Validity of Probit Functions For Domino Effect

Journal of Physics: Conference Series
Related content
- A case study of electrostatic accidents in
Study on the Rationality and Validity of Probit the process of oil-gas storage and
transportation
Models of Domino Effect to Chemical Process Yuqin Hu, Diansheng Wang, Jinyu Liu et
al.
Equipment caused by Overpressure - Photographs from ScieTech 2014
- Analysis of a Station Black-Out transient in

To cite this article: Dongliang Sun et al 2013 J. Phys.: Conf. Ser. 423 012002 SMR by using the TRACE and RELAP5
code
F De Rosa, C Lombardo, F Mascari et al.
View the article online for updates and enhancements.
This content was downloaded from IP address 168.194.162.217 on 05/03/2018 at 19:42

ScieTech 2013 IOP Publishing
Journal of Physics: Conference Series 423 (2013) 012002 doi:10.1088/1742-6596/423/1/012002
Study on the Rationality and Validity of Probit Models of

Domino Effect to Chemical Process Equipment caused by
Overpressure
Dongliang Sun1, Guangtuan Huang1, , Juncheng Jiang2, Mingguang Zhang2,

Zhirong Wang2
1
State Environmental Protection Key Laboratory of Environmental Risk Assessment and
Control on Chemical Process, School of Resources and Environmental Engineering, East
China University of Science and Technology, No. 130 Meilong Road, Xuhui District,
Shanghai 200237, China
2
Jiangsu Key Laboratory of Urban and Industrial Safety, Institute of Safety Engineering,
School of Urban Construction and Safety Engineering, Nanjing University of Technology,
No. 200 North Zhongshan Road, Nanjing 210009, Jiangsu, China
E-mail: dongliangsun@ecust.edu.cn
Abstract. Overpressure is one important cause of domino effect in accidents of chemical process
equipments. Some models considering propagation probability and threshold values of the domino
effect caused by overpressure have been proposed in previous study. In order to prove the
rationality and validity of the models reported in the reference, two boundary values of three
damage degrees reported were considered as random variables respectively in the interval [0,
100%]. Based on the overpressure data for damage to the equipment and the damage state, and the
calculation method reported in the references, the mean square errors of the four categories of
damage probability models of overpressure were calculated with random boundary values, and
then a relationship of mean square error vs. the two boundary value was obtained, the minimum of
mean square error was obtained, compared with the result of the present work, mean square error
decreases by about 3%. Therefore, the error was in the acceptable range of engineering
applications, the models reported can be considered reasonable and valid.
1. Introduction
Many types of equipment containing dangerous materials may exist in industrial installations: tanks. If
the equipment was damaged due to critical conditions of pressure, mechanical impacts or structural
cracking, a catastrophic sequence may rise, a sudden explosion, for instance. Then overpressure, heat
radiation, and many fragments can be generated and threaten other equipment erected in their vicinity,

To whom any correspondence should be addressed.
Published under licence by IOP Publishing Ltd 1

when the objects are destroyed, new accidents may take place, therefore, domino effect is resulted in and
make accident consequence more severe [1-11].
In recent years, a series of research work has identified that overpressure in explosion accident is one
important cause of domino effect to chemical process equipments. Some models considering propagation
probability and threshold values of the domino effect caused by overpressure have been proposed in
previous study [1-19]. With scarce data and some oversimplified assumption, these models almost only
related damage to peak static overpressure [1-8]. In the present study, in order to prove the rationality and
validity of the models developed by Zhang, two boundary values of three damage degrees were
considered as random variables respectively in the interval [0, 100%]. Based on the overpressure data for
damage to the equipment and the damage state, and the calculation method reported in the references, the
mean square errors of the four categories of damage probability models of overpressure were calculated
with random boundary value, and then a relationship of mean square error vs. the two boundary value
was obtained, in order to find the minimum of mean square error, and then it was compared with the
previous result, to prove whether the previous result was in the range of error allowed.
2. Overview of previous study
In previous work, based on damage phenomenon and relevant threshold data of peak static overpressure
from past accidents, many results were obtained, see Table 1 [1-8]: damage probability models of
overpressure generally went through three stages: Probit modelüEscalation thresholdsüProbit model.
So far, the models developed by Cozzani and Zhang were more advanced than others, especially the latter,
which was obtained with more sufficient data of damage caused by overpressure, and rational method of
probability analysis [5-8]. Based on the results of Cozzani, more data about damage to chemical process
equipment were gathered and analyzed, the damage degrees of the four categories of vessels (atmospheric,
pressurized, elongated, small) were divided into three classes respectively: DS1LI1, DS2LI2, DS2LI3, and
then entire probability range of 0–100% was also divided into three ranges: the range of 0–30% was
assumed to data in DS1LI1 damage state; the range of 30–70%was assumed to data in DS2LI2 damage
state; the range of 70–100% was assumed to data in DS2LI3 damage state. Finally, the damage
probabilities were calculated, quantitative relationship between damage probabilities and damage degrees
of equipment were built respectively for the four categories of vessels, and the mean square errors of the
four models decrease from 39.1% to 12.5%, the improvements of present models were evidenced through
comparison with literatures [5-8]. However, in the method of probability analysis above, the boundary
values of the three damage degrees 30% and 70% were set arbitrarily, if the boundary values were set to
other values, whether the mean square errors of the four models would be less than 12.5%, and whether
more reasonable and realistic models could be obtained?
Table 1. Probit models reported in the literatures.
Author Categorization of Damage Probability model Threshold values of
process plants peak static
overpressure
Eisenberg et al. [1] No categorization Y 23.8 2.92 ln'P
No categorization 'Pth = 36kPa

Fd 1 r rth
2 , r: distance from
Bagster and Pitblado [2] explosion center; rth: distance

from explosion center at which
static overpressure equals 'Pth
No categorization if 'P < 70kPa, Fd = 0; if 'P >
Khan and Abbasi [3]
70kPa, Y 23.8 2.92 ln'P
2
Atmospheric if 'P <'Pth, Fd = 0 'Pth = 7kPa

Gledhill and Lines [4]
Pressurized if 'P t 'Pth, Fd = 1 'Pth = 38kPa
Atmospheric Y 18.96 2.44 ln'P
Cozzani and Salzano Pressurized Y 42.44 4.33 ln'P
[5-7] Elongated Y 28.07 3.16 ln'P
Small Y 17.79 2.18 ln'P
Atmospheric Y 9.36 1.43 ln'P
Pressurized Y 14.44 1.82 ln'P

Zhang etal. [8]
Elongated Y 12.22 1.65 ln'P
Small Y 12.42 1.64 ln'P
Fd: failure probability; Y: probit values corresponding to failure probability; 'P: peak static overpressure (Pa).
3. Methods
3.1. Random variables

Based on the statement above, in order to prove the rationality of the models developed by Zhang, the
boundary values above were considered as random variables x, y, and supposed to followed uniform
distributions in the interval [0, 100%] respectively, and x < y:
x ui
(1)
y vi
where ui, vi are uniform random number respectively in [0, 1].
3.2. Damage probabilities

In each damage degree, the damage probabilities of the four categories of vessels caused by overpressure
can be obtained by the Eq. (2), whose initial form was built by Zhang in Ref. [8]:
'P
°x u , DS1LI 1
° ' Pmax 1
° 'P
P ®x y xu , DS 2 LI 2 (2)
° 'Pmax 2
° 'P
° y 1 y u , DS 2 LI 3
¯ 'Pmax 3
where P is probability value, 'P is overpressure value and 'Pmaxi (i = 1, 2, 3) is a threshold value of
overpressure in each damage state. The coefficient value x and y were decided based on some assumptions
stated below:
–In DS1LI1 damage state, damage probability value x was assumed to be correspondent to the highest
overpressure value ('Pmax1) of DS1LI1 damage state.
–In DS2LI2 damage state, damage probability value y was assumed to be correspondent to the highest
overpressure value ('Pmax2) of DS2LI2 damage state.
–In DS2LI3 damage state, damage probability value 100% was assumed to be correspondent to the
highest overpressure value ('Pmax3) of DS2LI3 damage state.
3
In Eq. (2), we can see that, 'P 'Pmax 1 1 , 'P 'Pmax 2 1 and 'P 'Pmax 3 1 , P follows uniform
distributions respectively in the interval [0, x], [x, y] and [y, 1], with respect to DS1LI1, DS2LI2 and
DS2LI3.
3.3. Mean square errors

Based on 'P values, damage degree for each kind of vessel reported in Table 2-5, the damage
probabilities can be calculated by Eq. (2) for each group of x and y, and then probit values can be
calculated from probability data. Therefore, probit models for different categories were obtained by least
square regression, as well as mean square error of the models. In a word, the mean square error of the four
categories of vessels could be calculated with x and y. If the mean square error less than 12.5% could be
obtained by building the relationship of mean square error vs. the two boundary value, more reasonable
models could be obtained.
Table 2. 'P values, damage phenomenon and degree for atmospheric vessels.
'P (kPa) Damage phenomenon Damage degree
5.17 Minor damage, cone roof tank (100% filled) DS1LI1
5.17 Minor damage, cone roof tank (50% filled) DS1LI1
6.10 1% structural damage of equipment DS1LI1
10.00 Failure of atmospheric equipment DS2LI2
14.00 Minor damage of atmospheric tank DS1LI1
18.70 Minor damage, floating roof tank (50% filled) DS1LI1
18.70 Catastrophic failure, cone roof tank (50% filled) DS2LI3
20.00 Deformation of atmospheric tank DS1LI1
24.00 20% structural damage of steel floating roof tank DS2LI2
25.00 Atmospheric tank destruction DS2LI3
27.00 Failure of steel vessel DS1LI1
42.51 Catastrophic failure, cone roof tank (100% filled) DS2LI3
136.00 Structural damage, low pressure vessel DS2LI3
136.05 Catastrophic failure, floating roof tank (50%filled) DS2LI3
136.05 Catastrophic failure, floating roof tank (100% filled) DS2LI3
136.10 99% structural damage of floating roof tank DS2LI3
137.00 99% damage (destruction) of floating roof petroleum tank DS2LI3
7.00 Collapse of atmospheric tank roof DS1LI1
7.00 Partial damage to atmospheric tank DS1LI1
10.00 Fixed roof tank damage DS1LI1
10.00 50% damage to atmospheric tank DS2LI2
20.00 Displacement of steel supports DS1LI1
20.00 100% damage, atmospheric tank DS2LI3
35.00 80% damage of process plant DS2LI3
21.00 Destruction of fixed roof atmospheric tank DS2LI3
42.51 Minor damage, floating roof tank (100% filled) DS1LI1
45.00 Catastrophic failure, floating roof tank DS2LI3
7.00 Failure of connection DS1LI1
4
20.00 Tubes deformation DS1LI1

22.10 Minor damage, pipe supports DS1LI1
37.42 Catastrophic failure, pipe supports DS2LI2
42.00 Tubes failure DS2LI2
Table 3. 'P values, damage phenomenon and degree for pressurized vessels.
30.00 Failure of pressure vessel DS1LI1
39.00 Structural damage to pressure vessel DS2LI2
39.12 Minor damage, pressure vessel horizontal DS1LI1
42.00 Pressure vessel deformation DS1LI1
52.72 Minor damage, tank sphere DS1LI1
53.00 Pressure vessel failure DS2LI2
53.00 Failure of spherical pressure vessel DS2LI2
55.00 20% structural damage of spherical steel petroleum tank DS2LI2
61.22 Catastrophic failure, pressure vessel horizontal DS2LI3
81.63 Minor damage, pressure vessel vertical DS1LI1
83.00 20% damage of vertical cylindrical steel pressure DS2LI2
88.44 Catastrophic failure, pressure vessel vertical DS2LI3
95.30 99% structural damage of vertical, steel pressure vessel DS2LI3
97.00 99% damage of vertical cylindrical steel pressure vessel DS2LI3
108.84 Catastrophic failure, tank sphere DS2LI3
108.90 99% structural damage of spherical, pressure steel vessel DS2LI3
110.00 99% damage of spherical steel petroleum tank DS2LI3
38.00 Partial damage of pressure vessel DS2LI2
70.00 Failure of pressurized storage sphere DS2LI2
Table 4. 'P values, damage phenomenon and degree for elongated vessels.
17.00 Minor damage, distillation tower and cylindrical steel DS1LI1
vertical structure. Failure of part of equipment
29.00 Distillation tower and cylinder steel vertical structure DS1LI1
35.71 Minor damage, fractionation column DS1LI1
38.00 Deformation of non-pressure equipment DS1LI1
42.52 Minor damage, extraction column DS1LI1
45.92 Catastrophic failure, fraction column DS2LI3
47.00 Failure of non-pressure DS2LI2
69.73 Catastrophic failure, extraction column DS2LI3
35.00 Damage to fractionating column DS1LI1
5
14.00 Minor damage of cooling tower DS1LI1

Table 5. 'P values, damage phenomenon and degree for small equipments.
25.30 Minor damage, reactor chemical DS1LI1
49.32 Minor damage, heat exchanger DS1LI1
59.52 Catastrophic failure, reactor chemical DS2LI3
59.52 Catastrophic failure, heat exchanger DS2LI3
76.53 Catastrophic failure, reactor cracking DS2LI3
81.63 Minor damage, pump DS1LI1
108.84 Catastrophic failure, pump DS2LI3
18.70 Minor damage, reactor cracking DS1LI1
4. Analysis of the rationality and accuracy of the models

By the calculation method discussed above, the mean square errors were obtained through a certain
number of simulations, and then a relationship of mean square error vs. x and y could be obtained by data
fitting. However, the accuracy of the regression depends on the number of the data, which is referred to
the number of the simulations. Finally, the relationship is expressed as follows and R = 0.9943:
0.05237 ln x 0.052176 tan y 0.027957 cos xy
0.000493 0.00211 0.00047 (3)
M
tan x tan y tan x cos y
where M is mean square error of the four obtained models. In the process of simulation, the coefficients were
set to a, b, c, d, e, and f (in Eq. (3), a = -0.05237, b = 0.052176, c = 0.027957, d = 0.000493, e = 0.00211, f =
-0.00047), they changed with the number of simulations, see Fig .1: a number of 1200 simulations could
provide a good convergence for the value of the coefficients.
6
Figure 1. The changes of the coefficients of the relationship with simulations.
12
100
25
68
11 78 2
4359
85
17 89 92 45 71
3
57 30 19 82
86
24 49 28
23
5848 52 21
63
9 2265 9915
56 87874
47
44
83 39
67734
62 41 76488
54
79
18
531
50
66
76
70721451
84
9896
1 81
40
2055
463
3774
328016
73
339735 69
6138
93 27
7526 67
9453
6013
10
91
42
95
36
90
29
7

Eq. (3) was a binary function, whose curve was shown in Fig. 2. Based on the trends of the curve, the
collection of x and y, in which the value of M was less than 12.5% was shown in the shadow of Fig. 3: in
this part, x < y, the blue curve was the point collection of x and y in which M=12.5%. In the shadow part
of Fig. 3, the minimum of M was 9.06%, compared with 12.5% obtained by Zhang, mean square error
decreased by about 3%. Therefore, the error was in the acceptable range of engineering applications
(generally 5%), the models developed by Zhang could be considered reasonable and valid.
5. Conclusion
The present work was to prove the rationality and validity of the probit models for damage to chemical
process equipment caused by overpressure:
–Damage probability models of overpressure generally went through three stages: Probit modelü
Escalation thresholdsüProbit model, and so far, the models developed by Cozzani and Zhang were more
advanced than others, especially the latter, which was obtained with more sufficient data of damage
caused by overpressure, and rational method of probability analysis.
–Damage probabilities of vessels, P followed uniform distributions respectively in the interval [0, x], [x, y]
and [y, 1], with respect to DS1LI1, DS2LI2 and DS2LI3.
–The relationship of mean square error vs. the two boundary value was built, and regression coefficient R
was more than 0.99.
–For building the relationship by data fitting, a number of 1200 simulations (1200 groups of x and y)
could provide a good convergence for the value of the coefficients of the equation.
–Based on the trends of the curve of the relationship erected by data fitting, the collection of x and y, in
which the value of M was less than 12.5% was obtained, the minimum of M was 9.06%, compared with
the result obtained by Zhang, mean square error decreased by about 3%. Therefore, the error was in the
acceptable range of engineering applications (generally 5%), the models developed by Zhang could be
considered reasonable and valid.
Acknowledgements
The financial support of China Postdoctoral Science Foundation-funded project (2012M520851), the
Shanghai Leading Academic Discipline Project (B506), National Natural Science Foundation of China
8
(No. 71001051, 50904037), and Research and Innovation Plan for Graduates of Colleges and Universities
in Jiangsu Province (CX09B_142Z) are gratefully acknowledged.
References
[1] Eisenberg N A, Lynch C J and Breeding R J 1975 Vulnerability model: a simulation system for
assessing damage resulting from marine spills, Rockville, MD, Enviro Control Inc., Report
CG-D-136-75
[2] Bagster D F and Pitblado R M 1991 Process Saf. Environ. Prot. 69 195–9
[3] Khan F I and Abbasi S A 1998 Process Saf. Prog. 17 107–23
[4] Gledhill J and Lines I 1998 Development of methods to assess the significance of domino effects
from major hazard sites, Health and Safety Executive, CR Report 183
[5] Cozzani V and Salzano E 2004 J. Hazard. Mater. A107 67–80
[6] Cozzani V and Salzano E 2004 J. Hazard. Mater. A107 81–94
[7] Cozzani V and Salzano E 2004 J. Loss Prevent. Process Ind. 17 437–47
[8] Zhang M G and Jiang J C 2008 J. Hazard. Mater. 158 280–6
[9] Antonioni G, Spadoni G and Cozzani V 2009 J. Loss Prevent. Process Ind. 22 439–49
[10] Cozzani V, Antonioni G and Spadoni G 2006 J. Loss Prevent. Process Ind. 19 463–77
[11] Cozzani V, Gubinelli G, Antonioni G, Spadoni G and Zanelli S 2005 J. Hazard. Mater. A127 14–30
[12] Darbra R M, Palacios A and Casal J 2010 J. Hazard. Mater. 183 565–73
[13] Kozanoglu B, Zjrate L, Gqmez-Mares M and Casal J 2011 J. Hazard. Mater. 197 104–8
[14] Palacios A and Casal J 2011 Fuel. 90 824–33
[15] Pavlova Y and Reniers G 2011 J. Hazard. Mater. 186 401–6
[16] Reniers G and Amyotte P 2012 J. Loss Prevent. Process Ind. 25 227–31
[17] Reniers G, Cuypers S and Pavlova Y 2012 J. Hazard. Mater. 209-210 164–76
[18] Reniers G L L, Audenaert A, Pauwels N and Soudan K 2011 J. Hazard. Mater. 186 779–87
[19] Reniers G L L, So’’rensen K and Dullaert W 2012 Reliab. Eng. Syst. Safe. 98 35–42

Validity of Probit Functions For Domino Effect

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Validity of Probit Functions For Domino Effect

Caricato da

Copyright:

Formati disponibili

Journal of Physics: Conference Series

Equipment caused by Overpressure - Photographs from ScieTech 2014

- Analysis of a Station Black-Out transient in

View the article online for updates and enhancements.

This content was downloaded from IP address 168.194.162.217 on 05/03/2018 at 19:42

Study on the Rationality and Validity of Probit Models of

Dongliang Sun1, Guangtuan Huang1, , Juncheng Jiang2, Mingguang Zhang2,

Published under licence by IOP Publishing Ltd 1

No categorization 'Pth = 36kPa

Bagster and Pitblado [2] explosion center; rth: distance

Atmospheric if 'P <'Pth, Fd = 0 'Pth = 7kPa

Cozzani and Salzano Pressurized Y 42.44  4.33 ln 'P

[5-7] Elongated Y 28.07  3.16 ln 'P

Small Y 17.79  2.18 ln 'P

Atmospheric Y 9.36  1.43 ln 'P

Pressurized Y 14.44  1.82 ln 'P

Small Y 12.42  1.64 ln 'P

3.1. Random variables

3.2. Damage probabilities

3.3. Mean square errors

20.00 Tubes deformation DS1LI1

14.00 Minor damage of cooling tower DS1LI1

4. Analysis of the rationality and accuracy of the models

Figure 1. The changes of the coefficients of the relationship with simulations.

Figure 2. The changes of the coefficients of the relationship with simulations.

Figure 3. The changes of the coefficients of the relationship with simulations.

Potrebbero piacerti anche

Cozzani and Salzano Pressurized Y 42.44 4.33 ln'P

[5-7] Elongated Y 28.07 3.16 ln'P

Small Y 17.79 2.18 ln'P

Atmospheric Y 9.36 1.43 ln'P

Pressurized Y 14.44 1.82 ln'P

Small Y 12.42 1.64 ln'P