Академический Документы
Профессиональный Документы
Культура Документы
A BSTRA C T
A model for subsea oil/gas production systems, being useful for carrying out
production regularity studies, is presented. Formulas for calculating
approximate (asymptotic) mean values of interesting reliability parameters
within this model are derived. The use of these formulas is exemplified and
discussed for a rather simple example. It is argued that the approximate
formulas should be used in an interaction with Monte Carlo simulation, giving
a powerful approach to the problem of assessing the availability (production
regularity) of a subsea production system.
1 INTRODUCTION
By planning of subsea production systems for oil/gas, there will be a main
objective to predict the operational features of a given concept. In particular,
it will be essential to evaluate/compare various
--layouts
- - m a i n t e n a n c e and intervention strategies
in order to find .the economically optimal solution (among those satisfying
the safety requirements).
For a subsea oil/gas production system, it will be of major interest to
A version of this paper was presented at Reliability '87, 14-16 April 1987, Birmingham, U K,
and is reproduced by kind permission of the organisers.
127
Reliability Engineering and System Safety 0951-8320/88/$03"50 Elsevier Applied Science
Publishers Ltd, England, 1988. Printed in Great Britain
128
P. Hokstad
129
Production
stations (PS)
PS module
(critical to all
production on template)
Wells
~Well
module,
(critical for the production
of a well only)
130
P. ttokstad
131
The PS and well availabilities will depend on the number of wells at the
actual PS. Thus the following parameters are also introduced:
Aw(n) = Availability of a well that is linked to a PS, having n wells in
total.
Aps(n) -- Availability of a PS, in total having n wells.
Thus Aw and Aps will be obtained by an appropriate weighting of the various
Aw(n ) values and Aps(n) values respectively.
132
P. ttokstad
(la)
(lb)
(lc)
k=l
(nk/N).Aw(nk)
(3)
133
Aps(nk)= [1 + B3(0)] s- 1
k=l
Aw(nk)
(4)
k=l
Finally
Acs = [-1 + B1]- 1
(5)
These availability measures, given by (3), (4) and (5), are rather fundamental,
and several other reliability parameters easily follow from these (see
Appendix 2). A discussion concerning the validity of the above formulas will
be given in the next Section. However, the formulas directly inform that the
really fundamental factors that influence the availability are the following:
- - T h e product of the total failure rate and the mean mobilization time (i.e.
2i.Wi) for the modules of
* well,
* PS, and
* CS
respectively.
- - T h e product of the total failure rate and the mean repair time (2i. Ri) for
* well,
* PS, and
* CS
respectively.
- - T h e total number of wells (N) and the way these are allocated to various
PS. (Thus the number of PS (that is s) of course also plays an important
role.)
This list should be no surprise. However, it should be stressed that the above
list should be considered to be the most fundamental factor to be taken into
consideration for a first rough reliability evaluation of the concept. (Observe
that there is at least one other candidate to this list, that is p = average
proportion of time that the whole PS is shut during the repair of a well
module. In the previous discussion it is assumed t h a t p = 1.0. Otherwise thep
value would enter the formulas.)
4 DISCUSSION OF T H E A S Y M P T O T I C F O R M U L A S
The assumptions leading to the formulas (3)-(5) will now be discussed. This
discussion together with the numerical investigations of the next Section will
rather clearly explain when these formulas should (or should not) be used.
There are two main factors restricting the applicability of these formulas.
134
P. Hokstad
Stationarity
For deriving (3)-(5) the system is assumed to be in a stationary (not time
dependent) state. This means that all variations during time should be
random, and thus that there is no systematic 'trend'. However, in an actual
situation, the system should probably start with no failures at time 0
(availability = I'0), and it will take some time before the availability
decreases to the asymptotic value that is provided by the formulas. Thus the
validity will depend both on how fast the system reaches stationarity, and on
the length of the life cycle of the field.
- - T h e 'speed' of convergence (towards the stationary state of the process)
will mainly depend on the failure distributions (the statistical
distributions of the time until failure for the various modules). If the
exponential distribution is used, stationarity will be reached rather
quickly. (But for the asymptotic formulas to be correct, there should in
fact be a certain probability for having failed modules also at time 0.)
Using a Weibull distribution with increasing failure rate instead, there
will usually be a longer time until failures start to occur and stationarity
is reached. The Weibull model being referred to here assumes that the
failure rate of the time until first failure is of the form
~(1) = ~ / ~ . ( / ~ I ) ~ - 1
'Multiple failures'
Occasionally two (or more) failures will occur close in time, and the repair
can be carried out during the same intervention. The effect of such events are
not accounted for in the formulas (3)-(5). Quite similarly the total repair time
for a 'multiple repair' on a template (PS) should typically be less than the sum
of the two repair times.
Thus, since actually some of the mobilization times and repair times will
be common for two or more failures, the availability obtained by the
135
approximate formula will be too low. Three factors are important for this
phenomenon:
--Failure rates. If the failure rates are high, there will be many multiple
errors, and the formulas become inaccurate.
--Mobilization and repair times. Long mobilization/repair times will also
tend to increase the number of multiple failures.
--Number of wells. Also if there is a very large number of units (wells),
there will be a tendency to experience multiple errors.
Observe that the influence of the factors listed above all imply that the
availability formulas represent lower limits of the actual values. However,
there is one exception: if the failure time distribution has a decreasing failure
rate, this might in the initial phase of the life cycle give a number of failures
that is higher than the asymptotic value (and thus give an availability that is
actually lower than the asymptotic value).
5 N U M E R I C A L E X A M P L E 'BASE CASE'
The use of the formulas derived in Section 3 will now be illustrated by some
numerical examples. An exponential model will first be assumed for the
failure times. Further the following numeric values are specified as the 'Base
Case' of the numerical investigations:
- - T o = total life time of the field = 10 years.
- - T h e r e are two production stations with two and four wells respectively
(thus s = 2, nl = 2,//2 = 4).
- - T h e mean mobilization time equals 90 days for rig, and 24 days for
DPS.
- - T h e total failure rates (summed over all modules) of CS, PS and wells
respectively are given in Table 1. It will be seen that all repair of CS is
carried out by a rig, giving a mean waiting time of 90 days. For PS
modules all repairs are by DPS, giving a waiting time of 24 days. For the
wells there is assumed to be one failure rate of 0.1 (yr-1), requiring
repair by rig, and one failure rate of 0.2 for failures that can be handled
by DPS (giving an average mean waiting time of (90 0-1 + 24 0.2)/
(0.1 + 0.2) = 46 days). So this value is obtained by taking an average over
the various well modules. Table A1 in Appendix 1 presents the input
values to the asymptotic formulas, based on the data in Table 1.
Some (200) simulations for the Base Case were carried out. It is seen in Table
2 that there is in fact a very good agreement between the values obtained
136
P. Hokstad
TABLE I
cs
PS
Well
365. PP"
365. R
0.06
0-20
0.30
90
24
46
10
6
14
365( W +
R)
1oo
30
60
from the simulations and the approximate formulas. The simulated (i.e.
'actual') values of Aw and Aps are as expected slightly higher than those
obtained from (3) and (4). But it is realized that stationarity must be reached
very fast (compared with the life time of 10 years), and also the total
frequency of failures must be rather small, to give a small n u m b e r o f multiple
repairs.
It should be observed that the simulations have been carried out under the
following two simplifying assumptions concerning the maintenance
strategy.
- - I n the case o f a 'multiple repair' at a PS (that is more than one module
repaired in the same intervention), the total repair time is reduced to
equal the repair time of that component requiring the longest repair.
- - T h e DPS is demobilized whenever a failure occurs that requires the use
of a rig. (Thus there are never two vessels mobilized at the same time.)
The first assumption implies that the difference between the result of the
formula and of the simulation is 'maximized' (a more realistic rule
concerning multiple repairs would give that the results for the formula and
the simulation become even closer).
It should, however, be stressed that for this example there is a rather low
number of multiple repairs, and the above two assumptions do in fact have a
very limited influence on the simulation results.
TABLE 2
Formula
Simulation
Aw
Avs
0'900
0.904
0"938
0.940
137
0"75
1 "00
1"25
1"50
1 "75
10
50
0"848
0"906
0"940
0-960
0'975
0'882
0'904
0'920
0-927
0-932
0"897
0"904
0"906
0'907
0"908
138
P. Hokstad
>1
krr 98
tI
2 YEARS
_J . 9 6
t~
94
5_1
n,,
.92
z
o
9
HU
.88
"7
i0 Y E A R S
50 Y E A R S
" 86
o_ . 8 4
.#5
i.'2s
SHRPE
i!5
i.bs
'
PARAMETER
'Multiple repairs'
Secondly the effect of multiple repairs is investigated. First the failure rates
are varied. Since the accuracy is quite good for the Base Case, there is no
reason to consider cases with lower failure rates. Thus in total the following
four cases are investigated:
Case
Case
Case
Case
1:
2:
3:
4:
Base Case.
All failure rates are twice the values of the Base Case.
All failure rates are three times the values of the Base Case.
All failure rates are four times the values of the Base Case.
Table 4 and Fig. 3 give the well availabilities obtained both from the formula
and the simulations for these four cases. It is observed that the accuracy is
surprisingly 'resistant' to increase in the failure rates. Even for Case 4 (having
unrealistically high failure rates), the value obtained from the formula is very
TABLE 4
Well Availability as a Function of the Failure Rates
Formula
Simulation
Case 1
Case 2
Case 3
Case 4
0.900
0"904
0"817
0"828
0-746
0'761
0"685
0702
139
>-
.9
fv
(Z
..J
.86
9
W
n/
.82
Z
0
.78
LATION
N-
U
Q
0
n~
n
FORMULA ~ ~
.74
.7
CRSE
Fig. 3.
1 2
2-4
4-8
6-12
8-16
Formula
Simulation
0.916
0-919
0-900
0.904
0.870
0.880
0'842
0.865
0.816
0.853
140
).P)---4
P. Hokstad
"
92
FF
IZ
_l
.9
(.9
W
n,, . 8 8
Z
o
b.-I
86
FU
I:Z:l "
O
FF
FOR.MULA
84
.82
1-5
2-4
3 3
Formula
Simulation
0.891
0.902
0.900
0.904
0.904
0.910
141
TABLE 7
Availability for Different Allocations of the Wells. (In total 20 Wells
divided between the various PS)
No. P S
No. wells
1-19
5-15
10-10
6-7-7
Formula
Simulation
0.780
0.842
0.823
0-855
0-841
0.863
0.870
0.878
5-5-5-5 4 ....
0.885
0.889
0.894
0'895
142
P. Hokstad
2.
3.
143
144
P. Hokstad
Central
station
(CS)
Production
stations (PS)
PS
Wells
Wells
S a t e l l i t e wells
Fig. 5.
145
and field data. This should be connected to a data base (which should be
another module).
--Various modules for carrying out several analyses should be available.
Relevant techniques are:
Thus the computerized tool should be flexible both with respect to model
specification and choice of analysis. In this way having several analysis
techniques available, using the same user interface, would give a powerful
tool in the process of for instance finding concepts that maximize production
regularity.
ACKNOWLEDGEMENT
This work is carried out with the financial support of the Norwegian Council
for Scientific and Industrial Research (NTNF, project SERA, 9204.16888).
REFERENCES
1. Hokstad, P. SUBSIM: A Computer Program for Reliability and Availability
Assessment of Subsea Oil/Gas Production Systems, SINTEF Report STF75
A86002, 1986.
2. Windebank, E. Reliability Engineering, 5 (1983), pp. 73-81.
3. Hulb~ekdal, J. M.Sc. thesis, Department of Mathematical Statistics, The
Technical University of Norway, Trondheim, 1987.
A P P E N D I X 1. I N P U T D A T A TO T H E N U M E R I C A L
CALCULATIONS
From Table 1 it follows (see eqn (1)), that for the Base Case
B 1 = B a = 6/365,
B3(n) = 0.3(46 + 14n)/365
The calculated values for B3(n ) and Aw(n) for n = 1,..., 20, are as given in
Table A1.
For the Base Case only n = 2 and n = 4 are needed, but to investigate the
influence of splitting 20 (or 6) wells between several PS, the other entries of
the table will be also needed.
P. Hokstad
146
TABLE AI
Input to the Numerical Calculations
n
B3(n)365
Aw(n)
B3(n)365
Aw(n)
1
2
3
4
5
6
7
8
9
10
18-0
22-2
26.4
30.6
34.8
39.0
43-2
47-4
51.6
55-8
0'923
0.913
0"904
0.894
0"885
0'876
0-867
0.858
0.850
0.841
11
12
13
14
15
16
17
18
19
20
60.0
64.2
68.4
72.6
76"8
81.0
85.2
89.4
93.6
97.8
0'833
0.825
0-817
0.809
0.802
0.794
0.787
0.780
0.773
0.766
A P P E N D I X 2. O T H E R RELIABILITY P A R A M E T E R S
Having assessed the average availability/production regularity of a concept,
as described in this paper, it is also possible to estimate several other
parameters. Just a couple of examples are given here.
l.
2.
For the Base Case this gives 6 x 0-3 x 10 0"9 = 16.2. This is almost
identical to the value obtained from the simulation (16-4 + 0.5).
Total repair time. First the number of PS failures, F2, and the number
of CS failures, Ft, are estimated in the same way as F 3. Then the total
repair time is estimated by
R-to T = F~R x + F 2 R z + F3R3