North Sea Flow Measurement Workshop

22-25 October 2019

Technical Paper

Parallel calibration of multiphase flow meters vs

separator

Stig Harald Hammer Gustavsen, Aker BP ASA

Torbjørn Selanger, Aker BP ASA
Therese Renstrøm, Expert Analytics AS & Cognite

1 Introduction

In recent years, the number of marginal oil fields put into production has increased. The
typical scenario is that these small fields are tied back to an existing installation that acts
as host. This way of arranging production of oil and gas in a production hub makes the
development of marginal fields economically viable. One challenge with this type of
arrangement is that as the owner structure gets more complex, the complexity in the
allocation system also increases. It is essential for a viable long-term collaboration that
the produced oil and gas revenue is accurately split according to owner fraction in a
transparent and robust manner.

Tying the production from small surrounding fields, often called 3 rd party fields, to a host
installation usually requires modifications of the process at the host installation. To
achieve accurate allocation measurements, the production from each license should
ideally be processed and measured isolated from the other licenses. This would require
enormous investments and is not realistic. Another approach is to have a dedicated inlet
separator for each license. This method provides good accuracy of production volumes,
but still requires relatively large investments in addition to space and weight reserves on
the host installation.

The cheapest, smallest and lightest solution usually involves using Multiphase flow
meters (MPFM) for allocation. The individual mass flow of the oil, gas and water phases
of the production fluids from each separate field is measured by a dedicated MPFM, and
allocation can be performed based on these measurements. Production from different
fields can then be processed with minimum modifications of the hosts processing
systems. The downside of using MPFMs for allocation is a reduction in measurement
accuracy. MPFMs have been shown to drift with varying flow conditions [1]. To reduce
the uncertainty in the MPFM measurements and ensure that the measurements are
representative for the present conditions, periodic calibrations of the meters is necessary.

The Alvheim Floating Production Storage, and Offloading (FPSO) installation produces oil
and gas from the Alvheim field and three 3rd party fields. Alvheim’s allocation regime
utilize, amongst other, topside MPFM measurements. The MPFM measurements are
calibrated against single phase meters downstream from an inlet separator. The
calibration method is well proven in terms of accuracy, but it leads to significant deferrals
in production. This is due to required rerouting of the flow during calibration from one of
the inlet separators to the other. The rerouting causes the load on the other separator to
exceed its maximum capacity and production must be reduced during the calibration
process.

1
 North Sea Flow Measurement Workshop
22-25 October 2019

Technical Paper

In this paper we will present a novel method for calibrating MPFMs. The method is
tailored to reduce production deferrals connected to Alvheim’s existing calibration
method. Our proposed scheme has been shown to reduce deferrals by more than 95%.

Emphasis in this paper will be put on the theoretical foundation of the method, results
from tests of the new method performed by Alvheim asset and an evaluation of the
uncertainty of this new method compared to the existing calibration method. We are
presenting a solution that is still in development, so we will include the missing parts of
the routine is described together with a rough roadmap for further improvements.

The work presented in this paper has been made possible through AkerBP’s [2]
digitalization initiative Eureka and Cognite’s [3] data platform: Cognite Data Fusion
(CDF). Within the CDF data storage repository, large amounts of historical data from
different system is contextualized and stored. In parallel to the work performed in the
Eureka project, the problem has also formed the basis for a master thesis that has been
executed at the University of South-Eastern Norway [4]. The results presented herein is
taken from both the master thesis and the work performed in the Eureka project.

2 Current MPFM calibration regime at Alvheim

The Alvheim field, situated in the central part of the North Sea, comprises the six
discoveries 24/6-2 (Kameleon), 24/6-4 (Boa), 25/4-7 (Kneler), 25/4-10 S (Viper), 25/7-
5 (Kobra) and 25/4-3 (Gekko). The water depth in the area is 120-130 meters. Alvheim
was first discovered in 1998, the plan for development and operation (PDO) was
approved in 2004 and production started in 2008. Alvheim is developed with subsea wells
tied to an FPSO. The Vilje (2008), Volund (2009) and Bøyla (2015) fields, referred to as
3rd party fields, are tied-back to Alvheim [5].

Table 1 – ownership allocation of fields produced by the Alvheim FPSO

Owners Field share [%]

Alvheim Vilje Volund Bøyla
Aker BP ASA 65 46.904 65 65
ConocoPhillips 20
Lundin 15 35 15
DNO North Sea AS 28.853
Vår Energi AS 20
PGNiG Upstream Norway 24.243

The owner structure of Alvheim is rather complex, see Tab.1. In total six Norwegian and
international oil companies own a share of the hydrocarbons produced from 32 wells.
Five owners in total have a stake in the produced hydrocarbons of the so-called 3rd party
fields and the fraction of ownership is unique for each field.

There are in total six incoming flowlines to the Alvheim FPSO vessel, whereof three are
producing from the Alvheim field and the other three are producing from different 3rd

2
 North Sea Flow Measurement Workshop
22-25 October 2019

Technical Paper

party field. The flowlines that have production solely from the Alvheim field are
conveniently routed to a dedicated inlet separator (the Alvheim separator) where the
production is measured. The 3rd party licenses’ flowlines are routed to a shared 3rd party
separator, as shown in Fig. 1. This means that the flow from each 3rd party field must be
monitored to ensure that the owners get their share of the produced oil and gas. Hence,
MPFMs are installed on the incoming 3rd party flowlines, see Fig. 1.

Fig. 1 – Simplified flowline and separator configuration on Alvheim FPSO.

According to the license agreement, the MPFMs shall be calibrated with an interval and
test time to be determined from operational experience to maintain appropriate
accuracy.
A calibration measurement entails that the three 3rd party lines are isolated in turn in the
3rd party separator to facilitate a direct comparison between the MPFM and the reference
measurements at the oil, gas and water outlet of the separator. This means that during a
calibration run, two 3rd party flowlines needs to be re-routed to the Alvheim separator.

In the existing calibration technique, three separate measurements performed in order to

determine three calibration coefficients. The resulting system of linear equations is:

𝑚𝑀𝑃𝐹𝑀1 𝑘1 = 𝑚𝑠𝑒𝑝1
𝑚𝑀𝑃𝐹𝑀2 𝑘2 = 𝑚𝑠𝑒𝑝2
(1)
𝑚𝑀𝑃𝐹𝑀3 𝑘3 = 𝑚𝑠𝑒𝑝3 ,

where 𝑘𝑖 are the calibration coefficients that are target for the measurement, 𝑚𝑀𝑃𝐹𝑀𝑖 are
accumulated masses of hydrocarbons measured by the MPFMs to be calibrated, and 𝑚𝑠𝑒𝑝𝑖
are the reference measurements, where 𝑖 𝜖 {1,2,3}, see Fig. 1. A separate set of calibration
coefficients are deduced for the gas, oil and water. The accumulated masses in Eq. (1) is

3
 North Sea Flow Measurement Workshop
22-25 October 2019

Technical Paper

calculated by integrating the mass flows, 𝑚̇, of the individual phases measured by the
single phase meters and the MPFMs over the calibration interval [𝑡0 , 𝑡1 ] as seen in Eq. (2):

𝑡1
𝑚𝑀𝑃𝐹𝑀𝑖 = ∫ 𝑚̇𝑀𝑃𝐹𝑀𝑖 𝑑𝑡
𝑡0
𝑡1
(2)
𝑚𝑠𝑒𝑝𝑖 = ∫ 𝑚̇𝑠𝑒𝑝𝑖 𝑑𝑡
𝑡0

Equation (1) gives a linear relationship between the accumulated mass in and out of the
separator, when the mass rates are accumulated over a sufficient time interval. Liquid
levels, pressure and temperature are monitored in the separator during the accumulation
period to ensure that the assumption of mass balance in and out of the system is valid.

See Figure 2 and Ref. [4] for a more detailed description of the current calibration
method at Alvheim.

4
 North Sea Flow Measurement Workshop
22-25 October 2019

Technical Paper

k-factor
Referance Referance
y
Subject gas
wtr gas FT FQI Kf_gas
oil gas x
FT FT FT FQI
FE
Gas
Referance
Gas Massflow Subject
FE Separator FQI
Massflows
Muliphase meter Oil k-factor
Calib ra tio n Sub ject y
Water Kf_oil
Referance Referance x

FT FQI

FE
Subject Water Referance Oil Referance
Massflow Massflow
Water k-factor
FQI FE
y
Kf_water
Conceptual overview of Traditional FT FQI x

Water
Calibration of MPFM vs Separator
Excluding: Flashing through PVTSim and Water content In Oil
Referance Referance

Fig. 2- Conceptual overview of the traditional calibration method in use at Alvheim [4]

5
 North Sea Flow Measurement Workshop
22-25 October 2019

Technical Paper

Figure 3 shows the development in calibration factors resulting from different MPFM
calibrations. For the selected sample intervals, MPFM 1, 2 and 3 has a standard deviation
of 9.4%, 5.6% and 16.2% respectively. Change in calibration factors for the oil phase are
normally less than 10%, but changes above 50% are observed in the most extreme
situations. A 5% error in measured oil production would for 2018 oil production from one
of the 3rd party fields amount to approximately 107 USD. It illustrates that the MPFMs are
drifting and that relatively frequent calibration is warranted to ensure that production
allocation is as correct as possible. At Alvheim FPSO the 3rd party flowline MPFMs are
normally being calibrated every month.

1.3

1.2

1.1
Oil calibration factor

1.0

0.9
MPFM 1
0.8
MPFM 2
0.7 MPFM 3
0.6

0.5

0.4
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
Calibration no

Fig. 3: Development of oil calibration factor for 3rd party flowline MPFMs on Alvheim.

Pro-rating is used for allocation of production on Alvheim. This paper focuses on the pro-
rating principle between 3rd party flowlines. The pro-rating factors A for the different
phases are calculated from accumulated mass measured by the topside MPFMs on 3 rd
party flowlines
𝑚𝑙𝑖𝑛𝑒1
𝐴𝑙𝑖𝑛𝑒1 = (3)
𝑚𝑙𝑖𝑛𝑒1 + 𝑚𝑙𝑖𝑛𝑒2 + 𝑚𝑙𝑖𝑛𝑒3

3 An alternative calibration method: parallel calibration

In the following, we will present an alternative to the present calibration method used at
Alvheim today. The main motivation for identifying a new method is, as mentioned in
Sec. 1, the limited capacity of the inlet separators at Alvheim. The necessary
simultaneous routing of in total five flowlines to one separator during the present
monthly calibrations leads to a total flow of fluids that exceeds the capacity of the
Alvheim separator. Reduction in production during the calibration events is therefore
necessary, resulting in deferred production. The overreaching goal of our new method is

6
 North Sea Flow Measurement Workshop
22-25 October 2019

Technical Paper

to reduce the need for rerouting and at the same time keep the uncertainties connected
to the calibration as small as possible.

As our new method is a generalization of the present one, we refer the reader’s attention
to Equation (4). Here we have a set of three linear equations, containing three
unknowns, i.e the calibration coefficients of interest. This is the simplest version of three
equations and three unknowns, where the solution is straight forward

𝑚𝑠𝑒𝑝1
𝑘1 =
𝑚𝑀𝑃𝐹𝑀1
𝑚𝑠𝑒𝑝2
𝑘2 = 𝑚𝑀𝑃𝐹𝑀2 (4)
𝑚𝑠𝑒𝑝3
𝑘3 = 𝑚𝑀𝑃𝐹𝑀3

As long as we have three linear independent equations, in our case, three unique
calibration measurements, the calibration coefficients can be determined. With unique we
mean linear independent, i.e as long as we do not perform measurements on the same
configuration of flowlines connected to the separator, with unchanged flow through the
pipes, the system will be solvable. One could envisage routing the lines with MPFM1 and
MPFM2 to the separator for a first calibration measurement, then MPFM3 and MPFM2, and
for a last run MPFM1 and MPFM3. An ensemble of possible configurations exists. All with
less need for rerouting. We have chosen to refer to our new calibration method as
parallel calibration since it opens up for calibration measurements where several lines are
connected to the separator.

At Alvheim the three 3rd party lines, Bøyla, Vilje and Volund, can be routed to the 3 rd
party separator. The relationship between accumulated flow through the MPFMs and the
separator is then:

𝑡1 𝑡1 𝑡1 𝑡1
∫ 𝑚̇𝑝,𝐵ø𝑦𝑙𝑎 𝑑𝑡 ∙ 𝑘𝑝,𝐵ø𝑦𝑙𝑎 + ∫ 𝑚̇𝑝,𝑉𝑖𝑙𝑗𝑒 𝑑𝑡 ∙ 𝑘𝑝,𝑉𝑖𝑙𝑗𝑒 + ∫ 𝑚̇𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 𝑑𝑡 ∙ 𝑘𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 =∫ 𝑚̇𝑝,𝑠𝑒𝑝 𝑑𝑡 , (5)
𝑡0 𝑡0 𝑡0 𝑡0

where 𝑝 ϵ {oil, gas, water} , 𝑚̇𝑝,𝑖 and 𝑘𝑝,𝑖 are the mass flow rates measured by the MPFM and
calibration coefficients connected to line i, where 𝑖 𝜖 {𝐵ø𝑦𝑙𝑎, 𝑉𝑖𝑙𝑗𝑒, 𝑉𝑜𝑙𝑢𝑛𝑑}. Lastly, 𝑚̇𝑝,𝑠𝑒𝑝
is the flow rate measured at the outlet of the 3 rd party separator. When expressing Eq.
(5) in terms of accumulated masses (see Eq. (2)), we get:

𝑚𝑝,𝐵ø𝑦𝑙𝑎 ∙ 𝑘𝑝,𝐵ø𝑦𝑙𝑎 + 𝑚𝑝,𝑉𝑖𝑙𝑗𝑒 ∙ 𝑘𝑝,𝑉𝑖𝑙𝑗𝑒 + 𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 ∙ 𝑘𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 = 𝑚𝑝,𝑠𝑒𝑝 (6)

We let Eq. (6) represent a calibration measurement. We need three separate calibration
measurements, as follows:

7
 North Sea Flow Measurement Workshop
22-25 October 2019

Technical Paper

𝑚𝑝,𝐵ø𝑦𝑙𝑎,1 ∙ 𝑘𝑝,𝐵ø𝑦𝑙𝑎 + 𝑚𝑝,𝑉𝑖𝑙𝑗𝑒,1 ∙ 𝑘𝑝,𝑉𝑖𝑙𝑗𝑒 + 𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑,1 ∙ 𝑘𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 = 𝑚𝑝,𝑠𝑒𝑝,1

𝑚𝑝,𝐵ø𝑦𝑙𝑎,2 ∙ 𝑘𝑝,𝐵ø𝑦𝑙𝑎 + 𝑚𝑝,𝑉𝑖𝑙𝑗𝑒,2 ∙ 𝑘𝑝,𝑉𝑖𝑙𝑗𝑒 + 𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑,2 ∙ 𝑘𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 = 𝑚𝑝,𝑠𝑒𝑝,2
(7)
𝑚𝑝,𝐵ø𝑦𝑙𝑎,3 ∙ 𝑘𝑝,𝐵ø𝑦𝑙𝑎 + 𝑚𝑝,𝑉𝑖𝑙𝑗𝑒,3 ∙ 𝑘𝑝,𝑉𝑖𝑙𝑗𝑒 + 𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑,3 ∙ 𝑘𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 = 𝑚𝑝,𝑠𝑒𝑝,3,

where each row represents one particular measurement.

At this point we introduce matrix notation for a more efficient representation of the
system of linear equations:

𝑚𝑝,𝐵ø𝑦𝑙𝑎,1 𝑚𝑝,𝑉𝑖𝑙𝑗𝑒,1 𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑,1 𝑘𝑝,𝐵ø𝑦𝑙𝑎 𝑚𝑝,𝑠𝑒𝑝,1

(8)
[𝑚𝑝,𝐵ø𝑦𝑙𝑎,2 𝑚𝑝,𝑉𝑖𝑙𝑗𝑒,2 𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑,2 ] [ 𝑘𝑝,𝑉𝑖𝑙𝑗𝑒 ]=[𝑚𝑝,𝑠𝑒𝑝,2 ]
𝑚𝑝,𝐵ø𝑦𝑙𝑎,3 𝑚𝑝,𝑉𝑖𝑙𝑗𝑒,3 𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑,3 𝑘𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 𝑚𝑝,𝑠𝑒𝑝,3
or as
𝑴𝑝 𝒌𝑝 = 𝒎𝑝 , (9)

where Mp is the matrix representation of the accumulated masses from the MPFMs, kp is
the vector of calibration factors and mp is the vector of accumulated masses from the
separator measurements.

To ensure that Eq. (9) is solvable and that the solution is unique, the rows of the square
matrix 𝑴𝑝 has to be linear independent. If so, the matrix is invertible and we obtain a
unique set of calibration coefficients:

𝒌𝑝 = 𝑴−1 𝑝 𝒎𝑝 (10)

We now have all the necessary requirements for a calibration measurement. In the next
section we will list the most promising ones. In the following, we will list some of the
possible calibration configurations that satisfies these requirements.

3.1 Possible calibration configurations

One obvious way to ensure that 𝑴𝑝 is invertible is to route a distinct combination of two
flowlines to the 3 party separator for each of the three necessary calibration runs. The
rd

system of equations becomes

𝑚𝑝,𝐵ø𝑦𝑙𝑎 𝑚𝑝,𝑉𝑖𝑙𝑗𝑒 0 𝑘𝑝,𝐵ø𝑦𝑙𝑎 𝑚𝑝,𝑠𝑒𝑝,1
(11)
[ 0 𝑚𝑝,𝑉𝑖𝑙𝑗𝑒 𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 ][ 𝑘𝑝,𝑉𝑖𝑙𝑗𝑒 ]=[𝑚𝑝,𝑠𝑒𝑝,2 ],
𝑚𝑝,𝐵ø𝑦𝑙𝑎 0 𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 𝑘𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 𝑚𝑝,𝑠𝑒𝑝,3

We will refer to this solution as the 2-2-2 solution. Here only one extra flowline would
need to be rerouted to the Alvheim separator per calibration run and the need for
reduction in the production during calibration is decreased significantly.

8
 North Sea Flow Measurement Workshop
22-25 October 2019

Technical Paper

Another possible solution that would lead to even less need for deferments is a 3-2-2
solution, where the total number of rerouted lines during the whole calibration is two.
There are three different versions of this solution:
𝑚𝑝,𝐵ø𝑦𝑙𝑎 𝑚𝑝,𝑉𝑖𝑙𝑗𝑒 0 𝑘𝑝,𝐵ø𝑦𝑙𝑎 𝑚𝑝,𝑠𝑒𝑝,1
(12)
[ 0 𝑚𝑝,𝑉𝑖𝑙𝑗𝑒 𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 ] [ 𝑝,𝑉𝑖𝑙𝑗𝑒 ] = [𝑚𝑝,𝑠𝑒𝑝,2 ],
𝑘
𝑚𝑝,𝐵ø𝑦𝑙𝑎 𝑚𝑝,𝑉𝑖𝑙𝑗𝑒 𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 𝑘𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 𝑚𝑝,𝑠𝑒𝑝,3
or

𝑚𝑝,𝐵ø𝑦𝑙𝑎 𝑚𝑝,𝑉𝑖𝑙𝑗𝑒 0 𝑘𝑝,𝐵ø𝑦𝑙𝑎 𝑚𝑝,𝑠𝑒𝑝,1

(13)
[𝑚𝑝,𝐵ø𝑦𝑙𝑎 0 𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 ] [ 𝑝,𝑉𝑖𝑙𝑗𝑒 ] = [𝑚𝑝,𝑠𝑒𝑝,2 ],
𝑘
𝑚𝑝,𝐵ø𝑦𝑙𝑎 𝑚𝑝,𝑉𝑖𝑙𝑗𝑒 𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 𝑘𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 𝑚𝑝,𝑠𝑒𝑝,3

or
𝑚𝑝,𝐵ø𝑦𝑙𝑎 0 𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 𝑘𝑝,𝐵ø𝑦𝑙𝑎 𝑚𝑝,𝑠𝑒𝑝,1
(14)
[ 0 𝑚𝑝,𝑉𝑖𝑙𝑗𝑒 𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 ] [ 𝑘𝑝,𝑉𝑖𝑙𝑗𝑒 ] = [𝑚𝑝,𝑠𝑒𝑝,2 ],
𝑚𝑝,𝐵ø𝑦𝑙𝑎 𝑚𝑝,𝑉𝑖𝑙𝑗𝑒 𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 𝑘𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 𝑚𝑝,𝑠𝑒𝑝,3

Observing the 3-2-2 versions, we see that one of the three flowlines, Bøyla, Vilje or
Volund is lined up towards the 3rd party separator for all the runs. In order to optimize
the use of separator capacity, the line that produces most should be the one that is left
in the 3rd party separator.

Other possibilities are the 1-1-2, 1-1-3 and 1-2-3, but none of these have particular
merit.

Is it possible to avoid the rerouting altogether? Is a 3-3-3 solution possible? That would
mean keeping all 3rd party and Alvheim lines in their respective separators during all
three calibration runs. Certainly, it is possible as long as three rows in Mp are linearly
independent. Let’s say that one performs three calibration runs on three consecutive
days. During normal operation, the average flowrates in the three lines are likely to be
quite stable. This means that all rows in our matrix will be linearly dependent and in
practice we have one equation and three unknowns. So infinitely many solutions. But if
the three calibration runs were to be performed in connection with for example well
integrity tests or other activities causing a change in the flowrate, this could be sufficient
to form a solvable system. It would in such a case be important to track the stability of
the solution by checking how well-conditioned the system is. Another concern could be
the representativeness of such a calibration if the flows are deviating from those at
normal operating conditions.

Lastly, we have the 1-1-1 solution of our system, which corresponds to the present
calibration method used at Alvheim:

9
 North Sea Flow Measurement Workshop
22-25 October 2019

Technical Paper

𝑚𝑝,𝐵ø𝑦𝑙𝑎 0 0 𝑘𝑝,𝐵ø𝑦𝑙𝑎 𝑚𝑝,𝑠𝑒𝑝,1

(15)
[ 0 𝑚𝑝,𝑉𝑖𝑙𝑗𝑒 0 ][ 𝑘𝑝,𝑉𝑖𝑙𝑗𝑒 ]=[𝑚𝑝,𝑠𝑒𝑝,2 ],
0 0 𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 𝑘𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 𝑚𝑝,𝑠𝑒𝑝,3

Solving the 1-1-1 system would lead to the well-known solution:

𝑚𝑝,𝑠𝑒𝑝,1
𝑘𝑝,𝐵ø𝑦𝑙𝑎 = 𝑚𝑝,𝐵ø𝑦𝑙𝑎
𝑚𝑝,𝑠𝑒𝑝,2
𝑘𝑝,𝑉𝑖𝑙𝑗𝑒 = (16)
𝑚𝑝,𝑉𝑖𝑙𝑗𝑒
𝑚𝑝,𝑠𝑒𝑝,3
𝑘𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 = ,
𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑

equivalent to Eq. (4). This shows clearly the parallel calibration method is simply a
generalization of the existing calibration.

3.2 Parallel calibration for random number of MPFMs

Our method works equally well cases when more than three MPFMs are targets for
calibration. Eq. (10) shows a general overview of a parallel calibration system with S
number of multiphase streams. T number of calibration runs are needed to build an
accumulated mass matrix Mp as shown in Eq.(17).

𝑚𝑝,1,1 𝑚𝑝,2,1 ⋯ 𝑚𝑝,𝑆,1

𝑚𝑝,1,2 𝑚𝑝,2,2 ⋯ 𝑚𝑝,𝑆,2
𝑴𝑝 ∈ ℝ𝑇×𝑆 =[ ⋮ ⋮ ⋱ ⋮ ] (17)
𝑚𝑝,1,T 𝑚𝑝,2,T ⋯ 𝑚𝑝,𝑆,T

Each calibration run in the matrix Mp is contains accumulated masses from the MPFMs
being calibrated. The reference measurement populates the vector mp as shown in Eq
(18)
𝑚𝑝,𝑟𝑒𝑓,1
𝑚𝑝,𝑟𝑒𝑓,2
𝒎𝑝 ∈ ℝ𝑇 = [ ⋮ ], (18)
𝑚𝑝,𝑟𝑒𝑓,𝑇

and the vector of the calibration coefficients

𝐤 p is
𝑘𝑝,1
𝑘
𝐤 p ∈ ℝ𝑆 = 𝑝,2 . (19)
⋮
[𝑘𝑝,𝑆 ]

When T = S, i.e the number of equations equals the number of unknown, and the rows of
𝑴𝑝 are linearly independent, the system is solvable and

10
 North Sea Flow Measurement Workshop
22-25 October 2019

Technical Paper

𝒌𝑝 = 𝑴−1
𝑝 𝒎𝑝 (20)

3.3 Measuring convergence of calibration factors

What most clearly distinguished the 2-2-2 and 3-2-2 from the traditional, or 1-1-1
method, is at what point in the calibration process the coefficients can be determined. As
seen from Eq. (4), each calibration run in the 1-1-1 method determines a calibration
coefficient for a particular line. This allows the metering technician to monitor the time-
evolution of the actual calibration coefficients and determine to complete the calibration
run when that coefficient is sufficiently stable, or in other words seems to have
converged.

In a parallel calibration, all measurements need to be completed before we have all

necessary information to solve the system. This means that the calibration coefficients
themselves cannot be used as a metric for convergence. We propose to use the ratio of
the MPFM measurements and the separator measurements instead. To give a specific
example, let’s look at the 2-2-2 method. In the first calibration measurement (first row)
one would monitor the “quasi”-calibration coefficients:
𝑚𝑝,𝐵ø𝑦𝑙𝑎 (21)
𝑘𝑞𝑢𝑎𝑠𝑖_𝐵ø𝑦𝑙𝑎 =
𝑚𝑝,𝑠𝑒𝑝,1

𝑚𝑝,𝑉𝑖𝑙𝑗𝑒
𝑘𝑞𝑢𝑎𝑠𝑖_𝑉𝑖𝑙𝑗𝑒 = (22)
𝑚𝑝,𝑠𝑒𝑝,1

Convergence of these parameters for all three calibration runs will indicates that the real
calibration coefficients will converge to a stable solution when the system is finally
solved. Surveillance of process parameters during calibration is of course very important
to ensure sufficient quality of the parallel calibration.

4 Results from parallel calibration

Two different approaches have been used to test the parallel calibration method on data
from Alvheim. First, we will present the results from time series using time-shifted time
series from traditional calibration. Thereafter, data from actual parallel calibration runs
has been used, where the Alvheim FPSO has configured fluid streams to generate
required data for the parallel calibration method.
The results given in this paper are generated without density adjustment and with a
simplified flashing calculation. Key aspects of the algorithm are covered in Appendix A
and [4].

11
 North Sea Flow Measurement Workshop
22-25 October 2019

Technical Paper

4.1 Synthetic parallel calibration

Three synthetic reference measurements have been generated by adding reference

measurements from combinations of two and two traditional calibrations. These
reference measurements populate the reference matrix mp. For each synthetic reference
measurement, the corresponding measurements from the MPFMs are inserted in the
accumulated mass matrix, Mp. The solution of the synthetic parallel calibration dataset is
shown in Figure 4. and compared to the solution where the calibration factor is found
using the traditional calculation method. The results show that the solution from parallel
and traditional calibration method matches. This is as expected since the calibration
factor is calculated from the exact same time series.

Fig. 4 calibration factor development of traditional vs synthetic parallel calibration of an

MFPM on Alvheim [4]

4.2 Results from field test of parallel calibration method

In early and late April 2019, measurements were done where streams were rerouted to
give real data for parallel calibration. This was done immediately after traditional
calibration enabling comparison of the methods. Figure 5 shows the resulting calibration
factor for one of the MPFMs determined from different test run configurations. As the
data is based on different time series, the resulting calibration factors are not expected
to be equal as instabilities in the process and MPFMs can give different results for the
different trials. Included in the results are also the calibration factor given by the
metering system and includes composition and density adjustment of the measurement.

12
 North Sea Flow Measurement Workshop
22-25 October 2019

Technical Paper

Resulting comparison of an MPFM

Current system Traditional Synthetic parallel Parallel 2x2x2 Parallel 3x2x2 1mnd later 3x2x2
1.7

1.6
Calibration factor

1.5

1.4

1.3

1.2

1.1

0.9

0.8
Oil Gas Water

Phase
Fig. 5 – Result comparison of one of the MPFMs at Alvheim FPSO during March/April
calibrations [4]

5 Uncertainty analysis of parallel calibration

It is beyond the scope of this paper to attempt a complete uncertainty analysis of the
calibration process (and probably impossible). MPFMs are known to have a rather
complex dependency of flow conditions, making it very difficult to estimate an
uncertainty associated with its measured values. Our approach to the uncertainty
analysis is to estimate the uncertainty of the parallel calibration method relative to the
traditional one. That is, we will investigate if parallel calibration introduces new sources
of uncertainty that influences the accuracy of the calculated calibration coefficients. As
our calibration method has become more complex, the correct treatment of the
propagation of errors is important. As we argued in Sec. 3.2, the parallel calibration
method can in theory be used for calibrating any number of MPFMs, N, in parallel.
Tracking how measurement errors are propagating through the matrix inversion of an (N
x N)-matrix and vector multiplication can be done analytically, but as N becomes large
Monte Carlo (MC) simulations would simplify the task. In our error analysis, we find both
analytical expression for the error and perform MC simulations.

We will in the following focus our analysis of the uncertainties connected by parallel
calibration by discussing two questions:

13
 North Sea Flow Measurement Workshop
22-25 October 2019

Technical Paper

1. Will one drifting/unstable MPFM lead to less accuracy in the measured calibration
coefficients of the two other MPFMs?
2. Will a difference in flowrate through the MPFMs calibrated in parallel lead to
increased uncertainty in the calibration factor of the low producing lines?

After presenting a theoretical analysis of the relative errors between the traditional
calibration and parallel calibration, we will take a look at historical data from Alvheim and
see how they relate to our analysis.

5.1 Systematic uncertainty

The first question will be answered by investigating how systematic uncertainty in the
measurements propagates through the calculations. Systematic uncertainty is here
defined as an uncertainty that is not statistical. A MPFM that reports a systematical
shifted measurement is assumed to stay shifted during the three calibration runs. We
make no assumptions about what causes the systematic errors.

Let us go through an example assume that a MPFM measuring the flow rate from the
Volund is particularly unstable and for the time interval of the calibration it systematically
reports a very low flow rate. Will this shift in measurement affect the calibration
coefficients of Vilje and Bøyla?

Constructing this scenario, we assume a set of “true”, unshifted calibration coefficients,

𝐤 ∗p . These are used to construct the corresponding reference measurements, 𝐦𝑠𝑒𝑝 . The
systematically shifted measurement from Volund is given by, 𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 − Δ𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑

We start by looking at the traditional 1-1-1 case:

𝑚𝑝,𝐵ø𝑦𝑙𝑎 0 0 𝑘𝑝,𝐵ø𝑦𝑙𝑎 𝑚𝑝,𝑠𝑒𝑝,1
(23)
[ 0 𝑚𝑝,𝑉𝑖𝑙𝑗𝑒 0 ][ 𝑝,𝑉𝑖𝑙𝑗𝑒 ]=[𝑚𝑝,𝑠𝑒𝑝,2 ].
𝑘
0 0 𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 − Δ𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 𝑘𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 𝑚𝑝,𝑠𝑒𝑝,3

We now express the 𝐦𝑠𝑒𝑝 vector in terms of 𝐤 ∗ and the total mass measured by the
MPFMs during one calibration run (one row in the matrix):
∗
𝑚𝑝,𝐵ø𝑦𝑙𝑎 0 0 𝑘𝑝,𝐵ø𝑦𝑙𝑎 𝑘𝐵ø𝑦𝑙𝑎 𝑚𝑝,𝐵ø𝑦𝑙𝑎
∗ (24)
[ 0 𝑚𝑝,𝑉𝑖𝑙𝑗𝑒 0 ][ 𝑘𝑝,𝑉𝑖𝑙𝑗𝑒 ]=[ 𝑘𝑉𝑖𝑙𝑗𝑒 𝑚𝑝,𝑉𝑖𝑙𝑗𝑒 ]
0 0 𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 − Δ𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 𝑘𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 ∗
𝑘𝑉𝑜𝑙𝑢𝑛𝑑 𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑

Solving this system gives the calibration coefficients:

∗
𝑘𝐵ø𝑦𝑙𝑎
𝑘𝑝,𝐵ø𝑦𝑙𝑎 ∗
𝑘𝑉𝑖𝑙𝑗𝑒 (25)
[ 𝑘𝑝,𝑉𝑖𝑙𝑗𝑒 ]= 𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 ∗
𝑘𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 𝑘𝑉𝑜𝑙𝑢𝑛𝑑
[𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 − Δ𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 ]

14
 North Sea Flow Measurement Workshop
22-25 October 2019

Technical Paper

We see as expected that a shift in the Volund MPFM measurements only gives an error in
the calibration coefficient of Volund. This is rather intuitive, since the calibration of the
meters are done separately.

Now, let’s look at the corresponding 2-2-2 case:

𝑚𝑝,𝐵ø𝑦𝑙𝑎 𝑚𝑝,𝑉𝑖𝑙𝑗𝑒 0 𝑘𝑝,𝐵ø𝑦𝑙𝑎 𝑚𝑝,𝑠𝑒𝑝,1
(26)
[ 0 𝑚𝑝,𝑉𝑖𝑙𝑗𝑒 𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 − Δ𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 ][ 𝑘𝑝,𝑉𝑖𝑙𝑗𝑒 ]=[𝑚𝑝,𝑠𝑒𝑝,2 ].
𝑚𝑝,𝐵ø𝑦𝑙𝑎 0 𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 − Δ𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 𝑘𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 𝑚𝑝,𝑠𝑒𝑝,3

Or, in terms of “real”, unshifted coefficients, 𝐤∗:

∗ ∗
𝑚𝑝,𝐵ø 𝑚𝑝,𝑉𝑖 0 𝑘𝑝,𝐵ø 𝑘𝐵ø 𝑚𝑝,𝐵ø + 𝑘𝑉𝑖 𝑚𝑝,𝑉𝑖
∗ ∗ (27)
[ 0 𝑚𝑝,𝑉𝑖 𝑚𝑝,𝑉𝑜 − Δ𝑚𝑝,𝑉𝑜 ][ 𝑘𝑝,𝑉𝑖 ]=[ 𝑘𝑉𝑖 𝑚𝑝,𝑉𝑖 + 𝑘𝑉𝑜 𝑚𝑝,𝑉𝑜 ],
𝑚𝑝,𝐵ø 0 𝑚𝑝,𝑉𝑜 − Δ𝑚𝑝,𝑉𝑜 𝑘𝑝,𝑉𝑜 ∗ ∗
𝑘𝐵ø 𝑚𝑝,𝐵ø + 𝑘𝑉𝑜 𝑚𝑝,𝑉𝑜

where {Bø, Vi, Vo} = {Bøyla, Vilje, Volund}. Solving this system analytically, gives
exactly the same solution as for the 1-1-1:
∗
𝑘𝐵ø𝑦𝑙𝑎
𝑘𝑝,𝐵ø𝑦𝑙𝑎 ∗
𝑘𝑉𝑖𝑙𝑗𝑒 (28)
[ 𝑘𝑝,𝑉𝑖𝑙𝑗𝑒 ]= 𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 ∗
𝑘𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 𝑘𝑉𝑜𝑙𝑢𝑛𝑑
[𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 − Δ𝑚𝑝,𝑉𝑜𝑙𝑢𝑛𝑑 ]

We have performed this exercise for the 3-2-2 calibration version as well. The result is
the same. We have also tested cases where several flowlines have systematic
uncertainties. The results are the same: systematic uncertainties in the measurements
are the same for the traditional calibration as for the new methods.

So, our answer to the first question 1) is that no, one drifting/unstable MPFM does not
lead to less accuracy in the measured calibration coefficients of the two other MPFMs.
In this sense, the parallel calibration method does not introduce additional systematic
uncertainties compared to the traditional method.

5.2 Statistical uncertainty

In the following, the statistical uncertainty is defined as a general, gaussian distributed

measurement error, distributed as:

1 (𝑥−𝜇)2
−
𝑃(𝑥) = e 2𝜎2
(29)
√2𝜋𝜎

where 𝜎 is the standard deviation and 𝜇 are expectation value. In the following analysis,
we have modified the standard calibration equation:

15
 North Sea Flow Measurement Workshop
22-25 October 2019

Technical Paper

𝑴𝑝 𝒌𝑝 = 𝒎𝑝 (30)

Now, we populate 𝑴𝑝 and 𝒎𝑝 with gaussian distributions instead of specific numbers. We

assume that in general 𝜎 ∝ 𝜇, i.e that the measurement uncertainty is proportional to the
size of the mass flow. As an example, 𝜎 = 𝑎𝜇 , where a is set to 0.03 for the MPFMs. This
number is randomly chosen, but since our focus is on the relative uncertainties between
the traditional and parallel calibration the absolute value of a is not important. The
distributions in 𝒎𝑝 , that is, the distributions of the measurements from the single phase
meters have been assigned a flow, Qs , dependent a parameter, a(Qs ). This only is done
to catch the behavior of the single phase meters when the flow, Q s , out of the separator
get very low, as can be the case at Alvheim when a 1-1-1 calibration is being performed.
In the Monte Carlo simulations, we produce versions of 𝑴𝑝 and 𝒎𝑝 by sampling from the
distributions and produce a 𝒌𝑝 by solving the system using a linear solver.
The resulting 𝒌𝑝 distributions from the 1-1-1, the 2-2-2 and the 3-2-2 versions are
compared. The simulations take measurement time and the relative flowrate through a
meter as input. The relative flowrate parameters are used to quantify the way
uncertainty in lines with small production is affected by being measured together with
lines with higher production. Figure 6 shows an example of results from simulations of
the distributions of the calibration factors for the MPFM connected to the Vilje flowline. As
the standard deviations indicates, doubling the measurement time will produce a
resolution as good as for the 1-1-1 calibration.

16
 North Sea Flow Measurement Workshop
22-25 October 2019

Technical Paper

Fig. 6 – Simulated calibration factor distributions for Vilje for the 1-1-1 and 2-2-2 version
and varying time intervals.

After looking into variations of measurement time or ratios of flow rates we conclude that
the answer to question 2) is: yes, if we assume that the individual measurement error
from each meter is dependent on flowrate, the parallel calibration method leads to an
increased uncertainty in the calibration factor of the low producing lines. But this effect
could be remedied by prolonging the calibration time periods over which the meters are
calibrated.

5.3 Historical data from Alvheim calibrations

Until this point in we have looked at theoretical error distributions of the MPFM and single
phase meters. We will now try to use historical data from calibration runs at Alvheim to

17
 North Sea Flow Measurement Workshop
22-25 October 2019

Technical Paper

investigate if the relative difference in uncertainty between a traditional 1-1-1 calibration

and the parallel versions 2-2-2 and 3-2-2. The analysis is based on time series from
four historical 1-1-1 calibrations. As described in Sec. 4.1, we produce synthetic time
series to test compare the time evolution of the calibration coefficients from the 1-1-1
calibrations with 2-2-2 and 3-2-2. To get a sense of the statistical fluctuation in the
calibrations, we find the standard deviation, 𝜎, of the calibration coefficients for the last
two hours of calibration. We chose to interpret this distribution as the statistical
distribution the calibration coefficients. At present we have done this analysis for the last
six months of calibrations. The preliminary results indicate that the statistical fluctuations
of the measurements are so small that there the difference between a 1-1-1 and a 2-2-
2 and 3-2-2 calibration is practically negligible.

6 Implementation of the parallel calibration method

When it comes to the implementation and execution of the calibration algorithm, having
access to and control over data quality from different Supervisory Control and Data
Acquisition (SCADA) systems is important. Possible implementations can be on the
metering SCADA level or through a data platform such as the Cognite Data Fusion
platform. Implementation on the SCADA level will require additional information from the
main vessel Safety and Automation system (SAS). The data from the SAS system is used
to identify calibration time windows based on valve positions and to document stability
during the calibration window.

Through the Cognite Data Fusion platform, which is used today, and the results of this
paper is based on, all necessary data is available. However, the integrity/data quality is
reduced as the timeseries values used are historical values with varying timesteps. This
is caused by data compression taking place in the systems feeding the CDF. To ensure
high fidelity, uncompressed data from each flow computer should be available via CDF.
The data should be immutable between the flow computer and the algorithms execution.
The implementation of the method could then be done on any machine with access to
CDF.

7 Density, composition and phase equilibrium

This section discusses the processing of the measured data that is done in the parallel
and traditional calibration algorithms. It is included as information of remaining work to
increase the accuracy of the parallel calibration method.

The traditional calibration method performs some processing of the flow measurements
using a thermodynamic software package. Pre-defined well fluids for each MPFM are used
as basis for this processing. For Alvheim FPSO, the processing can be divided in the
following main steps:
- Adjust composition data to match oil measured density.
- Adjust composition to match measured GOR.
- Flash fluid in MPFM and separator to common conditions for calculation of
calibration factors.
For single MPFM calibration, this processing is relatively straight forward. When more
than one flowline is introduced, the complexity increases. As an example, the density of

18
 North Sea Flow Measurement Workshop
22-25 October 2019

Technical Paper

the oil will now be a mix of the density from multiple incoming flowlines. Further, the
temperature and pressure in each MPFM and the separator will be different.

Future work will investigate resolving these challenges. Regarding allocation of densities
to the respective flowline, it is envisaged that this can be done by solving a system of
linear equations similar to how the calibration factors are calculated.

In Alvheim’s case, preliminary results have shown that the measured oil density is similar
to densities from the PVT simulation software and also from the lookup-tables in the
MPFM process computers. The difference is about 1%. Further, it appears that the
compositional adjustment done to match the measured GOR does not impact the
composition of the oil and gas phase. Pressure in the MPFMs and separator is also similar
making the effect of flashing moderate. Temperatures can vary from flowline to flowline.
Preliminary tests of the amount of flashing of oil shows that it is in the range 0.5 – 2%.
These factors will have some impact on the resulting calibration factors.

The results shown in this paper is based on a fixed flashing of 2.0 wt% of the oil mass to
the gas phase. It has been calculated based on approximate conditions at the MPFMs and
in the 3rd party separator. The amount of flashing has been estimated using UniSim
design process simulation software.

8 Other possible applications

Through the Third Party Access (TPA) agreement, the Norwegian government encourages
to utilize existing infrastructure on the NCS to develop new licenses. The parallel
calibration method supports this initiative by enabling reduced investments while
maintaining the health and accuracy of multiphase flow meters used for allocation
purposes. Alternatively, it can be used in combination with traditional calibration as a
verification method to increase or decrease the time between traditional calibrations,
thereby increasing allocation accuracy and/or reducing production deferrals, similar to
the Alvheim FPSO case.

The parallel calibration methodology is not limited to use with MPFMs, it can also be
applied for calibration of single-phase meters.

8.1 Process configurations concepts enabling parallel calibration

Key to the parallel calibration method is to have changes in flow rates between the trials
to enable solution of the system of equations. The easiest way to generate this change is
by closing upstream valves for one of the flowlines. Having two parallel separators where
flowlines are continuously rerouted between the separators is one possible configuration
to enable this for incoming flowlines. Another option is to use a compact/inline separator
in parallel to the inlet separator, where one flowline at a time is re-routed to this compact
separator. Both methods will give changes in flow rate that enables a semi-continuous
calibration or verification of the flow meters.

19
 North Sea Flow Measurement Workshop
22-25 October 2019

Technical Paper

If the production cannot be routed elsewhere it will normally have to be choked back to
generate the required change in flowrate for the parallel calibration technique. This is of
course undesirable. Another option to achieve change in flow rate could be to increase
the flow, e.g. through recirculation of fluids. As an example, oil and gas could be recycled
from downstream stages of the process plant to upstream of the MPFMs. This does
however imply that the flow meter is calibrated at a different rate and with a potentially
different fluid than what is normally flowing. It also requires installation of facilities for
routing and measuring of a stream upstream of the MPFM.

9 Notation

Bolded letters are matrices or vectors, the masses and calibration factors always have
subscripts to associate the values to the correct stream and phase.

ṁ mass flow Subscripts

m Accumulated mass in stream p phase [Oil, Gas, Water]
M Matrix of accumulated masses s Stream number / index
∈ ℝTxS S number of streams
m vector of accumulated- T Trial number / index
Number of trials
reference masses ∈ ℝT
n Index / index counter
k calibration factor [-]
ref Reference measurement
k calibration factor vector ∈ ℝS sub Calibration-Subject measurement
t time / timestamp / time-
segment Superscripts
-1 Matrix inverse
ℝ Denotes the matrix or vector
only consist of real values.

20
 North Sea Flow Measurement Workshop
22-25 October 2019

Technical Paper

10 Conclusion

The work in this paper demonstrates how calibration factors for flowmeters can be
determined while still having multiple streams routed towards the reference flow meter.
To be able to determine the calibration factor, a set of trials equal to the number of flow
meters being calibrated must be undertaken. Each flow meter being calibrated must be
routed towards the reference flow meter during one of the trials at least. All meters being
calibrated can be routed to the reference meter in all trials given that there is enough
difference in flow rate between each trial. For the MPFMs on Alvheim, variation in flow
rate is achieved by re-routing one of the streams at a time away from normal third-party
separator.
Where the method is used for calibration of multiphase meters, adjustment for difference
in process conditions between the MPFMs and the reference meter should be done to
increase the accuracy of the calibration. Further, adjustment of composition and
component to match measured GOR and densities could be done. This has not yet been
investigated in detail but will be the focus in future work. However, without any such
adjustment, calibration factors that is close to single flow meter calibration is achieved.
The error introduced when allocation is performed will be further reduced as the tests
performed shows that the calibration factors for the different meters are shifted in the
same direction relative to the single MPFM calibration method.

The uncertainty analysis shows that systematic uncertainty in one of the flow meters
being calibrated does not propagate to the other meters in the system. An analysis of the
statistical uncertainty demonstrates that this will increase compared to calibration of one
flow meter at a time. This can be compensated by increasing the duration of the trials.
Analysis of synthetic historical data shows that the statistical uncertainties connected to
the calibration measurements are so small, that their difference in statistical
uncertainties between a 1-1-1 and a 2-2-2 or 3-2-2 is negligible. It is important to note
that this conclusion is drawn based on synthetic data, and that more actual tests from
Alvheim will be important.

The concept of parallel calibration can open for simplified design solutions for tie-in of 3rd
party fields and/or an opportunity to reduce deferrals associated with calibration of
multiphase meters. The latter is the case at the Alvheim FPSO. It is also envisaged that
the method can have other similar applications that can both reduce investments
requirements, increase metering accuracy and/or reduce production deferrals.

11 Acknowledgments

We want to thank the offshore operations and onshore metering engineers on Alvheim,
whom has assisted us with field test runs giving us data to test the parallel calibration
method and supporting us with knowledge and information about the Alvheim metering
system and allocation algorithms, as well as challenging us on the subject of accuracy
and uncertainty of the method.

21
 North Sea Flow Measurement Workshop
22-25 October 2019

Technical Paper

12 References

[1] HANDBOOK OF MULTIPHASE FLOW METERING. https://nfogm.no/wp-

content/uploads/2014/02/MPFM_Handbook_Revision2_2005_ISBN-82-91341-89-
3.pdf
[2] Aker BP ASA, https://www.akerbp.com/
[3] Cognite AS, https://cognite.com/
[4] STIG HARALD HAMMER GUSTAVSEN “Parallel calibration of multiphase flow
meters (MPFM) based on measurements of phase streams in separators”, Master’s
Thesis 2019 Industrial IT and Automation. University of South-Eastern Norway,
Porsgrunn, May 2019, https://openarchive.usn.no/usn-
xmlui/handle/11250/2619328
[5] Norwegian Petroleum Directorate, Fact Pages.
https://npdfactpages.npd.no/Default.aspx?nav1=field&nav2=PageView%7cAll&na
v3=2845712&culture=nb-no

22
 North Sea Flow Measurement Workshop
22-25 October 2019

Technical Paper

Appendix A – Algorithm

The calibration figures are solved in a post calculation, since the all the trial windows is
needed to perform the calculation. The algorithm is written in the program language
python, and the stored datapoints are received from historical datapoints from the
Cognite Data Fusion repository (CDF), which collects data from Aker BP’s OsiSoft PI
system, that in turn collects data from both the control system and the measurement
system through OPC DA servers on each respective system onboard the Alvheim FPSO.
This data infrastructure entails a risk of the real-time datapoints stored historically may
not be of the same quality as the data inside the real-time embedded devices / flow
computers running on the SCADA level.

In the creation of the trials, initially the datapoints are all collected form the CDF through
the Cognite-SDK for the specific stream. Then a flashing of the multiphase stream is
executed, or if it is the reference stream the water in oil is removed from the oil stream
and added to the water stream. More details of how this is executed is shown in [2] as
referenced in the technical paper.

But each stream is collected in a trial collection, and each trial is collection of trials called
trials, which simplifies the calculations done where the data is structured in a manner to
combined data from each trial. The essential parts of the data structure explained and
shown in Figure 1 and is the data input into the calibration algorithm, but since each
stream is an object it contains also datapoints of intensive variables such as pressure,
temperature and soon densities and phase flow fractions.

In the algorithm the first thing done, and the most computationally expensive of the
entire calibration is where each phase in every trial is time synchronies, where the closes
datapoints in time is indexed toward each other in order to ensure that each phase in
each trial is using data, which has accumulated over the same time-period. This is due to
the spurious spreading of the time between each datapoint is different for each stream
and phase.

The next step is to create a frame of the Mp matrix, and is done by setting up a matrix
frame, containing information about which stream to be set into what cell in Mp. And
when the Mp matrix form is found, the system of equations can solve as shown in Figure
2 for a 2x2x2 calibration.

23
 North Sea Flow Measurement Workshop
22-25 October 2019

Technical Paper

OilMass:Accumulator
Bøyla:
MultiphaseStream:Stream GasMass:Accumulator

WtrMass:Accumulator

OilMass:Accumulator
Vilje:
MultiphaseStream:Stream GasMass:Accumulator
Trials:List[0]
WtrMass:Accumulator

OilMass:Accumulator
Volund:
MultiphaseStream:Stream
GasMass:Accumulator

WtrMass:Accumulator

OilMass:Accumulator
Separator:
SeparatorStream:Stream
GasMass:Accumulator

WtrMass:Accumulator

OilMass:Accumulator
Vilje:
MultiphaseStream:Stream GasMass:Accumulator

WtrMass:Accumulator

OilMass:Accumulator
Trials:List[1] Volund:
MultiphaseStream:Stream
Trials:List[] GasMass:Accumulator

WtrMass:Accumulator

OilMass:Accumulator
Separator:
SeparatorStream:Stream
GasMass:Accumulator

WtrMass:Accumulator

OilMass:Accumulator
Bøyla:
MultiphaseStream:Stream GasMass:Accumulator

WtrMass:Accumulator

OilMass:Accumulator
Trials:List[2] Volund:
MultiphaseStream:Stream
GasMass:Accumulator

WtrMass:Accumulator

OilMass:Accumulator
Separator:
SeparatorStream:Stream
GasMass:Accumulator

WtrMass:Accumulator

A Fig. 1 – Data structure / object diagram of trial data containing the data used to
perform the calibration of a 3x2x2 calibration of Third party Alvheim streams

24
 North Sea Flow Measurement Workshop
22-25 October 2019

Technical Paper

Trial 1 Trial 2 Trial 3

MPM Stream 1 MPM Stream 2 Separator Streams MPM Stream 1 MPM Stream 3 Separator Streams MPM Stream 2 MPM Stream 3 Separator Streams

Oil Gas Water Oil Gas Water Oil Gas Water Oil Gas Water Oil Gas Water Oil Gas Water Oil Gas Water Oil Gas Water Oil Gas Water
Cumualtive values

0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
1.0 1.0 1.0 1.0 1.0 1.0 2.0 2.0 2.0 1.0 1.0 1.0 1.0 1.0 1.0 2.0 2.0 2.0 1.0 1.0 1.0 1.0 1.0 1.0 2.0 2.0 2.0
2.1 2.1 2.1 2.1 2.1 2.1 4.1 4.1 4.1 2.1 2.1 2.1 2.1 2.1 2.1 4.1 4.1 4.1 2.1 2.1 2.1 2.1 2.1 2.1 4.1 4.1 4.1
3.4 3.4 3.4 3.4 3.4 3.4 6.4 6.4 6.4 3.4 3.4 3.4 3.4 3.4 3.4 6.4 6.4 6.4 3.4 3.4 3.4 3.4 3.4 3.4 6.4 6.4 6.4
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
5242 5242 5242 5242 5242 5242 10482 10482 10482 5242 5242 5242 5242 5242 5242 10482 10482 10482 5242 5242 5242 5242 5242 5242 10482 10482 10482
5245 5245 5245 5245 5245 5245 10490 10490 10490 5245 5245 5245 5245 5245 5245 10490 10490 10490 5245 5245 5245 5245 5245 5245 10490 10490 10490

Trials

Parallel calibration Algorithm

Oil Gas Water
MPM 1 MPM 2 MPM 3 referance
For each run For each run For each run
Trial 1 0

Trial 2 0
Mm Mm
Trial 2 0

Solve System of Solve System of

Solve System of equations equations equations

K-factors
Oil Gas Water
MPM1 MPM2 MPM3 MPM1 MPM2 MPM3 MPM1 MPM2 MPM3
Run
1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5
1.2 1.2 1.2 1.2 1.2 1.2 1.2 1.2 1.2
1.3 1.3 1.3 1.3 1.3 1.3 1.3 1.3 1.3
1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0
1.1 1.1 1.1 1.1 1.1 1.1 1.1 1.1 1.1
1.8 1.8 1.8 1.8 1.8 1.8 1.8 1.8 1.8
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
1.12 1.12 1.12 1.12 1.12 1.12 1.12 1.12 1.12
1.11 1.11 1.11 1.11 1.11 1.11 1.11 1.11 1.11
N 1.11 1.11 1.11 1.11 1.11 1.11 1.11 1.11 1.11

A Fig. 2 – Overview of the solving of the parallel calibration algorithm of 3 streams and
2x2x2 trial combination, where all phase streams in trials are time synchronized

The formation of the accumulated mass matrix Mp can also be solved with another cell
configuration, there is S! (factorial) number of trial combinations which can solve the system
of equations, and each can provide a collection of apparent k-factor developments. And
analysis of theses development is used to establish the resulting k-factor. When a
development of k-factors has stabilized a subset of theses stabilized data can then be used as a
statistical basis for calculation of a final k-factor, with pseudo-statistical stability values form
the subset.

Parallel calibration evaluation

For a quick overview of the data used by the algorithm the plots are set into a grid, with
the same dimensions as an augmented matrix of Mp and mp. The form of plotting gives

25
 North Sea Flow Measurement Workshop
22-25 October 2019

Technical Paper

an in-depth overview of the systems in question. Figure 3 shows an example of this

which is of the intensive variables of the system, which are of interest, such as the liquid
levels in the separator, water cut, pressure and temperature. By looking at the separator
stream on the first column this shows changing liquid oil levels and water cut out of the
separator, which give good indication of the process states during the time window of the
trial.

Fig. 3– Augmented matrix plot of Process conditions during trials, where the x axis is
successive raw datapoint during the trial, and therefore no numbers [2]

26