Engineering Failure Analysis 91 (2018) 354–369

Contents lists available at ScienceDirect

Engineering Failure Analysis

journal homepage: www.elsevier.com/locate/engfailanal

Failure analysis of a pressure vessel subjected to an internal blast

T
load
⁎
I. Barsouma, , S.A. Lawala, R.J. Simmonsb, C.C. Rodriguesb
a
Department of Mechanical Engineering, Khalifa University of Science and Technology, P.O. Box 2533, Abu Dhabi, United Arab Emirates
b
Department of Industrial and Systems Engineering, Khalifa University of Science and Technology, P.O. Box 2533, Abu Dhabi, United Arab Emirates

A R T IC LE I N F O ABS TRA CT

Keywords: The objective of the current work is to model a stainless steel (SA 316L) autoclave explosion and
Finite element analysis rupture that occurred during a research laboratory experiment designed to study the thermal
Failure locus decomposition of ammonium tetrathiomolybdate in the presence of dimethylsulfoxide (DSMO) in
Blast load the autoclave. A finite element analysis is conducted to better understand the cause of failure of
Rupture
the autoclave and with the objective to investigate whether the incident was caused by static
Overpressure
Pressure vessel
overpressure or an internal blast load. The empirical CONWEP blast loading model is used to
model the internal blast load. The constitutive behavior of the autoclave material is modelled
using the Johnson-Cook (JC) plasticity and material failure model, which both account for the
effect of strain rate and temperature. By conducting uniaxial tensile tests and tests on notched
ring specimens cut from the autoclave, the true stress-strain curve and the ductile failure locus of
the autoclave material are established, respectively, which are used to obtain the constants of the
JC plasticity and failure model, respectively. The result of the finite element analysis revealed
that a blast load from an equivalent TNT charge of 0.042 kg, which resulted from the decom-
position of DMSO at high temperature, predicted markedly well the structural response and
subsequent failure of the autoclave observed in the post-incident investigation.

1. Introduction

Pressure vessels [1] and autoclaves are commonly designed to resist overpressure to a certain extent. Autoclaves used to conduct
chemical experiments at high temperature and pressure are usually designed with a safety feature in the case of overpressure.
Commonly, either a rupture disc is part of the vessel or the bottom part of the autoclave is designed to shear off such that excess
pressure can be released, in addition to optional provision of pressure relief valve [2]. However, certain explosive chemical reactions
can proceed with a speed such that the shock wave created from the explosion may rupture the autoclave before any of the pressure
relief mechanisms can release the excess pressure.
Chemical explosions [3] involve rapid reaction of oxidizers and fuel components that comprise the explosive mixture. There is a
sudden release of energy with the explosion, which produces a shock wave that propagates outward from the center of the explosion.
Fig. 1 shows a typical form of a blast wave after a chemical explosion. The pressure pulse is often characterized by a sudden increase
in pressure from ambient pressure, Po to a peak incident overpressure Pso. The overpressure decays exponentially as shown in Fig. 1,
and generally decreases to the ambient pressure in time t0, referred to as the positive phase. This is followed by a negative pressure
phase which is often neglected in blast loading calculations as little data exists for this regime. The area under the curve for the

⁎
Corresponding author.
E-mail address: imad.barsoum@ku.ac.ae (I. Barsoum).

https://doi.org/10.1016/j.engfailanal.2018.04.037
Received 1 February 2018; Received in revised form 23 April 2018; Accepted 24 April 2018
Available online 30 April 2018
1350-6307/ © 2018 Published by Elsevier Ltd.
I. Barsoum et al. Engineering Failure Analysis 91 (2018) 354–369

Fig. 1. Blast wave profile (pressure vs. time history) of an explosion [4].

positive phase is known as the specific impulse is. The peak overpressure, positive phase duration, and specific impulse are of
particular interest in blast loading calculations.
In the current study, numerical modelling of an autoclave subjected to in internal blast load, resulting from an explosive chemical
reaction, is undertaken. The purpose is to simulate the failure scenario of an incident that occured in a chemical laboratory while
conducting an experiment to decompose ammonium tetrathiomolybdate (ATM) and carbon nanotubes using dimethyl-sulfoxide
(DMSO) as a solvent. Reactants in the ruptured autoclave, henceforth referred to as autoclave ‘A’, contained 33 mg ATM, 100 mg of
carbon nanotubes and 65 cm3 of DMSO as solvent. The reactants in the other autoclave, henceforth referred to as autoclave ‘B’, were
the same, but with 65 cm3 of water as solvent instead of the DMSO. Both experiments were conducted using Teflon®-lined reaction
autoclaves (125 ml capacity each) made of SA 316L stainless steel, manufactured by Parr Instrument Company [2]. The autoclaves
(model number 4750) are designed to operate below a temperature and pressure of 350 °C and 20 MPa, respectively. Both vessels
were equipped with means to attach a pressure relief valve, but this was not used during the experiment. The autoclaves were placed
in a bench top furnace where temperature was programmed to increase at a rate of 5 °C per minute from room temperature to a final
steady state temperature of 300 °C. The experiment was expected to run for 72 h, but an explosion occurred at approximately two
hours into the experiment. This resulted in the rupture of the autoclave ‘A’, causing significant damage to the furnace and the fume
hood as shown in Fig. 2, luckily without any injuries to laboratory staff as they were out of the lab when the explosion occurred.
DMSO is commonly used as a solvent in many chemical reactions [5–7], however it is also known to be thermally unstable due to its
low auto-ignition temperature of 215 °C and undergoes an exothermic reaction when decomposed. This is especially true when
heated above its decomposition temperature in the presence of an ammonia-generating compound, which can lead to a violent
exothermic reaction and a subsequent explosion.
Numerical analysis of equipment subjected to blast loads requires a detailed understanding of the blast phenomena, the dynamic
response of the material and subsequently the failure mode of the material involved. The Conventional Weapons Effect Program
(CONWEP) blast loading model, was developed for military purposes, and is a collection of conventional weapons effects calculations
from empirical relationships and curves found in TM5-855-1 [8]. The input parameters needed to model the blast load using the
CONWEP model include a defined amount of Trinitrotoluene (TNT) charge at a given distance from the source of explosion, type of
blast, detonation location, and surface identification for which pressure is applied. The CONWEP model also takes into consideration
the reflected pressure, and the total blast pressure is calculated using

Fig. 2. (a) Damage caused by the ruptured autoclave ‘A’ (containing DMSO as a solvent) to autoclave ‘B’ (containing water as a solvent) and (b)
damage caused to the furnace and the fume hood.

355
I. Barsoum et al. Engineering Failure Analysis 91 (2018) 354–369

P (t ) = Pi (t )[1 + cos α − 2(cos α )2 ] + Pr (t ) (cos α )2 for cos α ≤ 0

P (t ) = Pi (t ) for cos α > 0 (1)

where P(t) denotes the total overpressure experienced on the loading surface at a given point, α represents the angle of incidence, and
Pi(t) and Pr(t) represent, respectively, the incident and the reflected pressures, felt by the surface. One advantage of using the
CONWEP model is that the fluid medium does not have to be modelled explicitly to account for reflected pressure and this leads to
significant savings in computational time [9]. A number of studies have successfully used the CONWEP model to simulate blast
loading on solid plates and sandwich structures at varying standoff distances [10–13]. Palanivelu et al. [14] studied the effect of
external open blast load on empty metal cans and observed a good correlation between the numerical and experimental data. Prueter
[9] studied its application in a confined environment by examining the structural damage associated with an internal detonation in a
heat exchanger and showed that the structural damage could be predicted with reasonable accuracy. However, simulating internal
blast load in closed vessels has many challenges, and Duffey et al. [15] provide an extensive overview of techniques necessary for
evaluating response of a containment vessel subjected to detonation-induced dynamic pressure loading. Dong et al. [16] also in-
vestigated the interaction between internal blast loading and the dynamic elastic response of the containment vessel.
Due to the exothermic nature of an explosion resulting from a chemical reaction and the loading rate involved, a temperature and
strain-rate dependent constitutive model needs to be used to adequately capture the material's response. A well-known material
model which describes the material response at large strains, various strain rates, and elevated temperatures is the Johnson-Cook (JC)
plasticity model [17]. Several studies have utilized the JC plasticity model to investigate the structural response of materials for
various applications such as metal cutting, welding, ballistic impact, etc., as these processes are associated with large strains, high
strain rate, and generally high temperature [18–22]. Umbrello et al. [18] performed experimental and numerical studies using five
sets of material parameters obtained from previous studies and investigated their influence on the plastic behavior of SA 316L
stainless steel. The authors compared the results obtained from simulation with the experimental results and concluded that the
forces, chip morphology and residual stresses in the machined components are very sensitive to the JC material parameters.
In order to model the failure in the autoclave, the Johnson-Cook failure model [23] is commonly used for modelling failure at high
strain rates and incorporates the effects of temperature and stress triaxiality in the failure locus (e.g. fracture strain). Johnson and
Cook [23] have shown that the influence of temperature and strain rate on the fracture locus is negligible in some materials, but is
however strongly dependent on stress triaxiality. This conclusion has also been confirmed in other studies [24–27].
Hence, in the current study, numerical simulation is performed to investigate the effect of internal blast load in an autoclave, with
the purpose to simulate the incident described above. The CONWEP model is utilized to model the blast load, while the Johnson-Cook
material and failure models are used to describe the material behavior under these severe loading conditions.

2. Material modelling

2.1. Johnson-Cook plasticity model

The JC plasticity model [17] expresses the flow stress of the material as a function of plastic strain, strain rate and temperature
and is given by

σ = [A + Bεpn ][1 + C ln(ε˙p/ ε˙ 0 )](1 − θ∗m) (2)

where A is the yield stress at the reference temperature and reference strain, B is the coefficient of strain hardening, n is the strain
hardening exponent, C is the strain rate sensitivity coefficient, m is the thermal softening exponent, εp is the equivalent plastic strain,
εṗ is the equivalent plastic strain rate and ε0̇ is a reference strain rate. The dimensionless temperature is given by θ∗ = (θ − θr)/
(θm − θr), where θ is current temperature, θr is the reference temperature and θm is the melting temperature of the material.
The JC parameters utilized in this work to define material property for simulation of blast load are obtained from [28] after
comparing parameters obtained from the preliminary experiment presented in Table 1 with various JC parameters outlined in [18].

2.2. Johnson-Cook failure model

The ductile failure model considered in this work is based on continuum damage mechanics, where damage accumulation is
calculated based on the form proposed by Johnson and Cook [23],

Table 1
JC plasticity and failure model parameters for SA 316L.
Material JC plasticity model JC failure model

A B n εf C m ε0̇ D1 D2 D3
[MPa] [MPa] [−] [−] [MPa] [−] [s−1] [−] [−] [−]

SA 316L 284 913 0.609 0.51 0.1 0.85 200 0.0121 0.4241 −0.6284

356
I. Barsoum et al. Engineering Failure Analysis 91 (2018) 354–369

ε fp dε p
ω= ∫0 ε fp (T ) (3)
where ω is a phenomenological damage state variable, dε is the increment of accumulated plastic strain, εf (T) is the plastic strain to
p p

failure envelope, which is dependent on the stress triaxiality parameter T, strain rate and temperature and is given by
ε fp (T ) = (D1 + D2 exp (D3 T ))(1 + D4 ln(εṗ / ε0̇ ) )(1 + D5 θ∗) (4)
where D1, D2 and D3 are material parameters to define the dependence of the material failure strain to stress triaxiality, D4 is a strain
rate parameter and D5 is a temperature parameter. Failure is said to have occurred when total equivalent plastic strain reaches the
equivalent strain to fracture, i.e. when ω = 1.
Johnson and Cook [23] have shown in their study that the influence of temperature and strain rate on the material failure strain is
usually negligible, but is however strongly dependent on the stress triaxiality. This conclusion has also been confirmed in other
studies [24–27]. Therefore, in the current work the failure strain is assumed to only be dependent on stress triaxiality T (e.g.
D4 = D5 = 0) and hence Eq. (4) reduces to
ε fp (T ) = (D1 + D2 exp (D3 T )) (5)

3. Experiments

The aim of the experimental program in this study is to establish the true stress-strain behavior of the autoclave material, and its
ductile failure locus defined as the plastic strain versus the stress triaxiality at failure. These will be used to obtain the JC plasticity
and failure model material parameters in Eq. (2) and (4), which are used as input for the finite element model to simulate the failure
of the autoclave when subjected to an internal blast. In this section, the procedures to establish the material's hardening behavior, as
well as the failure envelope, are outlined.

3.1. Stress-strain curve of SA 316 L

The failed autoclave is manufactured of stainless steel SA 316 L, which has an ultimate tensile strength 560 MPa. A uniaxial tensile
test to establish the true axial stress-strain behavior of the autoclave material is conducted. The tensile specimens were manufactured
by cutting out smooth round bar specimens of 3 mm diameter and 60 mm gauge length from the thickness of a new autoclave, with
the tensile axis in the axial direction of the autoclave. Uniaxial tensile test at room temperature was conducted using a MTS Alliance
RF/150 tensile testing machine. In Fig. 3, the true stress-strain curve obtained from the uniaxial test is shown and is curve-fitted to
the JC plasticity model given by Eq. (2). The JC plasticity model parameters A = 284 MPa, B = 913 MPa and n = 0.609 are presented
in Table 1, where εf= 0.51 is the uniaxial failure strain.

3.2. Failure locus of SA 316 L

Here the procedure to establish the ductile failure locus in terms of strain versus stress triaxiality at failure for SA 316 L is
outlined. The ductile failure process in steels is mainly directed by nucleation, growth and coalescence of micro-voids [25,29]. The
initiation of ductile failure is highly affected by the stress state of the material, which is generally characterized by the dimensionless
stress triaxiality parameter T. Studies [24,25] have shown that the stress triaxiality has the largest influence on failure strain and is

Fig. 3. True axial stress-strain curve for SA 316L.

357
I. Barsoum et al. Engineering Failure Analysis 91 (2018) 354–369

Fig. 4. (a) Dimensions of the notched ring specimen, (b) the complete autoclave before, (c) the notched ring specimens and (d) assigned target
points during ring test.

defined as
σh
T=
σe (6)

where σh is the mean stress and σe is the von Mises effective stress [29].
Ring specimens were manufactured by cutting 5 mm rings from a new autoclave (Fig. 4(b)) using EDM machining, which was
used due to its capability of providing high precision wire cuts at a tolerance limit of 0.001 mm. Four different notch radii, i.e. r = 2,
4, 7 and 12.5 mm, were then manufactured on the outer side of the rings and three ring specimens of each notch size. This replication
was done to ensure repeatability and reproducibility of test results. In total, twelve ring specimens were manufactured and tested.
Variation in the notch radius on the ring specimens will give rise to different stress triaxiality levels. Fig. 4(a)-(c) show the dimensions
of the notched ring specimens, the autoclave from which the ring specimens were manufactured and the actual notched ring spe-
cimens, with a nominal thickness and width t = 5 mm outside diameter OD = 48 mm. The size of the notch is quantified by the
dimensionless parameter α = 4 r/t, which with the notch radii considered in this study gives α = 1.6, 3.2, 5.6 and 10.
The ring test fixture used in this study, as shown in Fig. 5, is a modified setup of the one proposed in [30,31]. The modification to
the original design is an addition of two pairs of support plates and a reduction in the geometry of the D-blocks to permit the
accommodation of the current specimen configuration. The existing test design consists of an upper and lower fixture, a pair of D-
blocks, and pins. The upper and lower fixtures are mounted to a conventional MTS machine. The notched ring tests were carried out
using an MTS Alliance RF/150 testing machine controlled via TestWorks software, as shown in Fig. 6, at quasi-static loading con-
ditions with a displacement rate of 0.1 mm/s. Instantaneous force data were recorded via TestWorks and synchronized with strain
data measured via a Tinius-Olsen video extensometer, which is based on Digital Image Correlation technique (DIC). The video

358
I. Barsoum et al. Engineering Failure Analysis 91 (2018) 354–369

Fig. 5. CAD assembly of the notched ring test apparatus with components.

Fig. 6. Experimental setup for the notched ring test.

extensometer system consists of a high resolution camera, as shown in Fig. 6, and a DIC post-processing software. Prior to testing, the
notched ring specimens are smoothened with sandpaper to remove manufacturing burrs and sharp edges. The internal surface of the
notched ring specimens and the D-blocks are lubricated with Amerglide PTFE lubricant to reduce friction. The outer surface of the
gauge region is sprayed with a matt white color and a matt black color to create an irregular speckled pattern, as shown in Fig. 4(d).
Target points are assigned, which are used to track the movement of the speckles and correlate back to the original non-deformed
pattern in order to obtain the strain. Here, six target points are assigned and engineering strain is measured between target point 1
and 4 (e14), between 2 and 5 (e25) and between 3 and 6 (e36). Fig. 7 shows a typical measurement of strains (e14, e25, e36) between the

359
I. Barsoum et al. Engineering Failure Analysis 91 (2018) 354–369

Fig. 7. Comparison between strains obtained experimentally for notch ring test.

assigned target points during a notch ring test. As can be seen, there is an insignificant difference between the strain components
measured, indicating that the tangential strain across the notch region is uniform and that no misalignment of the notched ring
specimen is present.
Finite element analysis (FEA) of the ring tests were carried out in parallel with the experimental tests using Abaqus/Standard
v6.13 to obtain the equivalent plastic strain at failure (εfp) and the average stress triaxiality at failure (Tf) in the center of the notch.
The exact dimensions of each notched ring specimen after manufacturing were determined and all tests were analyzed with aid of
FEA. A 3D model composed of 3D 20-node quadratic brick reduced integration elements (C3D20R) with mesh density of approxi-
mately 13,000 to 18,000 elements was used in the half symmetry model, depending on the notch configuration of the ring specimen
analyzed. A fine mesh with an element size of 0.6 mm was selected for the ring specimen, whereas an element size of 1 mm was
selected for the D-block. The geometry of the specimen exhibit symmetry along the X-Y and Y-Z plane as shown in Fig. 8(a), therefore
only one-fourth of it is modelled. The notched ring specimen (SA 316 L stainless steel) is modelled as elastic-plastic isotropic
hardening, obeying Eq. (2) and accounting for large deformations. The D-blocks (AISI 1080 carbon steel) are modelled as linear-
elastic since they are expected to only deform in the elastic regime. Young's modulus and Poisson's ratio 207 GPa and 0.3, respec-
tively, was used to define the elastic properties for both materials. A surface-to surface contact was specified between the D-blocks
and the ring specimen. The outer surfaces of the D-blocks and the inner surfaces of the specimen were assigned master and slave
contact interactions, respectively, and an isotropic tangential friction behavior with coefficient of friction ranging from
0.20 ≤ μ ≤ 0.35. Two reference nodes, each located at the mid-point of each of the D-blocks, were used to create a kinematic
constraint, which couples the inner surface of the pin-hole and thus mimics the motion of the pin in the D-block as observed during
the experiment. A fixed boundary condition was applied to the lower reference node, thereby constraining the node from transla-
tional and rotational motion, whereas the upper reference node was given a displacement boundary condition in the Y-direction, as
shown in Fig. 8(b).

Fig. 8. A ¼ FEA model showing (a) meshed ring specimen and D-block and (b) assembly showing applied interactions, constraints, loading and
boundary conditions.

360
I. Barsoum et al. Engineering Failure Analysis 91 (2018) 354–369

Fig. 9. Comparison between experimental and FEA of force versus strain response (F vs. e25) for the notched ring test with (a) α = 1.6, (b) α = 5.6
and (c) α = 10.

361
 I. Barsoum et al.

362
Fig. 10. Contour plot of equivalent plastic strain at notch region for notched ring specimen with α = 10.
Engineering Failure Analysis 91 (2018) 354–369
I. Barsoum et al. Engineering Failure Analysis 91 (2018) 354–369

Fig. 11. (a) Load F vs. deflection Δ and (b) stress triaxiality T versus normalized distance ξ along the center of the notch at various stages of loading
for ring specimen with α = 10.

In Fig. 9 the force F versus engineering strain e25 obtained experimentally and through FEA are shown for various notch para-
meters (α = 1.6, 5.6 and 10). A markedly good agreement between the experimental and FEA results is shown. It was observed both
in the FEA and the experiments that, beyond the maximum value of the axial force, the deformation becomes localized in the notch
region and the notch starts to increase in depth. As the deformation progresses a sudden drop in the force-strain responses in the ring
tests was noticed. In this study, the failure point is defined at a 5% drop below the maximum force value. At the failure instance, the
equivalent plastic strain and the stress triaxiality at the critical locations are obtained from the FEA results. A common evaluation
procedure to determine ductility in metals [24,25,32] is to consider the equivalent plastic strain at the center of the notch as the
strain to failure initiation, here denoted by εfp. The onset of ductile failure is most likely to occur at the location with the highest stress
triaxiality. As such, Fig. 10 shows the contour plot of equivalent plastic strain in the notch region of the notched ring specimen,
indicating localized plastic deformation and necking in the notch beyond the maximum force.
In Fig. 11(a), FEA results pertaining to force versus D-block displacement (F vs. Δ) are plotted and in Fig. 11(b), the stress
triaxiality versus a normalized path through the notch (ξ-axis in Fig. 10) at various stages of loading is plotted. Both the equivalent
plastic strain and the stress triaxiality are maximum at the center of the notch, as can be observed from Figs. 10 and 11, respectively.
Therefore, the subsequent local stress triaxiality and equivalent plastic strain values used to evaluate the specimen to determine the
failure locus of the material were obtained from this critical location.
Fig. 12(a) shows the evolution of the stress triaxiality T against equivalent plastic strain εp at the notch center for notch sizes
α = 1.6, 3.2, 5.6 and 10. As anticipated, triaxiality decreases with increasing α values. However, the equivalent strain to fracture
increases with increase in α value. Fig. 12(b) shows the evolution of equivalent plastic strain εp at the center of notch against
engineering tangential strain e25, indicating that εp increases with increasing values of α.
Similarly, with the experimentally-obtained true failure strain, the values of stress triaxiality at failure and the equivalent plastic
strain at failure at the center of the notch can be obtained from the FEA results to construct the ductile failure envelope for the
material (SA 316L) under consideration. However, the stress state at the notch center up to the failure point, as it is common practice
in the literature, can be better represented using the history-average value of stress triaxiality (Tf) [33]. It is considered to be a more
accurate representation than the instantaneous stress triaxiality at failure, since ductile damage is an accumulative damage process
that is driven by the stress triaxiality level present in the material over the loading history. By integrating the T vs. εp plots in
Fig. 12(a) upto point of failure (εfp), the average value of the stress triaxiality Tf can be given by

Fig. 12. (a) Evolution of stress triaxiality and (b) of equivalent plastic strain at the center of the notch.

363
I. Barsoum et al. Engineering Failure Analysis 91 (2018) 354–369

Fig. 13. Ductile failure locus for SA 316L.

1 ε fp
Tf =
ε fp
∫0 Tdε p
(10)

In Fig. 13, the resulting ductile failure locus εf vs. Tf for SA 316 L is given, where markers pertain to experimental data points and
p

red line is a least squares fit of the data points to the JC failure criterion given by Eq. (5) earlier. The least square fit yields JC failure
criterion stress triaxiality parameters D1 = 0.0121, D2 = 0.4241 and D3 = −0.6284, which are also listed in Table 1.
Now, with the stress-strain curve in Fig. 3 and the failure locus in Fig. 13, the failure of the autoclave can be studied by im-
plementing the material into a FEA model to simulate the autoclave subjected to an internal blast load.

4. Modelling the failure incident

In this section, the finite element analysis of the autoclave subjected to an internal blast load is presented. The finite element
model consists of two autoclaves placed close to each other, with the autoclave containing DMSO (autoclave ‘A’) being subjected to
an internal blast load while the other autoclave containing water as solvent is expected to receive an impact (autoclave ‘B’) from the
ruptured autoclave ‘A’ subjected to blast load. The two autoclaves (‘A’ and ‘B’) of the FEA model, which are placed 36 mm apart to
resemble realistic conditions, are shown in Fig. 14. The objective is to simulate the failure incident shown in Fig. 2. The results from
the FEA of the damages observed on the impacted autoclave and the profile of the ruptured autoclave are then compared with the

Fig. 14. The 3D geometry of the autoclaves (‘A’ and ‘B’) modelled.

364
I. Barsoum et al. Engineering Failure Analysis 91 (2018) 354–369

actual laboratory failure. Simulations of the autoclave subjected to blast load were done using Abaqus/Explicit v6.13. A full-scale 3D
model of the autoclaves shown in Fig. 14 was constructed, where autoclave ‘A’ is subjected to a blast load using the CONWEP blast
load model and autoclave ‘B’ is idle.
Elastic-plastic material properties and ductile failure criterion to predict the progressive damage in the material were assigned to
each part of the model in Fig. 14. The physical, thermal and mechanical properties for SA 316L are assigned to all the parts in the
model, e.g. Young's modulus of E = 210 GPa, Poisson's ratio of ν = 0.3, density of 7800 kg/m3 and a specific heat at constant volume
of 452 J/kg K. The plastic behavior of the material was defined based on the JC plasticity model with isotropic hardening given by Eq.
(2) with parameters presented in Table 1, which accounts for the effects of strain rate and temperature on the plastic behavior. The JC
plasticity model parameters A, B and n in Table 1 are obtained experimentally as outlined earlier, whereas the parameters C, m and ε0̇
pertaining to the strain-rate and temperature dependence of the plastic behavior, respectively, are obtained from the work by
Changeux et al. [28] and Umbrello et al. [18].
The cylindrical vessel of autoclave ‘A’, where the failure is anticipated to occur, was assigned a section with a failure criterion
corresponding to the JC dynamic failure model given by Eq. (5) with parameters D1, D2 and D3 given in Table 1. The remaining parts of the
autoclave ‘A’ and ‘B’ are modelled as elastic-plastic materials without any damage criterion, this in order to simulate the actual failure
scenario observed in Fig. 2. The JC dynamic failure model is based on the value of the equivalent plastic strain at element integration
points and failure is assumed to occur when the damage parameter exceeds unity. The damage parameter ω is defined in Eq. (3), where εfp
is the failure envelope for SA 316 L given by Eq. (5) and Fig. 13, defining the critical plastic equivalent strain as a function of the stress
triaxiality at failure initiation. Damage accumulation is then calculated in each element and at each load increment by evaluating the
integral in Eq. (3), which increases monotonically with increase in plastic deformation. Material failure is said to have occurred in the
element when the criterion is met, i.e. ω = 1, and the element is consequently deleted from the analysis.
A general contact interaction is defined between the inner surface of the sleeve and the outer surface of the split-ring, the bottom
of the top head and the top of the 125 ml cylindrical vessel, the inner surface of the split-ring closure and the outer surface of the top
head, and the inner surface of the split-ring closure and the outer surface of the cylinder. A tangential contact interaction property
with frictionless conditions is assumed. The internal blast load in autoclave ‘A’ is applied using the CONWEP model by defining an
incident wave interaction with an air blast definition. The detonation charge was initially defined arbitrarily as an equivalent mass of
trinitrotoluene (TNT) of 0.0001 kg. The source point for the detonation was defined at the center of the cylindrical vessel of autoclave
‘A’. A detonation time of 0 s was selected assuming an instantaneous detonation, while the TNT load was scaled gradually until a
magnitude scale factor of 420 was reached. The bolts were not modelled explicitly, but bolt pre-loads on the split-rings where applied
through reference points defined at the center of each bolt hole, which were coupled to the internal surface of the bolt holes. A
constant ambient temperature of 300 °C throughout the entire region of the autoclaves was specified, which corresponds to the
temperature during the laboratory experiment. All of the parts of the autoclave were meshed with 8-node linear brick, reduced
integration and hour glass control 3D stress elements C3D8R. The accuracy of blast simulation results are rather dependent on the
quality and size of the generated mesh. As such, a mesh-density sensitivity analysis was done by carrying out mesh refinement in
areas where the blast load and impact are expected to affect. A converged solution was arrived upon by using a mesh size of
0.001 mm, which corresponds to a total of 299,337 elements.
In addition to the internal blast load analysis, a separate simulation was conducted by applying a uniform static pressure load to
the internal surface of autoclave ‘A’ corresponding to a pressure level of about six times the maximum allowable working pressure
(MAWP = 20 MPa). The static pressure load case was considered in this study in order to assess whether the autoclave failure could
have been caused by a static pressure loading, e.g. over pressurized, rather than an internal blast.

5. Results

In this section, the results obtained from the finite element analysis of the autoclave subjected to blast load are presented. The FE
results are validated by comparing various geometric quantities with measurements made on the actual ruptured laboratory auto-
clave ‘A’ containing DMSO and the impacted autoclave ‘B’ adjacent to it containing water, both shown in Fig. 14.
The numerical simulation was approached by first investigating the location of the detonation source point of the blast load that
could result in a rupture profile similar to the one experienced on the actual laboratory-ruptured autoclave. After several iterations,
an approximate position was obtained and from this position an arbitrary initial amount of 0.0001 kg of TNT load was applied and
gradually increased through a scale factor to determine the minimum equivalent TNT load that could rupture the autoclave. Once the
minimum load needed to rupture the autoclave was determined, the profile of the ruptured autoclave obtained from the numerical
simulation was compared to the laboratory-ruptured autoclave by comparing geometric measurements (L1-L4, ϕ), as defined in
Fig. 15. The FEA results revealed that an equivalent TNT load of 0.042 ± 0.002 kg can produce a ruptured profile similar in size and
shape to the laboratory-ruptured autoclave, as depicted in Fig. 15.
In addition, at the failure incident the ruptured autoclave ‘A’ impact autoclave ‘B’ adjacent to it creating a noticeable dent, as
shown in Fig. 16, which is measured to further validate the blast load. The length, width and depth of the dent are measured on the
actual autoclave and compared with the FEA results pertaining to an equivalent TNT load of 0.042 kg. The findings are summarized in
Table 2, which show that the FEA results are in rather good agreement with the actual measurements for the equivalent TNT load
considered.
Fig. 17 shows the contour plots of the FEA results, depicting the frames of the simulated failure incident. Fig. 17(e) shows how the
ruptured autoclave A impacts autoclave B, leaving a visible dent on it. The increment numbers refer to the blast pressure profile in
Fig. 18.

365
I. Barsoum et al. Engineering Failure Analysis 91 (2018) 354–369

Fig. 15. Impacted autoclave ‘B’: (a) Actual and (b) FEA.

Fig. 16. Ruptured autoclave ‘A’: (a) Actual and (b) FEA.

Table 2
Measurements on blast loaded autoclave ‘A’ and impacted autoclave ‘B’, actual and FEA results (with equivalent TNT load of 0.042 kg).
Autoclave ‘A’ (internal blast loaded, containing DMSO) Autoclave ‘B’ (impacted, containing water)

Rupture Profile Actual Fig. 15(a) FEA Fig. 15(b) Dent Profile Actual Fig. 16(a) FEA Fig. 16(b)
L1 (mm) 82 72 Length (mm) 46 40
L2 (mm) 72 68 Width (mm) 14 22
L3 (mm) 86 85 Depth (mm) 3.6 3.1
L4 (mm) 94 91
ϕ (Deg) 9.3 8.6

366
I. Barsoum et al. Engineering Failure Analysis 91 (2018) 354–369

Fig. 17. Sequence of contour plots of both autoclaves throughout the simulation.

To further validate the FEA results, the extent of the change in thickness across the autoclave as a result of the pressure load was
also analyzed. The thickness across sections of the shape profile presented in Fig. 15 obtained from FEA was also compared to the
laboratory-ruptured autoclave. The smallest thickness measurement on the actual ruptured autoclave was 4.14 mm, whereas a
thickness of 4.10 mm was obtained from the FEA results, which compares remarkably well.
The pressure distribution over the surface area resulting from the applied CONWEP blast load is presented in Fig. 18. The shape is
typical of that observed in the literature for a chemical explosion as earlier shown in Fig. 1. The arrival time of the shock wave to the

367
I. Barsoum et al. Engineering Failure Analysis 91 (2018) 354–369

Fig. 18. Pressure-time curve obtained from numerical simulation.

surface of the autoclave is approximately 0.02 ms, which includes the time it takes the detonation wave to propagate through the
explosive charge. A peak pressure of approximately 2000 MPa was attained and this decays exponentially to the ambient pressure
with time. The integral of the pressure curve over the positive phase duration is known as the incident pressure impulse, which is
15.8 MPa-ms for the case presented.
Simulation to investigate possible rupture of the autoclave by static overpressure, rather than a blast load, resulting from a
gradual increase in pressure load within the autoclave was also performed. The result of such numerical simulation showed that the
autoclave fails at a static pressure load of 118 MPa, which is about six times the MAWP (20 MPa). However, the failure mode does not
resemble that observed on the laboratory-ruptured autoclave.
Hence, it is concluded that static overpressure could not have been the reason for the observed failure. The resulting failure must
have been caused by a blast load due to the exothermic decomposition of the DMSO solvent used in autoclave ‘A’. The hazards of
using DMSO as solvent in thermal reactions have been pointed out in the past [5,6]. This is further endorsed by the observation that
the dented vessel, which contained water as a solvent, did not burst and along with the good agreement between the FEA blast results
with the measurements on the ruptured and dented autoclaves.

6. Conclusions

The current study employed the use of finite element analysis to simulate the failure of an autoclave subjected to an internal blast
load. A 3D model of the autoclaves was developed and subjected to a blast load using the CONWEP blast load model. The CONWEP
model essentially calculates the blast wave parameters for a given explosive by using an equivalent TNT method. The results of the
numerical simulation were then validated by comparing with the measurements on the ruptured and dented autoclaves observed on
the actual case, which compared markedly well. In order to simulate the failure of the autoclave, a ductile failure locus of the
autoclave material SA 316L was established experimentally. The JC plasticity and dynamic failure models were used to represent the
material, which were derived from the experiments conducted.
It is found, based on the FEA results, that the observed failure is a result of an internal blast load, which arose from the de-
composition of DMSO at high temperature and is catalyzed by the presence of other compounds such as ammonia [5,6]. The result of
the analysis presented in this study gives a probable blast pressure based on TNT equivalency and the CONWEP model. In addition,
the analysis has been able to provide useful insight as to the failure of the autoclave and can be utilized to simulate similar failure
incidents involving internal blast loads.

Acknowledgements

The authors would like to acknowledge Abu Dhabi National Oil Company (ADNOC) for funding part of this work under the grant
RIFP- 15328-2015. The authors are also in debt to Dr. S. Alhassan at Khalifa University of Science and Technology for fruitful
discussion and to the reviewers for valuable feedback on the manuscript.

References

[1] ASME, "Alternative Rules," in ASME-BPVC, Section VIII-Div, 2, ed, (2011).

[2] Parr Instrument Company, Available https://www.parrinst.com.
[3] M. Goel, Blast: Characteristics, Springer, Loading and Computation—An Overview," in Advances in Structural Engineering, ed, 2015, pp. 417–434.

368
I. Barsoum et al. Engineering Failure Analysis 91 (2018) 354–369

[4] P.D. Smith, J.G. Hetherington, Blast and ballistic loading of structures, Digital Press, 1994.
[5] T.E. Larson, R. Zou, Chemical Safety: Dimethyl Sulfoxide Overpressurization Hazard, vol. 88, American Chemical Soc. 1155 16th St, NW, Washington, DC 20036
USA, 2010, pp. 4–6.
[6] T. Lam, T. Vickery, L. Tuma, Thermal hazards and safe scale-up of reactions containing dimethyl sulfoxide, J. Therm. Anal. Calorim. 85 (2006) 25–30.
[7] J. Hall, Loss Prev. Bull., vol. 114, December 1993.
[8] D. Hyde, Fundamentals of protective design for conventional weapons, CONWEP (Conventional Weapons Effects), TM 5-855, 1 1992.
[9] P.E. Prueter, Using explicit finite element analysis to simulate the structural damage associated with an internal detonation in a heat exchanger, ASME 2014
Pressure Vessels and Piping Conference, 2014 V002T02A014-V002T02A014.
[10] S.K. Lahiri, L. Ho, Simulation of rapid structural failure due to blast loads from conventional weapons (CONWEP), Proceedings of the NAFEMS World Congress,
2011.
[11] C. Dey, S. Nimje, Experimental and numerical study on response of sandwich plate subjected to blast load, Exp. Tech. (2014) 1–11, http://dx.doi.org/10.1111/
ext.12075.
[12] B. Cabello, Dynamic Stress Analysis of the Effect of an Air Blast Wave on a Stainless Steel Plate, Rensselaer Polytechnic Institute (2011).
[13] T.F. Henchie, S.C.K. Yuen, G. Nurick, N. Ranwaha, V. Balden, The response of circular plates to repeated uniform blast loads: an experimental and numerical
study, Int. J. Impact Eng. 74 (2014) 36–45.
[14] S. Palanivelu, W. Van Paepegem, J. Degrieck, K. De Wolf, J. Vantomme, D. Kakogiannis, et al., Study of blast load on recyclable empty metal cans, DYMAT-
International Conference on the Mechanical and Physical Behaviour of Materials under Dynamic Loading, 2009, pp. 709–715.
[15] T.A. Duffey, E.A. Rodriguez, C. Romero, Detonation-induced dynamic pressure loading in containment vessels, Pressure Vessel Research Council of the Welding
Research Council, vol. 477, WRC Bulletin, 2002.
[16] Q. Dong, Q. Li, J. Zheng, Interactive mechanisms between the internal blast loading and the dynamic elastic response of spherical containment vessels,
International Journal of Impact Engineering 37 (2010) 349–358.
[17] G.R. Johnson, W.H. Cook, A constitutive model and data for metals subjected to large strains, high strain rates and high temperatures, Proceedings of the 7th
International Symposium on Ballistics, 1983, pp. 541–547.
[18] D. Umbrello, R. M'saoubi, J. Outeiro, The influence of Johnson–Cook material constants on finite element simulation of machining of AISI 316L steel, Int. J.
Mach. Tools Manuf. 47 (2007) 462–470.
[19] C. Bonnet, F. Valiorgue, J. Rech, J. Bergheau, P. Gilles, C. Claudin, Development of a friction modelling method in dry cutting of AISI 316L austenitic stainless
steels, Int. J. Mater. Form. 1 (2008) 1211–1214.
[20] A. Svoboda, D. Wedberg, L.-E. Lindgren, Simulation of metal cutting using a physically based plasticity model, Model. Simul. Mater. Sci. Eng. 18 (2010) 075005.
[21] D. Zhou, W. Stronge, Ballistic limit for oblique impact of thin sandwich panels and spaced plates, International Journal of Impact Engineering 35 (2008)
1339–1354.
[22] E. Flores-Johnson, O. Muránsky, C. Hamelin, P. Bendeich, L. Edwards, Numerical analysis of the effect of weld-induced residual stress and plastic damage on the
ballistic performance of welded steel plate, Comput. Mater. Sci. 58 (2012) 131–139.
[23] G.R. Johnson, W.H. Cook, Fracture characteristics of three metals subjected to various strains, strain rates, temperatures and pressures, Eng. Fract. Mech. 21
(1985) 31–48.
[24] Y. Bao, T. Wierzbicki, On fracture locus in the equivalent strain and stress triaxiality space, Int. J. Mech. Sci. 46 (2004) 81–98.
[25] I. Barsoum, J. Faleskog, Rupture mechanisms in combined tension and shear—experiments, Int. J. Solids Struct. 44 (2007) 1768–1786.
[26] B. Erice, F. Gálvez, D. Cendón, V. Sánchez-Gálvez, Flow and fracture behaviour of FV535 steel at different triaxialities, strain rates and temperatures, Eng. Fract.
Mech. 79 (2012) 1–17.
[27] J. Trajkovski, R. Kunc, V. Pepel, I. Prebil, Flow and fracture behavior of high-strength armor steel PROTAC 500, Mater. Des. 66 (2015) 37–45.
[28] B. Changeux, M. Touratier, J.-L. Lebrun, T. Thomas, J. Clisson, High-speed shear tests for the identification of the Johnson-Cook law, 4 th International
ESAFORM Conference on Material Forming, 2001, pp. 603–606.
[29] J. Hancock, A. Mackenzie, On the mechanisms of ductile failure in high-strength steels subjected to multi-axial stress-states, J. Mech. Phys. Solids 24 (1976)
147–160.
[30] M.A. Al-Khaled, I. Barsoum, New ring specimen geometries for determining the failure locus of tubulars, J. Press. Vessel. Technol. 140 (2018) 011405.
[31] I. Barsoum, K.F. Al Ali, A procedure to determine the tangential true stress-strain behavior ofpipes, Int. J. Press. Vessel. Pip. 128 (2015) 59–68.
[32] Y. Bao, T. Wierzbicki, A comparative study on various ductile crack formation criteria, J. Eng. Mater. Technol. 126 (2004) 314–324.
[33] I. Barsoum, J. Faleskog, S. Pingle, The effect of stress state on ductility in the moderate stress triaxiality regime of medium and high strength steels, Int. J. Mech.
Sci. 65 (2012) 203–212.

369