Methodius University, Faculty of Electrical Engineering and Information Technologies, ul. Ruger Boskovic br. 18, POBox 574,
1000 Skopje, R. Macedonia

ABSTRACT
In electrical quantities metrology numerous examples of stochastic processes exist and a need for optimal solutions is posed. Here
stochastic genetic algorithm optimization is used for solving two typical metrological problems: 1) minimization of the metrological
parameters (final accuracy) in the design process of instruments and 2) predicting metrological reference standard’s time‐drift, i.e. re‐
calibration interval. The first case is the optimal metrological design of a combined instrument transformer and the second case is the
analysis of resistance standard time‐drift.

Section: RESEARCH PAPER

Keywords: optimization; genetic algorithm; electrical measurements; instrument transformer; resistance standard
Citation: Marija Cundeva‐Blajer, Application of optimization methods for improved electrical metrology, Acta IMEKO, vol. 5, no. 3, article 3, November 2016,
identifier: IMEKO‐ACTA‐05 (2016)‐03‐03
Section Editor: Franco Pavese, Italy
Received April 18, 2016; In final form April18, 2016; Published November 2016
Copyright: © 2016 IMEKO. This is an open‐access article distributed under the terms of the Creative Commons Attribution 3.0 License, which permits
unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited
Corresponding author: Marija Cundeva‐Blajer, e‐mail: [email protected]

1. INTRODUCTION practice either, since neither LS nor other data processing

methods can provide statistically consistent estimates for
In metrology among other problems two of them represent functions. Therefore, the key problem is to construct proper
general challenges [1]: extensions of regression models, which allow obtaining
1. Reduction of errors and uncertainties during the design consistent estimates for functions [2]. The proper model should
process of measurement instrumentation and devices; include additional information on data or data errors, which
2. Prediction of the time-behaviour of measurement makes it possible to construct consistent estimates. Confluent
instrumentation or standards, i.e. estimation of the estimates should be constructed in accordance with the type of
optimal recalibration interval. the extended model. However, the most important problem for
Construction of functions based on experimental data is a measurements is to study the accuracy characteristics of these
widespread and important procedure in measurement problems estimates, taking into account both random and systematic
(e.g. fitting of calibration curves for measuring instruments, or components of data errors. In [2], several ways for expansion of
deduction of equations for indirect measurement methods) [2]. regression models are outlined, and associated groups of
In these cases the quality of the empirical function is essential estimates are studied. The accuracy properties of estimates are
for ensuring the integral quality of measurement. The empirical considered, which may be compared with the corresponding
function must be sufficiently precise for practical purposes and characteristics of classical LS estimates. This is a heuristic
its estimates of the accuracy characteristics must be reliable. approach, but the optimisation does not work in this case. The
The traditional method for fitting of functions is the classical main criterion for deriving confluent methods is the
least squares (LS) method. The LS estimates are optimal only consistency as an asymptotic property of estimates. However,
under strict conditions upon errors of data, that is, within a confluent estimates are used in the cases of small or modest
classic regression model. So, when it does not hold in practice, volumes of data. Therefore, the accuracy characteristics of
some other methods are required, so-called confluent methods. various confluent estimates should be also compared with the
However, a general confluent model is not appropriate for classical LS fitting under practical conditions [2].

ACTA IMEKO | www.imeko.org November 2016 | Volume 5 | Number 3 |9

 In electrical measurements numerous examples of stochastic at the Ss. Cyril and Methodius University in Skopje [6] is used
processes exist, like the electrical quantities standards time- for solving two metrological problems:
behaviour, [1], [3], [4], [5], and [6]. a) Metrological optimal design of a combined instrument
In recent years the application of stochastic optimization transformer;
tools for the design of electrical devices (including electrical b) Predictions of resistance standards time drift for the
instrumentation) is evident [10]-[13]. determination of the optimal recalibration period, [16].
In electrical power distribution grids, state estimation is the
main part of the supervisory control and data acquisition 2. DESCRIPTION OF GENETIC ALGORITHM PROGRAM FOR
(SCADA) system. The aim is to determine an estimate of the METROLOGICAL OPTIMIZATION
system state based on a network model and the measurements
available. The network state consists of voltage bus magnitudes On one side the concept of a genetic algorithm is very
and angles, line flows, loads, transformer taps, active and simple as displayed in Figure 1, but on the other side it is a very
reactive power, power injections and generator outputs. In a powerful and robust computational tool.
typical state estimation scenario, the measurement data includes The GA implemented in the original computer program
active and reactive power flows, power injections and voltage developed at the Ss. Cyril and Methodius University in Skopje
magnitudes at the bus. In [3] a quasi-dynamic state estimation
method, based on a disturbance observer is proposed. Its aim is
to reconstruct electrical power (active and reactive) from
available measurements. On the distribution level, where a high
number of nodes is accompanied by a small number of
measurements, the typical approach is to replace missing data
with so-called pseudo-measurements. However, in this way the
state estimation quality depends on the reliability of the pseudo-
measurements. That means that in order to improve the
accuracy and reliability of the state estimation result, improved
pseudo-measurement models are needed. In [3] stochastic
models to obtain reliable pseudo-measurements and to improve
state estimation results are investigated. The autoregressive
moving average models (ARMA) to model the error of pseudo-
measurements are used. The employed disturbance observer
state estimation method is then able to improve the quality of
the pseudo-measurements based on assumed ARMA process
coefficients.
The optimal measurement theory arose as a branch of the
inverse problems theory [4]. The mathematical justification of
the algorithm for numerical restoration of the signal distorted
by inertia and resonances in measuring transducers is presented
in [4]. One of the directions for the development of this theory
is mathematical justification of restoration of stochastic signals,
in particular ”white noise”.
Design applications generally aim to minimise
manufacturing or material costs subject to performance criteria,
which is an inverse problem requiring optimisation techniques
[7]. What mathematical physics denotes as inverse problems is
the class of problems which are fundamental in measurement
theory and practice [8]. The main objective of such problems is
to develop procedures for acquiring information on object and
phenomena, accompanied by decreasing the distortion caused
by measuring instruments [8]. In [7] the implementation of
Levenburg-Marquardt, sequential quadratic programming,
Nelder-Mead, simulated annealing and practical swarm
optimisation for metrology and design with continuous models
is reported. In [10]-[13] the genetic algorithm (GA) as stochastic
optimization method is applied for optimal design.
Calibration consists of two stages [15]. In stage 1 a relation is
established between values provided by measurement standards
and corresponding instrument response values. In stage 2 this
relation is used to obtain measurement results from further
response values. Polynomials of various degrees, determined by
least squares, are extensively used as empirical calibration
functions in metrology [9]. In this contribution an original Figure 1. Genetic algorithm for solving problems of stochastic nature in
computer program with embedded genetic algorithm developed metrology of electrical quantities.

ACTA IMEKO | www.imeko.org November 2016 | Volume 5 | Number 3 |10

[10] starts with the initial units (chromosomes) stochastic  current measurement core (CMC) current error pi,
generation. Each chromosome is composed of the same  CMC phase displacement error δi.
number of genes. It is the initial population (generation). The CIT must comply with the standard IEC 61869-4 [14].
The size of the population is between 20 and 100. Iteration It comprises two transformer systems in complex
in the GA is the cycle of estimation, selection and reproduction electromagnetic coupling:
of the population. Each unit represents the possible optimal  a current measurement system with a regime close to a
solution. A possible solution for each unit is estimated short circuit in the secondary winding and
according to the following criterion: the value of the goal function
 a voltage measurement system with a regime of almost
and each solution is assigned an adequacy measure. According to
unloaded transformer, i.e. open-circuit regime.
the comparison of all adequate measures of the units, a decision
Both measurement systems are in a common housing with
is made which one of them will be allowed to form the next
strong electromagnetic influence which significantly contributes
generation and with which probability in the step of selection.
to the metrological properties of the whole device. The
The selection strategy is the “virtual roulette wheel”. All population
geometry of the CIT is displayed in Figure 2.
units are potentially capable for reproduction. However, whether
The goal during the optimal design process of the CIT is the
the unit will be selected depends on the value of the goal
minimisation of its metrological parameters pu, δu, pi and δi, as
function (a sorted ranking list is formed). Finally, the virtual
well as the VMC voltage error pu0,25 at 0.25 of VMC rated load,
roulette wheel is formed according to the assigned number in
according to the specifications of the IEC standard [10] as
the list. Conditionally, most of the “positions” in the wheel are
discussed and with initial results given in [11], [12]. Also, the
assigned to the best units, and the worst units have no
difference of the absolute values of the VMC voltage error at
“positions”. The selected units from the “virtual roulette
rated load and at 0.25 of the rated load should be as minimal as
wheel” are copied into a new generation ready for generic
possible. The minimization requirements of all the metrological
operation cross-over and mutation. The elitism function is
parameters could be contradictory. So, the process of
embedded, enabling automatic coping (cloning) of the best unit
metrological optimal design has to be performed by posing a
(according to the goal function value). Thus, the danger of
single objective function comprising them all. This is an
losing the best unit during the cloning process is avoided.
example of a multi objective optimisation problem. The
In the next step reproduction between pairs is performed,
optimization design of the 20 kV CIT, the case study in this
through simple copying of the units in the next generation. This
paper, is made through variation of eleven input constructional
is a process until the moment when cross-over and mutation
and geometrical variables of the CIT. The program GA
happen according to the previously defined probability. The
maximizes the objective function, therefore, the reciprocities of
selection of units exposed to the process of cross-over is
the errors are considered.
random. They are grouped and the spot of division of the genes
The objective function can be defined with a conservative
from the two parents is random again. Through the cross-over an
approach from a metrological point of view - the certain
exchange and recombination of the genetic material is achieved
measurement errors expressed through the absolute values are the
which forms a new chromosome with genes from the two
metrological objective function:
parents. The mutation is a change of a unit in the chromosome.
The mutation is a very important factor in the evolution, 1 1 1 1
f opt 1     (1)
because it introduces new genes in the process. This 1  p u  p u 0, 25 1  p u 1  p u 0, 25 1  p i
additionally increases the diversity of the units in the
population. Pairs of chromosomes (parents) are exposed to the as combination of the VMC voltage error pu at rated load, VMC
process of cross-over and the mutated units are returned in the voltage error pu0,25 at 0.25 of the rated load and CMC current error
population of the new generation. The cycle of estimation, pi at rated load. The phase displacement errors are constraints
selection and reproduction continues until a predefined during the optimization process in order to satisfy the achieved
acceptable behaviour level is achieved. If the termination
condition is the predefined generation number, then the best
unit in the last generation is the optimum. The population units
(chromosomes) in each next generation are more similar to
each other with convergence towards specific part of the search
space.

3. CASE STUDY NO. 1 ‐ GA OPTIMAL DESIGN OF AN

INSTRUMENT TRANSFORMER
Instrument transformers are of great metrological
significance for electrical power measurements, especially in the
field of legal metrology and trade of electrical energy. The
accurate metrological analysis of a 20 kV combined instrument
transformer (CIT) (voltage transformation ratio
20000 / 3 V : 100 / 3 V and current transformation ratio 100
A: 5 A) is possible only by numerical calculation of the
Figure. 2. Geometry of the instrument transformer
magnetic field distribution in the 3D domain [11]. The 1. VMC iron core
following four metrological parameters are relevant to the CIT: 2. CMC iron core
 voltage measurement core (VMC) voltage error pu, 3., 4. VMC windings
 VMC phase displacement error δu, 5., 6. CMC windings.

ACTA IMEKO | www.imeko.org November 2016 | Volume 5 | Number 3 |11

accuracy class of both measurement cores. The CIT is a highly
non-linear electromagnetic system, with mutual electromagnetic
influence of both cores which determines the metrological
performance of the device. The GA optimal design is coupled to
the results of the FEM-3D analysis of the magnetic field
distribution in the CIT as in [11], [12]. Some of the initial results
of the optimal design process with the goal function as defined in
(1) are given in [11], [12].
According to the Guide for Expressing the Uncertainty in
Measurements (GUM) [15], a statistical approach can be applied:
1
f opt 2  (2)
1  pu  pu 0,25 
2
 p u 2  p u 0, 25 2  p i 2
In the objective functions (1) and (2) the phase displacement
error u andi are constraints (a control criterion of the maximal
allowed values of the phase displacement errors for certain CIT Figure. 3. Genetic algorithm changes through the generations of the VMC
accuracy classes in the GA algorithm CIT mathematical model is primary winding current density.
embedded, as in [14]). The CIT optimal design starts with the
parameters derived by classical analytical design of the CIT based
on the T-equivalent circuit as a CIT model given in [11], [12] by
coupling with FEM-3D modelling and calculation of the main
electromagnetic characteristics. The chromosome comprises 11
genes (input variables-geometrical, constructional and electrical
parameters). The mapping range of the eleven GA input variables
is defined according to the previously derived results by FEM-3D
and the analytical transformer design, some most characteristic,
are given in Table 1.
The optimal solution is derived by the following genetic
parameters: cross-over probability 0.65, mutation probability 0.03,
population size 16, maximal number of generations 30000.
In Figures 3 to 5, the changes throughout the 30000
generations in the GA optimisation process of the current
densities in the CIT windings, as input variables (genes) are
comparatively displayed for both objective functions: conservative
approach fopt1 and statistical approach fopt2. Figure. 4. Genetic algorithm changes through the generations of the CMC
The main input variables in the calculation of the CIT secondary winding current density.
metrological parameters are the equivalent T-circuit parameters,
i.e. the winding’s resistances and the leakage reactances.
In Figures 6 to 12, changes throughout the 30000 generations
in the GA optimisation process of the CIT equivalent T-circuit

Table 1. Input variables for CIT optimal design (mapping range).

Input variable Minimum Maximum Initial value

(analytical)

VMC primary winding

23584 24000 24000
number of turns
VMC primary winding
2 1.5 3.0 2.04
current density [A/mm ]
VMC secondary winding
2 2.0 3.0 2.61
current density [A/mm ]
VMC magnetic core outside
183 193 185
length [mm]
Figure. 5. Genetic algorithm changes through the generations of the VMC
CMC secondary winding
115 125 120 secondary winding current density.
number of turns
CMC primary winding
2 1.0 1.6 1.36 parameters are comparatively displayed for both objective
current density [A/mm ]
functions: conservative approach fopt1 and statistical approach fopt2.
CMC secondary winding In Table 2 some of the optimal results (metrological
2 2.0 3.0 2.55
current density [A/mm ]
parameters, geometrical, constructional and electromagnetic)
CMC magnetic core outside
136 162 142
derived by both the goal functions (1)-conservative approach and
length [mm] (2)-statistical approach, are compared.

ACTA IMEKO | www.imeko.org November 2016 | Volume 5 | Number 3 |12

Figure. 6. Genetic algorithm changes through the generations of the VMC Figure. 10. Genetic algorithm changes through the generations of the VMC
primary winding resistance per turn. secondary winding leakage reactance per turn.

Figure. 7. Genetic algorithm changes through the generations of the VMC Figure. 11. Genetic algorithm changes through the generations of the CMC
secondary winding resistance per turn. primary winding leakage reactance per turn.

Figure. 8. Genetic algorithm changes through the generations of the CMC Figure. 12. Genetic algorithm changes through the generations of the CMC
secondary winding resistance per turn. secondary winding leakage reactance per turn.

Table 2. Comparison of the CIT metrological, electromagnetic and

constructive parameters derived by goal function with conservative and
statistical approach.

GA derived CIT Conservative Statistical

parameter approach approach
pu [%] ‐0.795 ‐0.625
pu0,25 [%] 0.795 0.855
‐7
pi [%] 0.0 10
δu [min] ‐69.43 ‐61.62
δi [min] ‐10.45 2.30
Bmu [T] 0.424 0.336
Bmi [T] 0.539 0.165
Figure. 9. Genetic algorithm changes through the generations of the VMC mFe [kg] 35.18 46.58
primary winding leakage reactance per turn. mCu [kg] 12.02 13.57

ACTA IMEKO | www.imeko.org November 2016 | Volume 5 | Number 3 |13

4. CASE STUDY NO. 2 ‐ GA FOR PREDICTION OF RESISTANCE calibration on 2007-06-14, sR7 and sR7exp are the respective
STANDARDS TIME DRIFT uncertainties. The optimal solution is for the following genetic
parameters: cross-over probability 0.65, mutation probability
The estimation of the optimal moment for calibration of the 0.03, population size 10, maximal number of generations 10000.
standards is a challenge for most of the metrological The resistance standard values in the five time points
laboratories [16]. The decisions are made based on the empirical derived from the process of genetic algorithm optimization are
experience of the laboratory staff, the behaviour history and the given in Table 4.
conditions of the standard [16]. The trend lines are usually The derived GA results after 10000 generations for the
extrapolated to predict the time-behaviour of the standards by Lagrange polynomial coefficients are given in Table 5.
using the classical least squares method, as in [5]. However, The value of the 1 Ω standard resistance in 2007 derived by
these models do not take into account the random nature of the the LP coupled with the GA optimization process is R2007
calibration results and uncertainties. =0.999992349 Ω.
In this contribution a new model of a time-drift trend line of The verification of the method is done through comparison
resistance standard derived by using the Lagrange polynomial of the theoretical standard resistance in 2010, with the
(LP), coupled with genetic algorithm optimization, is presented. experimentally derived value through calibration. The
The time points of the performed calibrations and the gained theoretical value of the standard resistance on 2010-06-14,
values of standard resistance are input data. The final model of gained by the LP trend line and through the GA process, is
the trend line using five input experimental values can be 0.9999916844, with difference of 7.7710-7 , which is the
reduced to a polynomial of fourth order: same order of uncertainty as the input variable with highest
R(t)=a0t 4 + a1t 3 + a2t 2 + a3t + a4 (3) uncertainty R1996.
The experimental input values of the 1 Ω resistance standard
where t is the time, R is the resistance and ai, i=0,...,4 are the LP are compared in Figure 13 to the resistance values in different
coefficients, depending on the values of the resistance at five time points derived by the Lagrange polynomial coupled to the
calibration moments. The LP model is derived on the basis of genetic algorithm optimization process.
the experimental data of calibrations of 1  standard in time The uncertainties of the theoretical values are compared to
intervals from 1996 to 2005 year as in Table 3. the measurement uncertainties of the experimental data and
The goal function of the optimization process by the genetic overlapping of the both uncertainties interval exists. The
algorithm (GA) is the minimum difference between the overlapping is satisfactory for the point in 2010, which is a
theoretical resistance value, derived by the LP, and the verification point of the methodology.
experimental value of the resistance at certain time moment
(2007-04-20). The original GA program maximizes the goal Table 4. Resistance standard values at five time points derived from the GA
function and the following goal function fopt is given in the optimization process.
optimization process:
R1996 [] R1998 [] R2000 [] R2003 [] R2005 []
1
f opt  (4)
1  R 6  R 6 exp 0.99999245 0.99999242 0.9999923 0.9999923 0.99999235

where R6=R2007 is the resistance of the standard derived by the

LP, determined by GA, at the time moment 20. 04. 2007 and Table 5. The GA output data for the LP model.
R6exp is the resistance of the standard derived by experimental
calibration on 2007-04-20. LP coefficient GA derived value
A verification of the gained model with known experimental a0 [Ω/month ]
4
0.00000000000
data is done (for the time point 2010-06-14). The verification a1 [Ω/month ]
3
0.00000000000
criteria k is:
‐0.00000000025
2
a2 [Ω/month ]
R 7  R 7 exp a3 [Ω/month] 0.00000000378
k (5) a4 [Ω] 0.999992438
s R 7  s R 7 exp
where R7=R2010 is the resistance of the standard derived by the
LP, determined by GA, at the time moment 2010-06-14 and
R7exp is the resistance of the standard derived by experimental

Table 3. The experimental GA input data for the LP model.

Date R [] Uncert. of R []

‐7
1996‐09‐26 0.999992300 ±510
‐8
1998‐10‐23 0.999992450 ±310
‐9
2000‐12‐28 0.999992330 ±610
‐9
2003‐04‐08 0.999992306 ±610
‐8
2005‐05‐09 0.999992353 ±110
‐8
2007‐04‐20 0.999992349 ±110
‐8 Figure. 13. Comparison of the GA derived LP time‐trend line and the
2010‐06‐14 0.999992461 ±110
experimentally derived time‐drift of 1 Ω resistance standard.

ACTA IMEKO | www.imeko.org November 2016 | Volume 5 | Number 3 |14

5. CONCLUSIONS uncertainty evaluation”, Series on Advances in Mathematics for
Applied Sciences-Vol.84, Advanced Mathematical and
In this paper the stochastic optimization method-genetic Computational Tools in Metrology and Testing IX, World
algorithm is applied for gaining a solution of two different, but Scientific Publishing, 2012, pp. 260-267.
challenging problems in metrology of electrical quantities: [8] K. K. Semenov, G. N. Solopchenko “Inverse Problems in
1. Reduction of errors and uncertainties during the design Theory and Practice of Measurements and Metrology”, Series on
process of an instrument transformer; Advances in Mathematics for Applied Sciences-Vol.86,
2. Prediction of the time-drift of a resistance standard. Advanced Mathematical and Computational Tools in Metrology
and Testing X, World Scientific Publishing Co. Pte. Ltd, 2015,
One original and universal GA software solution is used for pp. 330-339.
two problems of diverse nature. The derived results are [9] M. Cox, P. Harris “Polynomial Calibration Functions Revisited:
satisfactory and verified. It can be concluded that lots of Numerical and Statistical Issues”, Series on Advances in
options exist, i.e. the genetic algorithm further to be applied for Mathematics for Applied Sciences-Vol.86, Advanced
gaining other optimal solutions in different metrology areas and Mathematical and Computational Tools in Metrology and
problems of stochastic nature. Testing X, World Scientific Publishing Co. Pte. Ltd, 2015, pp. 9-
16.
[10] G. Cvetkovski, L. Petkovska, S. Gair, “Specific power as
REFERENCES
objective function in GA optimal design of permanent magnet
[1] M. Cundeva-Blajer, “Optimization Methods in Metrology of disc motor”, COMPEL: Int. Jour. for Computation and
Electrical Quantities”, Proc. of 21st IMEKO World Congress, Mathematics in Electrical and Electronic Engineering 29, Iss. 4,
Aug. 30-Sept. 4, 2015, Prague, Czech Republic, pp. 1903-1908. 2010, pp. 964-973.
[2] T. N. Siraya, “Regression and Confluent Algorithms and Models [11] M. Cundeva-Blajer, L. Arsov „Computer Aided Techniques for
For Construction Of Functions In Measurements”, Proc. of 21st Estimation and Reduction of Electromagnetic Measurement
IMEKO World Congress, Aug. 30-Sept. 4, 2015, Prague, Czech Devices Uncertainties“, Int. Journal of Metrology and Quality
Republic, pp. 2047-2051. Engineering, Vol. 2, EDP Sciences, Paris, 2010, pp. 89-97.
[3] N. Makarava, S. Eichstaedt, “Stochastic modelling for state [12] M. Cundeva-Blajer, L. Arsov “Influence of the Objective
estimation in medium-voltage electric grids”, Proc. of 21st Function in the Process of Optimal Design of an
IMEKO World Congress, Aug. 30-Sept. 4, 2015, Prague, Czech Electromagnetic Measurement System”, Int. Journal of Applied
Republic, pp. 1997-2001. Electromagnetics and Mechanics Vol. 39, IOS Press,
[4] A. L. Shestakov, G. A. Sviridyuk, A. V. Keller, “Optimal Amsterdam, 2012, pp. 705-710.
Measurements”, Proc. of 21st IMEKO World Congress, Aug. 30- [13] D. Filipovic-Grcic, M. Pljak, Z. Stih, “Optimization of the Paper-
Sept. 4, 2015, Prague, Czech Republic, pp. 2072-2281. Oil Insulation of High Voltage Inverse Current Transformers
[5] P. Mostarac, R. Malarić, H. Hegeduš “Extrapolation of Using Genetic Algorithm”, Proc. of 13th Int. Symp. on
Resistance Standard Values between Calibrations“, Conf. Proc. Electromagnetic Fields in El. Eng. ISEF’07, Prague, (paper ID
of the IX Nat. Conf. with Int. Part. ETAI, Ohrid, Macedonia, 065 on a CD ROM), 2007.
CD ROM Е2-2, 2009. [14] IEC 61869-4, Edition 1.0, 2013-11: Instrument transformers,
[6] M. Cundeva-Blajer, L. Arsov “Prediction of Resistance Standards Part 4: Additional requirements for combined transformers,
Time Behaviour by Stochastic Determination of Lagrange Geneva, 2013
Polynomial”, Series on Advances in Mathematics for Applied [15] Expression of the Uncertainty of Measurements in Calibration,
Sciences-Vol.84, Advanced Mathematical and Computational (GUM), JCGM 100 with member organizations (BIPM, IEC,
Tools in Metrology and Testing IX, World Scientific Publishing ILAC, OIML, ....), 1999.
Co. Pte. Ltd, 2012, pp. 102-109. [16] ILAC-G24:2007 / OIML D 10:2007 (E) Guidelines for the
[7] C. E. Matthews, L. Wright, X-S Yang “Metrology and Design determination of calibration intervals of measuring instruments,
with Continuous Models: tools for optimisation, sensitivity and 2007.

ACTA IMEKO | www.imeko.org November 2016 | Volume 5 | Number 3 |15