Professional Documents
Culture Documents
Abstract
Corrosion of large storage tank bottom has been identified as the main threat to tank security around the world.
Corrosion detection becomes the main approach to ensure the safety of large storage tanks. In this article, we propose a
two-stage tank bottom corrosion analysis and remaining useful life prediction framework. In the first stage, the causes and
mechanisms of tank bottom corrosion are analyzed, the evaluation metric system is established, and the fuzzy compre-
hensive evaluation method is used to evaluate the health status of tank bottom. The preliminary evaluation result in the
first stage determines whether a subsequent remaining useful life prediction is required. If the tank bottom is not in healthy
state, in the second stage, the remaining useful life prediction of the tank bottom is implemented by acoustic emission and
risk-based inspection. Both qualitative and quantitative evaluation are carried out to provide a basis for the operation and
maintenance decision of storage tanks.
Keywords
Tank bottom, metric system, fuzzy comprehensive evaluation, acoustic emission, remaining useful life
Creative Commons CC BY: This article is distributed under the terms of the Creative Commons Attribution 4.0 License
(http://www.creativecommons.org/licenses/by/4.0/) which permits any use, reproduction and distribution of the work without
further permission provided the original work is attributed as specified on the SAGE and Open Access pages (https://us.sagepub.com/en-us/nam/
open-access-at-sage).
2 International Journal of Advanced Robotic Systems
atmosphere storage tanks using MFL.7 Shi et al. introduced framework. The “Preliminary health evaluation” section
the main principles, measurement, and processing of MFL introduces the preliminary health evaluation. The “Residual
data. As the key point of a quantitative analysis of MFL life prediction based on AE and corrosion risk pace” section
detection, the identification of the leakage magnetic signal illustrates the accurate diagnosis. The “Case study” section
is also discussed. The advantages and disadvantages of presents a case study to verify our proposed framework. The
different identification methods are analyzed.8 Yu et al. “Conclusion” section concludes the article.
established the three-dimensional finite element model of
tank bottom, which is then used as a simulation tool to Overall framework for tank bottom health
study the propagation characteristics of AE signals. They
analysis
concluded that multipoint measurements can be used to
help understand the status of bottom plate.9 Gong and Hu In this article, a two-stage health state analysis method for
constructed an AE system to detect and evaluate tank bot- tank bottom is proposed. In the first stage, a preliminary
tom corrosion. A corrosion detection process was designed health evaluation of tank bottom is made: firstly, a scien-
and put in practice for qualitative evaluation.10 Jirarungsa- tific risk evaluation metric system for tank bottom corro-
tian and Prateepasen carried out AE experiments on differ- sion is established based on expert experience and failure
ent types of damage and analyzed the parameters of AE analysis. Then, the weights of evaluation metrics are deter-
signals.11 The experimental results were used for on-site mined by combining the analytic hierarchy process (AHP)
detection of atmospheric storage tanks. Liu et al. analyzed and the entropy method. Finally, fuzzy comprehensive eva-
the advantages and disadvantages of AE detection and luation is adopted to give health status. The preliminary
MFL detection and proposed a tank corrosion detection evaluation result in the first stage determines whether a
method by combining the above two methods.12 Based subsequent remaining useful life (RUL) prediction is
on the risk-based inspection (RBI) technology, Guo et al. required. If the tank bottom is in good condition, it is unne-
carried out risk assessment on 20 large atmospheric storage cessary to carry out AE testing so as to save tank mainte-
tanks. By setting different risk levels and implementing nance costs. If the tank bottom is not in good condition, the
maintenance methods and strategies for each storage tank second stage is carried out to make accurate diagnosis. We
correspondingly, the reliability of RBI technology was ver- establish a quantitative evaluation model based on the his-
ified.13 Park et al. monitored the corrosion state and AE torical experimental data collected through AE testing and
data of tank bottom in acidic soil and verified the consis- MFL testing. Then, the AE experiment is implemented and
tency of the AE technology and risk-based method.14,15 the result is used to estimate the corrosion rate and predict
Zhang et al. conducted an experiment on AE monitoring the RUL of storage tank. The overall framework for tank
of tank bottom in a laboratory condition. He analyzed the bottom health analysis is shown in Figure 1.
emission signals collected in the experiment and obtained
the AE characteristics during tank bottom corrosion pro- Preliminary health evaluation
cess under different working conditions. A corrosion detec-
tion mechanism was provided with improved reliability in Corrosion analysis
AE testing and evaluation field.16 Yu et al. claimed that in Preliminary health evaluation reflects the current operating
some circumstances, the types of corrosion AE signal of the status of tanks, hence it is extremely significant to establish
tank bottom are unknown. He thus proposed a weighted a system of evaluation metrics to give an accurate, credible,
fuzzy clustering recognition method to distinguish corro- and objective evaluation result, which is also a guarantee
sion types with each other. Nearest neighbor method was for the long-term effective operations of oil tanks. To estab-
used for optimizing initial clustering. The data redistribu- lish the risk metric system of tank floor corrosion, we first
tion is adjusted with the weighted distance between the summarize all types of tank floor corrosion and study the
gravity and center. The experimental result shows that the main influencing factors that cause corrosion.
detection accuracy increases by about 9%.17 The corrosion of tank floor mainly consists of two
Although some achievements have been made both in types: internal corrosion and external corrosion. Internal
research and in application, the AE method can only be corrosion mainly includes atmospheric corrosion and
used for qualitative evaluation. Other detection methods, liquid corrosion, and external corrosion mainly includes
such as MFL testing,18,19 can effectively quantitatively electrochemical corrosion and soil corrosion. The main
evaluate corrosion defects. However, it usually causes factors affecting the corrosion of tank floor are as follows:
unnecessary tank opening, which may result in over main-
tenance of storage tanks, affecting the normal operation. 1. Sedimentary water at the bottom of tank: Sedimen-
The deficits of both methods greatly encourage the pro- tary water mainly comes from produced fluid, rein-
posal of our health analysis framework. jection water, condensate water, and rainwater,
The rest of the article is organized as follows: The which is the main source of the internal liquid cor-
“Overall framework for tank bottom health analysis” sec- rosion. The corrosivity of sedimentary water usu-
tion gives an overview of our tank bottom health analysis ally differs in terms of its components. The main
Feng et al. 3
influencing factors include pH value, chloride con- positions of the bottom will be destroyed, resulting
centration, sulfur amount, salinity, and so on. The in oxygen concentration cell.
pH value determines the corrosion type of tank bot- 4. Mechanical factors: The fluctuation of the fluid in
tom, especially in strong acid or alkaline environ- the tank can make the bottom bear additional vibra-
ment, which usually result in a significant increase tion and impact loads, which will cause the vibra-
in corrosion rate.20 The chloride concentration is the tion damage of the tank bottom, leading to crevice
main factor affecting the corrosion rate of tank bot- corrosion and pitting corrosion.
tom.5 Sulfides can be decomposed into corrosive
substances that result in corrosion. The salinity is
mainly affected by the concentration of calcium and
Evaluation metric system
magnesium. When the concentration is too high, it
will promote the formation of local cell corrosion. Based on the detailed analysis of the corrosion mechanism
At the same time, calcium and magnesium can of tank bottom in the previous section, this section tries to
increase the conductivity of water, thus accelerating establish a scientific, effective, and standardized evaluation
the corrosion rate. metric system, as shown in Table 1.
2. Gas effect: Due to the constantly rising and falling of
liquid level in the tank, a certain amount of air will be
mixed into the tank. Hydrogen sulfide gas has the Evaluation method
greatest impact on tank bottom corrosion, which can Weight derivation and modification. After health evaluation
dissolve in water to form acid that has very high cor- model has been constructed, we need to assign weights for
rosivity.21 At the same time, hydrogen sulfide gas is different evaluation metrics. The weight determination of
the catalyst for cathodic reduction of tank bottom, evaluation metrics is an important part for reflecting the
which can promote the absorption of hydrogen in health status of tank bottom. AHP and entropy weight
reduction reaction and further deteriorate corrosion. method are widely used in similar problems. However, AHP
3. Production factors: In daily production, the opera- is strongly based on subjective opinions of experts, while
tion of the tank’s charging and discharging will entropy weight method does not work well when data are not
cause the fluctuation of the fluid in the tank, and sufficient. In this article, we applying Bayes theorem and
turbulence may form inside the tank. Turbulence combine these two methods to overcome their deficiencies.
can cause ups and downs of tank bottom, which can First, a priori value of metric weights are adopted through
abrade the corrosion-resistant coating on the inner AHP method, then the entropy information of feedback data
surface of tank bottom and form a corrosion is taken to calculate posterior value of metric weights.
source.22 At the same time, a series of cleaning If we model the weights of evaluation metrics as prob-
operations of the tank will continuously scour the ability distribution, we hope that we can estimate the dis-
tank bottom, and the membrane layer at some tribution through experts’ knowledge when the amount of
4 International Journal of Advanced Robotic Systems
data is insufficient. At the same time, with the accumula- where w~ ¼ ½w 1 ; w2 ; . . . ; wm represents the weight vector of
tion of data, the weights of evaluation metrics gradually the evaluation metrics, DirichletðÞ denotes the Dirichlet
adapt to the information in actual data. Finally, when the Distribution, and ~a is the distribution parameter. The dis-
amount of data is large enough, the weights of the evalua- tribution parameter is initialized proportionally to metric
tion metrics can fully reflect the amount of information weights derived from AHP method.
provided by each metric. There are mainly two reasons why we choose Dirichlet
The process of assigning weights can be understood as distribution:
continuous revision of the importance of each metric.
Assume that the weights of the evaluation metrics follow According to the P practical meaning of metric
m
Dirichlet distribution weights, we have i¼1 wi ¼ 1. Therefore, when
we choose joint distribution of the m metrics, the
aÞ
~*Dirichletð~
w ð1Þ sum of expectations of each random variable in joint
Feng et al. 5
distribution should be equal to 1. The attribute of metrics, the evaluation matrices of abstract levels are cal-
Dirichlet distribution perfectly satisfies the require- culated. Finally, we determine the health status of tank
ment for joint distribution. bottom based on the rule of maximum degree of member-
According to the hypothesis of Bayesian view, prob- ship. More detailed explanation of the method is demon-
ability itself is a random variable with prior and strated in the “Case study” section.
posterior distributions. When the condition likeli-
hood function can make the posterior distribution
has the same form with the prior distribution, we say
the distribution is the conjugate prior of the likeli-
Residual life prediction based on acoustic
hood function. Considering the conjugation of emission and corrosion risk pace
Dirichlet distribution and polynomial distribution, AE technology
we choose the likelihood function of polynomial
distribution and assume the prior distribution as AE detection technology is widely employed across the
Dirichlet distribution, then the calculation of prob- world because it can dynamically detect and analyze stor-
ability posterior distribution will have a good math- age tanks. Compared with other nondestructive methods,
ematical form. the special characteristic of AE is that it does not affect
daily production. When corrosion or leakage happens, the
After collecting a certain number of user feedback, we load of the tank will change, and the energy will be released
calculate the entropy weight of each evaluation metrics, in the form of elastic wave to generate AE signals. The
which is denoted as w ^ ¼ ½w ^ 1; w ^ m . If we believe
^ 2; . . . ; w sensors on the tank are set to receive the signals that exceed
that the entropy weights of user feedback data reflect the the threshold. The signals are then transformed into char-
probability of different results in a multi-distribution and acteristic parameters through feature extraction circuit.
assume that the amount of feedback data collected by users Among them, ring-down count can effectively reflect the
is , then the number of times each result occurs is intensity of AE activity, as well as the corrosion status. AE
w^ ¼ ½w ^ 1 ; w ^ m . According to the conjugate
^ 2 ; . . . ; w activity represents the frequency of AE events. It can be
distribution theorem, the posterior distribution of evalua- calculated by dividing the hit count of all channels per unit
tion metrics is another Dirichlet distribution time by the number of channels. The AE activity can be
employed to predict the RUL of tank bottom.
a þ wÞ
~*Dirichletð~
w ^ ð2Þ
In conclusion, the proposed weight derivation method
takes the following steps: Corrosion risk pace
In this article, the concept of risk is introduced to model
1. Assume we are assigning weights to m evaluation
tank bottom corrosion. Corrosion risk pace (CRP) is a sci-
metrics. Use AHP method to obtain their a priori
entific risk assessment method that makes use of the test
values w¼w 1; w
2; . . . ; w
m , respectively.
result of MFL testing. CRP represents the average corro-
2. Choose a proper coefficient l, then initialize the
sion rate of tank bottom, which can reflect the health status
distribution parameters as ~ a ¼ l w.
of the most active corrosion area. Let xi denote the corro-
3. Collect user feedback data and use entropy method
sion reduction at a certain location of tank bottom,
to calculate evaluation weight values w ^ ¼w
^ 1; w
^ 2;
FðhÞ ¼ CPDðxi hÞ denote the probability of corrosion
...;w^ m , respectively.
depth that exceeds h (mm), then ln FðhÞ ln h is defined
4. Modify evaluation metric weights:
as the risk curve of tank bottom corrosion. Since prelimi-
nary health evaluation has been made, we assume that the
ai þ w
~i storage tank is at the latter part of its life, and the gradient at
~i ¼
w ð3Þ
ai þ the tail of the curve is
@ ln F @F=F
D¼ ¼ ð4Þ
Health evaluation. After the weights of evaluation model are @ ln h @h=h
derived and modified, the fuzzy synthetic evaluation We can conclude from the above formula that
method is adopted to compute reputation. The values of
every evaluation metric are collected and calculated. Our 1 @F
@h ¼ h ð5Þ
health evaluation method is conducted as follows: First, for D F
both quantitative and qualitative metrics, the membership Assume that the corrosion increment of the region where
functions of every evaluation metric are figured out accord- corrosion depth is greater than h can be illustrated as
ing to a predetermined limitation. The single metric mem- follows
bership degree can be integrated to get the evaluating
matrix. Then, according to the weights of evaluation @F ¼ kFDt ð6Þ
6 International Journal of Advanced Robotic Systems
where k represents the corrosion rate, which can be consid- Table 2. Magnetic flux leakage detection data of experimental
ered to remain unchanged during the corrosion process. storage tanks.
From the above formulas, it can be concluded that the Thinning (mm) Cumulative counts Cumulative probability
corrosion depth increases along with time has the following
form 3.66 1 0.0000
3.60 3 0.0002
k t 3.54 4. 0.0003
hðtÞ ¼ h^ 1 e D ð7Þ .. .. ..
. .
2.46
.. 150 .. 0.0100
..
where h^ is the statistical expectation of corrosion depth . . .
near the current location. It is concluded that the corrosion 0.78 10,411 0.7001
rate of the current position is as follows 0.72 11,872 0.7983
0.66 13,253 0.8912
@hðtÞ k h^ 0.65 14,872 1
@t ¼ D ð8Þ
t¼0
Table 6. The AE data of testing tank bottom. 3. The quantitative model between AE activity and cor-
rosion rate of storage tanks is established through AE
Channel AE hits and MFL testing of experimental storage tanks.
1 453 4. Based on Chinese national standard, the RUL of
2 519 storage tanks is predicted by combining AE with
3 34 RBI technology.
4 699 5. A comprehensive case study is conducted to show
5 537 the feasibility of our approach.
6 484
7 392 In the future, we plan to collect more AE data and add
8 620
9 556
multiple features to build a more complicated model.
10 528 According to the characteristics of collected data, we are
going to apply deep learning or sequence learning mechan-
AE: acoustic emission. isms to our research, which is expected to improve the
accuracy of RUL prediction.
4. Chang JI and Lin CC. A study of storage tank accidents. in oil storage tank on the neutral sand under loading. Mater
J Loss Prevent Proc 2006; 19(1): 51–59. Trans 2006; 47(4): 1240–1246.
5. Du G, Jin S, Zhang C, et al. A study on acoustic emission 16. Zhang Y, Li C, Li W, et al. Experimental research on acoustic
technology for tank bottom corrosion inspection. Anti-Corros emission monitoring of dynamic corrosion on a simulated
Method M 2010; 57(6): 275–279. tank floor. In: Shen G, Wu Z, and Zhang J (eds) Advances
6. Liu S and Zhang G. An improved grey model for corrosion in acoustic emission technology. Springer proceedings in
prediction of tank bottom. Prot Met 2007; 43(4): 407–412. physics, Vol. 179, Cham: Springer, pp. 383–400.
7. Liu F, Guo X, Hu D, et al. Comprehensive inspection and 17. Yu Y, Cao H, Yang P, et al. Identification of acoustic
evaluation technique for atmospheric storage tanks. Nondes- emission signal of tank bottom corrosion based on
truct Test Eva 2010; 25(1): 45–59. weighted fuzzy clustering. In: Shen G, Wu Z, and Zhang
8. Shi Y, Zhang C, Li R, et al. Theory and application of mag- J (eds) Advances in acoustic emission technology.
netic flux leakage pipeline detection. Sensors 2015; 15(12): Springer proceedings in physics, Vol. 158, NY: Springer,
31036–31055. pp. 395–404.
9. Yu Y, Qian X, and Yang P. Research of tank bottom corro- 18. Amos D. The truth about magnetic flux leakage as applied to
sion acoustic emission simulation. In: Gongtian S, Wu Z, and tank floor inspections. Insight 1996; 38(10): 730–731.
Zhang J (eds) Advances in acoustic emission technology. 19. Amos DM. Magnetic flux leakage as applied to aboveground
Berlin, Germany: Springer, 2017, pp. 85–99. storage tank flat bottom tank floor inspections. Mater Eval
10. Gong K and Hu J. Online detection and evaluation of tank 1996; 54(1): 26–28.
bottom corrosion based on acoustic emission. In: Z Qu and J. 20. Whited T. Mitigation of soil-side corrosion on double-
Lin (eds) Proceedings of the international field exploration contained aboveground storage tank floors. Mater Perfor-
and development conference 2017, Chengdu, China, 21 Sep- mance 2011; 22: 7.
tember–22 September 2017, pp. 1284–1291. Singapore: 21. Soares CG, Garbatov Y, Zayed A, et al. Corrosion wastage
Springer. model for ship crude oil tanks. Corros Sci 2008; 50(11):
11. Jirarungsatian C and Prateepasen A. Pitting and uniform 3095–3106.
corrosion source recognition using acoustic emission para- 22. Mobin M, Malik AU, Al-Fozan S, et al. Corrosion failure of
meters. Corros Sci 2010; 52(1): 187–197. bottom plates of an aboveground storage tank. J Fail Anal
12. Liu F, Ding S, and Guo X. Magnetic flux leakage and acous- Prev 2007; 7(1): 18–22.
tic emission testing technique for atmospheric storage tanks. 23. GB/T 19801-2005. Non-destructive testing—acoustic emis-
In: 17th World conference on nondestructive testing. Shang- sion inspection—secondary calibration of acoustic emission
hai, China, 25–28 October 2008. sensors (Chinese national standard).
13. Guo B, Shen G, Zhang W, et al. Application of RBI technol- 24. NB/T 47013.9-2012. Non-destructive testing of pressure
ogy in atmospheric storage tanks. Press Vess Tech 2010; equipments—Part 9: acoustic emission testing (Chinese
27(4): 55–60. industry standard).
14. Park S, Kitsukawa S, Katoh K, et al. AE source and relation 25. GB/T 30578-2014. Risk-based inspection and evaluation for
between AE activity and rate of corrosion of oil tank bottom atmospheric pressure storage tanks (Chinese national
plate on acidic soils. Mater Trans 2005; 46(11): 2490–2496. standard).
15. Park S, Kitsukawa S, Katoh K, et al. Development of AE 26. API 653-2009. Tank inspection, repair, alteration, and
monitoring method for corrosion damage of the bottom plate reconstruction.