Simulation Model of Fragmentation Risk

*Mirko Djelosevic and Goran Tepic*

#### **Abstract**

In this chapter, a simulation model for fragmentation risk assessment due to a cylindrical tank explosion is presented. The proposed fragmentation methodology is based on the application of Monte Carlo simulation and probabilistic mass method. The probabilities of generating fragments during the explosion of the tank were estimated regardless of the available accident data. Aleatoric and epistemic uncertainty due to tank fragmentation has been identified. Generating only one fragment is accompanied by aleatoric uncertainty. The maximum fragmentation probability corresponds to the generation of two fragments with a total mass between 1200 *kg* and 2400 *kg* and is 17%. The fragment shape was assessed on the basis of these data and fracture lines. Fragmentation mechanics has shown that kinematic parameters are accompanied by epistemic uncertainty. The range of the fragments in the explosion of the tank has a Weibull distribution with an average value of 638 *m*. It is not justified to assume the initial launch angle with a uniform distribution, since its direction is defined by the shape of the fragment. The presented methodology is generally applicable to fragmentation problems in the process industry.

**Keywords:** Simulation, fragmentation, explosion, cylindrical tank, risk assessment

#### **1. Introduction**

The most common accidents with dangerous substances involve leaks, fires and explosions [1]. If these events are an integral part of the accident chain, then they are manifested through a domino effect [2, 3]. The main reason for fires in process plants is the presence of flammable vapors [4]. Fires lead to heating of process installations and increase of pressure in them creating conditions for explosions due to BLEVE effect [5]. Explosions of process equipment due to the domino effect imply a pronounced fragmentation effect [6]. Fragmentation action in the accident chain (domino effect) is characterized by the fact that it is both a cause and a consequence of explosions [7]. The fragmentation effect during the explosion of the tank is accompanied by a very pronounced uncertainty of geometric and kinematic parameters of the fragments [8]. Large-scale accidents that have occurred in recent times are the result of progressive technological developments, and the typical examples are Toulouse [9] and Neyshabur [10]. Fragmentation barriers are used as a form of protection against the fragmentation effect [11, 12]. The fragmentation effect and the mechanism of formation during the explosion of the tank are presented in [13]. The literature recognizes the basic geometric characteristics as the number and shape of fragments, and defines the kinematic parameters through the trajectory and velocity of the fragments [14]. An initial procedure for estimating the number and mass of a fragment of a LPG storage tank was proposed by Baker et al. [15]. This research served as a starting point for some recent studies in the field of tank fragmentation [16, 17]. Tank fragmentation analysis should identify potential hazards and risks in terms of protecting process equipment from accident escalation [18]. Fragmentation barriers are used as protection against the fragmentation effect and were originally used in nuclear plants [19]. There are several models in the literature according to which the impact energy of a fragment into the target zone is estimated based on the fragment velocity [20–22]. Accident data show that two or three fragments are usually formed during tank fragmentation [23]. It has been determined that the distinct presence of fire during the fragmentation of the tank usually gives one fragment [24]. Previous research for the number of generated fragments estimation has exclusively used the maximum entropy model [25]. This model is based on accident data and shows that explosions with more than five generated fragments are very rare accidents [26, 27].

Tank explosions with the BLEVE effect very rarely provide more than three generated fragments [28]. The implementation of the maximum entropy model is possible only if there are available accident data for this type of process equipment [29]. Fragmentation mechanics analyzes the flight of a fragment and for that purpose the literature sources state a simplified mathematical model [30]. This model is represented in all recent research and was originally proposed by Mannan [31]. Greater mass of the fragment corresponded to the higher initial kinetic energy or the initial velocity [32]. Risk assessment due to the action of fragments is very often estimated in the literature on the basis of kinetic energy of fragments [33]. This energy is usually defined by the percentage share of expansion energy [34]. Some recommendations suggest that this percentage ranges between 5% and 20% [35]. This procedure of determining the initial velocities of the fragments has significant deviations from the real values, so it can only be used for general estimation. The aerodynamic properties of the fragments have a great influence on its kinematic parameters and are reflected in the uncertainty of the shape [36]. Djelosevic and Tepic conducted a complex study of cylindrical tank fragmentation in terms of identifying aleatoric and epistemic uncertainty [37, 38]. The same authors established a correlation of geometric and kinematic parameters of fragments, stating the importance of simulation technique in fragmentation analysis. This chapter will present a general methodological framework for the study of tank fragmentation under the conditions of the BLEVE effect, which is characteristic of the gas industry [39]. The focus of this research is on the analysis of types of uncertainty that follow fragmentation parameters with special reference to fracture probabilities and elimination of conditions that introduce influential fragment sizes into the zone of epistemic uncertainty [40].

## **2. Fragmentation methodology**

Tank fragmentation implies physical separation of fragments from the tank construction itself. The basic feature of fragments is the kinetic energy they have just before hitting a target. Greater kinetic energy of the impact creates greater potential hazard as a result of the fragmentation effect of the explosion. Assessing the kinetic energy of fragments is a complex task which requires identification of geometric and kinematic parameters. The basic geometric parameters are the shape and mass of fragments, whereas the most important kinematic parameter is the initial velocity of a fragment. Geometric and kinematic parameters are not independent since the shape of a fragment affects its initial velocity and initial launch direction. Literature

resources on the assessment of geometric and kinematic parameters of fragments are scarce. Therefore, the assessment of these parameters in this paper was conducted using probabilistic and simulation techniques. The probabilistic approach will first be presented and thereafter the simulation analysis of fragmentation.

## **2.1 Probabilistic approach**

The probabilistic approach is used for the assessment of fragment shapes, number of generated fragments as well as for the identifaction of aleatoric and epistemic uncertainty. The effect of uncertainty in assessing geometric and kinematic parameters is extremely important because if a parameter has the same or approximately the same probability with the change of influential factors, then it is definitely accompanied by aleatoric uncertainty. Aleatoric uncertainty is typical of those parameters whose uncertainty cannot be eliminated. On the other hand, there is epistemic uncertainty. If a parameter has epistemic uncertainty, that kind of uncertainty can be eliminated by additional research procedures and the parameter can be subjected to deterministic principles. A typical example of epistemic uncertainty is the initial launch angle of a fragment. Literature resources show that any angle value has the same probability of occurence and thus this parameter is introduced into the zone of aleatoric uncertainty. We will show that this is not justified and that the intitial launch angle of a fragment is defined by the shape of the fragment, i.e. by potential fracture lines of a tank.

Probabilistic analysis of influential fragmentation factors is an efficient way of distinguishing between epistemic and aleatoric uncertainty. The probabilistic approach is based on the use of the probabilistic mass method which was originally developed by Djelosevic and Tepic [37, 38]. The main purpose of this method is the assessment of tank fragmentation probabilities on the basis of ideal values and the mass factor. Ideal fragmentation probabilities are assessed using statistical simulation on a sufficient number of samples. The precondition for a sufficient number of statistical samples is clear convergence of results, which, in this case, is achieved with more than 100,000 samples. Ideal fragmentation probabilities were obtained under the assumption of uniform stress state and strength (homogeneity) of the material. Ideal fragmentation probabilities are those that correspond to the explosion of a tank with a uniform stress state, and the values that refer to the specific number of generated fragments are presented in **Table 1**.

Since the actual stress state of a tank is not uniform, the ideal fragmentation probabilities have to be corrected. The correction factor used is the so-called mass factor (*fmass*) which represents the ratio between the mass of the part of the tank with the non-uniform stress state and the total mass of the tank. Mass factor values for typical cylindrical tanks with torispherical end caps range between 0.55 and 0.75. Greater values indicate greater uniformity of the stress state and vice versa. The effect of fire contributes to greater non-uniformity so the mass factor in this case has lower values. The assessment of mass factors requires division of the tank into segments. Cylindrical tanks should be devided into three segments irrespective of their construction type and size. The first segment comprises the cylindrical part between the supports and it is defined by length L (S1) according to **Figure 1**. The other two segments (S2 and S3) comprise the parts of the tank outside the supports


#### **Table 1.** *Ideal tank fragmentation probabilities.*

**Figure 1.** *Type of the tank and critical stress zones.*

and their lengths amount to [L – L(S1)]/2. Tank fracture can occur in individual segments (either S1 or S2 or S3) or in at least two segments (S1 and S2 or S1 and S3 or S2 and S3) or in all three segments (S1 and S2 and S3). Depending on the mentioned scenarios, fracture probabilities and conditional fracture probabilities resulting from tank fragmentation are derived. Fracture probabilities are associated with tank fragmentation within only one segment (either S1 or S2 or S3). Conditional fracture probabilities comprise fragments which were generated from at least two segments.

Fracture probabilities and conditional fracture probabilities according to the number of fragments are given in **Table 2**.

Conditional fracture probabilities for individual fragmentation scenarios (Sc1 … Sc8) were obtained on the basis of fracture probabilities values for segments S1, S2 and S3. This way the probabilistic mass method enabled easy assessment of fracture probabilities according to the number of generated fragments. The literature on the subject usually mentions the entropic model for this purpose where deviation of about 50% for the probability of the third fragment is observed. The entropic model is based on accident data fitting, whereas the probabilistic mass method is independent of accident data. A comparative analysis of fracture probabilities according to the number of generated fragments for the entropic model, the probabilistic mass method and accident data is presented in **Figure 2**.

*Simulation Model of Fragmentation Risk DOI: http://dx.doi.org/10.5772/intechopen.98955*


#### **Table 2.**

*Fracture probabilities and conditional fracture probabilities of tank fragmentation.*

**Figure 2.** *Comparative analysis of fracture probabilities.*

### **2.2 Fracture scenarios**

Fracture scenarios of fragmentation in tank explosions imply a qualitative and quantitative analysis which is performed using Failure Tree Analysis (FTA). The triggering event that leads to tank fragmentation is reaching the critical pressure. The major central events are fractures of segments 1, 2 and 3. Qualitative analysis offers eight potential scenarios (Sc1 … Sc8). All the scenarios imply tank fragmentation apart from scenario Sc5 which excludes the possibility of tank fracture when the critical pressure is reached. Quantitative analysis implies the assessment of fracture probabilities and conditional fracture probabilities. These probabilities were assessed by means of Monte Carlo simulation using the probabilistic mass method, and they are presented in **Table 2**.

A significant part of this research deals with the assessment of the mass of fragments generated during tank explosion. For that purpose, the mass of fragments is expressed via the percent share of the empty tank mass which amounts to 12.3 *t*. Simulation results with over 100,000 samples show that fragments whose sum of masses ranges between 10% and 20% of the empty tank have the greatest generation probability. In this concrete case, the mass of fragments is between 1,230 *kg* and 2,460 *kg*. The maximum fragmentation probability is observed when two fragments are generated in an explosion and it amounts to around 17%.

Generation of fragments with small mass (smaller than 1% of the total mass of the tank) and generation of more than six fragments are extremely rare. The distribution of probabilities for the generated fragments depending on their mass is presented in **Figure 3**.

The probabilities of generation of only one fragment for different masses have uniform distribution, which points to aleatoric uncertainty. This means that the shape and mass of only one fragment generated in an explosion cannot be predicted

**Figure 3.** *Distribution of probabilities for fragments with different mass.*

#### *Simulation Model of Fragmentation Risk DOI: http://dx.doi.org/10.5772/intechopen.98955*

with certainty. On the other hand, generation of two or more fragments is accompanied by epistemic uncertainty. This means that with adequately used methodology, the mass and shape of two or more generated fragments can be predicted with certainty with the probabilities shown in **Figure 3**.

### **2.3 The assessment of shape and mass of fragments**

The assessment of mass and shapes of fragments requires defining fracture lines, i.e. the lines along which tank fracture occurs**.** Fracture lines are zones of pronounced stress of the material and are the result of different construction conditions. Definition of fracture lines implies prior pressure analysis. In this case, the analysis was performed using the ANSYS software for the operating pressure of 16.7 *bar*. Fracture lines can spread in those zones whose pressure exceeds 130 *MPa*.

Accordingly, the investigated cylindrical tank has a total of 13 typical areas from which tank fragments can be generated (**Figure 4**). The most pronounced stress of the tank is on the transition from the cylinder to the end caps of the tank. This is the reason why during an explosion a fragment containing a larger or smaller part of the end cap is almost always generated.

Tank fragmentation is most often accompanied by generation of one, two or sometimes three fragments. Their typical shapes, mass, fracture zones and generation probabilities are given in **Table 3**. The most probable scenario in tank explosion is the one with the generation of two fragments with the sum of masses of around 1,200 *kg* or 2,255 *kg*.

**Figure 4.** *Stress state of the tank with fracture lines.*


**Table 3.** *Characteristic fragmentation forms of the tank.*

#### **3. Fragmentation mechanics**

Fragmentation mechanics involves modeling the flight of fragments created by a tank explosion. The basic characteristics of the fragments include geometric and kinematic parameters. We identified the geometric parameters in Section 2 of this chapter and they include number, mass and shape of the generated fragments. The main kinematic parameters include the initial velocity and the initial direction of the fragment launch. The literature does not provide information on the distribution of the range of fragments.

Therefore, it is not justified to assume that the range of the fragments is accompanied by a random distribution. Also, it is not justified to introduce assumptions about the uniformity of kinematic parameters of the fragments, just because we do not have enough available information about their behavior. Assuming a uniform distribution for some of the kinematic parameters, we enter an area of aleatoric uncertainty that does not allow an adequate assessment of fragmentation risk. The authors of this chapter start from the assumption that the generation of fragments does not follow a stochastic process, thus putting epistemic uncertainty in the foreground.

#### **3.1 Fragment flight model**

The flight of the fragment takes place under the influence of inertial, gravitational and aerodynamic forces (air resistance and lift force). The trajectory of the fragment uniquely determines the vector form of the equation of motion:

$$m\_{\hat{fr}} \cdot \overrightarrow{a}\_{\hat{fr}} = \overrightarrow{\dot{W}}\_D + \overrightarrow{\dot{W}}\_L + \overrightarrow{\dot{G}} \tag{1}$$

The air resistance force (*WD*) and the thrust force (*WL*) are defined by:

$$
\vec{W}\_D = -\left(\frac{1}{2}\rho\_v \mathbf{C}\_D A\_D v\_{\hat{fr}}\right) \cdot \vec{v}\_{\hat{fr}} \tag{2}
$$

*Simulation Model of Fragmentation Risk DOI: http://dx.doi.org/10.5772/intechopen.98955*

$$
\overrightarrow{\dot{W}}\_{L} = -\left(\frac{1}{2}\rho\_v \mathbf{C}\_L A\_L v\_{fr}\right) \cdot \overrightarrow{v}\_{fr} \tag{3}
$$

The flight of each of the fragments should be observed in the local coordinate system Oxyz and then the projections of the vector differential Eq. (1) read:

$$\boldsymbol{a}\_{\mathbf{x}\circ\mathbf{r}} = \dot{\boldsymbol{v}}\_{\mathbf{x}\circ\mathbf{r}} = \left(-k\_{\mathrm{D}}\boldsymbol{v}\_{\mathbf{x}\circ\mathbf{r}} - k\_{\mathrm{L}}\boldsymbol{v}\_{\mathbf{x}\circ\mathbf{r}}\right)\sqrt{\boldsymbol{v}\_{\mathbf{x}\circ\mathbf{r}}^{2} + \boldsymbol{v}\_{\mathbf{x}\circ\mathbf{r}}^{2}}\tag{4}$$

$$
\mathfrak{a}\_{\mathfrak{y}\circ \mathfrak{r}} = \dot{\mathfrak{v}}\_{\mathfrak{y}\circ \mathfrak{r}} = \mathbf{0} \tag{5}
$$

$$a\_{x\_{\!\!f\!r}} = \dot{v}\_{\!\!y \!r} = \left(k\_{L}v\_{x\_{\!\!y \!r}} - k\_{D}v\_{x\_{\!\!f\!r}}\right) \sqrt{v\_{x\_{\!\!f\!r}}^{2} + v\_{x\_{\!\!f\!r}}^{2}} - \mathbf{g} \tag{6}$$

Practically, the flight of the fragment is completely described by (4), (5) and (6), since there is no motion in the direction of the y-axis (that is why we observed the motion in the local coordinate system). The Taylor's series method was used to solve the coupled system of nonlinear differential equations. The ratio of the velocity components initially defines the initial launch angle of the fragment (**Figure 5**). This simply proves that the initial launch angle is not accompanied by a stochastic distribution.

#### **3.2 Initial conditions**

The initial conditions define the kinematic parameters at the initial moment (at the moment of the tank explosion). These conditions are necessary for initiating the procedure of numerical solution of differential equations and read:

$$\mathbf{x}\_{\hat{f}r}(t\_0) = \mathbf{x}\_0 \wedge \mathbf{z}\_{\hat{f}r}(t\_0) = \mathbf{z}\_0 \tag{7}$$

$$\upsilon\_{\mathbf{x}\circ\mathbf{f}}(t\_0) = \upsilon\_{\mathbf{x}o} \land \upsilon\_{\mathbf{z}\circ\mathbf{f}}(t\_0) = \upsilon\_{\mathbf{z}o} \tag{8}$$

$$
\dot{\upsilon}\_{x\circ\hat{r}}(t\_0) = \mathfrak{a}\_{xo} \wedge \dot{\upsilon}\_{x\circ\hat{r}}(t\_0) = \mathfrak{a}\_{xo} \tag{9}
$$

**Figure 5.** *Kinematic parameters due to tank fragmentation.*

#### *Simulation Modeling*

These values are used as the first step in the numerical procedure and they are of unknown magnitude at the moment. At this level we know that the initial velocity is not independent of the initial acceleration of the fragments. The components of velocity at any moment t read as follows:

$$v\_{xfr} = \sqrt{\frac{(-a\_x)}{k\_D + k\_L \cdot t \text{g} \,\phi}} \cdot \frac{1}{\sqrt[4]{1 + t \text{g}^2 \phi}} \tag{10}$$

$$v\_{x,fr} = \sqrt{\frac{(-a\_x)}{k\_D + k\_L \cdot \text{tg}\,\phi}} \cdot \frac{\text{tg}\,\phi}{\sqrt[4]{1 + \text{tg}^2\phi}}\tag{11}$$

The direction of the fragment velocity can be determined at any time during the flight of the fragment, including the initial moment using the expression:

$$\phi = \operatorname{arctg}\left(\frac{\upsilon\_{xfr}}{\upsilon\_{xfr}}\right) = \operatorname{arctg}\frac{\left(\frac{k\_l}{k\_D}\right) \cdot (-a\_x) + (-a\_x) - \mathbf{g}}{\left[ (-a\_x) + \left(\frac{k\_l}{k\_D}\right) \cdot a\_x + \mathbf{g} \right]} \tag{12}$$

In order to obtain the initial launch angle of a fragment, it is necessary to know the components of the initial acceleration. In the continuation of this chapter, we show how the initial acceleration occurs. It is important to point out that (12) shows an unjustified assumption about the uniformity of the initial acceleration. Thus, the initial launch angle as a kinematic parameter is classified in the category of epistemic uncertainty. Literature sources in this area do not use the initial acceleration parameter, although it is a way to remove uncertainty regarding a reliable fragmentation risk assessment.

#### **3.3 Defining the initial acceleration**

We have previously came to the conclusion that in order to define the initial velocity and the initial launch angle of a fragment, it is necessary to know the initial acceleration. Hence the idea and justification for introducing this kinematic parameter into fragmentation analysis. The initial acceleration is proportional to the force of pressure on the fragment. This force is created by the critical pressure *pcr* acting on the inner surface of the fragment. The proportion of explosive energy transferred to the fragment is limited by the action of critical pressure. This means that the proportion of explosive energy of the fragment transferred to the fragment depends on the tensile strength of the material.

The procedure for determining the initial acceleration is based on this assumption. The lower tensile strength gives less initial kinetic energy of the fragment. Fragment generation occurs when the von Mises's stress reaches a critical value under the action of internal tank pressure and is defined as:

$$
\sigma\_{cr} = \sqrt{\sigma\_{\text{x}}^2 + \sigma\_{\theta}^2 - \sigma\_{\text{x}}\sigma\_{\theta} + \frac{3}{2}(\sigma\_{\text{x}} - \sigma\_{\theta})^2} = 102 \cdot p\_{cr} \tag{13}
$$

Where the corresponding components of the von Mises's stress are given by:

$$\begin{aligned} \sigma\_{\mathfrak{x}} &= 101 \cdot p\_{cr} \\ \sigma\_{\theta} &= 105 \cdot p\_{cr} \end{aligned} \tag{14}$$

*Simulation Model of Fragmentation Risk DOI: http://dx.doi.org/10.5772/intechopen.98955*

Separation of fragments from the tank as a whole occurs when the critical stress reaches the value of tensile strength of the material *fm*, so the critical pressure is determined as *pcr* = *fm*/102. Accordingly, the initial acceleration of a fragment can be defined by:

$$a\_o = \frac{F}{m\_{fr}} = \frac{p\_{cr}A\_{fr}}{\rho \delta A\_{fr}} = \frac{f\_m}{102\rho\delta} = const \tag{15}$$

This proves that the initial acceleration has a constant value for a certain type of steel from which the tank is made. LPG transport and storage tanks have a constant wall thickness of 14 *mm*. The density of steel *ρ* is also constant and amounts to 7850 *kg*/*m*<sup>3</sup> .

Thus, by knowing the initial acceleration, we are able to define the initial velocity and the initial launch angle, as well as all other kinematic parameters at any time during the flight of the fragment. It should be borne in mind that the initial acceleration depends on the tensile strength of the material whose values are subject to variation with temperature changes. The change in the influence values on the range of the fragment is given in **Table 4**.

#### **3.4 Fragments range**

Based on a mathematical model that describes the flight of the fragment and the initial acceleration, we are able to determine the range of the fragments. For this purpose, all geometric and kinematic parameters will be classified into two groups. The first group consists of invariant parameters and their value is fixed. The second group consists of variable parameters, ie those whose value changes during the flight of the fragment. This group includes the coefficients of aerodynamic and thrust acceleration *kL* and *kD*. The invariant parameters are fully defined and their value is not subject to variation during the flight of the fragment (tank wall thickness, tensile strength and specific weight of the material). Variable parameters change during the flight of the fragment due to the rotation of the fragment or some other effects. Variable parameters are defined via the air resistance coefficient *CD* and the thrust coefficient *CL*.

The literature gives approximate values of these coefficients which depend on the shape of the fragment. Smaller fragments whose mass does not exceed a few hundred kilograms generally have the shape of a shell (shells are aerodynamic shapes), so they have a pronounced thrust effect. In order to estimate the range of the fragments, the fluctuation of variable parameters is performed, whereby different trajectories of the fragments are obtained. These trajectories enable the definition of the limit values of the coefficients *kL* and *kD*. Parabolic trajectories of


#### **Table 4.**

*Temperature influence on the influence values of the fragment range.*

**Figure 6.** *Distribution of the range of typical shaped fragments.*

small height (fragment reaches small range and small height) as well as pointed trajectories (fragments reach large height and relatively small range) are very rare and represent boundary cases for the selection of coefficients *kL* and *kD*. Fragment range estimation was realized by Monte Carlo simulation by processing 240 different samples for different fragmentation parameters. Fragments of up to a few hundred kilograms launched at an angle of up to 35<sup>0</sup> can be adequately represented by the Weibull's distribution with parameters 2.3 and 723.8 as shown in **Figure 6**.

The maximum probability of 6.5% corresponds to fragments with a range between 613 *m* and 663 *m*. This means that 6.5% of the total number of generated fragments will fall to the target between these distances.
