You are on page 1of 10

Proceedings of the ASME 2022 14th International Pipeline Conference

IPC2022
September 26–30, 2022, Calgary, Alberta, Canada

IPC2022-87259

A MORE EFFICIENT EFFECTIVE AREA METHOD ALGORITHM FOR CORROSION


ASSESSMENT (FASTER RSTRENG)

Jason Yan*, Dongliang Lu, Ian Khou, Shenwei Zhang

TC Energy Corporation
Calgary, Alberta, Canada

simulation technique where millions of simulations are


ABSTRACT* required.
Corrosion is one of the major threats to the safety and In addition, TC Energy has developed a more accurate and
structural integrity of oil and gas transmission pipelines. The precise corrosion assessment model, i.e., plausible profiles
corrosion threat is usually managed by regular in-line (Psqr) model, published in IPC 2018 and 2020, and reviewed
inspection (ILI). by industry experts through PRCI. The use of Psqr model needs
The effective area method (RSTRENG) is the most popular to generate hundreds of profiles for one corrosion anomaly and
corrosion assessment model to convert the measured corrosion repeatedly uses the effective area algorithm. This needs
size to predicted burst pressure. Given a detailed corrosion extensive computing efforts. When Psqr model was built into
measurement profile, the effective area method involves an the probabilistic assessment framework, the computing efforts
iterative process to find the minimum burst pressure. As stated were dramatically increased. Therefore, it is of great
in ASME B31G, “for a corroded profile defined by n importance to have a more efficient effective area algorithm to
measurements of depth of corrosion including the end points at facilitate the probabilistic assessment using Psqr model.
nominally full wall thickness, n!/2(n − 2)! iterations are A case study was used to demonstrate the efficacy of the
required to examine all possible combinations of local metal developed algorithm by comparing the computation time using
loss with respect to surrounding remaining material”, the different effective area algorithms. Engineers who are already
widely used effective area algorithm has at least an order of n- familiar with RSTRENG model, might jump to the Faster
square time complexity (O(n2)). As n increases, the RSTRENG section and Annex A directly.
computation time increases nonlinearly.
This paper reviewed the traditional RSTRENG algorithm INTRODUCTION
first, and demonstrated that it is not necessary to always loop Metal-loss corrosion continues to be one of the most
through all the combinations and check the corresponding burst significant threats to the safety and structural integrity of
pressure one by one. Because some combinations with pipeline systems. In-line inspection (ILI) using magnetic flux
shallower and shorter corrosion size are certainly not the final leakage (MFL) or ultrasonic (UT) tools has been recognized as
critical combination corresponding to the minimum burst the most effective means to manage integrity of energy
pressure. A more efficient algorithm (Faster RSTRENG) is pipelines. After ILI, operators need to identify the critical
proposed and presented in this paper, which can reduce the anomalies, determine appropriate response time, and whether
algorithm computation time significantly. pressure restriction (derate) is required. Repair decisions, such
While the impact of efficiency of the algorithm on the as recoat or sleeve, are made after excavations with field
integrity assessment process is insignificant if solely measured anomaly size.
conducting burst pressure calculation in a deterministic way, a For pressure-driven integrity decisions, the assessment
highly efficient algorithm is desired when assessing a large model used to convert measured corrosion size to predicted
volume of corrosion anomalies by the reliability-based burst pressure has the largest impact to the integrity decisions
(probabilistic) assessment method using Monte Carlo [1]. An accurate and precise assessment model is the key to
optimize operators’ decisions and ensure pipe safety [2, 3].
Corresponding author: jason_z_yan@tcenergy.com
* The effective area method [4] is one of the most popular
corrosion assessment model, commonly used and accepted by

1 Copyright © 2022 by ASME


regulators, operators and ILI vendors. It is originally published ILI assessment using Monte Carlo method [17, 18], millions of
in PRCI report PR-3-805 [4], and later included in ASME simulations are required for each anomaly. The computing
B31G [5]. The effective area method is evolved from earlier efforts are increased dramatically.
models. The chronology of the development is: In addition, to reduce the conservatism of the effective area
• NG-18 Ln-Sec, 1973 [6] method [2, 4], TC Energy has developed a new corrosion
• ASME B31G, 1984 [7] assessment model, i.e., plausible profiles (Psqr) model,
• Modified B31G (or known as 0.85dL), 1989 [4, 5] published in IPC 2018 and 2020 [19, 20], and reviewed by
• Effective Area, 1989 [4, 8-11] industry experts through a PRCI project [21, 22]. Psqr model
The effective area method has been adopted by various considers the circumferential separation between corrosion pits.
regulations and industry standards such as US Code of Federal Instead of the one “river bottom” profile used by the effective
Regulations (CFR, Title 49, Part 192 & 195), ASME B31.4 method, to predict the burst pressure more accurately and
[12], ASME B31.8S [13] and CSA Z662 [14]. precisely, Psqr model generates hundreds of profiles for each
For a given detailed corrosion measurement profile, the corrosion cluster and repeatedly uses the effective area
effective area method involves an iterative process to find the algorithm [2, 3, 15, 19-22]. Psqr model is discussed to be
minimum burst pressure. Therefore, it is usually implemented included in the next revision of ASME B31G [5] and CSA
by computer programs. As the original Microsoft DOS based Z662 [14].
computer program in PRCI PR-3-805 [4] developed by Battelle When performing probabilistic assessment, or using Psqr
Memorial Institute for American Gas Association (AGA), was model, or even performing probabilistic assessment by using
called RSTRENG (Remaining STRENGth), the effective area Psqr model, an ILI tally with 100 thousand reported corrosion
method is also well known as RSTRENG model in the pipeline clusters using millions of simulations might take millions of
industry today. minutes. Therefore, a more efficient effective area algorithm
As stated in Section 1.10 of ASME B31G [5], alternative becomes the key component to optimize the performance.
software other than the original RSTRENG disc is allowed to The following sections of this paper review the traditional
perform corrosion assessment if results are validated against RSTRENG algorithm first, demonstrate that it is not necessary
published benchmark test data such as [11]. Following PRCI to always loop through all the combinations and check the
PR-3-805 [4] and ASME B31G [5], the effective area method corresponding burst pressure one by one. Because some
has been rewritten as local or cloud-based modern computer combinations with shallower and shorter corrosion size are
software packages or spreadsheets by ILI vendors, regulators, certainly not the final critical combination corresponding to the
operators, and other companies in the industry (e.g., PRCI EC- minimum burst pressure. A more efficient algorithm (Faster
2-9 Psqr project spreadsheet [15], and the public available RSTRENG) is proposed in this study, which can reduce the
KAPA spreadsheet from Kiefner & Associates). algorithm computation time significantly.
When using effective method with detailed corrosion A case study comparing the computation time by using
profile inputs, currently most, if not all, of the software different effective area algorithms is presented at the end. For
packages or spreadsheets loop through and calculate the large corrosion clusters, with high resolution ultrasonic or laser
predicted burst pressure for all the possible sub-sections scan measurements, the more efficient algorithm could be 10-
(combinations) one by one. The minimum pressure of all 100 times faster.
combinations is used as the final output. Detailed steps are
discussed in the following sections. ASME B31G LEVEL 1 MODELS
As stated in ASME B31G [5], “for a corroded profile As all ASME B31G [5] family models share the common
defined by n measurements of depth of corrosion including the format, the original model published in 1984 [7], and the
end points at nominally full wall thickness, n!/2(n − 2)! modified B31G model (also known as 0.85dL model in CSA
iterations are required to examine all possible combinations of Z662 Clause 10.10.2.6 [14]) are reviewed here first before the
local metal loss with respect to surrounding remaining effective area method. When using the original B31G and
material”, the widely used effective area algorithm has at least modified B31G models, the corrosion size is only characterized
an order of n-square time complexity (O(n2)) (big O notation by the peak depth, dmax and total length, L. Therefore, they are
[16]). As n increases (e.g., longer corrosion or the same also called Level 1 models in ASME B31G [5] Section 2.2.
corrosion measured by higher resolution tools), the The predicted burst pressure of corroded pipe is:
computation time increases in a nonlinearly accelerating manor. 𝑃𝑏 = 𝑃𝑏𝑖𝑛𝑡𝑎𝑐𝑡 ∗ 𝑅𝑆𝐹 = 𝜎𝑓 ×
2𝑡
× 𝑅𝑆𝐹
𝐷
When conducting deterministic assessment using modern 𝐴 𝑑𝑚𝑎𝑥
1−
local computer or cloud-based servers, the effective area = 𝜎𝑓 ×
2𝑡
×
𝐴0
𝐴 = 𝜎𝑓 ×
2𝑡
×
1−𝑆𝐹×
1 𝑑
𝑡
(1)
algorithm will not impact the performance or user experience 𝐷 1−
𝑀𝐴0
𝐷 1− ×𝑆𝐹× 𝑚𝑎𝑥
𝑀 𝑡

significantly. For example, an MFL ILI tally with 100 thousand where Pb_intact is the predicted burst pressure of defect free
reported corrosion clusters, or hundreds of corrosion anomalies intact pipe. RSF stands for remaining strength factor, which is a
measured by UT ILI or field laser scan with detailed 3- real number between zero and one, defined as Pb/Pb_intact, and
dimensional (3D) maps could usually be assessed within used to characterize the severity of corrosion anomalies. RSF
minutes. But when performing reliability-based (probabilistic)

2 Copyright © 2022 by ASME


depends on corrosion size and pipe geometry only. D, t, f and scan, e.g., 2mm by 8mm, or 2mm by 1mm, is much higher than
y are the outside diameter, wall thickness, flow stress and yield MFL boxes. Therefore, for the same corrosion anomaly, UT or
stress, respectively. laser scan has more granular characterization of metal loss
As shown in Figure 1, A and SF are the corroded area and compared to MFL flat-bottom rectangular boxes.
shape factor, respectively. When Level 1 assessment methods As the pipe hoop stress is much larger than the longitudinal
are used, the corroded area is a function of the assumed shape stress and radial stress, the impact of corrosion on the pipe
factor, peak depth, dmax and total length, L. A0 is the reference capacity holding internal pressure is usually evaluated along the
area and defined as L × t. longitudinal direction. The “river bottom” profile is the
L commonly used 2D profile to represent the 3D corrosion
measurements, which is generated by connecting the deepest
measured points of each circumferential pipe slice and
dmax projecting them to the longitudinal cross section as shown in
t Figures 2(a) and 2(b). Psqr model [19, 20] generates multiple
A A0 2D plausible profiles instead of the most conservative “river
bottom” profile to better characterize the corrosion.
Figure 1 Corrosion Size and Shape
This paper uses the “river bottom” profile as an illustrative
M is bulging factor and also known as Folias factor [23], example. The same effective area algorithm could be used to
which is given by: convert any 2D corrosion profile to predicted burst pressure.
𝐿2 For MFL “river bottom” profiles, the effective area
𝑀𝑜𝑟𝑖𝑔𝑖𝑛𝑎𝑙 = √1 + 0.8 𝐷𝑡 (2a) algorithm creates new “segments” whenever depth changes.
or Therefore, each segment has uniform depth as shown in Figure
2
2(a). As the MFL reported boxes do not have the same length,
2 2 𝐿2
√1 + 0.6275 𝐿 − 0.003375 ( 𝐿 ) ≤ 50 so the length of MFL “river bottom” segments varies. Most of
𝑀={ 𝐷𝑡 𝐷𝑡 𝐷𝑡 (2b)
3.3 + 0.032
𝐿2 𝐿2
> 50
time, whether one segment with uniform depth (formed by one
𝐷𝑡 𝐷𝑡 MFL box) is allowed to be split into multiple shorter
The model settings are summarized in Table 1. The shape subsegments or not, has no impact to the predicted burst
factor is defined as A/(L×dmax), which is between zero and one. pressure using the effective area method. However,
The Level 1 models are generally conservative, because the counterexamples exist (see Annex B). When using Psqr model
assumed shape factors, either 2/3 or 0.85, are fixed rough [19, 20], the MFL 3D maps need to be meshed to evenly spaced
estimation of the metal loss area as shown in Figure 1. The grids, e.g., 5mm by 5mm or 1mm by 1mm. Therefore, the MFL
accuracy and precision of models were compared and discussed “river bottom” profiles could be converted to evenly spaced
in past studies [2, 15, 19, 24-27]. In addition, recently some rectangular metal-loss area “bars” with the same length, see
new equations are developed to estimate the burst pressure of Figure 2(d).
intact pipe [27, 28], which are beyond the scope of this paper. The UT or field laser scan measurement points are usually
Table 1: ASME B31G Family evenly spaced, so the measurement interval is equal to the
Model Flow Stress, f Shape Factor Bulging Factor
segment length and defined as x = xi+1 – xi, as shown in
Original B31G 1.1 × y 2/3 Eq. (2a) Figures 2(c) and 2(d). There are two ways to characterize the
Modified B31G, or measured depth, e.g., the trapezoid or rectangle bar assumption,
y + 10 ksi 0.85 Eq. (2b)
0.85dL method as indicated in Figures 2(c) and 2(d). For n measurement
Effective area, or
y + 10 ksi deff /dmax_Aeff Eq. (2b) or (4c) points, the total length of the corrosion is (n - 1) x, or n x,
RSTRENG
based on different assumptions. The area of the ith segment Ai is
(di + di+1) x / 2, or di x., where di is the ith measured depth.
CORROSION CHARACTERIZATION
As UT or field laser scan usually has at least hundreds of
Detected and measured corrosion anomalies could be
measurement points, the impact of this assumption on the
characterized not only by their peak depth (dmax), total length
predicted burst pressure is negligible. The trapezoid assumption
(L), and total width (W), but also the details within the
could always be converted to rectangle bar assumption by
corrosion anomalies (clusters), e.g., the depth distribution,
letting (di + di+1) / 2 be the ith measured depth based on the
overall shape, or called morphology.
rectangle bar assumption.
After ILI, MFL signals are usually converted to flat-bottom
This paper focuses on 2D corrosion profiles formed by
rectangular metal-loss boxes and reported in pipe tallies. A
evenly spaced uniform depth rectangular metal-loss area bars as
plotted example is shown in Figure 2(a). Figure 2(b) shows an
shown in Figure 2(d) hereafter without losing the generality.
example of corrosion anomaly measured by ultrasonic (UT)
tools or field laser scan. In general, the resolution of UT or laser

3 Copyright © 2022 by ASME


Ai

di+1
di
xi xi+1

(c)

Ai

di
xi xi+1

(a) (b) (d)

Figure 2: Corrosion Measurement Examples


A1 A2 A3 A4 A5

A6 A7 A8 A9 A10

Figure 3: Effective Area Method Combinations

𝑞
𝐿𝑐𝑜𝑚𝑏_𝑗 = ∑𝑖=𝑝 𝐿𝑖 , (1 ≤ 𝑝 ≤ 𝑞 ≤ 𝑛) (3b)
EFFECTIVE AREA METHOD (RSTRENG) 𝑗 = [2𝑛 − (𝑞 − 𝑝 − 1)]/2 × (𝑞 − 𝑝) + 𝑝, (1 ≤ 𝑝 ≤ 𝑞 ≤ 𝑛) (3c)
Consider a 2D profile with n uniform depth segments,
denoted as S1, S2, …, Si, … Sn. The corresponding depth and 𝑑𝑎𝑣𝑒_𝑗 = 𝐴𝑐𝑜𝑚𝑏_𝑗 /𝐿𝑐𝑜𝑚𝑏_𝑗 , (1 ≤ 𝑗 ≤ 𝑛(𝑛 + 1)/2) (3d)
length of each segment are d1, d2, …, di, …, dn and L1, L2, …, 𝐴𝑜_𝑐𝑜𝑚𝑏_𝑗 = 𝑡𝐿𝑐𝑜𝑚𝑏_𝑗 , (1 ≤ 𝑗 ≤ 𝑛(𝑛 + 1)/2) (3e)
Li, … Ln. It is noted that some segments could have zero depth. where i, j, p, q, n are all integers. Equations (3a) to (3e) indicate
Given n uniform depth segments, a total of n (n + 1) / 2 that all n (n + 1) / 2 combinations could be derived by varying p
sub-sections (or called combinations hereafter) could be and q.
generated by combining the adjacent segments. For example, a The effective area method (RSTRENG) checks the
profile with 4 segments could generate a total of 10 calculated burst pressure for all the possible segment
combinations as shown in Figure 3. combinations one by one, and the minimum value is used as the
It is noted that the number of combinations here is n (n + predicted burst pressure for the 2D profile.
1) / 2 instead of n (n - 1) / 2 in ASME B31G [5], because the 2𝑡
notation is different. ASME B31G [5] uses n measurement 𝑃𝑏 = min {𝑃𝑏_𝑗 } = 𝜎𝑓 × 𝐷
× min {𝑅𝑆𝐹𝑗 } (4a)
1≤𝑗≤𝑛(𝑛+1)/2 1≤𝑗≤𝑛(𝑛+1)/2
points to form n - 1 trapezoid segments. As described in the 𝐴𝑐𝑜𝑚𝑏_𝑗
𝑑𝑎𝑣𝑒_𝑗
previous section, they are mathematically equivalent. 1−
𝐴0_𝑐𝑜𝑚𝑏_𝑗 1−
𝑅𝑆𝐹𝑗 = = 𝑡
(4b)
The metal-loss area, length, average depth and reference 1−
1
×
𝐴𝑐𝑜𝑚𝑏_𝑗
𝑀𝑗 𝐴0_𝑐𝑜𝑚𝑏_𝑗
1−
1 𝑑𝑎𝑣𝑒_𝑗
𝑀𝑗
×
𝑡
area of the jth combination are denoted as Acomb_j, Lcomb_j, dave_j
and A0_comb_j, respectively, where j = 1, 2, …, n (n + 1) / 2. They 2
√1 + 0.6275 𝐿𝑐𝑜𝑚𝑏_𝑗 − 0.003375 (𝐿𝑐𝑜𝑚𝑏_𝑗 )
2 2
𝐿𝑐𝑜𝑚𝑏_𝑗 2
≤ 50
could be calculated by: 𝑀𝑗 = { 𝐷𝑡 𝐷𝑡 𝐷𝑡 (4c)
𝐿𝑐𝑜𝑚𝑏_𝑗 2 𝐿𝑐𝑜𝑚𝑏_𝑗 2
𝑞 3.3 + 0.032 > 50
𝐴𝑐𝑜𝑚𝑏_𝑗 = ∑𝑖=𝑝 𝑑𝑖 𝐿𝑖 , (1 ≤ 𝑝 ≤ 𝑞 ≤ 𝑛) (3a) 𝐷𝑡 𝐷𝑡

4 Copyright © 2022 by ASME


where f = y + 10 ksi as listed in Table 1. It is noted that as D, 1) / 2 = 16110 combinations. There are 179 and 178
t and f are all constant for all combinations. As shown in Eq. combinations with length equal to x and 2x, respectively.
(4a), it is more efficient to calculate the remaining strength The “river bottom” profile with 179 segments in Figure 4 is
factor (0 ≤ RSFj ≤ 1) for all combinations first and convert it to used as an example hereafter in this paper.
the final burst pressure after taking the minimum of all RSF. The average depth of the n (n + 1) / 2 combinations could
The average depth, length, and area of the combination be plotted against their combination length over the constant
corresponding to the minimum burst pressure are defined as segment length, x, i.e., the normalized length, k as shown in
effective depth, effective length and effective area of the Figure 5.
profile, and denoted as deff, Leff and Aeff, respectively, as the red
box shown in Figure 4. The shape factor is deff /dmax_Aeff, where
dmax_Aeff is the maximum depth within the effective area.

Figure 5: Average Depth versus Normalized Length, k


It is obvious that for the same k, larger average depth leads
Figure 4: Effective Area Example
to lower burst pressure. Therefore, for each k, only the
For a fixed length, the larger average depth predicts lower combination with the largest average depth (denoted as d<k>),
burst pressure. For a fixed depth, the longer corrosion leads to i.e., the points on the red line in Figure 5, are possible to be the
lower burst pressure. However, it is noted that usually the point corresponding to the effective length and effective depth
effective area does not cover the entire corrosion profile, i.e, the predicting the overall minimum pressure, i.e., the large red
effective length is usually shorter than the total length, as the point in Figure 5.
example shown in Figure 4. Because as the combination length ∑𝑗+𝑘−1 𝑑𝑖 ∆𝑥 1 𝑗+𝑘−1
{ ∑𝑖=𝑗 𝑑𝑖 } (5)
𝑖=𝑗
increases, the average depth could become shallower. The 𝑑 <𝑘> = max { }= max
1≤𝑗≤𝑛 – 𝑘 + 1 𝑘∆𝑥 1≤𝑗≤𝑛 – 𝑘 + 1 𝑘
combination with balanced length and average depth is the
critical one to be searched by the effective area (RSTRENG) However, the computation cost of Eq. (5) is high. If looping all
algorithm. possible k from 1 to n, i.e., using Eq. (5) n times, the overall
For a given corrosion profile with n segments, the algorithm still has the same O(n2) complexity as the traditional
RSTRENG algorithm needs to check all n (n + 1) / 2 RSTRENG algorithm. Equation (5) is equivalent to Eqs. (3a) to
combinations. A 10 times larger profile with 10n segments (3e). If only calculating the burst pressure or RSF for
generates around 100 times more combinations. The combinations on the red line of Figure 5, Eqs. 4(a) to 4(c) are
computation time increases nonlinearly. The widely used used n times instead of n (n + 1) / 2. But the overall computing
RSTRENG algorithm has at least an order of n-square time time saving is still minimal.
complexity. It is known as O(n2) algorithm in computer science It is observed that by definition d<k>kx is the maximum
discipline [16], which would be the bottleneck when metal-loss area of all combinations with length equal to kx.
performing systematic probabilistic assessment using Psqr For given two different normalized combination length k and
model [19, 20]. (k+k), where k is a positive integer and 1 ≤ k < (k+k) ≤ n,
when the combination length increases from kx to (k+k)x,
MORE EFFICIENT ALGORITHM (FASTER RSTRENG) the corresponding maximum area increases from d<k>kx to
The O(n2) RSTRENG algorithm could be improved. This d<k+k>(k+k)x. Denote dmin and dmax as the overall minimum
paper uses 2D corrosion profiles formed by evenly spaced and maximum of all segment depth d1, d2, …, di, …, dn. The
uniform depth rectangular metal-loss area bars as shown in following inequality holds:
Figure 4 to illustrate the approach. Other scenarios could be 𝑑 <𝑘+∆𝑘> (𝑘+∆𝑘)−𝑑 <𝑘> 𝑘
converted as discussed in previous sections. 𝑑𝑚𝑖𝑛 ≤ ∆𝑘
≤ 𝑑𝑚𝑎𝑥 (6)
Consider a 2D profile with n segments with depth d1, d2, i.e., the length weighted average depth for the increased area
…, di, …, dn and constant segment length, x. The total number should be between dmin and dmax. Rearranging Eq. (6) and
of combinations is n (n + 1) / 2. Within these n (n + 1) / 2 considering the overall dmin and dmax boundary, the following
combinations, there are n – k + 1 combinations with length inequality holds:
equal to kx, where k is an integer and 1 ≤ k ≤ n. For example, 𝑚𝑎𝑥 {
𝑑 <𝑘> 𝑘+𝑑𝑚𝑖𝑛 ∆𝑘
, 𝑑𝑚𝑖𝑛 } ≤ 𝑑<𝑘+∆𝑘> ≤ 𝑚𝑖𝑛 {
𝑑 <𝑘> 𝑘+𝑑𝑚𝑎𝑥 ∆𝑘
, 𝑑𝑚𝑎𝑥 } (7)
the profile shown in Figure 4 has n = 179 segments and n (n + 𝑘+∆𝑘 𝑘+∆𝑘

5 Copyright © 2022 by ASME


Similarly, let 1 ≤ (k-k) < k ≤ n, For this example with n = 179, it takes 66 steps to find the
𝑑 <𝑘> 𝑘−𝑑<𝑘−∆𝑘> (𝑘−∆𝑘) effective length equal to 132x (i.e., k = 132) as shown in
𝑑𝑚𝑖𝑛 ≤ ≤ 𝑑𝑚𝑎𝑥 (8)
∆𝑘 Figure 6(d). The computationally expensive Eq. (5) has been
𝑚𝑎𝑥 {
𝑑 <𝑘> 𝑘−𝑑 𝑚𝑎𝑥 ∆𝑘
, 𝑑𝑚𝑖𝑛 } ≤ 𝑑<𝑘−∆𝑘> ≤ 𝑚𝑖𝑛 {
𝑑 <𝑘> 𝑘−𝑑 𝑚𝑖𝑛 ∆𝑘
, 𝑑𝑚𝑎𝑥 } (9) used 66 times, and 179 – 66 = 113 times are saved compare to
𝑘−∆𝑘 𝑘−∆𝑘
<k>
the traditional effective area (RSTRENG) algorithm. See Annex
For any known k and d , Eqs. (7) and (9) could be used A for details.
to estimate the upper and lower bound of unknown d<k±k> for The final results, i.e., the predicted burst pressure, effective
all other normalized length, k±k. For the same length, the length, effective depth, effective area, shape factor, using
deeper corrosion leads to lower pressure. Therefore, the upper RSTRENG and Faster RSTRENG are identical because they
bound of d<k±k> is corresponding to the lower bound of burst have rigorously mathematical equivalency.
pressure. Denote k±k as k′, where 1 ≤ k′ ≤ n and k′ ≠ k. After
refactoring Eqs. (7) and (9), the derived upper bound of the
unknown d<k′> from the known d<k> is:
𝑑<𝑘> 𝑘−𝑑𝑚𝑖𝑛 (𝑘−𝑘′)
𝑚𝑖𝑛 { , 𝑑𝑚𝑎𝑥 } , 𝑖𝑓𝑘′ < 𝑘
𝑘′
𝑑 <𝑘′>
≤ <𝑘′>
𝑑𝑢𝑝𝑝𝑒𝑟 ={
𝑑 <𝑘> 𝑘+𝑑𝑚𝑎𝑥 (𝑘 ′ −𝑘)
(10)
𝑚𝑖𝑛 { , 𝑑𝑚𝑎𝑥 } , 𝑖𝑓𝑘′ > 𝑘
𝑘′
All the direct calculation for d<k> using Eq. (5) is
computationally expensive. However, if d<k> is calculated for
one specific k, Eq. (10) could be used to estimate the upper
bound of all the other unknown d<k′> for free. The computation (a)
cost of Eq. (10) is negligible compare to Eq. (5).
For the same example in Figure 4, let k = 90 first, after
checking all the n – k + 1 = 179 – 90 + 1 = 90 combinations
with normalized length equal to 90, d<90> = 49.6% × t is
calculated using Eq. (5). With known d<90>, the upper bound of
d<k′> for k′ = 1, 2, …, 89, 91, …, 179, could be calculated using
Eq. (10), which is plotted as the red line in Figure 6(a).
If the RSF (proportional to Pb) corresponding to the upper
bound of d<k′> and k′x using Eq. (4b) and (4c) is greater than
or equal to the currently known minimum RSF, it is obvious
that k′x is not the final effective length, and all the (b)
combination with normalized length equal to k′ could be
excluded from all further calculation. In Figure 6(a), as marked
by black crosses, all the combinations with the normalized
length, k′, less than or equal to 6 are excluded, because their
lower bound RSF values corresponding to the upper bound d<k′>
are still greater than the calculated RSF for k = 90.
It is observed from Figure 6(a) that the closer k′ to the
calculated k = 90, the lower estimated upper bound of d<k′> is. It
is more efficient and informative to choose the next k (e.g., 135
in Figure 6(b)) far away from the known calculated k = 90, and
the minimum and maximum possible values, i.e., 1 or n. (c)
As shown in in Figure 6(b), similarly, after calculating
d<135> using Eq. (5), the upper bound for all the other unknown
d<k′> could be updated using Eq. (10) for free again. The RSF
corresponding to k = 135 is less than the RSF for k = 90, so k =
135 is corresponding to the currently known minimum RSF
value shown as the large red point in Figure 6(b). Based on the
RSF for the upper bound of d<k′>, k′ = 88, 89, 91 could be
excluded from any further calculation. See Annex A for details.
This process could be repeated until all the possible integer
k, 1 ≤ k ≤ n, checked using Eq. (5) or excluded based on Eq.
(10). Figure 6(c) shows that after 15 steps, 15 d<k> values for (d)
Figure 6: Faster RSTRENG Algorithm Steps
different k have been calculated by Eq. (5), and the other 23 k
values have been excluded using Eq. (10).

6 Copyright © 2022 by ASME


DISCUSSION the same length, it is not guaranteed that the commonly used
A comprehensive case study has been conducted using 2D RSTRENG algorithm always predicts the minimum burst
profiles with various n values ranging from 2 to 10000. As pressure for the profile after checking the n (n + 1) / 2
shown in Figure 7, Faster RSTRENG algorithm proposed in combinations. A counterexample is presented in Annex B. It is
this paper takes much less time for large n (i.e., long or hard to solve the generic problem analytically, especially for
complex corrosion anomalies) compare to the traditional complex MFL reported clusters. Therefore, it is suggested to
RSTRENG. Faster RSTRENG algorithm is much more mesh the MFL reported corrosion clusters into evenly spaced
efficient than the traditional RSTRENG. The results generated small girds, e.g., 5mm by 5mm, before generating “river
by the two algorithms are identical. bottom” profiles, and calculate using RSTRENG or Faster
1400 RSTRENG algorithm.
1200 RSTRENG
Faster RSTRENG
Time (seconds)

1000
SUMMARY
800
When assessing corrosion threat, the effective area
600
400
(RSTRENG) model is commonly used to predict burst pressure
200
from known measured corrosion anomaly size. RSTRENG
0 model contains an iterative process to find the critical sub-
0 2000 4000 6000 8000 10000 section (combination) corresponding to the minimum burst
n segments = total length / segment length pressure. The commonly used RSTRENG algorithm has an
Figure 7: Case Study Showing Time Efficiency Benefit order of n-square time complexity (O(n2)). As n increases (e.g.,
It should be emphasized that Faster RSTRENG and longer corrosion or the same corrosion measured by higher
RSTRENG algorithms are rigorously mathematically resolution tools), the computation time increases nonlinearly.
equivalent. The calculated burst pressure results are always The traditional effective area algorithm is reviewed first in
identical. All the combinations (or called iterations) mentioned this paper. It is found that it is not necessary to always loop
in ASME B31G [5] and PRCI PR-3-805 [4] have been checked. through all the combinations and check the corresponding burst
The benefit of Faster RSTRENG comes from that some pressure one by one. A more effective algorithm (Faster
shallower and shorter combinations, certainly not the final RSTRENG) is proposed and described in detail. Faster
critical combination, are checked and excluded more efficiently RSTRENG is mathematically equivalent to the traditional
without actually calculating the corresponding burst pressure RSTRENG algorithm and always guarantees identical burst
one by one. pressure results as the traditional algorithm. All the
The key component of Faster RSTRENG is Eq. (10) combinations (or called iterations) mentioned in ASME B31G
derived from the inequality Eqs. (6) and (8), which could be [5] and PRCI PR-3-805 [4] are checked but in a more efficient
used to check and exclude combinations efficiently with way.
negligible computational cost. As the example shown in Figure A case study is conducted with input profiles formed by
6(d), a 2D profile with n = 179 is calculated using 66 steps, all various measurement points, n. It is proven that Faster
the other 113 normalized length k values are excluded using Eq. RSTRENG algorithm is much more efficient than the
(10). The 113 black cross markers in Figure 6(d) are traditional RSTRENG.
computational free, whereas the red line in Figure 5 is Faster RSTRENG algorithm reduces the computation cost
computationally very expensive and unnecessary. significantly and makes it possible to perform reliability-based
Faster RSTRENG algorithm presented in this paper is (probabilistic) assessment using Monte Carlo method. Millions
clearly not the “most” efficient algorithm. There is still room to of simulations could be performed using RSTRENG model, or
further improve this algorithm. For instance, the strategy to even the more accurate and precise Psqr model developed by
select the next k to calculate using Eq. (5) based on the existing TC Energy [19, 20]. Operators usually benefit from the
information, e.g., calculated d<k> values and current upper advanced data driven corrosion assessment when optimizing
bound distribution of d<k′> (see Annex A for details). But it is the mitigation works [2, 15].
proved to be good enough for general engineering purpose. As the effective area method is the fundamental of
Assume an MFL ILI reported corrosion cluster formed by corrosion assessment, the proposed algorithm in this study
m individual flat bottom rectangular boxes as shown in Figure potentially brings broad benefits to the pipeline industry.
2(a), and the generated “river bottom” has n segments with
uniform depth and various length, where n ≤ m. The current ACKNOWLEDGMENTS
common industry practice is to check the burst pressure for all Authors acknowledge the comments received from internal
n (n + 1) / 2 combinations and use the minimum as the final and external reviewers. Jason Yan would like to thank the
burst pressure for the corrosion cluster. However, if the n family support when working from home during the novel
segments are allowed to be split into shorter subsegments, as coronavirus (COVID-19) pandemic.
shown in Annex B, sometimes, burst pressure lower than the
minimum of the n (n + 1) / 2 combinations could be found. In
other words, if the n segments forming a 2D profile do not have

7 Copyright © 2022 by ASME


REFERENCES Proceedings of the 2020 13th International Pipeline Conference,
[1] Zhou, W., Gong, C. and Kariyawasam, S. “Failure pressure ratios IPC2020-9484, Calgary, Alberta, Canada. 2020.
and implied reliability levels for corrosion anomalies on gas [19] Zhang, S., Yan, J., Kariyawasam, S. Huang, T. and Al-Amin, M.,
transmission pipelines”, Proceedings of the 2016 11th “A More Accurate and Precise Method for Large Metal Loss
International Pipeline Conference, IPC2016-64383, Calgary, Corrosion Assessment”, Proceedings of the 2018 12th
Alberta, Canada, 2016. International Pipeline Conference, IPC2018-78233, Calgary,
[2] Kariyawasam, S., Zhang, S., and Yan, J., “A Data Driven Alberta, Canada, 2018.
Validation of a Defect Assessment Model and its Safe [20] Zhang, S., Yan, J., Kariyawasam, S. Huang, T. and Al-Amin, M.,
Implementation”, Proceedings of the 2020 13th International “Plausible Profile (Psqr) Corrosion Assessment Model:
Pipeline Conference, IPC2020-9690, Calgary, Alberta, Canada. Refinement, Validation and Operationalization”, Proceedings of
2020. the 2020 13th International Pipeline Conference, IPC2020-9448,
[3] Kariyawasam, S., Zhang, S., Yan, J., Kariyawasam, S. Huang, T. Calgary, Alberta, Canada, 2020.
and Al-Amin, M., “Improving Safety and Economy Through a [21] Kiefner, J.F., Cosham, A., Gao, M., Hopkins, P., Krishnamurthy,
More Accurate and Precise Burst Pressure Model”, Proceedings of R., Nessim, M., Nestleroth, B., and Rosenfeld, M., “Peer Review
22nd Joint Technical Meeting, Brisbane, Australia, 2019. Paper of the Plausible Profile (Psqr) Corrosion Assessment Model”,
No. 31. PRCI Report, Project Number EC-2-9, Contract Catalog No.
[4] Kiefner, J. F., and Vieth, P. H., “A Modified Criterion for PR218-183607-Z01, September 24, 2019.
Evaluating the Remaining Strength of Corroded Pipe”, PRCI https://www.prci.org/141865.aspx
Report Catalog No. L 51609, Contract Catalog No. PR-3-805, [22] Kiefner, J. F., “Peer Review of the Plausible Profile Corrosion
Battelle Columbus Div., December 22, 1989. Assessment Model”, Proceedings of the 2020 13th International
[5] ASME. Manual for Determining the Remaining Strength of Pipeline Conference, IPC2020-9254, Calgary, Alberta, Canada,
Corroded Pipelines, A Supplement to ASME B31 Code for 2020.
Pressure Piping. ASME B31G-2012, American Society of [23] Folias, E.S., “The Stresses in a Cylindrical Shell Containing an
Mechanical Engineers, New York, 2012. Axial Crack”, International Journal of Fracture Mechanics, Vol.
[6] Kiefner, J. F., Maxey, W. A., Eiber, R. J., and Duffy, A. R., “Failure 1(2), pp. 104-113, 1964.
Stress Levels of Flaws in Pressurized Cylinders”, Progress in [24] Chauhan, V., Brister, J. and Dafea, M. “A Review of Methods for
Flaw growth and Fracture Toughness Testing. American Society Assessing the Remaining Strength of Corroded Pipelines”, US
for Testing and Materials STP, Philadelphia, pp.461-481, 1973. Dept. of Transportation (DOT), Pipeline and Hazardous Materials
[7] ASME. Manual for Determining the Remaining Strength of Safety Administration (PHMSA), Report No. 6781., 2009.
Corroded Pipelines, A Supplement to ASME B31 Code for [25] Kiefner, J. F., and Vieth, P. H., “Pipeline Defect Assessment – A
Pressure Piping. ASME B31G-1984, American Society of Review and Comparison of Commonly Used Methods”, PRCI
Mechanical Engineers, New York, 1984. Report, Contract Catalog No. PR-218-05404, 2011.
[8] Kiefner, J. F., and Vieth, P. H., “Evaluating Pipe – 1: New Method [26] Zhou, W., and Huang, G.X., “Model error assessments of burst
Corrects Criterion for Evaluating Corroded Pipe”, Oil and Gas capacity models for corroded pipelines”, International Journal of
Journal, August 6, 1990. Pressure Vessels and Piping, Volumes 99–100, Pages 1-8, 2012.
[9] Kiefner, J. F., and Vieth, P. H., “Evaluating Pipe – Conclusion: “PC [27] Zhu, X-K, “A Comparative Study of Burst Failure Models for
Program Speeds New Criterion for Evaluating Corroded Pipe”, Oil Assessing Remaining Strength of Corroded Pipelines”, Journal of
and Gas Journal, August 20, 1990. Pipeline Science and Engineering, 2021.
[10] Vieth, P.H. and Kiefner, J.F. “Database of Corroded Pipe Tests”, [28] Zhou, W, and Huang, G., “Model Error Assessment of Burst
PRCI Report, Contract Catalog No. PR-218-9206, 1993 Capacity Models for Defect-Free Pipes”, Proceedings of the 2012
[11] Kiefner, J. F., Vieth, P. H., and Roytman, I., “Continued Validation 9th International Pipeline Conference, IPC2012-90334, Calgary,
of RSTRENG”, PRCI Report Catalog No. L51749, Contract Alberta, Canada, 2012.
Catalog No. PR 218-9304, December 20, 1996.
[12] ASME. Pipeline Transportation Systems for Liquids and Slurries.
ASME B31.4-2019, American Society of Mechanical Engineers,
New York, 2019.
[13] ASME. Gas Transmission and Distribution Piping Systems.
ASME B31.8-2020, American Society of Mechanical Engineers,
New York, 2020.
[14] CSA. Oil and gas pipeline systems. CSA Z662-2019, Canadian
Standards Association, 2019.
[15] Kariyawasam, S., Zhang, S., Yan, J., Kariyawasam, S. Huang, T.,
Al-Amin, M., and Gamboa, G. “Plausible Profiles (Psqr) Model
for Corrosion Assessment”, Technical report submitted to PRCI,
Project Number EC-2-9, August 9, 2019.
https://www.prci.org/141865.aspx
[16] Cormen, T. H., Leiserson, C. E., Rivest, R. L., and Stein, C.,
Introduction to Algorithms. MIT Press, 2009.
[17] Owen, A., Monte Carlo Theory, Methods and Examples. McBook,
2013.
[18] Yan, J., Zhang, S., Kariyawasam, S., Lu, D., and Matchim, T.
“Reliability-based Crack Threat Assessment and Management”,

8 Copyright © 2022 by ASME


ANNEX A
FASTER RSTRENG ALGORITHM

Algorithm 1: Bulging factor or Folias factor, Eq.(2b) or (4c), M (D, t, L) 8: current_max_d_ave ← current_d_ave
Input: D, t and L (corrosion length, segment or combination length) 9 loop
1: if L ^ 2 <= 50 × (D × t) then 10: return current_max_d_ave
2: M ← (1 + 0.6275 × L ^ 2 / (D × t) - 0.003375 × L ^ 4 / (D × t) ^ 2) ^
0.5 Algorithm 6: Faster RSTRENG, n segments
3: else Input: D, t, y, 2D corrosion profile formed by n segments with same length
4: M ← 3.3 + 0.032 × L ^ 2 / (D × t)
(x) and uniform depth, {d1, d2, …, dn}
5: return M
1: Let unchecked_possible_k_list be an array from 1 to n # initialize
2: d_max ← max {d1, d2, …, dn}
Algorithm 2: Remaining strength factor, Eq.(4b), RSF (D, t, L, d) 3: d_min ← min {d1, d2, …, dn}
Input: D, t, L and d (depth, segment depth or combination average depth) 3: current_upper_d_ave ← an array with n elements equal to d_max
1: if L = 0 or d = 0 then # perfect pipe 4: current_min_RSF ← 1 # initialize
2: return 1 5: while unchecked_possible_k_list.size > 0 do
3: else 6: if this is the first step then
4: M ← M (D, t, L) # Folias factor 7: current_k ←
5: return (1 – d / t) / (1 – d / t / M) unchecked_possible_k_list [unchecked_possible_k_list.size // 2]
# start from the middle point of the list
8: else
Algorithm 3: RSTRENG, n segments, O(n2) complexity
9: last_k ← current_k
Input: D, t, y, {L1, L2, …, Ln} and {d1, d2, …, dn}, (length and depth of each 10: # get the next k to check, it should be far away from the last k, and
segment, as arrays) not close to 1 or n
1: loop through all n segments to get the area of each segment current_k ← get_next_k (unchecked_possible_k_list, last_k)
2: current_min_RSF ← 1 # initialize 11: remove current_k from unchecked_possible_k_list
3: for p = 1 to n do 12: d_ave_max_k ← get_d_ave_max_k (current_k, {d1, d2, …, dn})
4: for q = p to n do # loop through all n(n+1) combinations # based on Eq.(5)
5: get the combination length, combination area and average depth 13: current_RSF ← RSF (D, t, k × x, d_ave_max_k)
from the pth segment to the qth segment inclusive # Eqs.(3a) to (3e) 14: if current_RSF < current_min_RSF then
# When using Eqs. (3a) and (3b), always remember the area and 15: current_min_RSF ← current_RSF # update the minimum
length from the last loop to avoid duplicate summation, otherwise 16: if unchecked_possible_k_list.size > 0 then
Algorithm 3 will become O(n3) complexity 17: current_upper_d_ave ←
6: current_RSF ← RSF (D, t, current combination length and average update_upper_bound (current_upper_d_ave, d_ave_max_k,
depth) current_k, d_max, d_min) # based on Eq. (10)
7: if current_RSF < current_min_RSF then 18: for all uncheck_k in unchecked_possible_k_list do
8: current_min_RSF ← current_RSF # update the minimum 19: current_RSF ← RSF (D, t, uncheck_k × x,
9: loop current_upper_d_ave [uncheck_k])
10: loop 21: if current_RSF >= current_min_RSF then
11: return 2 × t × (y + 10ksi) / D × current_min_RSF # convert RSF to 22: remove uncheck_k from unchecked_possible_k_list
Pb, unit conversion as needed 23: loop
24: loop
Algorithm 4: Get the next k to check, this algorithm has room to further 25: return 2 × t × (y + 10ksi) / D × current_min_RSF # convert RSF to
improve. get_next_k (unchecked_possible_k_list, last_k) Pb, unit conversion as needed
Input: unchecked_possible_k_list, last_k
1: split unchecked_possible_k into 2 arrays, left and right
left contains all elements in unchecked_possible_k less than last_k
right contains all elements in unchecked_possible_k greater than last_k
both left and right are in ascending order
2: if left.size > right.size then
return left[left.size // 2] # return the middle point from the left array
3: Else
4: return right[right.size // 2] # return the middle point from the right
array

Algorithm 5: get the maximum of all average depth for combinations with
length equal to kx, Eq.(5). get_d_ave_max_k (k, {d1, d2, …, dn})
Input: k, depth of n segments {d1, d2, …, dn}
1: current_max_d_ave ← min {d1, d2, …, dn} # initialize
2: for j = 1 to n – k + 1 do
3: current_d_ave ← 0 # initialize
4: for i = j to j + k - 1 do
5: current_d_ave ← current_d_ave + di / k
6: loop
7: if current_d_ave > current_max_d_ave then

9 Copyright © 2022 by ASME


ANNEX B
AN MFL ILI CLUSTER COUNTEREXAMPLE

Figure B1(a) shows a magnetic flux leakage (MFL) in-line remaining strength factor (RSF) equal to 0.8688 as shown in
inspection (ILI) reported corrosion cluster example, which is Figure B1(e).
formed by two flat bottom rectangular individual corrosion However, if it is allowed to split the 2 segments into
boxes. The two boxes are 150 mm × 100 mm × 10% wall shorter subsegments, e.g., Combination #4 shown in Figure
thickness (WT) and 50 mm × 50 mm × 40% WT, respectively. B1(c), a lower RSF value could be found. Figure B1(e) shows
The 2D “river bottom” profile has 2 segments (see Figure that the RSF value changes as a function of the length of
B1(b)). A total of 2 × (2 + 1) / 2 = 3 combinations could be Combination #4. The minimum RSF value of the cluster is
generated from the 2 segments profile as shown in Figure 0.8665 with effective length = 103 mm and effective depth =
B1(c). After checking all 3 combinations, Combination #3 with 24.56%WT.
length = 150 mm and average depth = 20%WT has the lowest


① ② ①
10%WT ① ② ④
50 mm × 40%WT 100 mm × 10%WT ④
103 mm × 24.6%WT
50 mm × 40%WT 100 mm × 10%WT ③
103 mm × 24.6%WT
40%WT ② ③

Combination Length
(a) ③ ④ (d)
③ ④ ①

150 mm × 20%WT 103 mm × 24.6%WT ③
150 mm × 20%WT 103 mm × 24.6%WT ③

(b) (c) (e)


Figure B1: An MFL ILI Example

10 Copyright © 2022 by ASME

You might also like