Journal of Natural Gas Science and Engineering 22 (2015) 313e320

Contents lists available at ScienceDirect

Journal of Natural Gas Science and Engineering

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

Use of data reconciliation: A strategy for improving the accuracy in gas

flow measurements
Elcio Cruz de Oliveira a, b, *, Maurício Nogueira Frota b, Gisele de Oliveira Barreto a, b
a
Petrobras Transporte S.A., Project Management, Rio de Janeiro, 20091-060 RJ, Brazil
b
Posgraduate Programme in Metrology, Metrology for Quality and Innovation, Pontifical Catholic University of Rio de Janeiro, 22453-900 RJ, Brazil

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

Article history: The economic impact that results from the reliability of measurements associated with natural gas (flow
Received 10 October 2014 rate and fluid properties) and caveats related to custody transfer contracts demands vigilant control of
Received in revised form the net balance in the delivery systems. The methodology for data reconciliation has proved to be an
10 December 2014
effective tool to reduce uncertainties associated with measurements used in the calculations of the net
Accepted 11 December 2014
Available online 18 December 2014
balance in distribution networks such as gas pipelines. The intrinsic nature of the calculation algorithm,
founded on the redundancy of measurements, qualifies the technique for increasing confidence in the
measurement, thus reducing the individual uncertainty associated with each physical magnitude capable
Keywords:
Data reconciliation
of affecting the measurement.
Unaccounted for gas This Brazilian gas pipeline study discusses the adequacy of the data reconciliation technique. The
Improving the accuracy proposed technique proves to be very effective as it generates lower uncertainties than those obtained by
Gas net balancing traditional techniques: the level of 1% associated with the accountability of the unaccounted for gas was
Gas flow measurements reduced to less than 0.3% when the data reconciliation methodology was used.
© 2014 Elsevier B.V. All rights reserved.

1. Introduction operating and handling techniques, imposing challenges on their

Although oil had been avoided by ancient civilizationsd a heavy
waste that exhibits strong and viscous odord, early records confirm
that natural gas was discovered in Iran between 6000 and 2000 BC
and that the Persians had already used it to record the “eternal fire”
symbolizing the adoration of their profound religious convictions.
Since 900 BC, natural gas has been extracted in China with the help
of bamboo poles introduced into wells 1000 m deep. However, it
was only in the year 211 BC that the raw material was used to
accelerate the drying of salt stones.
Motivating multimillionaire contracts for their commercializa-
tion, international technical cooperation and even wars between
nations that dispute their exploitation, oil and natural gas are
perceived by society in general as strategic energy inputs. Their
commercialization, whether for lubrication or energy fuels, re-
quires rigorous quality controls that impose challenges to ensure
transport with safety and environmental care. Distinct in their
composition, their transport and use require special care, specific
* Corresponding author. Petrobras Transporte S.A., Project Management, Rio de
Janeiro, 20091-060 RJ, Brazil.
E-mail address: elciooliveira@petrobras.com.br (E.C.d. Oliveira).
exploratory phases (Amui, 2010).
More specifically, as far as natural gas is concerned, diverse
factors contribute to elect it as an attractive alternative less
polluting energy: the growth in demand for energy; opportunities
created by the competitive international energy market, and the
high costs associated with the production and transport of oil. Its
use has also been disregarded by the consciousness that oil con-
sumption to generate electricity uncontestably constitutes a rele-
vant portion of the consumer's ecological footprint.
Experience has shown that the control of measurements
dsensitive to different measuring techniquesd is not a simple
task. The use of inadequate measuring methods, either for non-
compliance with critical length established in measurement
norms, incorrect installation and lack of calibration of the
equipment are among factors that severely may compromise the
quality of measurement. The reliability of measurement results
depends on strict requirements related to the installation of the
measurement system and characteristics of the flow to be
measured (level of turbulence, velocity profiles upstream of the
meter, variations in fluid properties). Inaccuracies associated with
measurements certainly compromise the confidence in the audit
processes associated with the transfer of custody, particularly in

http://dx.doi.org/10.1016/j.jngse.2014.12.008
1875-5100/© 2014 Elsevier B.V. All rights reserved.
314 E.C.d. Oliveira et al. / Journal of Natural Gas Science and Engineering 22 (2015) 313e320
the case of volume flow measurement in gas pipelines. From the
economic point of view, inaccuracies in measurements can reach
intolerable amounts as easily estimated when one considers that
enormous volumes of gas are transported (about 100 million
cubic meters of natural gas are transported daily in Brazilian
pipelines). The accuracy of measurement is dependent on the
measuring technique, the capacity of the person responsible for
the supervision, the calibration of the meter and the frequency of
the measuring process. When operated correctly, ultrasonic and
orifice plate meters are able to provide acceptable levels of
accuracy.
The growing trend for the use of cleaner fossil fuels is the result
of the global consciousness that any alternative utilized in the
generation of electricity, one way or another, has a detrimental
impact on the environment. From this perspective, natural gas
should replace other fossil fuels in the energy matrix as its com-
bustion generates lower levels of pollutant gas emissions and res-
idues. However, the rational use of natural gas depends on an
efficient process to make its safe and economic distribution viable.
Despite the advances in gas pipelines (increasingly complex and
operated in compliance with modern legislation founded on the
concepts of transport logistics and appropriate regulation), the
adequate bookkeeping for the so-called unaccounted for gas re-
quires precise measurements whose accuracy and control of asso-
ciated uncertainties still remain as metrological challenges to be
overcome (Arpino, 2014). Orifice plate, turbine or ultrasound
technology flowmeters continue to be the techniques most
employed for measurements of gas flow, accepted by regulators as
an appropriate alternative for custody transfer. The inherent
properties of natural gas (e.g. low density and high volatility of its
components) facilitate leaks and losses of volume that could result
in serious economic impacts (on the order of billions of dollars) for
the investors. This is the reason why the control of gas transported
(conservation of mass) is so critical. Oliveira and Aguiar (2009) and
Bagajewicz (2003) showed that the utilization of the technique
called Data Reconciliation (DR) dthe use of measurement redun-
dancy to reduce the uncertainty associated with measurementsd
made an important contribution to the metrological control of
unaccounted for gas in gas pipelines. Here, redundancy is the
replication of critical components of a system with the intention of
increasing its reliability.
Whenever the laws of conservation are violated, random or
systematic errors are introduced to compromise the process of
measurement (Narasimhan and Jordache, 2000; Ozyurt € and Pike,
2004). In this context, the technique of data reconciliation dan
approach based on the statistical tool that considers the restrictions
imposed on the processd does in fact improve confidence when
measurements are analyzed on a global basis. Data reconciliation
incorporates redundant information to compensate and eliminate
random errors, thereby reducing uncertainties associated with
measurement. Considering the heteroscedastic behavior associated
with measurements to be reconciled, this approach deals with
multivariate nonparametric locally weighted least squares regres-
sion (Alhaj-Dibo et al., 2008; Mene ndez et al., 1998). The data
reconciliation technique enables several redundant measurements
to reconcile experimental data in one single value, yielding the so-
called reconciled data. Oliveira and Aguiar (2009) have shown in
previous work that controlled redundancy leads to lower levels of
uncertainty when compared to what it would have been obtained
in the absence of redundant data. Data reconciliation is based on
the conjecture that if a gross error resulting from an apparent
measurement bias is present in some measurement or if a note-
worthy process leak was not accounted for in the model con-
straints; reconciled data may be very inaccurate.
Implementation of the data reconciliation technique requires
that:

The reference model be defined based on the mathematical
premise that it is perfect, with no systematic errors. In other
words, random errors are usually distributed and independent
(Mansour and Ellis, 2008). An “objective function Fob ” is utilized
to evaluate the resulting difference between the data produced
by the model and the experimental data;

The formulated “objective function Fob ” be optimized by making
use of a multivariate distribution that models this routine,
whose results are expressed based on weightings attributed to
the experimental measurements of dispersion. In this work, the
use of uncertainties associated with the measurements to sup-
port the results is proposed. In this stage it is necessary to
minimize or maximize the probability of meeting experimental
measures;

The parameters be evaluated. Ramamurthi et al. (1993) sug-
gested the evaluation by using the maximum likelihood esti-
mation (MLE).
The data reconciliation approach is then applied to calculate the
reconciled values and their associated uncertainty bands, therefore
allowing detection and exclusion of the related gross errors.
Whenever the uncertainty band falls within an unacceptable range,
the experimental data will no longer be considered as the final
results but rather used to calculate the reconciliated values,
ensuring that there will be no partial overlap of their uncertainty
bands.
The objective of this work is to apply the approach of data
reconciliation to approve the analysis of the net balance in gas
pipelines. Brazilian gas pipelines utilize the tolerance criteria of 1%
as the social indicator to evaluate the quantity of unaccounted for
gas. On making use of the data reconciliation approach, it is ex-
pected that with this indicator incorporated the index could ach-
ieve the international standard of 0.3%.
2. Methodology
Measuring loops and inherent inaccuracies associated with
computer data processing and its storage usually contribute to the
relatively high uncertainties and unacceptable tolerances associ-
ated with measurement control in pipelines. Because all product
transactions associated with gas-pipeline network depend on flow-
rate measurements, control of the associated uncertainties is
absolutely critical. Assessment of the overall performance of a
pipeline system requires control and monitoring of the energy and
mass balances and a rigorous evaluation of potential leakages. Be-
sides the economic impact associated with leaks, their control
prevents possible environmental damage and increases the avail-
ability and maintainability of the pipeline network. Generally
speaking, the operation of a gas-pipeline network is evaluated and
controlled on the basis of the operating data (e.g. volume flow rate;
internal pressure and gas temperature) allowing that pressure drop
inside the pipeline and compressibility of the natural gas be
calculated.
2.1. Data reconciliation: the concept
Data reconciliation and gross error detection is thoroughly dis-
cussed by Narasimhan and Jordache (2000). According to the au-
thors, both are achieved by exploiting the redundancy property of
measurements. The technique improves the accuracy of process
data by adjusting the measured values to satisfy process constraints
while the amount of adjustment made to the measurements is
minimized whenever the random errors in the measurements are
 E.C.d. Oliveira et al. / Journal of Natural Gas Science and Engineering 22 (2015) 313e320 315

expected to be small.
Data reconciliation is a technique for validating the measure-
ε ¼ bZ e  Z c c (3)
ments by extracting the precise and reliable information from the
gross measurements that can reflect the intrinsic state of industrial Where ε is the error between the experimental data and the
processes. The process utilizes information and mathematical reconciled values.
methods to produce a single consistent set of data capable of rep- Substituting (3) in (1):
resenting the most probable operation of the process. The tech-
nique is strongly based on the concept of redundancy to correct the  
minimum possible results of measurements aiming to satisfy re- 1 1 T 1
PðXÞ ¼
1=2 exp  2ðεÞ V ðεÞ (4)
strictions inherent to the process. An intelligent reconciliation of
ð2pÞN detV
the values converted requires that the data dusually originating
from the physical measurements that incorporate associated
errorsd are statistically validated thereby guaranteeing that the
process of reconciliation is capable of eliminating gross errors to PðεÞ ¼ PðZ e  Z c Þ (5)
produce useful and reliable data.
Gross error detection and data reconciliation convey embodied  
1 1 e c T 1 e c
techniques complementing each other. When applied together, PðXÞ ¼
1=2 exp  2ðZ  Z Þ V ðZ  Z Þ (6)
identify and eliminate gross errors, therefore improving accuracy of ð2pÞN detV
measured data. Data reconciliation and gross error detection, when
applied together, contributes for reducing error by exploiting the PðZ e Þ needs to be maximized:
redundancy property that is inherent to measurements.
The reconciliation of collected operational data minimizes the
overall uncertainty associated with the closing up of the mass and maxbPðZ e  Z c Þc ¼ Fob
2
energy balances. Satisfactory use of redundant data minimizes the 
quantity of measured variables required to obtain a solution; i.e. 6 1
¼ max4
1=2 exp
optimize the use of measured variables associated with the ð2pÞN detV
respective available variances and co-variances. The methodology 3
allows measured variables derived from redundant measured 
1 e 7
 ðZ  Z c Þ V 1 ðZ e  Z c Þ 5
T
values to be consistently used to estimate true values. From a sta- (7)
2
tistical point of view, the data reconciliation technique increases
the accuracy of single measurements; treat the data by means of a
redundant network of information that takes into account the
actual state of the pipeline system, therefore improving measure- 2 3
 
ment tolerances that may lead to substantial economic benefit. 6 1 1 7
 ðZ e  Z c Þ V 1 ðZ e  Z c Þ 5
T
Fob ¼ max4In
1=2
N 2
ð2pÞ detV
(8)
2.2. The statistical modeling for data reconciliation
 
1
Fob ¼ max  ðZ e  Z c Þ V 1 ðZ e  Z c Þ
T
The normal multivariate distribution denotes the multidimen- (9)
sional model commonly used to describe variations in experi- 2
mental data, whose probability density function P(X) may be All constants in Equations (8) and (9) are considered to be
expressed by the diagonal matrix of co-variance V: negligible.
 
1 1 T 1 j k
PðXÞ ¼
1=2 exp  ðX  mÞ V ðX  mÞ (1)
Fob ¼ max  ðZ e  Z c Þ V 1 ðZ e  Z c Þ
T
2 (10)
ð2pÞN detV
Equations (10) and (11) can be considered similar because to
Based on the following parameters: minimize Fob is the same that maximize the probability to find
3
2 2 3 2 3 experimental measures (Narasimhan and Jordache, 2000).
X1 m1 s21 s212 / s21N
6 X2 7 6 m2 7 6 s2 s22 / s22N 7 j k
X¼6
4« 5
7 m¼6
4« 5
7 V ¼6
4 21
7
« 5 Fob ¼ min ðZ e  Z c Þ V 1 ðZ e  Z c Þ
T
« « 1 (11)
XN mN s2N1 s2N2 / s2N
Rewriting equation (11) for the condition: mc1 ¼ mc2 ¼ / ¼ mcN ;
s2irepresent the variances of diagonal matrix V. that is, the reconciliation of N results of the same property m can
The probability density function reflects a region of confidence then be expressed as:
for N points on the curve when:
2 3 2 3 2 3
me1 mc1 me1  mc1
PðXÞ ¼ constant ¼ ðX  mÞ V T 1
ðX  mÞ (2) 6m 7e 6m 7c 6m  m 7
e c
Ze ¼ 6
4« 5
2 7 Zc ¼ 6
4« 5
2 7 Ze  Zc ¼ 6
4«
2 1 7
5
Only random errors are considered in this process. That is, the
meN mcN mN  mc1
e
experimental data ðX or Z e Þ and the reconciled data ðm or Z c Þ
must essentially result from the uncertainties in the experimental For non-correlated quantities, Equation (11) assumes the
data: mathematical form:
316 E.C.d. Oliveira et al. / Journal of Natural Gas Science and Engineering 22 (2015) 313e320

reconciliation concept and the reason why line packs are always
22 3T 2 2 31 2 e 33
me1  mc1 s1 0 / 0 m1  mc1 positive.
66 m  m 7 6 0
e c
0 7 6 e c 77
Fob ¼ min6 6 2 2 7 6 s22 / 7 6 m2  m2 77 Experimental uncertainties used: GTR and GTE (1.5%); GCOM
44 « 5 4 « « 1 « 5 4« 55 and DEMP (3.0%).
mN  mcN
e
0 0 / s2N meN  mcN Taking the ideal situation expressed by GPOE ¼ 0 as a basis, the
unaccounted for gas (GNC) may be calculated by (16) and (17):
(12)

2 GNC ¼ GTR  GTE  GCOM  DEMP (16)

6 e 
Fob ¼ min6 c e c e c
4 m1  m1 m2  m2 /mN  mN
GNC
GNCð%Þ ¼  100 (17)
2  2 32 33 GTR
1 s1 0 / 0 me1 mc1
6 0  76 me 77
1 s22 / 0  mc2
6
4 « 54 «
76 2 77
55 (13)
« 1 «
 2 e c
0 0 / 1 sN m N  mN

Accordingly, the objective function assumes the form:

"
2 e
2 e
2 #
me1  mc1 m2  mc2 mN  mcN
Fob ¼ min þ þ/þ 4. Results and discussion
s21 s22 s2N
(14) Table 1 describes volumes of natural gas transported in the
pipeline system during a seven-day period.
As redundancy exists in the measurements, the process model
constraints are considered by Equation (18):
3. Experimental analysis

The UrucueManaus Brazilian gas pipeline located in the GTRc ¼ 0:96  GTRc þ 0:01  CGOM c þ 0:03  DEMP c (18)
northern region of Brazil started its operation in 2009. Based on the The constraints may be described by the expressions:
experimental data collected along a seven-day period (Feb 1, to Feb
7, 2013) and taking into account that the pipeline is 662 km long
(383 km of 20 inches diameter pipe and 279 km of 18 inches GTEc ¼ 0:96  GTRc (19)
diameter pipe) and that the average volume flow rate is 4.26
million m3 per day, the data reconciliation analysis can be devel-
GCOMc ¼ 0:01  GTRc (20)
oped as follows.
Fig. 1 depicts the simplified mathematical model proposed:
The data collected were treated by means of the data reconcil- DЕМ P c ¼ 0:03  GTRc (21)
iation technique, in compliance with the balance equation (15):
General equation (14), when expressed in terms of GTRc , as-
GTR ¼ GTE þ GCOM þ GPOE þ DEMP (15) sumes the following forms:
In this expression: "
ðGTRe  GTRc Þ ðGTEe  GTEc Þ ðGCOM e  GCOM c Þ

GTR: Total gas received, measured by an orifice plate meter; Fob ¼ þ þ
s2GTRe s2GTEe s2GCOMe

GTE: Total gas delivered, measured by a four-path ultrasonic #
flow meter; DEMP e  DEMP c
þ

GCOM: Fuel gas, measured by micro turbine flow meters; s2DEMP e

GPOE: Total gas lost in the operation, estimated by equations of
(22)
state from differential pressure, pressure, temperature, diam-
eter of the pipeline and gas composition;

DEMP: Variation in line packing, assessed by comparing ma-
nometers readings.

Natural gas can be temporarily stored in the pipeline system

itself, through a process called line packing (line pack). Generally, Table 1
gas volumes can vary due to temperature and pressure changes on Volumes of natural gas.
two consecutive days. However, as discussed above, experimental GTRe (m3) GTEe (m3) GCOMe (m3) DEMPe (m3) GNC (%)
data are adapted to facilitate readers to understand the data
Day 1 2674969 2560900 25145 80249 0.32
Day 2 2677264 2570173 40000 80318 ¡0.49
Day 3 3922694 3765786 39227 112888 0.12
Day 4 3917279 3760588 19333 117518 0.51
Day 5 2755777 2645546 37001 81999 ¡0.32
Day 6 2711803 2603330 38111 81354 ¡0.41
Day 7 2676892 2569816 26212 70035 0.40

The experimental values of the unaccounted for gas that appear in bold correspond,
Fig. 1. Simplified architecture of a gas pipeline. in absolute terms, to a level that exceeds 0.3%.
 E.C.d. Oliveira et al. / Journal of Natural Gas Science and Engineering 22 (2015) 313e320 317

2 2 2 2 2 3
ðGTRe Þ  2  GTRe  GTRc þ ðGTRc Þ ðGTEe Þ  2  GTEe  GTEc þ ðGTEc Þ
6 2
þ 2
þ7
6 s e s e 7
6 GT R GT E 7
6 7
6 ðGCOMe Þ2  2  GCOMe  GCOMc þ ðGCOM c Þ2 7
6 þ 7
Fob ¼6 7 (23)
6 s 2 7
6 GCOM
e 7
6 7
6 2 2 7
4 ðDEMP Þ  2  DEMP  DEMP þ ðDEMP Þ
e e c c 5
s2 e
DEM P

2 ðGTRe Þ2  2  GTRe  GTRc þ ðGTRc Þ2 3

þ
6 s2 e 7
6 GT R 7
6 7
6 7
6 ðGTE Þ  2  GTE  0:96  GTR þ 0:96  ðGTR Þ
e 2 e c 2 c 2 7
6 þ 7
6 7
6 s2 e 7
6 GT E 7
Fob ¼ 6 7
6 2 2 7
6 ðGCOM e Þ  2  GCOMe  0:01  GTRc þ 0:012 ðGTRc Þ 7
6 þ7
6 s2 7
6 GCOM
e 7
6 7
6 7
4 ðDEMP e Þ  2  DEMP e  0:03  GTRc þ 0:032  ðGTRc Þ 5
2 2

s2 e
DEM P
(24)

2 2  GTRe þ 2  GTRc 1:92  GTEe þ 1:8432  GTRc 3

þ þ
6 s2 e s2 e 7
vFob 6 GT R GT E 7
¼6
6
7 (25)
vGTR c c7
4 0:02  GCOM e þ 0:0002  GTRc e
0:06  DEMP þ 0:0018  GTR 5
þ
s2 e s2 e
GCOM DEM P

0 12
2
B C
2 B s2GTRe C
UGTR c ¼ B  UGTRe C
@ 2 þ 1:8432 þ 0:0002 þ 0:0018 A
s2GTRe s2GTEe s2GCOMe s2DEMPe

0 12
1:92
B C
B s2GTEe C
þB  UGTEe C
@ 2 þ 1:8432 þ 0:0002 þ 0:0018 A
s2GTRe s2GTEe s2GCOMe s2DEMPe

0 12
0;02
B C
B s2GCOMe C
þB  UGCOMe C
@ 2 þ 1:8432 þ 0:0002 þ 0:0018 A
s2GTRe s2GTEe s2GCOMe s2DEMPe

2GTRe þ 1:92GTE
e
þ 0:02GCOM þ 0:06DEMP
e e
0 12
c s2GTRe s2 s2 s2
GTR ¼ GTEe GCOMe DEMP e
(26) 0;06
2 þ 1:8432 þ 0:0002 þ 0:0018 B C
s2GTRe s2GTEe s2GCOMe s2DEMPe B s2DEMPe C
þB  UDEMP eC (27)
@ 2 þ 1:8432 þ 0:0002 þ 0:0018 A
By applying the applicable concepts described in the Guide to s2GTRe 2
sGTEe 2
sGCOMe
2
sDEMPe
the expression of uncertainty in measurement (GUM, 1995), one
obtain:
318 E.C.d. Oliveira et al. / Journal of Natural Gas Science and Engineering 22 (2015) 313e320

Similarly, according as GTRc , GTEc , GCOMc , DEMP c and their

associated uncertainties, the following two expressions can be
written:

2:084GTRe þ 2GTE
e
þ 0:02GCOM þ 0:06DEMP
e e

c s2GTRe s2 s2 s2
GTE ¼ 2:17
GTEe GCOM e DEMPe
(28)
s2GTRe
þ s22 þ 0:00024
s2
þ 0:0019
s2 Fig. 2. GTR: experimental values and reconciled values.
GTEe GCOM e DEMP e

0 12
2:084
B C
2 B s2GTRe C
UGTE e ¼ B 2:17  U GTReC
@ 2 þ 22 þ 0:00024 þ 0:0019 A
s s GTRe
s2 s2 GTEe GCOMe DEMP e

0 12
2
B C
B s2GTEe C Fig. 3. GTE: experimental values and reconciled values.
þ B 2:17  UGTE eC
@ 2 þ 22 þ 0:00024 þ 0:0019 A
s s s2
GTRe
s2 GTEe GCOMe DEMP e

0 12
0:02
B C
B s2GCOMe C
þ B 2:17  UGCOM eC
@ 2 þ 22 þ 0:00024 þ 0:0019 A
s s s2
GTRe
s2 GTEe GCOMe DEMP e

0 12
0:06
B C
B C
s2DEMPe
þ B 2:17  UDEMP e C (29)
@ 2 þ 22 þ 0:00024 þ 0:0019 A
s s s2
GTRe
s 2
GTEe GCOMe DEMP e

Fig. 4. GCOM: experimental values and reconciled values.

200GTRe þ 192GTE þ 2GCOM

e
þ 6DEMP
e e

s2GTRe s2 s2 s2
GCOMc ¼ 2000
GTEe GCOMe DEMP e
(30)
s2GTRe
þ 18432
s2
þ s2 2 þ s218
GTEe GCOM e DEMP e

Table 2
Reconciled volumes of natural gas and respective associated uncertainties.

GTRc UGTRc GTEc UGTEc GCOM c UGCOMc DEMP c UDEMPc

Day 1 2655989 25201 2547996 24177 26542 252 79625 756

Day 2 2741998 26149 2630383 25085 27400 261 82199 784 Fig. 5. DEMP experimental values and reconciled values.
Day 3 3908172 37079 3749262 35571 39055 371 117164 1112
Day 4 3297510 32479 3163932 31164 32958 325 98873 974
Day 5 2810158 26742 2695794 25654 28081 267 84244 802
Day 6 2772377 26405 2659537 25331 27704 264 83111 792
Day 7 2629583 24969 2522707 23955 26278 250 78835 749
 E.C.d. Oliveira et al. / Journal of Natural Gas Science and Engineering 22 (2015) 313e320 319

Table 3 Table 4
Relative uncertainties, reconciled and minimized. New calculations for GNC based on reconciled values.

UGTRc UGTEc UGCOMc UDEMPc GTR (m3) GTE (m3) GCOM (m3) DEMP (m3) GNC (%)

0.95 0.95 0.95 0.95 Day 1 2674969 2560900 26542 80249 0.27
0.95 0.95 0.95 0.95 Day 2 2677264 2570173 27400 80318 ¡0.02
0.95 0.95 0.95 0.95 Day 3 3922694 3765786 39227 112888 0.12
0.98 0.98 0.98 0.98 Day 4 3297510 3163932 32958 98873 0.05
0.95 0.95 0.95 0.95 Day 5 2755777 2645546 28081 81999 0.01
0.95 0.95 0.95 0.95 Day 6 2711803 2603330 27704 81354 ¡0.02
0.95 0.95 0.95 0.95 Day 7 2676892 2569816 26212 78835 0.08

while those that result from operational measurement were

0 12
reduced by a factor of three.
200
B C As may be seen, there is no overlap, even partially, between the
2 B s2GTRe C
UGCOM c ¼ B  UGTRe C experimental values and the reconciled values (highlighted in black
@20000 þ 18432 þ 2 þ s218 A
2
sGTRe2
sGTEe s2GCOMe DEMP e circle), as demonstrated in Figs. 2e5. This behavior generates
doubts in the experimental results; gross errors are suggested,
0 12
beyond the expected random errors.
192
B C After substituting those values considered suspect with those
B s2 e C
þ B20000 18432 GTE  UGTEe C reconciled (shown in bold), the unaccounted for gas may then be
@ 2 þ 2 þ 2 þ s218 A
sGTRe sGTEe s2GCOMe DEMPe recalculated (data in Table 4). As can be observed, the application of
the data reconciliation approach was capable of bringing the values
0 12
of GNC within the criteria proposed as a minimum limit of 0.3%,
2
B C which confirms that the objectives proposed in making use of the
B s2 C
þ B20000 18432GCOM 2  UGCOMe C
e

@ 2 þ 2 þ 2 18 A data reconciliation technique to increase the confidence in mea-

sGTRe sGTEe sGCOMe
þ s2
DEMPe surements were fully achieved.
0 12
6 5. Conclusions
B C
B s2 e C
þ B20000 18432DEMP 2  UDEMP e C (31)
@ 2 þ 2 þ 2 þ s218 A As shown, detection of errors in a multivariable system is not an
sGTRe sGTEe sGCOMe DEMPe
easy task even when the source of the abnormal values is known. If
a gross error that may result from an extreme observation is not
evidently defensible, discard of unacceptable values must always
66:667GTRe þ 64GTE
e
þ 0:6667GCOM þ 2DEMP
e e

s2GTRe s2 s2 s2 be based on statistical analysis. The data reconciliation approach

GDEMP c ¼ 2217:78
GTEe GCOMe DEMP e
(32) used depicts the process to identify and classify similar clusters of
s2GTRe
þ 2048
s2
þ s0:222
2 þ s2 2
GTEe GCOM e DEMPe inputs and outputs as a strategy to eliminate gross errors inherently
imbedded in the results. The analysis undertaken considered a real
0 12 set of seven-day real data collected at the Urucu-Manaus pipeline
66:667
B C to illustrate that the reconciliation data approach was robust
2 B s2GTRe C
UDEMP c ¼ B  UGTRe C enough to identify and exclude gross errors that eventually has a
@2217:78 þ 2048 þ 0:222 þ s2 2 A
2
sGTRe 2
sGTEe s2GCOMe DEMP e
detrimental effect in measurement systems. The study also
confirmed that the data reconciliation results do not depend on
0 12 data only, but also on the process architecture.
64
B C Contrasting to the common perception, this paper highlights
B s2GTEe C
þ B2217:78 2048  UGTEe C that measurement redundancy should not be perceived as an
@ 2 þ 2 þ 0:222 þ s2 2 A
sGTRe sGTEe s2GCOMe DEMP e
additional cost, but rather as an attractive strategy for improving
system accuracy. Redundant data, even of lower accuracy, may
0 12 satisfactorily be used to enhance the overall accuracy by generating
0:6667
B C reconciled values, therefore confirming that the correct use of the
B s2GCOMe C
þ B2217:78 2048  UGCOMe C reconciliation methodology may conveniently replace the need for
@ 2 þ 2 þ 0:222 þ s2 2 A
sGTRe sGTEe s2GCOMe DEMP e
additional experimental work.
The application of the data reconciliation approach proved to be
0 12 a useful and powerful tool to evaluate the unaccounted for gas,
2
B C therefore capable of explaining deviations associated with net
B s2DEMPe C
þ B2217:78 2048 0:222  UDEMP e C (33) balances in a natural gas pipeline. Within best international prac-
@ 2 þ s2 þ s2 þ s2 2 A
s GTRe GTEe GCOMe DEMP e
tices, the confidence level associated with gas pipeline metrology
confirm that 0.3% seems to be a realistic criteria to express the
The data in Table 2 and the data in Figs. 2 to 5 describe the unaccounted for gas in natural gas pipelines.
volumes reconciled and the respective uncertainties associated
with these volumes (all expressed in m3). References
Based on the results of Table 3 and on the data from the figures
that follow, it is observed that the reconciled uncertainties are Alhaj-Dibo, M., et al., 2008. Data reconciliation: a robust approach using a
contaminated distribution. Control Eng. Pract. 16, 159e170.
minimized. Note that the uncertainties derived from the custody Amui, S., 2010. Oil and Gas Executives to: Exploration Areas, Drilling and Comple-
transfer measurements were reduced from by more than 50%, tion of Wells and Production of Hydrocarbons. Intercie^ncia, Rio de Janeiro.
320 E.C.d. Oliveira et al. / Journal of Natural Gas Science and Engineering 22 (2015) 313e320
Arpino, F., Dell’Isola, M., Ficco, G., Vigo, P., 2014. Unaccounted for gas in natural gas
transmission networks: prediction model and analysis of the solutions. J. Nat.
Gas. Sci. Eng. 17, 58e70.
Bagajewicz, M.J., Cabrera, A.E., 2003. Data reconciliation in gas pipeline systems.
Ind. Eng. Chem. Res. 42 (22), 5596e5606.
Guide to the Expression of Uncertainty in Measurement (GUM), 1995. ISO.
Mansour, M., Ellis, J.E., 2008. Methodology of on-line optimization applied to a
chemical reactor. Appl. Math. Model 32, 170e184.
Mene ndez, A., Biscarri, F., Gomez, A., 1998. Balance equations estimation with bad
measurements detection in a water supply net. Flow. Meas. Instrum. 9,
193e198.
Narasimhan, S., Jordache, C., 2000. Data Reconciliation and Gross Error Detection:
an Intelligent Use of Process Data. Gulf Publishing Company, Houston.
Oliveira, E.C., Aguiar, P.F., 2009. Data reconciliation in the natural gas industry:
analytical applications. Energy Fuel 23, 3658e3664.
€
Ozyurt, D.B., Pike, R.W., 2004. Theory and practice of simultaneous data reconcili-
ation and gross error detection for chemical processes. Comput. Chem. Eng. 28,
381e402.
Ramamurthi, Y., et al., 1993. Control-relevant dynamic data reconciliation and
parameter estimation. Comput. Chem. Eng. 17, 41e59.
 Fuel 300 (2021) 120936

Contents lists available at ScienceDirect

Fuel
journal homepage: www.elsevier.com/locate/fuel

Full Length Article

Data reconciliation applied to the conformity assessment of fuel products

Elcio Cruz de Oliveira a, b, *, Felipe Rebello Lourenço c
a
Posgraduate Metrology Programme, Rio de Janeiro Catholic University, Marquês de São Vicente Street, 225, Gávea, Rio de C, RJ, Brazil
b
Technology Management, PETROBRAS TRANSPORTE S.A, Rio de Janeiro, RJ, Brazil
c
Programa de Pós-Graduação em Fármaco e Medicamentos, Departamento de Farmácia, Faculdade de Ciências Farmacêuticas, Universidade de São Paulo, Brazil

A R T I C L E I N F O A B S T R A C T

Keywords: The discrepant measurement results regarding the conformity assessment of diesel and gasoline fuels may lead to
Data reconciliation commercial dispute between producers (refinery) and consumers (storage terminal). The aim of this work was to
Conformity assessment apply data reconciliation in order to solve the discrepancy between producers’ and consumers’ measurements
Measurement uncertainty
and allow to make decisions regarding the conformity/non-conformity assessment. Diesel and gasoline samples
Risk of false decisions
were analyzed in respect to total sulfur content and flash point and for sulfur content and final boiling point,
Diesel and gasoline
Producers (refinery) respectively. Discrepant results between consumers’ and producers’ measurements were subject to data recon­
Consumers (storage terminal) ciliation. Data reconciliation was perform using a MS Excel spreadsheet. The sheet was divided into five sections:
‘Specification limits’, ‘Producer’s measurement’, ‘Consumer’s measurement’, ‘Reconciled measurement’, and
‘Histograms’. Data reconciliation was useful to solve the discrepancy between producer’s and consumer’s
measurements in most of the studied cases, since provided reconciled values inversely weighted as function of
measurement uncertainty. In addition, data reconciliation provided a reduced measurement uncertainty.

1. Introduction Therefore, non-compliance with the specifications indicated by the

Diesel oil and gasoline are petroleum-derived liquid fuels that are
widely used worldwide, as they meet the demands of cargo and pas­
senger transport, industry, power generation, construction machinery
and agricultural machinery, locomotives and vehicles for private use.
Brazilian Agency for Petroleum, Natural Gas and Biofuels (ANP) is
the body responsible for the inspection of the quality of automotive fuels
in the country and indicates both the test methods used and the speci­
fication limits allowed for their commercialization. For the S-10 diesel,
considering sulfur mass fraction and flash point, the specification limits
are 10 mg kg− 1 (maximum) and 38 ◦ C (minimum), respectively, estab­
lished by Resolution ANP n◦ 50, of 23 December 2013. For the gasoline,
considering mass fraction of sulfur and final boiling point, the maximum
specification limits are 50 mg kg− 1 and 215 ◦ C, respectively, established
by ANP Resolution 807 of January 23, 2020.
Knowledge about the sulfur content of combustible materials is
essential for process control, corrosion and environmental issues. Flash
point brings information concerning shipping, transport and safety
regulations to define flammable and fuel products. Final boiling point
(FBP) assesses the formation of excessive deposits in the combustion
chamber, formation of varnishes and sludge in the engine.
ANP can make a fuel not in accordance for commercialization; which
highlights the need for the methodology for assessing conformity to the
specification to be very robust; otherwise, it may lead to non-compliance
with regulations without taking into consideration possible environ­
mental damage, in case of false conformity decisions [1].
In Brazil, one of the main transportation modals of fuel products
between terminals and refineries is by pipelines. This transport logistics
chain can generate disputes between the involved parties, mainly when
the test results do not converge in relation to the conformity assessment.
Data Reconciliation (DR) system has the role of ensuring data con­
sistency, using the redundancy of the measured variables and a statis­
tical measurement model to increase the reliability of the reconciled
result, in addition to reducing measurement uncertainty.
Analytical applications, using DR, related to natural gas matrix
showed to be suitable in order to solve problems related to a charac­
terization study, a dispute between supplier and customer, an operation
diagnosis in industrial unit, and an among day homogeneity study [2].
Still in natural gas industry, DR proved to be very effective in
reducing measurement uncertainty from 1% to 0.3% associated with
the accountability of the unaccounted the analysis of the net balance
in gas pipelines [3].

* Corresponding author at: Rio de Janeiro Catholic University, Marquês de São Vicente Street, 225, Gávea, 22453-900 Rio de Janeiro, RJ, Brazil.
E-mail address: elciooliveira@puc-rio.br (E.C. de Oliveira).

https://doi.org/10.1016/j.fuel.2021.120936
Received 23 December 2020; Received in revised form 21 April 2021; Accepted 23 April 2021
0016-2361/© 2021 Elsevier Ltd. All rights reserved.
E.C. de Oliveira and F.R. Lourenço Fuel 300 (2021) 120936

A simulation tool program package for a mixed alcohol synthesis parameters of the fuels are within and out the tolerance limits,
reactor applies the method of least squares for data reconciliation. The respectively.
results showed that the established model was very accurate [4]. The particular consumers’ or producers’ specific risks for diesel and
A techno-economic analysis of a CO2 capture process using fixed bed gasoline analysis were estimated based on measured values and their
reactors with a microencapsulated solvent was carried out and the respective uncertainties, assuming normal distribution [1]. Calculations
optimization step was solved by DR approach, by means of a sequential were implemented in a MS Excel spreadsheet available as supplemental
quadratic programming in aspen custom modeler [5]. material.
A Generalized Least Squares (GLS) based optimal reconciliation In this study, the concept of guard bands is used in order to provide
method for hierarchical time series was proposed to obtain the distribute acceptance/rejection limits that guarantee reduced risks of false con­
reconciliation in day-ahead wind power forecasting following aggregate formity decisions [16]. A guard band (g) is a multiple of standard un­
consistency constraints [6]. certainty (u) that is added to or subtracted from the specification limits
A data reconciliation model, using robust estimators, based on (lower and/or upper specification limits, LSL and/or USL, respectively)
cooperative working equations applied to aeroengines was proposed. in order to minimize the producer’ (LSL − g and/or USL + g) or con­
The results show fair performance and that the errors of the measured sumers’ (LSL +g and/or USL − g) risks. Assuming that the distribution of
and unmeasured data can be reduced by means of DR, without being the likely values of the measurand is approximately normal, a value of
affect by gross errors [7]. g = 1.64u will provide a risk of false conformity decisions of up to 5%.
DR was applied to correct mass balance deviations by means the
minimization of the experimental error in liquid–liquid equilibrium 4. Data reconciliation
prediction of ternary systems related to biodiesel production process
[8]. Since that, both random and systematic errors are found in all pro­
The technique of data validation and reconciliation (DVR), a tech­ cess measurements, probably, the conservation laws are disobeyed. DR
nology that uses mathematical models to process information, was is an excellent alternative in order to adjust two or more experimental
conducted to distributed control systems of thermal processes. The re­ redundant measurements to a unique value, called here as reconciled
sults show DVR was able of detecting the stability in industrial thermal value, based on restrictions imposed on the process [17].
processes [9]. This reconciled value, due to the fact of depending on the relation­
DVR was proposed as an alternative allocation method for error ships between process variables, represents more accurate the “true
correction and gross error detection in multiphase production systems. value”; moreover, DR approach minimizes the experimental errors.
The results proved that DVR is more robust than the traditional methods Considering that the multivariate normal models experimental data
and that is an excellent alternative to oil & gas industry, whose use is still fluctuations, the probability density function (PDF) can be defined by
very limited in this area [10]. Eq. (1):
Recently, Brazilian researchers presented a methodology for deter­ [ ]
mining the tolerance limits of critical properties in some fuels, based on P(X) = (
1 1
exp − (me − mc )T V − 1 (me − mc ) (1)
N )1/2
the reproducibility, on the difference between the averages of all (2πdetV) 2
acceptable results of the consumer and producer and on the specifica­
With the following parameters:
tions derived from the Brazilian National Agency for Oil, Natural Gas
and Biofuels [11]. However, they did not use the DR concept nor ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
particular consumers’ or producers’ specific risks based on band guard me1 mc1 2 2 2
⎢ ⎥ ⎢ ⎥ ⎢ U11 U12 ⋯ U1N ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
bands. ⎢ me ⎥
⎢ ⎥
⎢ mc ⎥
⎢ ⎥
⎢
⎢
2
U21 2
U12 ⋯ 2
U2N ⎥
⎥
The aim of this study is to apply the DR approach in order to evaluate me = ⎢ 2 ⎥mc = ⎢ 2 ⎥V = ⎢ 2 ⎥
⎢ ⋮ ⎥ ⎢ ⋮ ⎥ ⎢ Ui12 Ui22 ⋱ UiN ⎥
to the conformity to specifications of some physicochemical properties ⎢
⎣ e ⎦
⎥ ⎢
⎣ c ⎦
⎥ ⎢
⎣ 2 2 2
⎥
⎦
of fuel products. Here, both the refinery and storage terminal test results mN mN UN1 UN2 ⋯ UNN
are taken into account in order to obtain a unique reconciled value that
is more reliable than each individual ones; moreover, the measurement Where, me are the experimental values and mc is the reconciled value
uncertainty is minimized, which establishes acceptance limits (guard the variance–covariance matrix is based on the measurement uncer­
bands) that guarantee a reduced risk of false conformity decisions. tainty, U.
N points of equal probability density are found in the region of
2. Materials and methods greatest reliability, Eq. (2):

P(X) = constant = (me − mc )T V − 1 (me − mc ) (2)

2.1. Diesel and gasoline analysis
Based on DR premises, the bias between m and m is calculated by
e c

Diesel S-10 (called as “diesel”) samples were analyzed for total sulfur Eq. (3):
content [12] (high tolerance limit of 10 mg/kg) and flash point [13]
Bias = ε = [me − mc ] (3)
(low tolerance limit of 38 ◦ C). Gasoline-ethanol blend 26% v/v (called as
“gasoline”) samples were tested for sulfur content [14] (high tolerance Therefore, Eq. (1) becomes Eq. (4):
limit of 50 mg/kg) and final boiling point [15] (high tolerance limit of [ ]
1 1 T −1
215 ◦ C). All diesel and gasoline samples were analyzed by both refinery P(X) = ( ) exp − (ε ) V (ε) (4)
1/2 2
(Brazilian oil and gas industry - called as “producer”) and also storage (2πdetV)N
terminal (called as “consumer”). Discrepant results between consumers’
In the objective function, Fob , P(me ) must be maximum, i.e., P(ε)
and producers’ measurements were subject to data reconciliation.
must be the greatest value. In other words, Eqs. (5) to (9):
3. Consumers’ or producers’ risk estimation max[P(ε)] = Fob
⎡ ⎤
[ ]
Here, the consumer risk is considered as approval of an out of = max⎣(
1 1 e c T − 1
)1/2 exp − 2(m − m ) V (m − m ) ⎦
e c

specification fuel, the producer risk as rejection of a fuel within speci­ (2πdetV)N
fication and the particular specific consumers’ and producers’ risk are (5)
previously estimated when the measured values of the physicochemical

2
E.C. de Oliveira and F.R. Lourenço Fuel 300 (2021) 120936

Fig. 1. Producers’ and consumers’ measurement (measured values and respective expanded uncertainties) of total sulfur content [12] (high tolerance limit of 10 mg/
kg) and flash point [13] (low tolerance limit of 38 ◦ C) for diesel analysis.

⎡ ⎤ ⎡⎛⎡
[ ] ⎤ ⎞T ⎡ ⎤− 1 ⎡ ⎤⎤
1 1 e
Fob = max⎣ln ( )1/2
c T − 1
− (m − m ) V (m − m ) ⎦
e c
(6) ⎢⎜⎢ me1 c
− m ⎥⎟ ⎢ 2
U11 2
U12 ⋯ 2
U1N ⎥ ⎢ me1
− m ⎥⎥c
(2πdetV)N 2 ⎢⎜⎢ ⎥⎟ ⎢ ⎥ ⎢ ⎥⎥
⎢⎜⎢ e ⎢ me − mc ⎥⎥
⎢⎜⎢ m2 − mc ⎥ ⎟ ⎢
⎥⎟ ⎢
2
U21 2
U22 ⋯ 2
U2N ⎥
⎥ ⎢ 2 ⎥⎥
[ ] Fob = min⎢ ⎜
⎢⎜⎢⎢ ⎥⎟ ⎢ ⎥ ⎢ ⎥⎥
⎢⎜⎢ ⎥ ⎟ ⎢ Ui12 Ui22 2
⋱ UiN ⎥ ⎢ ⎥⎥
1 ⋮ ⎥⎟ ⎢ ⎥ ⎢ ⋮ ⎥⎥
Fob = max − (me − mc )T V − 1 (me − mc ) (7) ⎢⎝⎣
⎢ e c ⎦⎠ ⎣ 2 2 2 ⎦ ⎣ e ⎥
c ⎦⎥
2 ⎢ mN − m UN1 UN2 ⋯ UNN mN − m ⎥
⎢ ⎥
[ ]
Fob = max − (me − mc )T V − 1 (me − mc ) (8) (9)
⌈ ⌉ In this study, since there are no correlations between the input
Fob = min (me − mc )T V − 1 (me − mc ) (9) quantities, because the quantities are independent, the elements off the
After the reference model (process model) is chosen, mcproducer = main diagonal are zero in the covariance-variance matrix, V, Eq. (10):
mcconsumer , the
objective function, Fob, is optimized by a weighed routine.

⎡ ⎡ ⎤ ⎤
⎡ ⎤
⎢ ⎢ 1/U 2 0 ⋯ 0 ⎥ me − mc ⎥
⎢ ⎢ 11 ⎥⎢ 1 ⎥⎥
⎢ ⎢ ⎥⎢ ⎥⎥
⎢[ ]⎢ 0 2
1/U22 ⋯ 0 ⎥⎢ me − mc ⎥⎥
⎢ ⎢ ⎥⎢ 2 ⎥⎥
Fob = min⎢ me1 − mc me2 − mc ⋯ meN − mc ⎢ ⎥⎢ ⎥⎥ (10)
⎢ ⎢ ⋮ ⋮ ⋱ ⋮ ⎥⎢ ⎥⎥
⎢ ⎢ ⎥⎢ ⋮ ⎥⎥
⎢ ⎢ ⎥⎣ e c ⎦⎥
⎢ ⎣ 0 0 ⋯ 2
1/UNN ⎦ mN − m ⎥
⎢ ⎥
⎢ ⎥

Here, there is a single reconciled value for the same measurand, mc1 =
mc2 = ⋯ = mcN , Eq. (9) becomes Eq. (10), considering the quantities are Algebraically, Fob becomes from Eqs. (11) to (13), reaching the
not correlated: reconciled value:
⎡ ⎤ ⎡ ⎤ [( )2 ( e )2 ( e )2 ]
me1 − mc m2 − mc mN − mc
e
⎢ m1 ⎥
c
⎢ m1 = m
c
⎥ Fob = min 2
+ 2
+⋯+ 2
(11)
⎢ ⎥ ⎢ ⎥ U11 U22 UNN
⎢ me ⎥ ⎢ mc = mc ⎥
⎢ ⎥ ⎢ ⎥
me = ⎢ 2 ⎥mc = ⎢ 2 ⎥
⎢ ⋮ ⎥ ⎢ ⋮ ⎥ ∂Fob
⎢ ⎥ ⎢ ⎥ = minFob
⎣ e ⎦ ⎣ c ⎦ ∂mc
mN mN = mc [ ( ) ( ) ( )]
− 2 × me1 − mc 2 × me2 − mc 2 × meN − mc
= 2
− 2
− ⋯− 2
=0
U11 U22 UNN
(12)

3
E.C. de Oliveira and F.R. Lourenço Fuel 300 (2021) 120936

Fig. 2. Producers’ and consumers’ measurement (measured values and respective expanded uncertainties) of sulfur content [14] (high tolerance limit of 50 mg/kg)
and final boiling point [15] (high tolerance limit of 215 ◦ C) for gasoline analysis.

me1 me me ∑ mei Since the expanded uncertainties of all measurands studied here are
+ U22 + ⋯U2N
m = c
2
U11 22 NN
=∑
Ui2
(13) derived from many standard uncertainty sources, the degrees of freedom
1
2
U11
+ U12 + ⋯ + U12 1
Ui2 associated with them reach the infinite; otherwise, one recommends
using the standard uncertainty for compensating different degrees of
22 NN

The reconciled expanded uncertainty, Umc , can be analytically freedom.

calculated based on the Law of Propagation of Uncertainty, Eqs. (14) and Brazilian Agency for Petroleum, Natural Gas and Biofuels accepts
(15): different test methods to analyse the sulfur mass fraction of fuels prod­

⎛ ⎞2 ⎛ ⎞2 ⎛ ⎞2
1 1 1
⎜ 2 ⎟ ⎜ 2 ⎟ ⎜ 2 ⎟
(14)
U11 U22 UNN
Um2 c = ⎜
⎝ 1
× ume1 ⎟ ⎜
⎠ +⎝ × ume2 ⎟ ⎜
⎠ +⋯+⎝ × umeN ⎟
⎠
2
U11
+ U12 + ⋯ + U12 1
2
U11
+ U12 + ⋯ + U12 1
2
U11
+ U12 + ⋯ + U12
22 NN 22 NN 22 NN

ucts. Generally, the producer uses ASTM D 5453 (Ultraviolet Fluores­

cence) and the consumer uses ASTM D7039 (Monochromatic
√̅̅̅̅̅̅̅̅̅̅ Wavelength Dispersive X-Ray Fluorescence Spectrometry). Although
U mc = ∑ 1
1
(15) Brazilian Regulation lets both test methods, there is a bias between them
U 2i [18].
Seventeen diesel samples showed discrepant producers’ and con­
Here, the reconciled value, Eq. (16), and its reconciled expanded
sumers’ measurements for total sulfur content [12] (high tolerance limit
uncertainty, Eq. (17), depend on exclusively of the producer (pro) and
of 10 mg/kg). Considering the analyses of flash point [13] (low toler­
consumer (con) values and their respectively uncertainties:
ance limit of 38 ◦ C), discrepant producers’ and consumers’ measure­
∑ mei mepro e
+ mU2con ments were obtained for 32 diesel samples. A summary of producers’
and consumers’ measurements of diesel samples are showed in Fig. 1.
U2i U2pro
m =∑c
1
= 1
con
(16)
U2i U2pro
+ U12 Seventeen and twelve gasoline samples showed discrepant pro­
con

√̅̅̅̅̅̅̅̅̅̅ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ ducers’ and consumers’ measurements for sulfur content [14] (high
1 1 tolerance limit of 50 mg/kg) and final boiling point [15] (high tolerance
Umc = ∑1 = (17)
U 2 U
1
2 + U2
1 limit of 215 ◦ C), respectively. Producers’ and consumers’ measurements
of gasoline samples for sulfur content and final boiling point are showed
i pro con

in Fig. 2.
5. Results and discussion
Producers’ and consumers’ measurements were metrologically
compared, using the following equation: z =
Firstly, the samples are analysed by the producer and sent by a ⃒
⃒ e
⃒/√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
e ⃒
pipeline to the consumer. Both labs are certified by ISO 9000 and ⃒mpro − mcon ⃒ umpro
e 2 + umcon e 2 . Considering all the 78 pair of producers’
accredited by ISO 17025. and consumers’ measurements, based on criteria ISO 17043 [19], 67%

4
E.C. de Oliveira and F.R. Lourenço
Fig. 3. Histograms for producer’s measurement, consumer’s measurement and reconciled measurement, specification and guard band limits for total sulfur content
in diesel. Legend: x-axis in the unit of measured value and y-axis as relative frequency.
(52 pair of measurements) were consistent (z ≤ 2.0), 13% (10 pair of
measurements) were questionable (2.0 < z ≤ 3.0), and 20% (16 pair of
measurements) were inconsistent (z > 3.0).
In addition, producers’ and consumers’ measurements were metro­
logically compar67ed to the reconciliated measurements, using the
⃒ ⃒/√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
equation: z = ⃒me − mc ⃒
i ume 2 + umc 2 , where i = producer (pro) or
i
consumer (con). About 78% (61 measurements) of producers’ mea­
surements were consistent (z ≤ 2.0), 5% (4 measurements) were ques­
tionable (2.0 < z ≤ 3.0), and 17% (13 measurements) were inconsistent
(z > 3.0). Likewise, about 86% (67 measurements) of consumers’ mea­
surements were consistent (z ≤ 2.0), 6% (5 measurements) were ques­
tionable (2.0 < z ≤ 3.0), and 8% (6 measurements) were inconsistent
(z > 3.0) [19].
Particular specific consumers’ or producers’ risks for diesel and
gasoline analysis were estimated for all producers’ and consumers’
measurements, using a MS Excel spreadsheet. Regarding the diesel
analysis, all producers’ measurements were within the tolerance limits
for total sulfur content [12] (high tolerance limit of 10 mg/kg) and flash
point [13] (low tolerance limit of 38 ◦ C). On the other hand, all con­
sumers’ measurements were found to be out of tolerance limits for both
total sulfur content and flash point. Consumers’ and producers’ risks for
total sulfur content in diesel samples were found from 0.0 to 8.0% and
from 0.0 to 35.7%, respectively. For flash point analysis of diesel, the
consumers’ and producers’ risks were found to be between 0.0 and
50.0% and between 0.0 and 49.8%, respectively.
For gasoline analysis, all producers’ measurements were within the
tolerance limits for sulfur content [14] (high tolerance limit of 50 mg/
kg) and final boiling point [15] (high tolerance limit of 215 ◦ C). On the
other hand, all the consumers’ measurements for both sulfur content and
final boiling point were out of the tolerance limits. Consumers’ and
producers’ risks for sulfur content in gasoline samples were found to be
between 0.0 and 39.8% and between 0.8 and 50.0%, respectively. For
5
Fuel 300 (2021) 120936
final boiling point analysis of gasoline, the consumers’ and producers’
risks were found to be from 1.4 to 8.7% and from 22.8 to 50.0%.
The discrepant measurement results lead to a commercial dispute
between producers (refinery) and consumers (storage terminal). More­
over, the increased consumers’ and producers’ risks do not allow to
make decision for conformity or non-conformity of diesel and gasoline
samples. Thus, data reconciliation may be helpful since it takes into
account both producers’ and consumers’ measurements. In addition,
data reconciliation provides a reduced measurement uncertainty asso­
ciated to final result, reconciled value.
Data reconciliation was perform using a MS Excel spreadsheet. The
sheet was divided into five sections: ‘Specification limits’, ‘Producer’s
measurement’, ‘Consumer’s measurement’, ‘Reconciled measurement’,
and ‘Histograms’.
In section ‘Specification limits’, it should be provided the lower and
upper specification limits (in cells C8 and E8, respectively), if applicable,
the specification limit(s) type (in cell G8 – ‘Interval limits’, ‘Minimum
limit’, or ‘Maximum limit’), and the type of Guard band to be used as
decision rule (in cell G11 – ‘Consumer’s risk’, ‘Producer’s risk’, or
‘Shared risk’). In addition, a comparison of producer’s and consumer’s
results regarding the conformity/non-conformity assessment is provided
in cell C11 (‘Coincident results’ or ‘Discrepant results’).
Measured values (cells C17 and C29), expanded uncertainty (cells
E17 and E29), units (cells G17 and G29), and coverage factors (cells I17
and I29) for both procedure’s and consumer’s measurements should be
provided into sections ‘Producer’s measurement’ and ‘Consumer’s
measurement’. Based on measured values and the respective expanded
uncertainty, risk type and risk values are provided in cells C20 and F20
for ‘producer’s measurement’ and C32 and F32 for ‘consumer’s mea­
surement’). In addition, the ‘acceptance/rejection/specification’ lower
and/or upper limits (cells C23 and E23 for ‘producer’s measurement’
and C35 and E35 for ‘consumer’s measurement’) are calculated ac­
cording to the type of guard band (cell G20 and G32, for ‘consumer’s
E.C. de Oliveira and F.R. Lourenço
Fig. 4. Reconciled measurement (reconciled measured values and respective
reconciled expanded uncertainties) of total sulfur content [12] (high tolerance
limit of 10 mg/kg) and flash point [13] (low tolerance limit of 38 ◦ C) for
diesel analysis.
measurement’ and ‘producer’s measurement’, respectively). When the
measured value is within the limits, then the consumer’s risk is esti­
mated. On the other hand, when the measured value is out of the
specification, them the producer’s risk is estimated. The ‘acceptance/
rejection/specification’ lower and/or upper limits are calculated as the
specification limit values minus or plus the guard band width (L =
1.64 × u). Guard band width is zero (L = 0) when ‘shared risk’ guard
band choice is selected, which may lead to risk of false decisions up to
50%. Finally, conformity statement is provided in cells G23 and G35
(‘Conform to specification’ or ‘Not conform to specification’).
In section ‘Reconciled measurement’, reconciled measurement and
reconciled expanded uncertainty are calculated and provided in cells
C41 and E41, respectively. Measurement unit and coverage factor are
provided in cells G41 and I41, respectively. Based on reconciled mea­
surement and reconciled expanded uncertainty, risk type and risk value
are provided in cells C44 and F44, respectively. The ‘acceptance/rejec­
tion/specification’ lower and/or upper limits (cells C47 and E47,
respectively) are calculated according to the type of guard band (cell
G44). Reconciled conformity statement (‘Conform to specification’ or
‘Not conform to specification’) is provided in cell G47.
Histograms for producer’s measurement, consumer’s and reconciled
measurements, as well as the specification and guard band limits, are
provided in ‘Histogram’ section.
An example of discrepant results obtained for total sulfur content in
diesel was used to validate the MS Excel spreadsheet. Producer’s and
consumer’s measurements were found to be (7.7 ± 0.7) mg/kg (k = 2)
and (10.5 ± 0.5) mg/kg (k = 2), respectively. Considering the upper
specification limit (10 mg/kg) for total sulfur content in diesel, one
concludes that the producer’s measurement is conforming to specifica­
tion while the consumer’s measurement is not conforming to specifi­
cation (discrepant results). The consumer’s risk for producer’s
measurement was found to be 0.0%, while the producer’s risk for
6
Fuel 300 (2021) 120936
Fig. 5. Reconciled measurement (reconciled measured values and respective
reconciled expanded uncertainties) of sulfur content [14] (high tolerance limit
of 50 mg/kg) and final boiling point [15] (high tolerance limit of 215 ◦ C) for
gasoline analysis.
consumer’s risk measurement was found to be 1.3%. Applying DR
approach, the reconciled measurement was found to be (9.6 ± 0.4) mg/
kg (k = 2). Since the reconciled value represents both the producer’s and
consumer’s values, a shared risk decision rule was assumed. The con­
sumer’s risk estimated for reconciled measurement was found to be
4.1%. Thus, one concludes that reconciled measurement is conforming
to specification limits for total sulfur content in diesel. The histograms
for producer’s measurement, consumer’s measurement, and reconciled
measurement, as well as the specification and guard band limits, are
provided in Fig. 3.
A shared risk decision rule [20] was assumed for conformity
assessment of diesel and gasoline. Considering the reconciled values, 11
(65%) diesel samples comply to the tolerance limit for total sulfur
content samples, while 6 (35%) samples do not comply. Twenty (62%)
diesel samples comply to the tolerance limit for flash point and 12 (38%)
samples does not complies to the tolerance limits, based on the recon­
ciled values. Regarding the reconciled values for gasoline analysis, 4
(33%) gasoline samples comply to the tolerance limit for sulfur content,
while 8 (67%) samples do not comply. All 17 (100%) gasoline samples
comply to the tolerance limit for final boiling point, considering the
reconciled values. Reconciled measured values and respective expanded
uncertainties for diesel and gasoline analysis are showed in Figs. 4 and 5,
respectively.
6. Conclusions
DR associated to the guard band concept proved a good approach to
evaluate the conformity of fuel products.
Data reconciliation was useful to solve the discrepancy between
producer’s and consumer’s measurements in most of the studied cases,
since provided reconciled values inversely weighted as function of
measurement uncertainty. In addition, data reconciliation provided a
reduced measurement uncertainty.
E.C. de Oliveira and F.R. Lourenço
CRediT authorship contribution statement
Elcio Cruz de Oliveira: Conceptualization, Data curation, Formal
analysis, Funding acquisition, Investigation, Methodology, Project
administration, Resources, Supervision, Validation, Visualization, Soft­
ware, Writing - original draft, Writing - review & editing. Felipe Rebello
Lourenço: Conceptualization, Data curation, Formal analysis, Funding
acquisition, Investigation, Methodology, Project administration, Re­
sources, Supervision, Validation, Visualization, Software, Writing -
original draft, Writing - review & editing.
Declaration of Competing Interest
The authors declare that they have no known competing financial
interests or personal relationships that could have appeared to influence
the work reported in this paper.
Acknowledgments
This work was supported by FAPESP – Fundação de Amparo à Pes­
quisa do Estado de São Paulo (Brazil) (2019/16206-3). This study was
financed in part by the Coordenação de Aperfeiçoamento de Pessoal de
Nível Superior – Brasil (CAPES) – Finance Code 001.
Appendix A. Supplementary data
Supplementary data to this article can be found online at https://doi.
org/10.1016/j.fuel.2021.120936.
References
[1] Oliveira EC, Lourenço FR. Chemosphere, 2021, 263, 128265 (https://doi.org/
10.1016/j.chemosphere.2020.128265).
[2] Oliveira EC, Aguiar PF. Data reconciliation in the natural gas industry: analytical
applications. Energy Fuels 2009;23(7):3658–64. https://doi.org/10.1021/
ef9001428.
[3] Oliveira EC, Frota MN, Barreto GO. Use of data reconciliation: a strategy for
improving the accuracy in gas flow measurements. J Nat Gas Sci Eng 2015;22:
313–20. https://doi.org/10.1016/j.jngse.2014.12.008.
[4] Weber G, Di Giuliano A, Rauch R, Hofbauer HD. A simulation model for a mixed
alcohol synthesis reactor and validation of experimental data in IPSEpro. Fuel
Process Technol 2016;Part 1(141):167–76.
Fuel 300 (2021) 120936
[5] Kotamreddy G, Hughes R, Bhattacharyya D, Stolaroff J, Hornbostel K,
Matuszewski M, et al. Process modeling and techno-economic analysis of a CO2
capture process using fixed bed reactors with a microencapsulated solvent. Energy
Fuels 2019;33(8):7534–49.
[6] Bai L, Pinson P. Distributed reconciliation in day-ahead wind power forecasting.
Energies 2019;12(6). https://doi.org/10.3390/en12061112.
[7] Yang X, Yang Q, Dong W. Aeroengine data reconciliation model based on
cooperative working equations. Energy 2019;186:115914. https://doi.org/
10.1016/j.energy.2019.115914.
[8] Noriega MA, Narváez PC. UNIFAC correlated parameters for liquid-liquid
equilibrium prediction of ternary systems related to biodiesel production process.
Fuel 2019;249:365–78. https://doi.org/10.1016/j.fuel.2019.03.124.
[9] Szega M. Methodology of advanced data validation and reconciliation application
in industrial thermal processes. Energy 2020;198:117326. https://doi.org/
10.1016/j.energy.2020.117326.
[10] Badings TS, van Putten DS. Data validation and reconciliation for error correction
and gross error detection in multiphase allocation systems. J Petrol Sci Eng 2020;
195:107567.
[11] de Oliveira EC, de Santana Maia LA, da Costa LG, Lourenço TC. Validation of new
methodology for the definition of tolerance limits of critical properties in fuels
handled in terminals and pipelines—giveaway: compliance with the specification
(2018), Accreditation and Quality Assurance, 23, pp. 365–369 https://doi.org/
10.1007/s00769-018-1353-5.
[12] ASTM D7039-15a. Standard test method for sulfur in gasoline, diesel fuel, jet fuel,
kerosine, biodiesel, biodiesel blends, and gasoline-ethanol blends by
monochromatic wavelength dispersive X-Ray fluorescence spectrometry. West
Conshohocken, PA: ASTM International; 2020.
[13] ASTM D93-18. Standard test methods for flash point by Pensky-Martens closed cup
tester. West Conshohocken, PA: ASTM International; 2018.
[14] ASTM D5453-19a. Standard test method for determination of total sulfur in light
hydrocarbons, spark ignition engine fuel. Diesel engine fuel, and engine oil by
ultraviolet fluorescence. West Conshohocken, PA: ASTM International; 2019.
[15] ASTM D86 e 20. Standard test method for distillation of petroleum products and
liquid fuels at atmospheric pressure. West Conshohocken, PA: ASTM International;
2020.
[16] Theodorou D, Zannikos F. The use of measurement uncertainty and precision data
in conformity assessment of automotive fuel products. Measurement 2014;50:
141–51.
[17] Srinivasan S, Billeter J, Narasimhan S, Bonvin D. Data reconciliation for chemical
reaction systems using vessel extents and shape constraints. Comput Chem Eng
2017;101:44–58. https://doi.org/10.1016/j.compchemeng.2017.02.003.
[18] De Oliveira EC, Costa LG, Caspistrano LC, Tiroel LCO, Paixão RA. Interlaboratory
comparison of sulfur mass fraction in gasoline: MWDXRF spectrometry versus UV
fluorescence (2019). Pet Sci Technol, 37 (7), pp. 812–820. https://doi.org/
10.1080/10916466.2019.1566255.
[19] International Organization for Standardization. ISO 17043:2010 – conformity
assessment – general requirements for proficiency testing. 1st ed. Geneva: ISO;
2010.
[20] JCGM 106, 2012. Joint committee for guides in Metrology. Evaluation of
measurement data and the role of measurement uncertainty in conformity
assessment. Available at. http://www.bipm.org/utils/common/documents/jcgm/
JCGM_106_2012_E.pdf. (Accessed 3 July 2020).