Professional Documents
Culture Documents
Ipc2022 86833
Ipc2022 86833
Ipc2022 86833
tte
IPC2022
nd
September 26-30, 2022, Calgary, Alberta, Canada
ee
R
ea
d-
O
IPC2022-86833
nl
y
C
op
y
A QUANTITATIVE RISK ASSESSMENT FRAMEWORK FOR NATURAL GAS
STORAGE WELLS
Thomas Dessein1, Brent Ayton1, Alex Fraser1, Shawn Smith1, Mari Shironishi2, Travis Sera2
1
Integral Engineering, Edmonton, AB, Canada
2
SoCalGas, Los Angeles, CA, USA
ABSTRACT NOMENCLATURE
A quantitative framework for risk assessment of natural gas 𝑐 consequences of a failure
storage wells, including the wellhead and connected piping, has
𝐶 conditional expression, which is either true or false
been developed to assess SoCalGas’s underground gas storage
sites in California. The approach has been developed to meet and 𝑐𝑉𝐶 historical incident rate calibration factor for vehicle
exceed the risk assessment requirements of API RP 1171 collisions
(incorporated by reference in U.S. regulation 49 CFR 192.12) 𝐷𝑣𝑖→𝑠𝑖 damage due to the impact between vehicle vi and
and the recent changes to the California Code of Regulations. the wellhead or associated piping section si
Further, several of the recommendations made in the recent
PHMSA study, “Risk Assessment and Treatment of Wells” 𝐹𝑉𝐶 annual frequency of failures (failures/well-year) due
(2021), have been addressed and incorporated. to vehicle collisions
𝑓𝑣𝑖𝑠,𝑠𝑖 factor to account for barriers that reduce the
The framework uses a dynamic fault tree to aggregate the probability of a collision between a vehicle and the
barrier failure rates from over 80 potential failure mechanisms wellhead or associated piping section si
to quantify the combined probability of an accidental release to 𝐼[ ] ‘indicator function’ which equals 1 if the
the atmosphere and the resulting hazards. The modular conditional expression within the [ ] is true and 0 if
architecture allows operators to use threat-specific models with it is false.
differing levels of sophistication. Additionally, the framework
accounts for the interplay between barriers to failure in a well, 𝐿𝑒𝑥𝑝,𝑠𝑖 length of the wellhead or associated piping section
the benefits of continuous monitoring, and the effect that si exposed to the potential for a vehicle collision
wellhead spacing, cement quality, and well-inflow performance 𝑁 number of Monte Carlo simulation trials
have on the expected consequences. p probability of a failure
Priority to develop quantitative models is given to the 𝑃𝐶 probability of failure for well component C
threats that potentially have high probability or high (ex. flange, hanger, tubing, casing)
consequences and to threats with effective mitigation options, 𝑃𝑟 probability of an accidental release event
such as corrosion and external interference threats. For threats
with very low consequences or likelihood of occurrence, simple r risk of a failure
models with conservative assumptions are typically sufficient 𝜎𝑣𝑚 von Mises stress
because the contribution to overall risk is low. This process
𝜎𝑎𝑥𝑖𝑎𝑙 stress component in the axial direction
minimizes the overall analysis complexity and allows mitigations
to be tailored to the higher-risk threats. 𝜎ℎ𝑜𝑜𝑝 stress component in the hoop direction
1
Contact author: tdessein@integraleng.ca
Completions
Reservoir &
for the interplay between barriers to failure in a well, the benefits
Connected
Wellhead
Caprock
Piping
of continuous monitoring, and the effect that wellhead spacing,
Well
Threat Categories
cement quality, and well-inflow performance have on the
expected consequences.
Corrosion • • •
The threats to gas storage wells and the risk framework are
described in Section 2, with Section 2.3 describing the Erosion • • •
constraints and requirements considered when developing the Manufacturing Defects • • •
risk framework to model these threats. The process to model the
Installation Defects • • •
overall well failure rates is described in Section 3, the approach
used to model failure rates of individual barriers in a well is Mechanical Damage • • •
described in Section 4, and the consequence assessment models Equipment Failure • • •
are described in Section 5. The framework can quantify the
External Interference • • • •
benefits of integrity management activities, such as an erosion
monitoring program, continuous pressure monitoring of the Well Intervention • •
annuli, and adding protections to minimize risk from vehicle Incorrect Operations & Maintenance • • • •
collisions. A more in-depth discussion of the use of this
framework for corrosion management is provided in the Weather & Outside Forces • • • •
accompanying paper, IPC2022-86794 [7], and for external Geological Uncertainty •
interference threats is in IPC2022-86734 [8].
TABLE 1: THREAT CATEGORIES AND AFFECTED
2. BACKGROUND SYSTEM COMPONENTS
2.1 Threat Identification
Both applicable regulations in California – API RP 1171 and
the recent changes to the California Code of Regulations –
require operators to develop a program to manage risk that
includes identifying potential threats to the storage operation
[1, 2, 3]. API RP 1171 includes a high-level description of
common potential threats. While the CalGEM regulations do not
explicitly enumerate potential threats, many are mentioned
throughout the regulation. An operator may supplement these to
identify potential threats more comprehensively.
Atmospheric
flange Release
hanger
OR
Release path 1 Release path 2 Release path 3 …
PAND
PAND
1 2 1 2
Wellhead
Tubing Casing Packer Flange
Leak
leak corroded leak leak
tubing
AND
casing
Hanger Flange
packer leak leak
FIGURE 1: STORAGE WELL SHOWING A SUBSET OF A dynamic fault tree can be represented equivalently by a
POSSIBLE LEAK PATHS directed graph [6], as shown in Figure 3. This form of the fault
tree is better suited to automation and provides a clear visual
If the barrier failure probabilities are assumed to be small, representation of the release flow paths in the context of a well.
the probability of a release for the example shown in Figure 1 The sequencing constraints (PAND gates), which only reach
can be approximated2 by Equation 2. failure if the input components fail in a prescribed order, are
reflected in the directed graph by the open circles surrounding
𝑃𝑟 ≈ 𝑃𝑡𝑢𝑏𝑖𝑛𝑔 ∙ 𝑃ℎ𝑎𝑛𝑔𝑒𝑟 ∙ 𝑃𝑓𝑙𝑎𝑛𝑔𝑒 + 𝑃𝑝𝑎𝑐𝑘𝑒𝑟 ∙ 𝑃𝑐𝑎𝑠𝑖𝑛𝑔 + 𝑃𝑓𝑙𝑎𝑛𝑔𝑒 (2) the node. Integral has developed a solution that can
automatically generate a directed graph from the information on
If the tubing-casing annulus is monitored for changes in
a well completion diagram or from structured data stored in a
pressure, the mode of failure of the casing affects the sequencing
WellView software database.
of events that can lead to a release. For a burst failure of the
casing to occur, the inner barrier (tubing or packer) needs to fail
tubing hanger flange
after the casing has corroded to a point where it would not handle
the full pressure. Otherwise, the rise in pressure in the annulus
between the tubing and casing can be detected, and the operator
can perform a well workover to remedy the leak before a release packer casing
to the atmosphere can occur. The same concept applies to wells Release
with multiple casing strings.
flange
2
The probability of either of two independent events, A or B, occurring can be answer is 0.0199. This simplification is useful for the purpose of discussion, but
calculated as P(A or B) = P(A) + P(B) – P(A and B). For small probabilities this in practice we apply the first equation which subtracts the union of the
can be approximated by P(A or B) ≈ P(A) + P(B). For example, if the probability probabilities and extend this for cases with more than two independent events.
of each event is 0.01, the approximation yields P(A or B) = 0.02 while the exact
4.1 Approach
4.2 Corrosion
The annual failure rate from the applicable threats to each
barrier in the well is modelled using combinations of the As a well ages, metal loss on the casing can grow, increasing
following quantitative methods. the probability of a failure from corrosion. Inspection and repair
programs manage this probability by reducing uncertainty in the
• Structural reliability models use physics-based casing condition and repairing significant metal loss anomalies.
probabilistic models to describe the failure mechanism and Multiple modes of failure are typically considered when
use statistical distributions to directly capture the assessing corrosion. A corrosion defect can fail by a small leak
uncertainty in measurements of degradation, the expected when a small portion of the metal loss goes through the casing
loading, material properties, the rate of degradation, and wall but the surrounding casing is strong enough to contain the
the accuracy of the model (model error). Examples include internal pressure, or by burst when the growth of the corrosion
the corrosion and erosion growth and burst models. defect degrades the pressure capacity of the casing to a point
• Historical failure rates are derived from historical failures where it can no longer contain the internal pressure. These limit
recorded in industry-wide datasets. With the low rates of states are shown in Figure 4 (adapted from [9]).
storage failures, these rates require pooling data from many
operators and typically only represent an average rate of
failure. This approach can be applied to well intervention
related failures.
• Fault tree analysis captures the relationships between
Leak
lower-level events that can combine to lead to a higher-
level failed state using Boolean logic (AND & OR gates). Wall
Thickness
Examples include vehicle collisions and excavation
damage. No Leak
Anomaly Depth
Reported Depth Simulation Depth Future Measured Depth Depth Adjusted in Repair Check
Burst Depth
Defect Depth
Defect Depth
Defect Depth
Burst Depth
3rd Inspection
3
This question was also studied in a recent project jointly funded by PRCI and However, the physical tests in the project did not consider the effect of axial
PHMSA [12]. The study found that for localized corrosion, where the uncorroded loading and only assessed metal loss with burst failure pressures greater than
sections surrounding the metal loss can provide constraint of the plastic 5,700 psi, well beyond the limits of normal storage operating pressures.
deformations, the von Mises ultimate tensile stress criterion can be exceeded.
FIGURE 6: EXAMPLE CORROSION ASSESSMENT AS A The estimated failure rate due to vehicle collisions for a
FUNCTION OF DEPTH 5 YEARS AFTER THE MOST given well construction and aboveground context is directly
RECENT INSPECTION proportional to the vehicle traffic rates. The frequency of
scheduled and unscheduled activities involving the use of
vehicles for each well, as well as estimates of the number of
vehicles involved, mass, and speed of those vehicles for each
activity, were used with site traffic networks to determine the
frequency of traffic passing by each well, grouped by speed and
mass. An example of a representative traffic network, colored by
(a) PIR Model (b) WMPS Model – Vertical Flame (c) WMPS Model – Horizontal Flame
FIGURE 10: COMPARISON OF HEAT AFFECTED ZONES
The WMPS model can reflect the effects of non-vertical The model to assess the knock-on failure potential at a gas
fires, lift-off, and shielding from buildings and the site storage well site consists of:
topography, as described in Smith et al. [21]. Flame lift-off refers
• The weighted multiple point jet fire model described in
to the distance between the release location and the visible flame
Section 5.2 above.
envelope, as shown in Figure 11. The released gas can have
• A transient heat transfer model, which models the
outlet speeds exceeding several hundred feet per second,
temperature and pressure changes within wells and other
resulting in a lift-off distance of more than 50 ft from the gas
pressurized equipment near a jet fire in response to the
travelling upward for only a small fraction of a second (~0.02s)
incident radiation predictions from the jet fire model.
before enough oxygen is incorporated for combustion to occur.
• An equipment failure model, which accounts for the
decreasing material strength capacity as temperature
increases. This model uses the pressure and temperature
predictions from the heat transfer model to determine when
or if a failure of the wellhead components, lateral piping,
or flanges will occur.
• A knock-on timeline model, which combines the three
Lift-off
models above to simulate the progression of knock-on
failures over time following an initial ignited release.
An example simulation is shown in Figure 12 with an
FIGURE 11: WEIGHTED MULTIPLE POINT SOURCE ignited small release on the wellhead that is modelled to progress
(WMPS) MODEL WITH LIFT-OFF to a full rupture of the wellhead. A small leak also occurred
during the simulation on the adjacent well, but the surface safety
5.3 Knock-on Failures valve (SSV) was triggered and prevented the propagation of
knock-on failures to the rest of the well. The initial release is
If wells are closely spaced together, an ignited release from shown in time step 0, the knock-on to the adjacent well occurs in
one well can lead to the failure of an adjacent well, commonly step 9 (see blue arrow), a full wellhead rupture occurs in step 12,
referred to as the domino effect or a knock-on failure. Reports and the adjacent well shut-in section is fully emptied in step 16
from the California Council on Science and Technology and after the SSV has triggered in step 13.
Lawrence Berkeley National Laboratory note the potential for an
ignited well rupture to lead to multiple wellhead failures as well
as the use of risk assessments to identify optimal interventions to
prevent such low-probability high-consequence events [22, 23].
These reports were incorporated into the California Department
of Conservations’ recent rulemaking on gas storage. In addition,
a relatively small, ignited release has the potential to cause heat
damage to the wellhead, reducing its pressure capacity and
ultimately leading to a full rupture of the wellhead.