Príloha A -...
87
82 Príloha A
Transcript of Príloha A -...
82
Príloha A
83
Predslov k prílohe A
Prílohou A dizertačnej práce je článok v zahraničnom karentovanom časopise „Process Safety
and Environmental Protection“ s názvom The role of a commercial process simulator in
computer aided HAZOP approach, ktorý bol publikovaný v roku 2017 v čísle 107. V článku
je v úvode spísaná podrobná literárna rešerš súčasného stavu automatizácie hodnotenia
nebezpečenstva a v ďalších častiach článku sú predstavené základná štruktúra softvéru, princípy
analýzy simulačných dát a aplikácia softvéru na dve prípadové štúdie.
Z dôvodu transformácie A4 formátu do B5 formátu je prílohou prepis článku. Pre plnú verziu
článku v publikovanej verzii a s formátovaním daného časopisu je čitateľovi odporučené stiahnuť
článok online – http://dx.doi.org/10.1016/j.psep.2017.01.018.
84
The role of a commercial process simulator in
computer aided HAZOP approach
Ján Janošovský, Matej Danko, Juraj Labovský, Ľudovít Jelemenský*
Institute of Chemical and Environmental Engineering, Slovak University of Technology,
Bratislava, Slovakia
Abstract
Process safety is one of the key pillars of sustainable industrial development. In combination
with the increasing use of computer aided process engineering, the demand for an appropriate
model-based safety analysis tool capable to identify all hazardous situations leading to a major
accident has increased. Commercial process simulators are equipped with extensive property
databases and they employ high accuracy mathematical models providing the capability to
simulate real behavior of a process operated within the area of the mathematical model validity.
The main focus of this work is to improve standard hazard identification methods by the
combination of hazard and operability (HAZOP) study and process simulation in commercial
process simulator Aspen HYSYS. Software tool consisting of modules for computer simulation
and complex analysis of simulation data will be proposed. The developed tool was applied to
modern chemical productions exhibiting strong nonlinear behavior, where proper prediction of
consequences can be very difficult. In the first case study, hazard identification in continuous
glycerol nitration employing user-dependent analysis is presented. Mathematical methods of
simulation data analysis independent of the user is demonstrated in the second case study of
ammonia synthesis. Possibilities and limitations of the proposed tool are revealed and discussed
in this work.
Keywords: computer aided hazard identification; HAZOP study; Aspen HYSYS
1. Introduction
Several serious industrial accidents (e.g. Flixborough, Seveso, Bhopal and Tianjin disasters)
in the past have underlined the importance of loss prevention approach in chemical industry.
Dynamic development of industry has not only resulted in more efficient and profitable chemical
productions, but also in the increase of plants complexity as well as the variety of chemicals and
processes used in the plant. In addition, the majority of modern processes exhibit strong nonlinear
behavior. Therefore, the task of identifying potential sources of hazards has become more
complex (De Rademaeker et al., 2014). In combination with the ever growing use of computer
85
aided process engineering, the demand for an appropriately detailed safety analysis tool capable
to identify all hazardous situations leading to a major accident has increased. The safety point of
view should be implemented not only in the design stage of any chemical plant, but also during
the entire plant life cycle. Identification of all possible fault propagation paths is thus, for
example, the key feature of proper design of control systems (Leveson and Stephanopoulos, 2014;
Parmar and Lees, 1987; Seider et al., 2014).
Actual trend in computer aided loss prevention is to improve standard hazard identification
methods by employing mathematical modeling and process simulation in commercially available
simulators. Commercial process simulators are equipped with extensive property databases and
utilize high accuracy mathematical models thus providing the capability to simulate real behavior
of a process operated within the area of the mathematical model validity. Model-based hazard
identification also benefits from the fact that mathematical modeling of the analyzed process is
usually employed as a part of process design and optimization activities, e.g. optimization of
biorefinery downstream processes employing SimSci PRO/II (Corbetta et al., 2016), design of
hydrocarbons separation unit using Aspen Plus (de Riva et al., 2016) and Aspen HYSYS
supported design of syngas production proposed by Sunny et al. (2016). If the use of process
simulators is well implemented in the company policy, successful adoption of safety extensions
for these simulators is more likely.
Model-based approach was applied not only in hazard identification, but also in reliability
engineering (Favarò and Saleh, 2016) and quantitative risk assessment (Labovský and
Jelemenský, 2011; Qi et al., 2014). While mathematical modeling in these areas is accepted by
the safety engineering community, model-based hazard identification is still subject of discussion
because of model validity and its input parameters uncertainty (Labovská et al., 2014; Švandová
et al., 2009). Published model-based tools vary in the complexity of mathematical models and
simulation data evaluation methods. The complexity of mathematical models depends on whether
they were constructed specifically for the analyzed system or the developed tool employed a
group of mathematical models, e.g. a commercial process simulator. Although the computing
time increases with the increase of model complexity, several efforts were made towards
shortening the time required for the solution of large nonlinear systems, e.g. utilization of parallel
computing (Danko et al., 2015; Labovský et al., 2015). The simulation data can be evaluated
manually, automatically or by a combination of both ways. The majority of published works
benefited from the robustness and complexity of the hazard and operability (HAZOP) study that
belongs to the most used process hazard analysis procedures worldwide and is recognized as an
86
effective and accurate hazard identification method in chemical industry (Dunjó et al., 2010;
Kletz, 2001).
Eizenberg et al. (2006) combined a standard HAZOP study and process simulation in
MATLAB in order to develop a software tool for better understanding of hazards for the safety
education process. Similar approach was adopted in the work of Li et al. (2010). The examined
system was a three-phase hydrogenation in an intensified stirred continuous reactor and the
simulation results of hazardous cases generated based on the HAZOP principles were presented.
HAZOP principles were also applied in the safety assessment based on parametric sensitivity
analysis of the key operating parameters in a hydrogen production unit (Ghasemzadeh et al.,
2013).
Previously mentioned works focused on safety analysis based on a specific mathematical
model of the unit under review. A disadvantage of such an approach is its limited application. If
the safety analysis of another unit was required, it was necessary to decompose the current
mathematical model and to form and validate a new set of equations describing the behavior of
the new unit. Therefore, this approach is not suitable for the development of a universal model-
based hazard identification tool. This limitation can be eliminated by involving the use of a
commercial process simulation software with predefined and prevalidated sets of unit operations
commonly used in industry. In this case, the safety analysis of different units in a plant requires
only switching between the generally prepared mathematical models.
A successful combination of the HAZOP study and simulation in Aspen Plus in the case study
of biodiesel production was presented by Jeerawongsuntorn et al. (2011). Alternatives including
standard and reactive distillation were analyzed for the purpose of the decision-making process
improvement and safety instrumented system implementation. The K-Spice software was used
for process simulation followed by the HAZOP analysis in the work of Enemark-Rasmussen et
al. (2012). Results of the simulated deviations were recorded, evaluated and ranked according to
the severity of deviations determined by the sensitivity measure. Tian et al. (2015) introduced the
dynamic simulation-based HAZOP (DynSim-HAZOP) methodology employing dynamic
simulation in process simulators such as Aspen Dynamics, Aspen Plus and Aspen HYSYS to
perform model-based safety analysis of an extractive distillation column and an ammonia
synthesis plant. Both Jeerawongsuntorn et al. (2011) and Tian et al. (2015) used monitoring of
user defined threshold values (e.g. auto-ignition temperature or maximum allowed liquid level in
the separator) in the simulation data evaluation. Enemark-Rasmussen et al. (2012) partially
automated the process of data evaluation by quantifying the deviation effects and their ranking
according to the sensitivity measure, i.e. comparing the change of the selected parameter
87
(temperature, pressure …) to the change of the deviated parameter. The proposed ranking system
allowed the elimination of deviations with negligible impact on the process. Systematic approach
combining advantages of previously mentioned works applied in process simulation in Aspen
HYSYS was proposed by Janošovský et al. (2016a) and it was further analyzed (Janošovský et
al., 2016b).
Principal objective of this paper is to summarize issues with the developing computer aided
hazard identification tool based on process simulation in the Aspen HYSYS environment. Two
case studies focused on modern continuous productions exhibiting strong nonlinear behavior with
various levels of complexity are presented. In the first case study, hazards of glycerol nitration in
a continuously stirred tank reactor are identified and evaluated. The presence of the multiple
steady states phenomenon in an ammonia synthesis reactor with a preheating system and the
related numerical complications are discussed in the second case study.
2. Model-based HAZOP tool
Aspen HYSYS v8.4 simulation environment was selected as the commercial simulation tool.
Aspen HYSYS is a powerful engineering software tool for steady state and dynamic modeling
designed for continuous processes consisting of multiple process units especially in gas and oil
industry (AspenTech, 2015). All information necessary for the description of physico-chemical
properties of individual components and their mixtures are contained in fluid packages stored in
the Aspen HYSYS library. Its extensive size allows more effective search for the solution of
complex mathematical models by providing an accurate estimation of all necessary model
parameters and an appropriate numerical solver. Components not available in the Aspen HYSYS
library can be specified as “Hypo Components” by entering the required properties (density,
boiling point etc.) in the program. Although various studies found negligible differences between
the process simulations in Aspen Plus and Aspen HYSYS (Øi, 2012; Smejkal and Šoóš, 2002),
the use of Aspen HYSYS in model-based hazard identification is scarce.
Deviations observed by a conventional HAZOP study are generated by a simple logic
combination of guide words (more, less, none etc.) with process parameters (temperature,
pressure, flow etc.). Such information is insufficient for mathematical modeling. The model-
based HAZOP study requires not only the existence of the deviation but also its value and, in case
of dynamic simulation, also its duration. Therefore, determination of the deviation range is added
to the process of standard HAZOP deviation generation. In the final deviation list, deviations
characterized as “higher temperature” or “lower flow” are not satisfactory; but instead, deviations
like “temperature higher by 10%” or “flow lower by 10%” (in case of dynamic simulation, “flow
88
lower by 10% lasting for one hour”) are to be given. This fundamental demand for deviation
quantification in the model-based HAZOP study provides a promising way of at least partial
elimination of the disadvantages of the conventional HAZOP study such as the possibility of
overlooking hazardous events especially when they have never been observed before.
The amount of data necessary to be handled during a hazard identification process
geometrically increases when the dimension of size is assigned to conventional HAZOP
deviation. This task requires a robust software solution with well-arranged data structure to enable
the expert HAZOP team to smoothly process and visualize complex relations between various
process parameters. The increase in the amount of data being processed and the complexity of
relations is even more significant when time dimension of a deviation is considered. As
summarized in Introduction, although dynamic process simulation provides more detailed insight
into potential hazard and operability problems resulting from process nonlinearity, the majority
of possible industrial accidents can be, at least partially, revealed employing just the steady state
simulation approach. For the simplicity and explanatory purpose, only possibilities of steady state
simulation in the Aspen HYSYS environment are further discussed in this paper.
Methodology of the proposed software tool consists of a module utilizing Aspen HYSYS and
a module for the application of HAZOP principles (Fig. 1). A unique software dictionary and
infrastructure have been developed to ensure a reliable connection between the simulator and the
HAZOP module. First, the access to Aspen HYSYS and the availability of an active simulation
case and its flowsheet are tested. After the connection is established, information about individual
units and streams is transferred and stored in the internal database. In the next step, parameters of
individual streams available for the HAZOP study are highlighted for the user. The user is then
able to select the desired parameters and to create their deviations using the default value range
or creating a user specified value range. Only the application of quantitative guide words is
currently considered. From the software engineering point of view, the use of qualitative guide
words in computer aided approach is limited because of practically infinite possibilities of their
interpretation. When the deviation list is complete, it is saved. When the process simulation is
launched, deviations from the list are sent to the Aspen HYSYS environment and simulated one
by one. This process is schematically illustrated for steady state simulation in Fig. 2. Generated
deviations are stored in form of data containing four levels: <type> <ID> <parameter> <deviation
value>. Process footprints contain information also in four-level structure <type> <ID>
<parameter> <value after deviation>. In case of dynamic simulation, additional levels quantifying
the dimension of time have to be introduced. After each simulation, steady state found by the
Aspen HYSYS solver is stored in the internal database in form of a process footprint containing
89
all necessary information about the individual units and streams. List of process footprints
presents de facto the list of HAZOP consequences in form of pairs <deviation> <process
footprint>.
After the last deviation from the list is simulated, evaluation of the simulation data takes place.
Consequences of each simulated deviation are investigated and hazardous events or operability
problems are recognized. The investigation procedure presents predefined methods of analysis,
e.g. parametric sensitivity analysis or user-defined critical values monitoring. Detected
significant consequences are formulated in an HAZOP-like report which serves as a supporting
process hazard analysis for the human expert HAZOP team. The scope and flexibility of the
proposed methodology were demonstrated on two case studies focused on modern nonlinear
processes frequently used in chemical industry.
Fig. 1 – Methodology of the proposed software tool
90
Fig. 2 – Operations carried out by the simulation module with insight into the data storage
methodology
3. Results and discussion
3.1. Case study 1 – Continuous glycerol nitration in stirred tank reactor
Nitroglycerin belongs to chemical compounds widely used in pharmaceutic industry (Boden
et al., 2012) and as propellant ingredients (Pichtel, 2012). One of the most common industrial
ways of nitroglycerin production is glycerol nitration. The nitration of glycerol is usually
91
expressed as an esterification reaction of glycerol and nitric acid shown in Eq. (1) with the
appropriate reaction kinetic model (Eq. (2)). Nitric acid is in the industrial manufacture of
nitroglycerin present in form of mixed acid, a solution of water, nitric acid and sulfuric acid,
which acts as a dehydrating agent. Kinetic data (Table 1) and design conditions (Table 2) have
been adopted from the work of Lu et al. (2008).
𝐶3𝐻5(𝑂𝐻)3 + 3𝐻𝑁𝑂3 → 𝐶3𝐻5(𝑂𝑁𝑂2)3 + 3𝐻2𝑂 (1)
𝑟𝑛𝑖𝑡𝑟𝑎𝑡𝑖𝑜𝑛 = 𝐴1𝑒−𝐸𝑎,1𝑅𝑇 𝐶𝐺
𝑎𝑐𝑁𝑏 (2)
To calculate the physico-chemical properties of pure components and their mixtures, the
Wilson equation of state with parameters taken from the Aspen HYSYS library was selected. In
this case study, mixed acid and pure glycerol were fed in a CSTR with internal cooling coils
containing brine as the cooling medium. The output stream formed a heterogeneous liquid
reaction mixture. Mathematical model of CSTR in the Aspen HYSYS environment consists of
Aspen HYSYS models of “Continuously Stirred Tank Reactor” and “Cooler” (Fig. 3). Additional
vapor stream as the second output stream from CSTR was required to authorize the simulation of
CSTR in Aspen HYSYS. Models of “Continuously Stirred Tank Reactor” and “Cooler” were
connected by heat flow Q and therefore, the heat transfer rate was not taken into account. In order
to simulate real behavior of the examined system, the following issues had to be considered:
coefficients of pre-defined polynomial expansion in the Aspen HYSYS library to calculate
nitroglycerin heat capacity as a function of temperature were mistaken and their modification in
order to correct the polynomial correlation according to the observed behavior of nitroglycerin
(Lu et al., 2008; Sućeska et al., 2010) had to be done. After this correction, the calculated heat
removal in CSTR was not in an agreement with the plant operating data due to the
underestimation of the heat of mixing in the predefined calculation method. Parameters of
calculation were thus corrected to satisfy the relationship between the heat of dilution and the
composition of the reaction mixture correlated by Lu et al. (2008).
Table 1
Kinetic parameters of glycerol nitration
Variable Value Unit
𝐴1 9.78 × 1022 l−1.052mol1.052min−1
𝐸𝑎,1
𝑅 14674.04 K
a 0.935 Dimensionless
b 1.117 Dimensionless
92
Table 2
Design parameters of glycerol nitration
Stream Mass flow
[kg/h]
Temperature
[°C]
Mass fraction
𝐶3𝐻5(𝑂𝐻)3 𝐻𝑁𝑂3 𝐻2𝑆𝑂4 𝐶3𝐻5(𝑂𝑁𝑂2)3 𝐻2𝑂
glycerol 63.6 19 1.00 0.00 0.00 0.00 0.00
mixed_acid 311.6 19 0.00 0.51 0.49 0.00 0.00
product 375.2 15 0.00 0.08 0.41 0.41 0.10
vapor_output 0.0 15 0.00 0.08 0.41 0.41 0.10
Fig. 3 – Glycerol nitration model in Aspen HYSYS environment
Several hazardous events occurred during the nitration reaction and product storage due to the
thermal instability of nitroglycerin (Lu and Lin, 2009; Lu et al., 2008; Pichtel, 2012).
Recommended temperature of the mixture in the reactor is 20 °C and thus the probability of
thermal decomposition of nitroglycerin resulting in the runaway effect at above 30 °C is very
high (Astuti et al., 2014). Therefore, appropriate safety assessment is needed to recognize
potentially dangerous deviations leading to runaway situations. A HAZOP study utilizing the
proposed software tool was performed. Deviated parameters for this case study were: flow of the
stream “brine_in”, i.e. heat removal in unit “CSTR100”, and temperature, flow and composition
of input streams “glycerol” and “mixed_acid”. The absolute and relative deviations were defined
as:
𝑎𝑏𝑠𝑜𝑙𝑢𝑡𝑒 𝑑𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛 = 𝑝𝑎𝑟𝑎𝑚𝑒𝑡𝑒𝑟 𝑣𝑎𝑙𝑢𝑒 𝑎𝑡 𝑓𝑎𝑖𝑙𝑢𝑟𝑒 𝑠𝑡𝑎𝑡𝑒 −
𝑑𝑒𝑠𝑖𝑔𝑛 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑡ℎ𝑒 𝑝𝑎𝑟𝑎𝑚𝑒𝑡𝑒𝑟 (3)
93
𝑟𝑒𝑙𝑎𝑡𝑖𝑣𝑒 𝑑𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛 = 𝑎𝑏𝑠𝑜𝑙𝑢𝑡𝑒 𝑑𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛
𝑑𝑒𝑠𝑖𝑔𝑛 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑡ℎ𝑒 𝑝𝑎𝑟𝑎𝑚𝑒𝑡𝑒𝑟×100 (4)
The relative deviation range for this case study was set from – 30% to + 30% with the step of
1%. The evaluation of simulation data was focused on monitoring of the reactor temperature, i.e.
user specified critical value of a selected parameter. The reactor temperature was assumed to be
equal to the temperature of stream “product”. If the simulated deviation caused exceeding of the
safety limit (temperature of 30 °C in the reactor), the deviation was highlighted for user and
classified as dangerous.
Visualized effect of the analyzed deviations is depicted in Figs. 4-6. As it can be seen, safety
constraints were exceeded in case of cooling system and glycerol flow control failure. Fig. 4
shows the effect of the heat removal deviation on the reactor temperature. When the cooling
system failure caused heat removal decrease of more than 11 %, the “product” temperature and
temperature in the CSTR100 exceeded the critical value of 30 °C and runaway would have
occurred. A similar effect was observed and is plotted in Fig. 5, where the effect of “glycerol”
parameters deviation is shown. When mass flow of stream “glycerol” was increased, temperature
in CSTR100 increased. Above the mass flow of approximately 71 kg/h (absolute deviation of 7.6
kg/h and relative deviation of 12 %), the safety constrain was exceeded. On the other hand, the
“glycerol” temperature deviation had negligible effect on process safety as well as the effect of
“mixed_acid” parameters deviation (Fig. 6).
Fig. 4 – Effect of heat removal deviation in CSTR100 on the temperature in CSTR100 (bold
red line – runaway conditions, X mark – design point, empty circle – last numerical solution in
Aspen HYSYS environment, where reaction rate was calculated)
17 18 19 20 21 22 23 24 250
10
20
30
40
50
60
Safety constraint
Design point
Reaction switched off
Te
mp
era
ture
in C
ST
R1
00
(°C
)
Heat removal in CSTR100 (kW)
94
Fig. 5 – Effect of mass flow (a) and temperature (b) of stream “glycerol” on the temperature
in CSTR100 (bold red line – runaway conditions, X mark – design point, empty circle – last
numerical solution in Aspen HYSYS environment, where reaction rate was calculated)
Fig. 6 – Effect of mass flow (a), temperature (b) and composition (c) of stream “mixed_acid”
on the temperature in CSTR100 (X mark – design point)
It is necessary to point out numerical problems regarding the solution in case of process
variables deviation - heat removal in CSTR100, “glycerol” temperature and “mixed_acid”
composition and mass flow (empty circles in Figs. 4-6). These problems cause significant
decrease of the temperature in CSTR100 (below – 100 °C) because the reaction was switched off
60 62 64 66 68 70 72 74 76 780
10
20
30
40
50
13 14 15 16 17 18 19 20 21 22 23 24 2513
14
15
16
17
Safety constraint
Design point
Reaction switched off
Tem
pera
ture
in
CS
TR
100 (
°C)
Glycerol mass flow (kg/h)
a
Design point
b
Tem
pera
ture
in
CS
TR
100 (
°C)
Glycerol temperature (°C)
260 280 300 320 340 360 380 400 4209,5
10,0
10,5
11,0
11,5
12,0
12,5
13,0
13,5
14,0
14,5
15,0
15,5
16,0
16,5
13 14 15 16 17 18 19 20 21 22 23 24 258
10
12
14
16
18
20
0,40 0,45 0,50 0,55 0,60 0,65 0,70 0,75
10
11
12
13
14
15
16
Design point
a
Te
mp
era
ture
in
CS
TR
10
0 (
°C)
Mixed_acid mass flow (kg/h)
Design point
b
Te
mp
era
ture
in
CS
TR
10
0 (
°C)
Mixed_acid temperature (°C)
Design point
c
Te
mp
era
ture
in
CS
TR
10
0 (
°C)
Mass fraction of nitric acid in mixed_acid (-)
95
by the Aspen HYSYS solver. Inability to find a satisfactory steady state was attributed to the
presence of the heterogeneous liquid phase and values of reaction temperature near the freezing
point. Aspen HYSYS was not capable of modeling the effect of solidification. Another important
observation is the existence of steady states found above the critical temperature. Aspen HYSYS
was unable to detect runaway effect in the CSTR100 because of the selected mathematical reactor
model that takes into account only user defined chemical reactions. Thus, the reaction kinetics of
nitroglycerin decomposition was not taken into account. Steady states found above 30 °C by the
Aspen HYSYS solver were only hypothetical steady states contrary to the real behavior of the
reaction mixture above 30 °C.
Although previously stated problems limit the extent of safety analysis, the developed
software tool successfully identified hazardous events and generated a HAZOP-like report (Table
3) based on simulation data analysis using a user-specified critical value of the selected parameter.
Results of the conducted HAZOP study were in good agreement with safety analysis done by Lu
et al. (2008).
Table 3
HAZOP-like report on glycerol nitration
Deviation Consequence
Heat removal in CSTR100 lower by 11% Possible runaway
“glycerol” mass flow higher by 12% Possible runaway
3.2. Case study 2 – Ammonia synthesis
Ammonia has wide application in chemical industry, e.g. in the production of fertilizers,
cleaning agents or explosives. It is produced by heterogeneously catalyzed hydrogenation of
nitrogen in the gaseous phase (Eq. (5)). The reaction rate was given by Froment et al. (2010) and
modified to include the effect of higher activity of modern industrial catalysts as presented in Eq.
(6). Kinetic parameters (Morud and Skogestad, 1998) are summarized in Table 4.
𝑁2 + 3𝐻2 ↔ 2𝑁𝐻3 (5)
𝑟ℎ𝑦𝑑𝑟𝑜𝑔𝑒𝑛𝑎𝑡𝑖𝑜𝑛 =𝛽
𝜌𝑐(𝐴2𝑒
−𝐸𝑎,2𝑅𝑇 𝑝𝑁2
𝑝𝐻21.5
𝑝𝑁𝐻3− 𝐴−2𝑒
−𝐸𝑎,−2𝑅𝑇
𝑝𝑁𝐻3
𝑝𝐻21.5 ) (6)
96
Table 4
Kinetic parameters of ammonia synthesis
Variable Value Unit
𝐴2 1.79 × 104 kmol1m−3bar−1.5h−1
𝐸𝑎,2
𝑅 10475.10 K
𝐴−2 2.57 × 1016 kmol1m−3bar0.5h−1
𝐸𝑎,−2
𝑅 23871.06 K
𝛽 4.75 Dimensionless
One of the most used industrial ways of ammonia synthesis is its continuous production in an
adiabatic fixed-bed catalytic reactor. Fixed-bed reactors usually consist of several beds connected
in series with feed quenching between the beds. The purpose of feed quenching is to achieve
optimal temperature profile throughout the whole reactor system. The goal of this case study was
to reproduce the strongly nonlinear process behavior that occurred in German ammonia plant in
1989 and was well documented by Mancusi et al. (2000). The cause of the unusual behavior of
the reaction mixture was the presence of the steady state multiplicity phenomenon in ammonia
synthesis (Laššák et al., 2010; Mancusi et al., 2000; Morud and Skogestad, 1998; Pedernera et
al., 1997). Although identification of paths between stable and unstable steady states generally
requires complex mathematical methods such as bifurcation and continuation analyses (Labovský
et al., 2006; Labovský et al., 2007), AspenTech software was successfully used to obtain
approximate locations of stable steady states of an industrial acetic acid dehydration system (Li
and Huang, 2011). The ammonia plant under review consisted of one reactor separated to three
segments and a feed preheater. The reactor system was extended by a separation unit consisting
of a refrigeration system and a flash separator (Fig. 7). Design values of the key operating process
variables (Table 5) were specified by Janošovský et al. (2015). Operating pressure was 20 MPa
and the total reactor volume was 31.52 m3 with the diameter of 0.9 m.
Fig. 7 – Ammonia plant in Aspen HYSYS environment
97
Table 5
Design parameters of ammonia synthesis
Stream Mass flow
[103 kg/h]
Temperature
[°C]
Mole fraction
N2 H2 NH3
fresh feed 252 250 0.239 0.719 0.042
4 127 250 0.239 0.719 0.042
quench1 58 250 0.239 0.719 0.042
quench2 35 250 0.239 0.719 0.042
1a 127 424 0.239 0.719 0.042
R101out 185 520 0.215 0.645 0.140
R102out 220 530 0.210 0.630 0.160
R103out 252 525 0.209 0.625 0.166
6 252 436 0.209 0.625 0.166
liquid ammonia 50 8 0.003 0.015 0.982
The Peng-Robinson equation of state with parameters from the internal Aspen HYSYS library
was used to calculate the properties of gaseous mixtures. Reaction types pre-defined in Aspen
HYSYS did not satisfactorily correlate reaction kinetics in Eq. (6). Therefore, HYSYS extension
containing the proposed kinetics was registered through the customization procedure. According
to previous studies (Honkala et al., 2005; Lísal et al., 2005), the Aspen HYSYS model of “Plug
Flow Reactor” was selected as the appropriate mathematical model for ammonia synthesis reactor
performance simulation. Model of the feed preheating system was built from the Aspen HYSYS
model of “Heat Exchanger”. The overall heat transfer coefficient was calculated to approach the
observed behavior of the preheater (Morud and Skogestad, 1998). Although no material recycle
was present, the “Recycle” unit had to be applied because of the energy recycle (heat generated
by the reaction was partially transferred from stream “R103out” to stream “4” in the preheater).
In this form, mathematical model of ammonia synthesis in the Aspen HYSYS environment was
verified and ready for HAZOP analysis.
For this case study, results of the HAZOP study applied for the operating pressure and the
temperature of “fresh feed” are discussed. Operating pressure relative deviation was set from –
50% to + 100% with the step of 5% (equal to the operating pressure absolute deviation of 1 MPa).
Parameters of “fresh feed” were deviated in the range of relative deviations from – 30% to + 30%
with the step of 2% (equal to the temperature absolute deviation of 5 °C). The effect of the “fresh
98
feed” temperature and the operating pressure on the temperature profile of the reaction mixture
is plotted in Fig. 8. Step change of the temperature of streams “R103out”, “R102out” and
“R101out” was clearly caused by steady state multiplicity. If the temperature of “fresh feed” was
decreased by 18 % or the operating pressure was lowered by more than 25 %, the reactive system
was shifted towards lower solution branch, where the reaction rate was practically equal to zero.
Even after the deviated parameter was set back to the design value, the reactive system remained
in the steady state with low reaction conversion. To restore the original design point, new reactor
start-up was required. Solution branches composed of stable steady states were plotted. However,
the Aspen HYSYS solver was incapable of finding the position of unstable steady states during
the simulation.
Unlike the first case study, definition of critical values by the user was not employed. On the
contrary, mathematical methods of analysis independent from the user have been developed. Fig.
9 shows possible graphical output of such an analysis when applied on the data set plotted in Fig.
8b. The curves in Fig. 9 were constructed as follows: the starting position of the analysis was the
design point (on the higher solution branch). Operating pressure was first increased and then
decreased; i.e. only the shift from the higher to the lower solution branch is described by these
curves. Parameters used in the analysis are defined by Eqs. (3), (4) and (7).
As depicted, a small change of one parameter caused a significant change of another one, e.g.
operating pressure absolute deviation of −6 MPa (–30 %) resulted in a sudden decrease of the
“R103out” temperature by more than 250 °C (50 %) (Fig. 9a,b). This phenomenon was more
significantly exposed in the parametric sensitivity analysis (Fig. 9c), where the peak of the
“R103out” temperature sensitivity to the operating pressure was clearly identified in the region
of the operating pressure absolute deviation of −6 MPa. It was possible to automatically detect
nonstandard behavior of the analyzed reactive system by monitoring these step changes. This
approach is particularly applicable to strongly nonlinear processes characterized by a rapid or a
step change of certain process parameters, e.g. systems exhibiting steady state multiplicity.
Despite the inability to simulate real behavior of the reactive system in the region of unstable
steady states, the proposed software tool satisfactorily simulated sudden changes of operating
conditions between the higher and lower solution branch. Location of the found stable steady
states was in strong agreement with the solution diagrams obtained by Laššák et al. (2010) and
Mancusi et al. (2000). Operability problems corresponding to those observed in industrial
ammonia plants were detected and summarized in a HAZOP-like report (Table 6). In addition,
the HAZOP study was accelerated by implementation of mathematical methods for partially
automated identification of hazards and operability problems.
99
Fig. 8 – Effect of „fresh feed“ temperature (a) and operating pressure (b) on the temperature
of streams “R103out” (black square), “R102out” (red circle) and “R101out” (blue triangle)
(design point – thick square)
Fig. 9 – Effect of operating pressure deviation on the „R103out“ temperature
𝑠𝑒𝑛𝑠𝑖𝑡𝑖𝑣𝑖𝑡𝑦 𝑜𝑓 𝑐𝑜𝑛𝑠𝑒𝑞𝑢𝑒𝑛𝑐𝑒 𝑡𝑜 𝑑𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛 = 𝑑(𝑐𝑜𝑛𝑠𝑒𝑞𝑢𝑒𝑛𝑐𝑒)
𝑑(𝑑𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛) (7)
Table 6
HAZOP-like report on ammonia synthesis
Deviation Consequence
Operating pressure lower by 70 % Operability problem – reaction conversion too low
“fresh feed” temperature lower by 18 % Operability problem – reaction conversion too low
150 200 250 300 350 400100
200
300
400
500
600
10 15 20 25 30 35 40100
200
300
400
500
600
R103out
R102out
R101out
Design point
R103out
R102out
R101out
Design point
Str
ea
m t
em
pe
ratu
re (
°C)
Fresh feed temperature (°C)
a
Operating pressure (MPa)
b
-10 -5 0 5 10 15 20
-300
-250
-200
-150
-100
-50
0
50
-50 -25 0 25 50 75 100
-60
-50
-40
-30
-20
-10
0
10
-10 -5 0 5 10 15 20
0
50
100
150
200
250c
Con
se
qu
en
ce
-
Ab
so
lute
ch
an
ge
of
R10
3o
ut
tem
pe
ratu
re (
°C)
Absolute deviation of operating pressure
(MPa)
b
Con
se
qu
en
ce
-
Rela
tive c
ha
ng
e o
f R
10
3o
ut
tem
pe
ratu
re (
%)
Relative deviation of operating pressure
(%)
Se
nsitiv
ity o
f co
nse
qu
en
ce
to d
evia
tio
n (
°C/b
ar)
Absolute deviation of operating pressure
(MPa)
a
100
4. Conclusions
Process hazard analysis supported by a commercial process simulator can be a very powerful
tool for the analysis of unexpected operating conditions resulting from nonlinear process
behavior. In this paper, construction of a mathematical model in the Aspen HYSYS environment
and steady state simulations applied to two case studies were discussed. Hazards and operability
problems identified were in good agreement with industrial practice. Process simulations in
Aspen HYSYS benefited from the pre-defined property fluid packages and mathematical models
of frequently used unit operations. However, several modifications of the pre-defined models
were necessary in order to simulate real process behavior. In the first case study, alterations in
the built-in model parameters and interaction with the user were crucial to recognize valid results
of computer simulations. The Aspen HYSYS built-in solver was unable to find steady state
solutions in the whole range of deviations. In the second case study, a set of robust mathematical
methods for proper HAZOP analysis of the reactive system exhibiting steady state multiplicity
were developed, but the region of unstable steady states remained unidentified.
Resulting from the general simulation environment of Aspen HYSYS, the developed tool can
be adapted to other chemical processes while maintaining its reliability and accuracy. Universal
application of the proposed tool presents a promising way of more effective and less time
consuming procedure of hazard identification based on HAZOP principles. The presented output
reports can serve as a guide for human HAZOP expert team or in the design and operation phase
of modern chemical plants.
Robustness and applicability of numerical methods currently employed in the proposed
software tool are strictly limited by the Aspen HYSYS built-in solver capability. Future research
will be focused on the elimination of revealed disadvantages of Aspen HYSYS modeling and on
the expansion of the range of software application towards dynamic simulations. The use of
Aspen HYSYS will be extended by developed modules for the simulation of modern industrial
production and separation units and by utilization of numerical procedures optimized for process
safety engineering purposes. With such modules, more effective analysis of complex fault
propagation paths and in-depth investigation of deviation–consequence interactions will be
achievable.
Acknowledgment
This work was supported by the Slovak Scientific Agency, Grant No. VEGA 1/0749/15 and
the Slovak Research and Development Agency APP-14-0317.
101
Nomenclature
a reaction order of glycerol
A pre-exponential factor
b reaction order of nitric acid
c molar concentration, l.mol-1
Ea activation energy, J.mol-1
p partial pressure, bar
r reaction rate
R universal gas constant, J.K-1.mol-1
Greek symbols
β enhancement factor
ρc catalyst bulk density, kg.m-3
References
AspenTech, 2015. Aspen Technology Engineering Products - Aspen HYSYS.
<http://www.aspentech.com/products/aspen-hysys/> (accessed 22.7.2016).
Astuti, E., Supranto, Rochmadi, Prasetya, A., Ström, K., Andresson, B., 2014. Kinetic
Modeling of Nitration of Glycerol. Mod. Appl. Sci. 8(2), 78-86.
Boden, W.E., Finn, A.V., Patel, D., Frank Peacock, W., Thadani, U., Zimmerman, F.H.,
2012. Nitrates as an Integral Part of Optimal Medical Therapy and Cardiac
Rehabilitation for Stable Angina: Review of Current Concepts and Therapeutics. Clin.
Cardiol. 35(5), 263-271.
Corbetta, M., Grossmann, I.E., Manenti, F., 2016. Process simulator-based optimization
of biorefinery downstream processes under the Generalized Disjunctive Programming
framework. Comput. Chem. Eng. 88, 73-85.
Danko, M., Labovský, J., Janošovský, J., Labovská, Z., Jelemenský, Ľ., 2015. Utilization
of parallel computing in chemical engineering. Acta Chim. Slovaca 8(2), 146-151.
De Rademaeker, E., Suter, G., Pasman, H.J., Fabiano, B., 2014. A review of the past,
present and future of the European loss prevention and safety promotion in the process
industries. Process Saf. Environ. Prot. 92(4), 280-291.
de Riva, J., Ferro, V.R., Moreno, D., Diaz, I., Palomar, J., 2016. Aspen Plus supported
conceptual design of the aromatic–aliphatic separation from low aromatic content
naphtha using 4-methyl-N-butylpyridinium tetrafluoroborate ionic liquid. Fuel Process.
Technol. 146, 29-38.
Dunjó, J., Fthenakis, V., Vílchez, J.A., Arnaldos, J., 2010. Hazard and operability
(HAZOP) analysis. A literature review. J. Hazard. Mater. 173(1–3), 19-32.
Eizenberg, S., Shacham, M., Brauner, N., 2006. Combining HAZOP with dynamic
simulation—Applications for safety education. J. Loss Prev. Process Ind. 19(6), 754-
761.
102
Enemark-Rasmussen, R., Cameron, D., Angelo, P.B., Sin, G., 2012. A simulation based
engineering method to support HAZOP studies, in: Iftekhar, A.K., Rajagopalan, S.
(Eds.), Computer Aided Chemical Engineering. Elsevier, pp. 1271-1275.
Favarò, F.M., Saleh, J.H., 2016. Toward risk assessment 2.0: Safety supervisory control
and model-based hazard monitoring for risk-informed safety interventions. Reliab. Eng.
Syst. Saf. 152, 316-330.
Froment, G.F., Bischoff, K.B., De Wilde, J., 2010. Chemical Reactor Analysis and
Design, 3rd Edition. John Wiley & Sons, Inc., New Jersey, USA.
Ghasemzadeh, K., Morrone, P., Iulianelli, A., Liguori, S., Babaluo, A.A., Basile, A.,
2013. H2 production in silica membrane reactor via methanol steam reforming:
Modeling and HAZOP analysis. Int. J. Hydrogen Energy 38(25), 10315-10326.
Honkala, K., Hellman, A., Remediakis, I.N., Logadottir, A., Carlsson, A., Dahl, S.,
Christensen, C.H., Nørskov, J.K., 2005. Ammonia Synthesis from First-Principles
Calculations. Science 307(5709), 555-558.
Janošovský, J., Danko, M., Labovský, J., Jelemenský, Ľ., 2016a. Investigation of
nonlinear behaviour of chemical reactors using Aspen HYSYS as a useful tool for
model-based hazard identification. In: Proceedings of 4th International Conference on
Chemical Technology: ICCT 2016.
Janošovský, J., Labovský, J., Jelemenský, Ľ., 2015. Ammonia synthesis fundamentals
for a model-based HAZOP study. Acta Chim. Slovaca 8(1), 5-10.
Janošovský, J., Labovský, J., Jelemenský, Ľ., 2016b. Automated Model-based HAZOP
study in process hazard analysis. Chem. Eng. Trans. 48, 505-510.
Jeerawongsuntorn, C., Sainyamsatit, N., Srinophakun, T., 2011. Integration of safety
instrumented system with automated HAZOP analysis: An application for continuous
biodiesel production. J. Loss Prev. Process Ind. 24(4), 412-419.
Kletz, T.A., 2001. Hazop and Hazan. IChemE, Rugby, UK.
Labovská, Z., Labovský, J., Jelemenský, Ľ., Dudáš, J., Markoš, J., 2014. Model-based
hazard identification in multiphase chemical reactors. J. Loss Prev. Process Ind. 29(0),
155-162.
Labovský, J., Danko, M., Janošovský, J., Labovská, Z., Jelemenský, Ľ., 2015. Chemical
engineering simulation on parallel computers. In: Proceedings of 3rd International
Conference on Chemical Technology: ICCT 2015.
Labovský, J., Jelemenský, L., 2011. Verification of CFD pollution dispersion modelling
based on experimental data. J. Loss Prev. Process Ind. 24(2), 166-177.
Labovský, J., Jelemenský, Ľ., Markoš, J., 2006. Safety analysis and risk identification
for a tubular reactor using the HAZOP methodology. Chem. Pap. 60(6), 454-459.
Labovský, J., Švandová, Z., Markoš, J., Jelemenský, Ľ., 2007. Model-based HAZOP
study of a real MTBE plant. J. Loss Prev. Process Ind. 20(3), 230-237.
Laššák, P., Labovský, J., Jelemenský, Ĺ., 2010. Influence of parameter uncertainty on
modeling of industrial ammonia reactor for safety and operability analysis. J. Loss Prev.
Process Ind. 23(2), 280-288.
Leveson, N.G., Stephanopoulos, G., 2014. A system-theoretic, control-inspired view and
approach to process safety. AlChE J. 60(1), 2-14.
Li, S., Bahroun, S., Valentin, C., Jallut, C., De Panthou, F., 2010. Dynamic model based
safety analysis of a three-phase catalytic slurry intensified continuous reactor. J. Loss
Prev. Process Ind. 23(3), 437-445.
103
Li, S., Huang, D., 2011. Simulation and analysis on multiple steady states of an industrial
acetic acid dehydration system. Chin. J. Chem. Eng. 19(6), 983-989.
Lísal, M., Bendová, M., Smith, W.R., 2005. Monte Carlo adiabatic simulation of
equilibrium reacting systems: The ammonia synthesis reaction. Fluid Phase Equilib.
235(1), 50-57.
Lu, K.-T., Lin, P.-C., 2009. Study on the stability of nitroglycerine spent acid. Process
Saf. Environ. Prot. 87(2), 87-93.
Lu, K.-T., Luo, K.-M., Yeh, T.-F., Lin, P.-C., 2008. The kinetic parameters and safe
operating conditions of nitroglycerine manufacture in the CSTR of Biazzi process.
Process Saf. Environ. Prot. 86(1), 37-47.
Mancusi, E., Merola, G., Crescitelli, S., Maffettone, P.L., 2000. Multistability and
hysteresis in an industrial ammonia reactor. AlChE J. 46(4), 824-828.
Morud, J., Skogestad, S., 1998. Analysis of instability in an industrial ammonia reactor.
AlChE J. 44(4), 888-895.
Øi, L.E., 2012. Comparison of Aspen HYSYS and Aspen Plus simulation of CO2
Absorption into MEA from Atmospheric Gas. Energy Procedia 23, 360-369.
Parmar, J.C., Lees, F.P., 1987. The propagation of faults in process plants: Hazard
identification for a water separator system. Reliab. Eng. 17(4), 303-314.
Pedernera, M., Borio, D.O., Porras, J.A., 1997. Steady-state multiplicity in cocurrently
cooled autothermal reactors. AlChE J. 43(1), 127-134.
Pichtel, J., 2012. Distribution and Fate of Military Explosives and Propellants in Soil: A
Review. Appl. Environ. Soil Sci. 2012, 33.
Qi, G., Wang, D., Chen, Y., Xin, H., Qi, X., Zhong, X., 2014. The application of kinetics
based simulation method in thermal risk prediction of coal. J. Loss Prev. Process Ind.
29, 22-29.
Seider, W.D., Soroush, M., Arbogast, J.E., Oktem, U.G., 2014. Design for Process
Safety – A Perspective, in: Mario R. Eden, J.D.S., Gavin, P.T. (Eds.), Computer Aided
Chemical Engineering. Elsevier, pp. 795-800.
Smejkal, Q., Šoóš, M., 2002. Comparison of computer simulation of reactive distillation
using aspen plus and hysys software. Chem. Eng. Process. 41(5), 413-418.
Sućeska, M., Mušanić, S.M., Houra, I.F., 2010. Kinetics and enthalpy of nitroglycerin
evaporation from double base propellants by isothermal thermogravimetry.
Thermochim. Acta 510(1–2), 9-16.
Sunny, A., Solomon, P.A., Aparna, K., 2016. Syngas production from regasified
liquefied natural gas and its simulation using Aspen HYSYS. J. Nat. Gas Sci. Eng. 30,
176-181.
Švandová, Z., Labovský, J., Markoš, J., Jelemenský, Ľ., 2009. Impact of mathematical
model selection on prediction of steady state and dynamic behaviour of a reactive
distillation column. Comput. Chem. Eng. 33(3), 788-793.
Tian, W., Du, T., Mu, S., 2015. HAZOP analysis-based dynamic simulation and its
application in chemical processes. Asia-Pac. J. Chem. Eng. 10(6), 923-935.
104
Príloha B
105
Predslov k prílohe B
Príloha B dizertačnej práce je článok v zahraničnom karentovanom časopise „Computers &
Chemical Engineering“ s názvom Software approach to simulation-based hazard
identification of complex industrial processes, ktorý je pozvanou rozšírenou verziou
konferenčného príspevku a bol po revízii prijatý do tlače. V článku je detailne predstavená
štruktúra softvéru spojená s ukážkou aplikácie na komplexný model prevádzky na výrobu
amoniaku. Súčasťou článku je i podrobná definícia dizajnového zámeru s uvedením hodnôt
kľúčových procesných parametrov. Výsledky bezpečnostnej analýzy sú zobrazené
prostredníctvom užívateľského rozhrania softvéru.
Z dôvodu transformácie A4 formátu do B5 formátu je prílohou prepis článku. Plná verzia
článku bude prístupná online po jeho uverejnení.
106
Software approach to simulation-based hazard
identification of complex industrial processes
Ján Janošovský, Matej Danko, Juraj Labovský, Ľudovít Jelemenský*
Institute of Chemical and Environmental Engineering, Slovak University of Technology,
Bratislava, Slovakia
Abstract
Process safety is of major importance in chemical industry. Numerous activities have targeted
the modification of conventional risk assessment strategies by computer aided approach. In this
paper, a software tool for Hazard and Operability (HAZOP) study based on process simulation
is presented. Individual components of the proposed software tool are described and the principal
methodology of their function is explained. As the simulation engine, commercial process
simulator Aspen HYSYS was employed. Proposed tool was applied to a case study of an ammonia
synthesis plant based on an existing plant. Hazardous events and operability problems in the
syngas purification unit and ammonia synthesis loop have been detected and reported. The steady
state multiplicity phenomenon in the ammonia synthesis loop has also been successfully
identified. Based on simulation data evaluation performed in the semi-automatic manner by the
proposed tool, a HAZOP-like report containing HAZOP deviations and their causes and
consequences was generated.
Keywords: computer aided hazard identification, HAZOP study, Aspen HYSYS process
simulation
1. Introduction
The use of smart software solutions in loss prevention is the inevitable future of industrial
practice. Risk assessment of modern complex plants has become unbearable for human expert
teams using only conventional process safety analysis methods. One of the possible novel
approaches utilizes process simulations as a tool improving process hazard identification
techniques (Cameron et al., 2017; Khan et al., 2015; Pasman, 2015). Simulation-based hazard
identification employing appropriate mathematical models can significantly reduce the risk of
overlooking process hazards that are consequences of complicated fault propagation paths or that
have never been observed before.
Software tools for hazard identification automation based on thorough process simulations
require suitable simulation environment and process hazard analysis (PHA) methodology. There
are two principal options for simulation environment. The first one employs mathematical models
107
developed specifically for the analyzed unit (Cui et al., 2015; Danko et al., 2017; Eizenberg et
al., 2006; Ghasemzadeh et al., 2013; Labovský et al., 2008; Li et al., 2010). Own mathematical
model provides full control over equations solving methods and thus of the process simulation
results. To such a mathematical model, extensions to include advanced physico-chemical
phenomena can be easily added. However, all necessary model parameters for proper model
preparation have to be gathered from literature or experimentally measured. Another
disadvantage is the limited application because of model specificity. In case of hazard
identification of another unit, a new set of equations describing the new unit has to be formed and
verified. One of the possible solutions is to create simplified mathematical models where mass,
energy and information flows are interconnected in form of functional models (de la Mata and
Rodríguez, 2012; Rodríguez and de la Mata, 2012; Rossing et al., 2010). However, this approach
is more suitable for the development of qualitative dynamic models, especially in an early design
phase when quantitative models are not yet available. If complex and potentially hazardous
behavior is expected, more detailed mathematical models and advanced steady state and dynamic
simulations should be performed to simulate the complicated fault propagation paths (Wu et al.,
2015). The second principal option for simulation environment is the use of commercial process
simulators to partially eliminate the previously mentioned drawbacks. Commercial process
simulators utilize groups of predefined and properly verified mathematical models of unit
operations widely used in industrial manufacturing processes and they usually have access to
extensive property databases (e.g. The National Institute of Standards and Technology NIST
database and The Design Institute for Physical Properties DIPPR database). Therefore, various
parts of a chemical plant can be simulated by simply adding or removing available components
and unit operations in the process flowsheet. In recent years, a variety of commercial process
simulators has been successfully applied in the procedure of hazard identification, e.g. Aspen
Plus (Jeerawongsuntorn et al., 2011), Aspen HYSYS (J. Janošovský et al., 2017a), Aspen
Dynamics (Berdouzi et al., 2017, 2016; Tian et al., 2015) and K-Spice (Enemark-Rasmussen et
al., 2012). Disadvantages of this approach include the lack of mathematical models of modern
hybrid systems (such as reactive separation processes) and usually insufficient capability of the
built-in solvers to investigate complex nonlinear process behavior (such as steady state
multiplicity). For both simulation environments, crucial factor of correct prediction of process
behavior in unusual situations is the precise verification of the selected mathematical model and
its validity for a wide range of process parameters (Laššák et al., 2010; Švandová et al., 2009).
Appropriate automated determination of potential process hazards and operability problems
is a challenging task. As a PHA methodology suitable to be implemented into a hazard
108
identification software tool, various conventional techniques have been studied. Three main
approaches have recently been investigated in scientific literature: PHA based on Hazard and
Operability (HAZOP) study (Labovská et al., 2014), PHA based on Failure Mode and Effects
Analysis (FMEA) (Chen et al., 2014) and PHA based on their combination (Giardina and Morale,
2015; Seligmann et al., 2012). Both HAZOP and FMEA are the most commonly used PHA
methods in industry by the world safety experts. In addition, their methodologies can be easily
converted into computer logic. As a consequence, in recent years, research focus areas especially
in HAZOP improvement were strongly linked to safety analysis automation (De Rademaeker et
al., 2014; Dunjó et al., 2010).
Commercial process simulators provide reliable mathematical models and simulation
environment fitting hazard identification requirements; the HAZOP study represents a robust
PHA technique with principles readily incorporable into software solutions. In our work, process
simulation in Aspen HYSYS is combined with a software tool for hazard identification based on
the HAZOP methodology. A plant for ammonia production was selected as a case study for
software demonstration because of its well-known nonlinear behavior and comprehensive records
of historical accidents in ammonia production and storage.
First part of this paper is dedicated to the description of software modules and their features.
In the second part, the ammonia plant in its design stage is reviewed employing simulation-based
hazard identification provided by the proposed software tool. In this part, complex interaction of
causes-deviations-consequences is studied in detail through steady state simulations to
thoroughly investigate potential hazardous events and operability problems as well as to obtain
profound understanding of process nonlinearity. Such an analysis of chemical plants in their
design stage represents an inevitable step before any layer of protection is added, e.g. process
control configuration. Although only steady state simulations are considered, sufficient
knowledge of nonlinear process behavior is achievable. To conclude the hazard identification
procedure, a HAZOP-like report is generated. A separate section is dedicated to the discussion
on process simulation limitations in Aspen HYSYS and completeness of the performed analysis.
2. Software structure
Individual components of the proposed software tool are schematically depicted in Fig. 1 (Ján
Janošovský et al., 2017b). It consists of two main parts, Process simulation module and
Simulation data analysis module, and two auxiliary components Simulation engine for process
simulation and Database engine for storage of HAZOP deviations and consequences. The choice
of Simulation engine is optional. The only fundamental requirement for an eligible simulation
tool is the possibility of external communication with the simulation environment to input data
109
(HAZOP deviation) and to access simulation outputs (HAZOP consequence). In this case study,
Aspen HYSYS was selected because of its preferability as a commercial simulator for processes
in oil and gas industry (AspenTech, 2017). Despite the frequent use of Aspen HYSYS as a
suitable simulation environment (Ahmad et al., 2012; Aspelund et al., 2010; Perederic et al.,
2015; Pirola et al., 2017), its use in computer aided hazard identification is scarce. The Database
engine serves as a communication bridge between two main modules. In the proposed software
tool, the SQLite database engine is currently employed (SQLite, 2017).
Fig. 1. Schematic description of the proposed software tool components
2.1. Process simulation module
In this module, actual communication with the process simulator is provided. After the
software startup procedure, the user can select a simulation file with prepared mathematical
model of the plant under review. While establishing the connection between our tool and the
process simulator, data containing relevant information on process streams and unit operations
are extracted and stored in multilevel form: first level – stream type, second level – stream
identification number, third level – stream parameter, fourth level – design value. This data
container is in our software dictionary referred to as the “Footprint” (example depicted in Fig. 2).
Following the HAZOP dictionary, “Footprint” with data from the original simulation file
represents the design intent.
Design intent is accessed by the HAZOP team via graphical user interface (GUI), where the
HAZOP team can easily generate desired HAZOP deviations by combining available process
parameters with guide words for the selected HAZOP node. HAZOP node in the internal
dictionary of the proposed software tool is defined as any material or energy stream, or unit
operation. Quantitative guide words NONE, MORE, LESS and REVERSE are currently
implemented. Qualitative HAZOP deviations are implemented only to a certain extent by
combining quantitative guide words MORE and LESS with process parameters describing
material streams composition (mole and mass fractions and mole and mass component flows).
110
After a HAZOP deviation in its conventional form is created, its extension by desired range
of parameter value follows. The HAZOP team has to associate every generated HAZOP deviation
with its value range because of the mathematical modeling principle, i.e. fundamental
requirement of specified process parameters values in order to perform calculations for process
simulation. All HAZOP deviations in their finalized forms are stored in a similar manner as
depicted in Fig. 2 except that a special identification number is assigned to each HAZOP
deviation. In other words, one particular HAZOP deviation in our software tool environment
contains a unique deviation identification number and data specifying the deviated process
parameter: first level – stream type, second level – stream identification number, third level –
stream parameter, fourth level – deviation value (absolute or relative change from the design
intent).
Fig. 2. Data storage in the form of “Footprint”
Every generated HAZOP deviation is simulated individually. When the Aspen HYSYS solver
is set to holding mode and simulation case is in steady state, information containing the deviated
process parameter and deviation value is sent to the Aspen HYSYS environment. Process
parameters are not directly changed to prevent conflict between the set and the calculated values
of process parameters caused by the sequential character of Aspen HYSYS modeling, where each
operation unit is represented by a set of mathematical equations and an appropriate numerical
algorithm and is simulated individually in an order determined by the Aspen HYSYS solver. For
this purpose, an apparatus for process parameter manipulation is constructed within the Aspen
HYSYS simulation case. For a material stream deviation, the manipulation apparatus consists of
four auxiliary material streams, Mixer and Cooler unit operations (Fig. 3). The target material
stream is attached to the manipulation apparatus (specifically to the “Mixer” unit). Material
stream “TEST_1” is used for flow alterations, material stream “TEST_THIEF” is used for
composition alterations and Cooler unit serves for pressure, temperature and vapor fraction
111
alterations. With this approach, the apparatus output material stream “TEST_3” is de facto
cloning the target material stream except for the particular process parameter selected for the
HAZOP deviation. This configuration enables the change of any desired process parameter
without violating the Aspen HYSYS natural calculation process.
Fig. 3. Process parameter manipulation apparatus for a material stream in the Aspen HYSYS
environment
After simulating a particular HAZOP deviation, process steady state is captured in form of
the “Footprint”. Before the next HAZOP deviation is simulated, the examined process is set back
to the initial state (design intent) and only then, the next HAZOP deviation can be simulated. This
is a standard calculation approach. In strongly nonlinear systems, a different approach has to be
applied. Unlike Aspen Plus, Aspen HYSYS does not allow the user to formulate straightforward
initial guess for parameters values of individual unit operations. However, the Aspen HYSYS
solver automatically uses the current steady state to estimate parameter values in a new steady
state. This fact can be exploited to facilitate the Aspen HYSYS calculation process. If the
calculation of a new steady state using standard calculation approach does not converge to a
solution, the proposed software tool switches to a specific calculation approach to overcome
failure in the numerical solving procedure. In this approach, the simulation is not automatically
reset after every HAZOP deviation. Instead, HAZOP deviations of the same process parameter
are sorted by the deviation value in ascending or descending order and simulated sequentially,
i.e. after simulating a HAZOP deviation, steady state is kept and used as the starting point for
steady state calculation of the following HAZOP deviation. This is particularly advantageous in
case of steady state multiplicity where correct calculation of different solution branches using
only the predefined Aspen HYSYS solver options is complicated (Li and Huang, 2011). If the
calculation of a new steady state still fails, the proposed software tool automatically lowers the
step between two consecutive deviations (specific calculation approach with deviation step
adjustment). The whole calculation process is schematically depicted in Fig. 4.
112
Fig. 4. HAZOP deviation simulation order in the proposed software tool (rounded squares
represent corresponding steady states and roman numerals indicate the order of actions)
Simulation data required for the evaluation process are stored using the database engine and
therefore is the communication with an external process simulator no longer necessary. After the
simulation of all desired HAZOP deviations, the connection with the process simulator can be
terminated and data evaluation procedure follows.
2.2. Simulation data analysis module
The simulation data analysis module provides complex evaluation of simulated process steady
states corresponding to HAZOP deviations, i.e. instances of “Footprint” assigned to particular
HAZOP deviations and severity ranking of identified hazardous events and operability problems.
Following the HAZOP terminology, this module serves as a tool for HAZOP consequences
analysis. As mentioned above, the simulation data evaluation process does not require live
connection with a process simulator that allows running both software modules simultaneously
thus supporting parallel HAZOP deviations simulation and HAZOP consequences analysis.
At first, instances of “Footprint” labeled as correctly simulated steady states are decomposed
and investigated. The investigation procedure consists of several steps employing numerical
algorithms optimized for hazard identification purposes. Generally, two fundamentally different
hazard identification approaches are applied. The first one is based on predefined advanced
numerical methods not requiring user intervention. Methods such as runaway conditions
identification, parametric sensitivity analysis, steady state multiplicity investigation, etc., are
progressively applied to simulation data to automate identification of hazardous events and
operability problems. These methods were further discussed in our previous works (Danko et al.,
2018, 2016; Janošovský et al., 2015). The second implemented hazard identification approach
113
involves user intervention and enables the adjustment of hazard assessment by defining the
process- or equipment-specific threshold values that can be either determined by technical
specifications such as the maximum pressure in pressure relief system or empirically determined
by operating staff such as allowed maximum and minimum level in a storage tank, empirically
determined hazardous temperature limits in a reactor, etc. Process parameters for which threshold
values are set are monitored and exceeding of these threshold values is recorded.
Hazardous events and operability problems identified by the investigation mechanisms are
ranked either automatically or by the user (HAZOP teams usually rank consequences based on
the company safety policy, e.g. taking into account severity and frequency of events) and
transformed into simplified HAZOP-like report consisting of corresponding HAZOP deviations,
their causes and their potential HAZOP consequences. Results of the HAZOP analysis can be
accessed also through several visualization techniques via GUI. For explanatory purposes,
HAZOP consequences visualization and evaluation will be further demonstrated on a case study.
3. Case study
Ammonia, one of the most versatile compounds in chemical industry, is conventionally
produced in a fixed-bed reactor system by the synthesis of hydrogen and nitrogen at elevated
pressure and temperature. In the presented case study, steam reforming of natural gas with
supplementary syngas purification was modeled as the source of hydrogen. Parameter values and
unit operations for these processes were taken from operational data of an ammonia production
plant located in the Slovak Republic. The nitrogen manufacturing process was not simulated in
this case study. The reactor system was composed of a heat exchanger for preheating a part of the
reactor system feed stream and an adiabatic fixed-bed catalytic reactor. A simplified
mathematical model of the separation unit was also included in the simulated ammonia synthesis
plant. The whole ammonia production plant model constructed for process simulation in the
Aspen HYSYS environment is depicted in Fig. 5.
3.1. Design intent
In the first step, desulfurized natural gas was mixed with high pressure water steam and
preheated in heat exchanger E-100. In standard plant operation, molar ratio of water steam to
natural gas was controlled and set to the desired value of 2.8. Natural gas composition was
calculated from weighted averages stated by the natural gas distributor in the Slovak Republic
(SPP Distribucia, 2016). The presence of carbon dioxide and hydrocarbons higher than the C4
fraction in natural gas was neglected. The mixture from heat exchanger E-100 was led to the
primary reforming reactor. After primary reforming, air was added to the reaction mixture and
secondary reforming took place. Primary and secondary reformers were modeled in Aspen
114
HYSYS as “Gibbs Reactor” with specified equilibrium reactions. Reaction set for the primary
reformer consisted of seven reactions (reactions of carbon monoxide, methane, ethane, propane,
i-butane and n-butane with water) and reaction set for the secondary reformer consisted of four
reactions (oxidation reactions of methane, carbon monoxide and hydrogen) (Häussinger et al.,
2000). Equilibrium data were calculated either from temperature correlations (reactions of carbon
monoxide and methane with water) (Abbas et al., 2017) or from values of the Gibbs free energy
(reactions of higher hydrocarbons with water and oxidation reactions). Design parameters of
primary and secondary reformers are summarized in Table 1.
Fig. 5. Ammonia production plant in Aspen HYSYS environment
Table 1
Design intent of primary and secondary reforming processes – selected process parameters
HAZOP node
Parameter NaturalGas LiveSteamHP S3 AIR S5
Temperature [°C] 25.0 350.0 809.7 800.0 969.3
Pressure [kPa] 3 548 12 000 3 378 3 500 3 228
115
Mass flow [103 kg/h] 31.9 95.8 127.7 52.6 180.3
Mole fraction of methane 0.9575 0.0 0.0822 0.0 0.0069
Mole fraction of ethane 0.0267 0.0 0.0 0.0 0.0
Mole fraction of propane 0.0065 0.0 0.0 0.0 0.0
Mole fraction of i-butane 0.0010 0.0 0.0 0.0 0.0
Mole fraction of n-butane 0.0011 0.0 0.0 0.0 0.0
Mole fraction of nitrogen 0.0072 0.0 0.0014 0.7900 0.1170
Mole fraction of carbon dioxide 0.0 0.0 0.0592 0.0 0.0458
Mole fraction of carbon monoxide 0.0 0.0 0.0644 0.0 0.1060
Mole fraction of water 0.0 1.0000 0.3720 0.0 0.2919
Mole fraction of oxygen 0.0 0.0 0.0 0.2100 0.0
Mole fraction of hydrogen 0.0 0.0 0.4208 0.0 0.4324
After secondary reforming, the reaction mixture was cooled down and led to high temperature
conversion (HTC) and low temperature conversion (LTC) reactors to transform carbon monoxide
to carbon dioxide by a water gas shift reaction (also occurring in primary reforming but with
lower conversion). Both reactors were modeled as “Gibbs Reactor”. Between HTC and LTC
reactors, heat exchanger E-102 was installed for additional cooling. Design parameters of HTC
and LTC processes are summarized in Table 2. Heat exchanger E-103 served for partial
condensation of the reaction mixture after the LTC step. Remaining steam with the water content
of over 99.8 mol. % was removed in “Separator” unit SEP-1.
Table 2
Design intent of HTC and LTC processes – selected process parameters
HAZOP node
Parameter S6 S7 S10 SynGas
Temperature [°C] 360.0 432.7 40.0 40.0
Pressure [kPa] 3 218 3 098 2 978 2 878
Mass flow [103 kg/h] 180.3 180.3 180.3 138.3
Mole fraction of methane 0.0069 0.0069 0.0069 0.0085
Mole fraction of nitrogen 0.1170 0.1170 0.1170 0.1439
Mole fraction of carbon dioxide 0.0458 0.1210 0.1484 0.1822
Mole fraction of carbon monoxide 0.1060 0.0307 0.0033 0.0041
Mole fraction of water 0.2919 0.2166 0.1892 0.0032
Mole fraction of hydrogen 0.4324 0.5078 0.5352 0.6581
116
To prevent catalyst poisoning, carbon monoxide, carbon dioxide and water were removed
from syngas produced in the steam reforming section (material stream SynGas) in the “syngas
purification” part of the plant. First, carbon dioxide was scrubbed in absorption column CO2 ABS
modeled as “Absorber Column”. An amine solution was used as the absorption agent and it was
regenerated in distillation column CO2 DES. After the absorption step, temperature of syngas
was increased in heat exchanger E-106 and carbon monoxide and residual carbon dioxide were
removed by methanation. In the methanizer modeled as “Gibbs Reactor”, carbon monoxide and
carbon dioxide reacted with hydrogen at 337 °C and were transformed into methane with water
as the byproduct. Output stream from the methanizer (material stream S28) containing ca. 80 mol.
% of hydrogen was then compressed to 20 MPa in a two-stage compressor (K-101 and K-102)
with cooling units and separators for water removal. Compression pressure ratio in K-101 and K-
102 was 2.8 and 2.5, respectively. Before entering the synthesis loop, hydrogen gas was
optionally dried in an absorber where a small amount of liquid ammonia served as the absorption
agent. Design parameters of key material streams in the “syngas purification” section of the
analyzed plant are shown in Table 3.
Table 3
Design intent of syngas purification – selected process parameters
HAZOP node
Parameter S13 S28 S36 Hydrogen
Temperature [°C] 41.6 337.1 160.2 250.0
Pressure [kPa] 2 878.0 2 878.0 20 000.0 20 000.0
Mass flow [103 kg/h] 57.0 57.0 56.0 63.2
Mole fraction of methane 0.0104 0.0156 0.0157 0.0153
Mole fraction of nitrogen 0.1763 0.1781 0.1793 0.1704
Mole fraction of carbon dioxide 0.0 0.0 0.0 0.0
Mole fraction of carbon monoxide 0.0050 0.0 0.0 0.0
Mole fraction of water 0.0026 0.0078 0.0011 0.0
Mole fraction of hydrogen 0.8057 0.7985 0.8039 0.7642
Mole fraction of ammonia 0.0 0.0 0.0 0.0501
Hydrogen gas, pure nitrogen and material stream S50, stream recycled from “ammonia
separation unit” were mixed together to create the feed stream (material stream Feed NH3) for
the ammonia synthesis reactor. The adiabatic fixed-bed catalytic reactor was composed of three
quenching sections, in which individual splits of material stream Feed NH3 were led to achieve
117
optimal temperature profile in the reactor. These quenching sections of the reactor were modeled
as three “Plug Flow Reactor” units (unit operations R101, R102 and R103 in the scheme). The
role of heat exchanger E-111 was to preheat material stream Main Feed 1, major part of the feed
stream, with reactor output stream, material stream R103out. As intended by design, temperature
of material stream Main Feed 1 was increased from 250 °C to 416.9 °C while the reactor output
stream was cooled from 540.9 °C to 455.3 °C. Ammonia synthesis loop design parameters and
its key operating parameters (Table 4) were determined by operation conditions from actual
historical accident well documented by Morud and Skogestad (1998). Configuration of the
ammonia synthesis loop as well as construction design of the reactor were thoroughly discussed
in our previous work (Janošovský et al., 2016). The simplified ammonia separation unit consisted
of two separators with the low-temperature separator operated at – 30 °C. Ammonia content in
the final product stream (material stream Product) was above 98 mol. %.
Table 4
Design intent of ammonia synthesis loop – selected process parameters
HAZOP node
Parameter Feed NH3 R103out S50 Product
Temperature [°C] 250.0 540.9 250.0 -30.0
Pressure [kPa] 20 000.0 20 000.0 20 000.0 20 000.0
Mass flow [103 kg/h] 296.1 296.1 205.0 81.4
Mole fraction of methane 0.0080 0.0089 0.0060 0.0135
Mole fraction of nitrogen 0.2209 0.1843 0.2081 0.0035
Mole fraction of hydrogen 0.7595 0.6623 0.7859 0.0022
Mole fraction of ammonia 0.0116 0.1445 0.0 0.9808
4. Application of the proposed software tool to the case study – results and discussion
Strongly nonlinear behavior of the ammonia synthesis loop causing steady state multiplicity
is well known and it was previously identified in our work focused on the HAZOP study of a
simplified reactor model (J. Janošovský et al., 2017a) where a change of the reactor pressure and
temperature from the design intent caused shifting between higher and lower steady state
branches. The goal of the following HAZOP analysis was to determine whether the plant design
allows propagation of the deviation in different sections of the plant to the actual ammonia
synthesis loop and to analyze these fault propagation paths.
118
4.1. Generation of simulation scenarios
A set of logic HAZOP deviations from the design intent was generated in GUI of the Process
simulation module (Fig. 6) and evaluated using steady state simulations. For explanatory
purposes, the range of HAZOP deviations was kept uniform. Upper limit of parameter deviations
was set to 30 %, lower limit of parameter deviations was set to –30 % and the value of default
step change was set to 1 %. Deviations in one particular HAZOP node were grouped into
simulation scenarios that were analyzed in the Simulation data analysis module. Table 5 presents
a list of selected HAZOP deviations whose analysis will be described in detail. Potential causes
of individual simulation scenarios were generated manually based on the actual conventional
HAZOP study report for an ammonia synthesis plant.
All simulation scenarios were analyzed using visualization mechanisms employed in the
Simulation data analysis module GUI. An example of such visualization with the description of
individual GUI parts is shown in Fig. 7. In the list of HAZOP nodes and their parameters available
for HAZOP deviation analysis, green spheres represent list items for which HAZOP deviations
were completely simulated, red spheres represent list items for which no simulations were carried
out and green-red spheres represent list items containing other list items and only for a part of
them, HAZOP deviations were completely simulated. Individual types of HAZOP consequences
visualization and evaluation will be further demonstrated by particular simulation scenarios. In
the figures associated with simulation scenarios, only the window for the currently open figure is
depicted because of the figure limitations.
Table 5
HAZOP deviations list with assigned simulation scenarios
HAZOP
node Parameter
Guide
word
Simulation
scenario Potential causes
S10
mass flow MORE
SS-1
Malfunction of steam reformers,
Malfunction of HTC reactor,
Malfunction of LTC reactor,
Malfunction of heat exchanger
E-103
LESS
temperature MORE
LESS
pressure MORE
LESS
mole fraction of
hydrogen
MORE
LESS
Nitrogen mass flow MORE
SS-2 Failure of nitrogen supply LESS
119
Feed
NH3
temperature MORE
SS-3
Malfunction of ammonia
synthesis temperature control,
Malfunction of ammonia
synthesis pressure control
LESS
pressure MORE
LESS
Fig. 6. Selection process of HAZOP deviations in GUI of the proposed software tool.
Fig. 7. Example of HAZOP deviation analysis in GUI of the proposed software tool.
120
4.2. Simulation scenario SS-1
For the analysis of HAZOP consequences caused by possible malfunction of HTC and LTC
reactors and heat exchanger E-103 used as a partial condenser or by parameter deviations after
steam reforming, HAZOP deviations of material stream S10 (HAZOP node) representing the
entry of the “syngas purification” section were generated (Fig. 8). Simulation scenario SS-1 was
composed of HAZOP deviations of four process parameters of material stream S10. For
parameters temperature, pressure and mole fraction of hydrogen, HAZOP deviations were
successfully simulated in the whole predefined range. For mass flow deviations, the Aspen
HYSYS solver was capable to successfully converge to a steady state in the range from + 16 %
to – 30 % change from the design intent. For values of deviations “mass flow of material stream
S10 higher by 17 % and more than design intent”, at least one unit operation was not correctly
solved by the Aspen HYSYS solver which resulted in missing parameter values for one or more
streams.
Fig. 8. Ammonia production plant scheme detail of deviations place of origin for simulation
scenario SS-1
121
One of the deviation visualization options is the analysis of one parameter of all monitored
HAZOP nodes for one particular value of a HAZOP deviation (Fig. 9). Individual material
streams represent x-axis of the presented figure. Material streams L1, L2, L3, L4 and L6 are not
visible in the Aspen HYSYS process scheme (Fig. 5). These liquid streams were added to every
“Gibbs Reactor” because of the computation process for “Gibbs Reactor” unit as due to internal
definition of unit operations, one liquid and one vapor outlet stream have to be attached to
properly calculate the reaction rate. Their flows were equal to zero because they were added only
for the authorization of the “Gibbs Reactor” simulation in the Aspen HYSYS environment.
Therefore, these streams were hidden in the Aspen HYSYS process scheme. On the y-axis,
relative temperature change from the design intent is plotted. Relative parameter change is
defined as the difference between the actual value of the parameter in the steady state
corresponding to the selected HAZOP deviation and the design intent value of the parameter
divided by the design intent value of the parameter. The most significant effect on the temperature
of material streams was simulated in case of temperature deviation (Fig. 9). Temperature of
material streams directly connected to the “Separator” unit SEP-1 (S11 and SynGas) copied the
behavior of the deviated temperature of material stream S10. Excluding these streams, relative
temperature change of all material streams from the design intent did not exceed 2 %. Similar or
smaller temperature changes were simulated for HAZOP deviations of other parameters of
material stream S10. However, temperature increase in “Separator” unit SEP-1 caused ineffective
water removal from material stream S10. When the temperature of material stream S10 increased
from the design intent value of 40 °C to the deviated value of 52 °C, mass flow of water in material
stream SynGas increased from the design intent value of 573.8 kg/h to a new value of 1 042.9
kg/h (Fig. 10). In other words, for the HAZOP deviation “temperature of material stream S10
higher by 30 % than design intent”, water content in material stream SynGas increased by more
than 80 %. Increased water content causes operability problems with carbon dioxide scrubbing
in absorption column CO2 ABS and thus the amount of amine solution used as the absorption
agent and operating parameters of its regeneration have to be properly adjusted. Therefore, the
HAZOP deviation “temperature of material stream S10 higher by 30 % than design intent” was
labeled as operability problem recommended for supplementary analysis. Simulated steady states
for the generated HAZOP deviations were not classified as hazardous events. As it is clear,
HAZOP nodes close to the deviation’s place of origin, material stream S10, were affected most
significantly.
122
Fig. 9. Relative temperature change of all material streams from the design intent for HAZOP
deviation “temperature of material stream S10 higher by 30 % than design intent”.
123
Fig. 10. Mass flow of water in material stream SynGas for HAZOP deviations “temperature
of material stream S10 higher than design intent” and “temperature of material stream S10 lower
than design intent” in the whole simulation range.
4.3. Simulation scenario SS-2
Simulation scenarios SS-2 and SS-3 (Fig. 11) were generated to analyze deviations in the
immediate vicinity of the “ammonia synthesis loop” feed as it has been observed to be a frequent
source of potentially hazardous consequences (Labovská et al., 2014; Morud and Skogestad,
1998). In case of simulation scenario SS-2, steady state was successfully calculated for every
value of HAZOP deviation of material stream Nitrogen (HAZOP node) mass flow change from
the design intent in the predefined range from + 30 % to – 30 %. Fig. 12 shows the temperature
change of all material streams from the design intent for the worst case – HAZOP deviation “mass
flow of material stream Nitrogen lower by 30 % than design intent”. As it is evident, only material
streams in the “ammonia synthesis loop” and in the “ammonia separation unit” were affected. For
further analysis of the “ammonia synthesis loop”, material stream R103out (outlet stream from
the fixed-bed reactor) was selected as representative material stream demonstrating the behavior
in the reactor system. Temperature in the reactor system decreased (Fig. 13) as a result of
improper molar ratio of hydrogen to nitrogen in material stream R103in that deviated from the
design intent value of 3.5 up to 8.5 causing unsuitable reaction conditions and a decrease of
ammonia production. For values of the molar ratio of hydrogen to nitrogen in the reactor inlet
stream above four, reaction conditions were considered problematic for the reactor operation.
This value was simulated first for the HAZOP deviation “mass flow of material stream Nitrogen
124
lower by 10 % than design intent”. Therefore, HAZOP deviations “mass flow of material stream
Nitrogen lower by 10 % and more than design intent” were labeled as operability problems.
Simulated steady states were not classified as hazardous events.
Fig. 11. Ammonia production plant scheme detail of deviations place of origin for simulation
scenarios SS-2 and SS-3
125
Fig. 12. Relative temperature change of all material streams from the design intent for
HAZOP deviation “mass flow of material stream Nitrogen lower by 30 % than design intent”.
126
Fig. 13. Temperature of material stream R103out for HAZOP deviations “mass flow of
material stream Nitrogen higher than design intent” and “mass flow of material stream Nitrogen
lower than design intent” in the whole simulation range.
4.4. Simulation scenario SS-3
Simulation scenario SS-3 was created to analyze temperature and pressure deviations in the
“ammonia synthesis loop” (Fig. 11). The HAZOP node for this simulation scenario was material
stream Feed NH3. Steady state was successfully calculated for every value of HAZOP deviation
in the predefined range. For the worst cases – HAZOP deviations “temperature of material stream
Feed NH3 lower by 30 % than design intent” and “pressure of material stream Feed NH3 lower
by 30 % than design intent”, relative temperature change is depicted in Fig. 14. As it is analogous
to simulation scenario SS-2, only material streams in the “ammonia synthesis loop” and in the
“ammonia separation unit” were affected. In case of temperature deviation, temperature of
material stream R103out decreased by more than 65 % from the design intent value. In case of
pressure deviation, temperature of material stream R103out decreased by more than 50 % from
the design intent value. Similar temperature decrease was simulated also for material streams
R101out and R102out representing outlet streams from the two sections of the fixed-bed reactor.
Such a dramatic temperature drop indicated the reaction termination, which was confirmed by a
material streams composition analysis revealing the drop of ammonia mass flow in material
stream R103out by more than 99 %. Sudden reaction termination was caused by the multiple
steady states phenomenon (Mancusi et al., 2000; Morud and Skogestad, 1998). For correct
calculation of different solution branches using only predefined Aspen HYSYS solver options,
127
our software tool provided two sets of simulations. First, HAZOP deviations were sorted by the
deviation value in the ascending order and simulated sequentially. The Aspen HYSYS simulation
case file was then reset back to the design intent. Then, HAZOP deviations were sorted by the
deviation value in the descending order and simulated sequentially (specific calculation approach
described in Fig. 4). Using this approach, mapping of solution branches for ammonia synthesis
represented by material stream R103out was enabled. Simulated steady states for HAZOP
deviations sorted in the ascending order (Fig. 15) are clearly different from the simulated steady
states for HAZOP deviations sorted in the descending order (Fig. 16). Similar behavior was
simulated also for pressure deviation. This phenomenon was precisely detected by the parametric
sensitivity analysis. Results of the parametric sensitivity analysis for the descending order of
simulations are depicted in Fig. 17. On the x-axis, relative deviated parameter change from the
design intent Drel defined as the difference between the deviated value of the parameter
corresponding to the selected HAZOP deviation and the design intent value of the parameter
divided by the design intent value of the parameter is plotted. On the y-axis, sensitivity parameter
s numerically calculated as the ratio of the difference between monitored parameter values to the
difference between the deviated parameter values for two consecutive HAZOP deviations is
plotted. Abnormally high peaks of sensitivity parameter s were detected and the corresponding
HAZOP deviations “temperature of material stream Feed NH3 lower by 25 % and more than
design intent” and “pressure of material stream Feed NH3 lower by 28 % and more than design
intent” were labeled as operability problems because of the practically zero ammonia production
reaction rate. Switching between two steady state branches is usually associated with possibly
hazardous parameter oscillations. For example, in Germany, 1989, temperature oscillations
destructive for the catalyst were observed as a consequence of pressure decrease in the ammonia
synthesis reactor (Morud and Skogestad, 1998). Such unpredictable rapid temperature changes
can cause mechanical disruption of the equipment and pipe couplings resulting in gas leakage.
Even small leaks of ammonia and hydrogen can be dangerous because of their toxicity and
flammability, respectively. More detailed evaluation of probability and severity of these events
requires further risk assessment. Therefore, these HAZOP consequences were labeled also as
potentially hazardous events.
128
Fig. 14. Relative temperature change of all material streams from the design intent for
HAZOP deviation “temperature of material stream Feed NH3 lower by 30 % than design intent”
(a) and “pressure of material stream Feed NH3 lower by 30 % than design intent” (b).
129
Fig. 15. Temperature of material stream R103out for HAZOP deviation “temperature of
material stream Feed NH3 higher than design intent” and “temperature of material stream Feed
NH3 lower than design intent” in the whole simulation range simulated in the ascending order.
Fig. 16. Temperature of material stream R103out for HAZOP deviation “temperature of
material stream Feed NH3 higher than design intent” and “temperature of material stream Feed
NH3 lower than design intent” in the whole simulation range simulated in the descending order.
130
Fig. 17. Parametric sensitivity analysis for the temperature of material stream R103out as a
function of the relative change of deviated parameter (temperature of material stream Feed NH3)
from the design intent in the descending order.
4.5. HAZOP table generation
Based on the results of computer simulation of the generated simulation scenarios presented
in previous sections, a preliminary HAZOP table was generated using our proposed software tool
(Fig. 18). The HAZOP table contains only a list of HAZOP deviations and their causes and
consequences. HAZOP deviations severity classification and formulation of recommendations
have to be completed by a human HAZOP expert team implementing specific company safety
policy.
Fig. 18. Example of a HAZOP table generated by the proposed software tool
4.6. Completeness of the steady state multiplicity identification using Aspen HYSYS
The proposed software tool was able to identify multiple potential operability problems and
hazards caused by nonlinear character of the studied process using only numerical methods
provided by the built-in Aspen HYSYS solver. Because of the model scale and complexity as
well as Aspen HYSYS solver capabilities, only steady state simulations were carried out. With
131
the help of these simulations, the proposed tool successfully recognized the presence of two
steady state solution branches for the ammonia synthesis loop (Figs. 15 and 16). However, as
already mentioned, transition between these two solution branches is usually characterized by
parameter oscillations that can be correctly predicted only by implementing dynamic simulations.
With the use of the built-in Aspen HYSYS solver, dynamic simulations of the considered
ammonia plant were not possible due to the convergence issues in Aspen HYSYS “Dynamics
Mode”. In our previous works (Labovská et al., 2014; Laššák et al., 2010), a thorough analysis
with the implementation of dynamic simulations combined with the Hopf bifurcation and
continuation algorithm successfully revealed the existence of parameter oscillations and positions
of the Hopf bifurcation points (Fig. 19) for an analogous configuration of the ammonia synthesis
loop. The obtained solution diagram in Fig. 19 included also loci of unstable steady states.
Difference between temperature values for steady state solution branches in Fig. 19 compared to
Figs. 15 and 16 is caused by a discrepancy in fresh feed flow and composition for the respective
ammonia synthesis loop configurations.
Fig. 19. Solution diagram of outlet temperature for three sections of the fixed-bed ammonia
reactor (black line – 1. section, red line – 2. section, blue line – 3. section) as a function of fresh
feed temperature (diamond – design intent, empty circle – Hopf bifurcation point, full square –
limit point, dashed line – unstable steady states) as presented by Labovská (2014)
Compilation of the whole solution diagram using only built-in Aspen HYSYS solver options
is significantly limited. As it was demonstrated, adjusted calculation approaches incorporated into
the proposed software tool allowed the identification of steady state solution branches (solid lines
132
in Fig. 19). However, it was not possible to calculate other crucial parts of the solution diagram.
To complement hazard identification of the presented ammonia plant and to ensure completeness
of the steady state multiplicity analysis, implementation of advanced numerical solving methods
into the proposed software tool seems to be pivotal.
5. Conclusions
The objective of our research is to enable process safety engineers to take advantage of process
simulators already implemented in modern chemical plants thus improving risk assessment
methods utilizing simulation-based HAZOP study software solution. Individual software
components providing the generation of HAZOP deviations and evaluation of simulated
consequences were introduced. As an external simulation engine, Aspen HYSYS was employed
as a representative commercial process simulator frequently used in chemical industry.
As a case study for software application, an industrial ammonia production plant based on an
existing plant located in the Slovak Republic was created in the Aspen HYSYS simulation
environment. The model consisted of syngas production via steam reforming, hydrogen
separation from syngas and an ammonia synthesis loop with a simplified ammonia separation
unit. Several simulation scenarios reflecting HAZOP deviations occurring in different parts of the
analyzed plant were generated. Complex fault propagation paths leading to hazardous events or
operability problems were identified in a semi-automatic manner using advanced mathematical
procedures, e.g. parametric sensitivity analysis and user set process-specific threshold values such
as the definition of improper molar ratio of reactants. The phenomenon of multiple steady states
was successfully detected by the proposed software tool and two solution branches for ammonia
synthesis loop were identified. The sequential character of Aspen HYSYS modeling enables
complex analysis of events only following HAZOP deviations, not leading to them. Therefore,
potential causes of corresponding HAZOP deviations had to be generated manually. To conclude
the performed simulation-based HAZOP study, a HAZOP-like report was generated for
individual HAZOP deviations.
The proposed software approach provides prediction of process behavior and it can be a very
useful tool not only in the plant operation but also plant design phase. Hazardous events and
operability problems are identified considering the HAZOP deviation size, which represents an
upgrade to the conventional HAZOP study where usually only the existence of a deviation is
considered and the complex fault propagation paths are difficult to predict. The software output
reports can also be used as a preliminary information basis for human HAZOP expert teams
performing a conventional HAZOP study.
133
Acknowledgments
Funding: This work was supported by the Slovak Scientific Agency [Grant No. VEGA
1/0659/18] and the Slovak Research and Development Agency [Grant No. APP-14-0317]. The
authors are also grateful for support from the project Science and Technology Park STU [Grant
No. ITMS26240220084], co-financed from the European Regional Development Fund.
References
Abbas, S.Z., Dupont, V., Mahmud, T., 2017. Kinetics study and modelling of steam methane
reforming process over a NiO/Al 2 O 3 catalyst in an adiabatic packed bed reactor. Int. J.
Hydrogen Energy 42, 2889–2903. doi:10.1016/j.ijhydene.2016.11.093
Ahmad, F., Lau, K.K., Shariff, A.M., Murshid, G., 2012. Process simulation and optimal
design of membrane separation system for CO2 capture from natural gas. Comput. Chem. Eng.
36, 119–128. doi:10.1016/j.compchemeng.2011.08.002
Aspelund, A., Gundersen, T., Myklebust, J., Nowak, M.P., Tomasgard, A., 2010. An
optimization-simulation model for a simple LNG process. Comput. Chem. Eng. 34, 1606–1617.
doi:10.1016/j.compchemeng.2009.10.018
AspenTech, 2017. Aspen Technology Engineering Products - Aspen HYSYS [WWW
Document]. URL http://home.aspentech.com/products/engineering/aspen-hysys (accessed
9.6.17).
Berdouzi, F., Olivier-Maget, N., Gabas, N., 2016. Using Dynamic Simulation for Risk
Assessment: Application to an Exothermic Reaction. Comput. Aided Chem. Eng. 38, 1563–1568.
doi:10.1016/B978-0-444-63428-3.50265-4
Berdouzi, F., Villemur, C., Olivier-Maget, N., Gabas, N., 2017. Dynamic Simulation for Risk
Analysis: Application to an Exothermic Reaction. Process Saf. Environ. Prot. 113, 1563–1568.
doi:10.1016/j.psep.2017.09.019
Cameron, I., Mannan, S., Németh, E., Park, S., Pasman, H., Rogers, W., Seligmann, B., 2017.
Process hazard analysis, hazard identification and scenario definition: Are the conventional tools
sufficient, or should and can we do much better? Process Saf. Environ. Prot. 110, 53–70.
doi:10.1016/j.psep.2017.01.025
Cui, X., Mannan, M.S., Wilhite, B.A., 2015. Towards efficient and inherently safer
continuous reactor alternatives to batch-wise processing of fine chemicals: CSTR nonlinear
dynamics analysis of alkylpyridines N-oxidation. Chem. Eng. Sci. 137, 487–503.
doi:10.1016/j.ces.2015.06.012
134
Danko, M., Frutiger, J., Jelemenský, Ľ., Sin, G., 2017. Monte Carlo Based Framework to
Support HAZOP Study. Comput. Aided Chem. Eng. 40, 2233–2238. doi:10.1016/B978-0-444-
63965-3.50374-3
Danko, M., Janošovský, J., Labovský, J., Jelemenský, Ľ., 2018. Fault propagation behavior
study of CSTR in HAZOP. Chem. Pap. 72, 515–526. doi:10.1007/s11696-017-0314-5
Danko, M., Janošovský, J., Labovský, J., Jelemenský, Ľ., 2016. Mathematical modelling and
stability investigation of multiple steady states in chemical reactors in the interest of automated
process safety analysis tool. Proc. 4th Int. Conf. Chem. Technol. ICCT 2016.
de la Mata, J.L., Rodríguez, M., 2012. HAZOP studies using a functional modeling
framework. Comput. Aided Chem. Eng. 30, 1038–1042. doi:10.1016/B978-0-444-59520-
1.50066-X
De Rademaeker, E., Suter, G., Pasman, H.J., Fabiano, B., 2014. A review of the past, present
and future of the European loss prevention and safety promotion in the process industries. Process
Saf. Environ. Prot. 92, 280–291. doi:10.1016/j.psep.2014.03.007
Dunjó, J., Fthenakis, V., Vílchez, J.A., Arnaldos, J., 2010. Hazard and operability (HAZOP)
analysis. A literature review. J. Hazard. Mater. 173, 19–32. doi:10.1016/j.jhazmat.2009.08.076
Eizenberg, S., Shacham, M., Brauner, N., 2006. Combining HAZOP with dynamic process
model development for safety analysis. Comput. Aided Chem. Eng. Volume 21, 389–394.
doi:10.1016/S1570-7946(06)80077-5
Enemark-Rasmussen, R., Cameron, D., Angelo, P.B., Sin, G., 2012. A simulation based
engineering method to support HAZOP studies. Comput. Aided Chem. Eng. 31, 1271–1275.
doi:10.1016/B978-0-444-59506-5.50085-7
Ghasemzadeh, K., Morrone, P., Iulianelli, A., Liguori, S., Babaluo, A.A., Basile, A., 2013.
H2 production in silica membrane reactor via methanol steam reforming: Modeling and HAZOP
analysis. Int. J. Hydrogen Energy 38, 10315–10326. doi:10.1016/j.ijhydene.2013.06.008
Giardina, M., Morale, M., 2015. Safety study of an LNG regasification plant using an FMECA
and HAZOP integrated methodology. J. Loss Prev. Process Ind. 35, 35–45.
doi:10.1016/j.jlp.2015.03.013
Häussinger, P., Lohmüller, R., Watson, A.M., 2000. Hydrogen, in: Ullmann’s Encyclopedia
of Industrial Chemistry. Wiley-VCH Verlag GmbH & Co. KGaA.
doi:10.1002/14356007.a13_297
Chen, Z., Wu, X., Qin, J., 2014. Risk assessment of an oxygen-enhanced combustor using a
structural model based on the FMEA and fuzzy fault tree. J. Loss Prev. Process Ind. 32, 349–357.
doi:10.1016/j.jlp.2014.10.004
135
Janošovský, J., Danko, M., Labovský, J., Jelemenský, Ľ., 2017a. The role of a commercial
process simulator in computer aided HAZOP approach. Process Saf. Environ. Prot. 107.
doi:10.1016/j.psep.2017.01.018
Janošovský, J., Danko, M., Labovský, J., Jelemenský, Ľ., 2017b. Smart software system
solution for model-based hazard identification of complex industrial processes. Comput. Aided
Chem. Eng. 40, 1225–1230. doi:10.1016/B978-0-444-63965-3.50206-3
Janošovský, J., Labovský, J., Jelemenský, L., 2016. Automated Model-based HAZOP study
in process hazard analysis, Chemical Engineering Transactions. doi:10.3303/CET1648085
Janošovský, J., Labovský, J., Jelemenský, Ľ., 2015. Ammonia synthesis fundamentals for a
model-based HAZOP study. Acta Chim. Slovaca 8, 5–10. doi:10.1515/acs-2015-0002
Jeerawongsuntorn, C., Sainyamsatit, N., Srinophakun, T., 2011. Integration of safety
instrumented system with automated HAZOP analysis: An application for continuous biodiesel
production. J. Loss Prev. Process Ind. 24, 412–419. doi:10.1016/j.jlp.2011.02.005
Khan, F., Rathnayaka, S., Ahmed, S., 2015. Methods and models in process safety and risk
management: Past, present and future. Process Saf. Environ. Prot. 98, 116–147.
doi:10.1016/j.psep.2015.07.005
Labovská, Z., Labovský, J., Jelemenský, Ľ., Dudáš, J., Markoš, J., 2014. Model-based hazard
identification in multiphase chemical reactors. J. Loss Prev. Process Ind. 29, 155–162.
doi:10.1016/j.jlp.2014.02.004
Labovský, J., Švandová, Z., Markoš, J., Jelemenský, Ľ., 2008. HAZOP study of a fixed bed
reactor for MTBE synthesis using a dynamic approach. Chem. Pap. 62, 51–57.
doi:10.2478/s11696-007-0078-4
Laššák, P., Labovský, J., Jelemenský, Ĺ., 2010. Influence of parameter uncertainty on
modeling of industrial ammonia reactor for safety and operability analysis. J. Loss Prev. Process
Ind. 23, 280–288. doi:10.1016/j.jlp.2009.10.001
Li, S., Bahroun, S., Valentin, C., Jallut, C., De Panthou, F., 2010. Dynamic model based safety
analysis of a three-phase catalytic slurry intensified continuous reactor. J. Loss Prev. Process Ind.
23, 437–445. doi:10.1016/j.jlp.2010.02.001
Li, S., Huang, D., 2011. Simulation and analysis on multiple steady states of an industrial
acetic acid dehydration system. Chinese J. Chem. Eng. 19, 983–989. doi:10.1016/S1004-
9541(11)60081-5
Mancusi, E., Merola, G., Crescitelli, S., Maffettone, P.L., 2000. Multistability and hysteresis
in an industrial ammonia reactor. AIChE J. 46, 824–828. doi:10.1002/aic.690460415
136
Morud, J., Skogestad, S., 1998. Analysis of instability in an industrial ammonia reactor.
AIChE J. 44, 888–895.
Pasman, H.J., 2015. Risk Analysis and Control for Industrial Processes - Gas, Oil and
Chemicals: A System Perspective for Assessing and Avoiding Low-Probability, High-
Consequence Events. Elsevier Science.
Perederic, O.A., Pleşu, V., Iancu, P., Bumbac, G., Bonet-Ruiz, A.-E., Bonet-Ruiz, J., Muchan,
B., 2015. Simulation and process integration for tert-amyl-methyl ether (TAME) synthesis.
Comput. Chem. Eng. 83, 79–96. doi:10.1016/j.compchemeng.2015.05.020
Pirola, C., Galimberti, M., Comazzi, A., Bozzano, G., Hillestad, M., Manenti, F., 2017.
Integrated reactor staging and plant optimization of a Biomass-To-Liquid technology. Comput.
Chem. Eng. 106, 719–729. doi:10.1016/j.compchemeng.2017.03.028
Rodríguez, M., de la Mata, J.L., 2012. Automating HAZOP studies using D-higraphs.
Comput. Chem. Eng. 45, 102–113. doi:10.1016/j.compchemeng.2012.06.007
Rossing, N.L., Lind, M., Jensen, N., Jørgensen, S.B., 2010. A functional HAZOP
methodology. Comput. Chem. Eng. 34, 244–253. doi:10.1016/j.compchemeng.2009.06.028
Seligmann, B.J., Németh, E., Hangos, K.M., Cameron, I.T., 2012. A blended hazard
identification methodology to support process diagnosis. J. Loss Prev. Process Ind. 25, 746–759.
doi:10.1016/j.jlp.2012.04.012
SPP Distribucia, 2016. Natural gas composition and emission factor [WWW Document].
URL http://www.spp-distribucia.sk/en_distribution-network/en_natural-gas-composition-and-
emission-factor (accessed 12.1.16).
SQLite, 2017. SQLite Database Engine [WWW Document]. URL https://www.sqlite.org/
(accessed 2.2.18).
Švandová, Z., Labovský, J., Markoš, J., Jelemenský, Ľ., 2009. Impact of mathematical model
selection on prediction of steady state and dynamic behaviour of a reactive distillation column.
Comput. Chem. Eng. 33, 788–793. doi:10.1016/j.compchemeng.2008.07.004
Tian, W., Du, T., Mu, S., 2015. HAZOP analysis-based dynamic simulation and its
application in chemical processes. Asia-Pacific J. Chem. Eng. 10, 923–935. doi:10.1002/apj.1929
Wu, J., Lind, M., Zhang, X., Jørgensen, S.B., Sin, G., 2015. Validation of a functional model
for integration of safety into process system design. Comput. Aided Chem. Eng. 37, 293–298.
doi:10.1016/B978-0-444-63578-5.50044-X
137
Príloha C
138
Predslov k prílohe C
Prílohou C dizertačnej práce je konferenčný príspevok z „15th International Symposium on
Loss Prevention and Safety Promotion“ s názvom Smart software system solution for model-
based hazard identification of complex industrial processes, ktorý bol osobne prezentovaný
formou prednášky. Chronologicky sa jedná o prvé predstavenie matematických metód
implementovaných do Modulu analýzy simulačných dát s dôrazom na predstavenie trojrozmernej
analýzy interpretovanej na zjednodušenom modeli syntézy amoniaku.
Z dôvodu transformácie A4 formátu do B5 formátu je prílohou prepis článku. Pre plnú verziu
článku v publikovanej verzii a s formátovaním daného zborníka je čitateľovi odporučené stiahnuť
článok online – https://doi.org/10.3303/CET1648085 (článok je voľne dostupný).
139
Automated Model-based HAZOP Study in
Process Hazard Analysis
Ján Janošovský, Juraj Labovský, Ľudovít Jelemenský*
Institute of Chemical and Environmental Engineering, Slovak University of Technology,
Bratislava, Slovakia
In the age of chemical processes operated at extreme pressures and temperatures it is
necessary to perform a detailed process safety analysis. Hazard and operability study (HAZOP)
is one of the most used and highly efficient techniques for the identification of potential hazards
and operability problems. Model-based HAZOP study is based on the implementation of a
detailed mathematical model of chemical productions. Possibilities and limitations of model-
based HAZOP study using Aspen HYSYS are discussed in this work. Results of numerical
simulations are collected and directly transformed into standard HAZOP tables. Advantages and
disadvantages of the presented software tool are shown in the process hazard analysis of a
chemical production with strong nonlinear behavior, an ammonia synthesis reactor system.
Introduction
Chemical industry together with other industrial sectors is subject to modernisation and
automation. These considerable changes push the chemical processes towards extreme operating
conditions (pressure, temperature, etc.). Therefore, conventional hazard analysis methods may
not be sufficient anymore. There are several process safety analysis techniques such as checklist
(CL), what-if (WI) analysis, failure modes and effects analysis (FMEA), fault tree analysis (FTA),
and hazard and operability (HAZOP) study. HAZOP study is currently recognised as one of the
most used and frequently modified process safety analysis methods (e.g. determination of the
required safety instrument level (Dowell, 1998), blended hazard identification (BLHAZID)
methodology (Seligmann et al., 2012), HAZOP analysis based on structural model
(Boonthum et al., 2014), innovative LOPA-based methodology with integration of a HAZOP
study (Argenti et al., 2015), etc.). HAZOP study is a highly sophisticated technique of hazard
identification based on monitoring process parameter deviations. The process parameter
deviation is created by combination of the guide word (more, less, etc.) and process parameter
(temperature, pressure, flow, etc.). The principal task of HAZOP analysis is to investigate the
causes, propagation and consequences of process variable deviations. The standard output of
"hazoping" is a list of all possible deviations, their causes and consequences, installed levels of
140
protection and recommendations for process safety improvement. (Kletz, 1997) The main
disadvantages of conventional process safety analysis methods including HAZOP are their time
consuming character, cost requirements and the demand for experienced and skilled human expert
team including safety engineers and process engineers for successful execution of process hazard
analysis.
Development of computer technology has created new possibilities to eliminate or reduce the
disadvantages of conventional process hazard analysis techniques. As indicated by the literature
review presented by Dunjó et al. (2010), approximately 40 % of HAZOP-related research is
focused on HAZOP automation. It is impossible to completely eliminate the presence of a human
expert team in the HAZOP execution process, but there are several attempts to create a robust
support tool that is able to automate some of the procedures necessary to perform a HAZOP study.
There are two basic approaches in HAZOP automation: knowledge-based and model-based.
Knowledge-based approach, dominant in the 20th century, uses large knowledge databases
containing information about the failure mode, causes and consequences of various process units
and/or pieces of equipment. Typical knowledge-based expert systems are e.g. HAZOPExpert, a
HAZOP automation tool developed by Venkatasubramanian and Vaidhyanathan (1994), projects
of OptHAZOP, TOPHAZOP and EXPERTOP (Khan and Abbasi, 2000), integration of
knowledge-based and mathematical programming approaches for process safety verification by
Srinivasan et al. (1997) or Automatic Hazard Analyzer (AHA) - an expert system based on multi-
model approach presented by Kang et al. (1999).
The model-based approach has gained more attention and importance in the 21st century. This
approach is based on the implementation of a detailed mathematical model of chemical
productions. The main benefit of using the model-based hazard analysis tool is the possibility of
complex overview of the analysed process limited only by the reliability of the mathematical
model. Mathematical modeling allows the user to consider not only the presence of the deviation,
but also its value and duration. Several benefits of model-based HAZOP automation are
demonstrated by combining the HAZOP technique with dynamic simulations employing
MATLAB in the work of Eizenberg et al. (2006) and in the article about the integration of human-
machine interface with automated HAZOP analysis using Aspen Plus® version 2006.5 proposed
by Jeerawongsuntorn et al. (2011). Principal issues of safety analysis utilizing mathematical
modeling have been discussed by Molnár et al. (2005) and several attempts of combining a
HAZOP study with mathematical models of process equipment based on this article were made.
Articles about mathematical model of chemical reactors as a useful complement in the HAZOP
study (Švandová et al., 2005) have been published. The use of a mathematical model of a
141
chemical reactor in safety analysis using the HAZOP methodology is also presented in the work
of Labovský et al. (2007a). The research of this team was later focused on HAZOP studies of real
chemical plants with nonlinear behavior, e.g. a MTBE plant (Labovský et al., 2007b). The
possibilities of model-based hazard identification in chemical reactors were summarised and
discussed by Labovská et al. (2014). Reliability of model-based HAZOP is strongly dependent
on the selection of an appropriate mathematical model and its parameters describing the
physicochemical behavior of individual components and their mixtures. These issues were
investigated by Laššák et al. (2010).
The goal of the presented paper is to propose the fundamentals of combining model-based
HAZOP analysis with the simulation environment optimised for process safety engineering
including the use of Aspen HYSYS v8.4 process modelling. Aspen HYSYS provides one of the
most extensive property databases and the possibility of data transfer between external software
and internal simulation environment. It will be shown, that presented software tool facilitates and
accelerates the execution of a process safety analysis. After the process deviation effect is
simulated, the simulation data are collected and processed applying the principles of the HAZOP
methodology. It will be demonstrated that the presented automated model-based HAZOP tool is
able to handle and examine processes operated in parametrically sensitive region that can lead to
process hazards and operating problems such as the ammonia synthesis reactor incident in 1989
(Morud and Skogestad, 1998).
Software methodology
The proposed software tool consists of two separate parts with shared classes and databases
(Figure 1). The first part is used for the actual simulation of the analysed system. In this software
module, the connection of our tool with the Aspen HYSYS simulation environment is established
and when the simulation case is open and active, individual streams and operation units are
checked for the possibility of performing a HAZOP study. After this primary control, the user
can select the desired material streams and create process variable deviations. When the final list
of process variable deviations is created, the simulation section is initiated. The list is transferred
to the internal database where deviations are stored.
After this procedure, the user is allowed to start the simulation. The information containing
the stream identification number, its parameter and deviation value is sent from the analysis tool
to the internal environment of Aspen HYSYS v8.4, where the process simulation is done. After
each simulation of variable deviation, the footprint of the current process state is created. This
footprint contains values of important process variables such as temperature, pressure, flow,
composition etc. of each stream and the live reference to the HYSYS environment. When all
142
selected process variable deviations are investigated, the investigation part of the presented
HAZOP analysis tool is finished.
Figure 1: Schematic description of the presented automated model-based HAZOP analysis
tool
The second part of our analysis tool is designed for the simulation data analysis. The depth of
the analysis is optional and depends on the user's choice. In this part, the shared internal database
belonging to the analysed system is loaded. Process footprints for each simulated deviation are
decomposed and user is able to investigate the system response to the deviated parameter.
Examination questions can be formulated as: Is the reaction terminated? Is the runaway effect
possible? Which stream is the most parametrically sensitive? Is there a possibility of unexpected
vapor fraction occurrence in the process? After the analysis, a HAZOP-like report is generated.
The possibilities and limitations of both modules of our analysis tool are demonstrated on one
example. The demonstrational example is a case study of a complex ammonia synthesis reactor
system focused on the ammonia synthesis nonlinearity and providing insight into the robustness
and advantages of our automated model-based HAZOP analysis tool.
Application to a case study - ammonia synthesis unit
Ammonia synthesis unit is well known for its nonlinear character documented e.g. by Morud
and Skogestad (1998). The mathematical model of an ammonia production plant consists of
fixed-bed reactor divided into three separate beds, a feed preheater and a refrigeration system
with a vapor-liquid separator (Figure 2). The feed stream is transferred to the splitter where the
feed is divided into four outlet streams. One outlet stream is led to the feed preheater as the cool
medium and three other outlet streams are parts of fresh feed quenching between each bed to
achieve the optimal temperature profile. As the hot medium in the feed preheater, the product
143
stream leaving the reactor system is used. The cooled product stream is led into the refrigeration
unit with a phase separator, where the products are separated into two phases, gaseous purge and
liquid ammonia. Relevant parameters of the selected material streams and their design values are
presented in Table 1.
Figure 2: Ammonia synthesis unit in Aspen HYSYS v8.4 simulation environment
Table 1: Design parameters of the ammonia synthesis unit
Stream name fresh
feed
4 quench
1
quench
2
1a R101out R102out R103out 6 liquid
ammonia
Temperature
[°C]
250 250 250 250 424 520 530 525 436 8
Pressure
[MPa]
20 20 20 20 20 20 20 20 20 20
Mass flow
[10³ kg/h]
252 127 58 35 127 185 220 252 252 50
Mole fraction
Nitrogen 0.239 0.239 0.239 0.239 0.239 0.215 0.210 0.209 0.209 0.003
Hydrogen 0.719 0.719 0.719 0.719 0.719 0.645 0.630 0.625 0.625 0.015
Ammonia 0.042 0.042 0.042 0.042 0.042 0.140 0.160 0.166 0.166 0.982
The conventional process hazard analysis of this chemical plant can be carried out using a
standard HAZOP study. The human expert team should hold meetings where the possibilities and
consequences of every potential process variable deviation have to be considered. This part of
"hazoping" is the most time consuming part. Some of the team members can adapt incorrect
conclusions and overlook potentially hazardous consequences resulting from the lack of
experience. Also some consequences can be overlooked completely because of insufficient
knowledge of the examined process (like the Seveso disaster) or too complicated failure
propagation. With the use of our analysis tool, these inconveniences are significantly reduced.
Our model-based HAZOP tool directly helps the human expert team with answering the majority
144
of the important questions. After selecting the target streams, their parameters and the deviation
range, the deviation list is generated and the process simulation takes place. Then, simulation data
are evaluated with the assistance of the human expert team. For this case study, the effect of
temperature, pressure and composition of each material stream was investigated. Because of the
huge amount of the simulation data, the stream “fresh feed” and its parameter temperature were
selected for the demonstration in the next section.
Results and discussion
Effect of feed temperature on simulated ammonia synthesis unit
Temperature in the fixed-bed reactor affects the overall conversion of reactants and the total
production of ammonia respectively. The reaction in every bed of the reactor was carried out at a
constant pressure of 20 MPa. The feed temperature deviation list was generated using the classic
HAZOP approach of combining guide words "more" and "less" with the parameter "temperature"
resulting in classic HAZOP deviations "higher feed temperature than design value" and "lower
feed temperature than design value". The developed model-based HAZOP analysis tool required
to consider the value of the parameter deviation 𝐷𝑃 defined in Eq(1), where 𝑃𝑂 is the original
non-deviated value of the parameter 𝑃 and 𝑃𝑁 is the new value of parameter 𝑃 after the
deviation is applied to the process.
𝐷𝑃 =𝑃𝑁−𝑃𝑂
𝑃𝑂×100 (1)
The range of feed temperature deviation 𝐷𝑇 was from + 30 % to - 30 % in this case study,
which means that the feed temperature varied from 175 °C to 325 °C. Figure 3a shows the effect
of the feed temperature on the “R103out” (see Figure 2) temperature, the absolute response.
Figure 3b presents the effect of the feed temperature deviation on the “R103out” temperature
deviation, the relative response. In Figure 3c, the effect of the feed temperature on the rate of
“R103out” temperature change defined as the change of the “R103out” temperature divided by
the corresponding change of the feed temperature is depicted.
As shown in Figure 3, an increase of the feed temperature causes gradual increase of the
reactor outlet temperature. Since the exothermic reaction is enhanced at lower temperatures, when
the temperature in the reactor system was decreased, the overall ammonia production increased.
However, a step change (a decrease of ca. 60 %) of the operating temperature of “R103out” in
the region of the feed temperature deviation of around 18 % was observed. This step decrease of
the reactor outlet temperature indicates the system switch to a steady state with lower reaction
rates. The shift between different steady states led to the termination of the reaction and a new
145
reactor start-up was required. The phenomnenon of multiple steady states in the ammonia
synthesis similar to that one described by Morud and Skogestad (1998) was observed.
Figure 3: Effect of feed temperature deviation on the temperature of stream “R103out” (a –
absolute response, b –relative response, c –rate of “R103out” temperature change, see Figure
2)
Similar behavior was exhibited by the operating temperatures of the reactor beds outlet and
inlet streams and the feed preheater outlet and inlet streams. For the analysis of the overall process
response, the effect of the feed temperature deviation on the temperatures of material streams in
the process is depicted in Figure 4. The color intensity is dependent on the parameter deviation
value according to the color scale in the right section of Figure 4. The temperature change of
streams "quench1", "quench2", "quench3" and "4" directly mirrored the feed temperature
deviation, because these streams are outlet streams from a splitter where the only input stream is
the fresh feed. Temperature of streams "7", "purge" and "liquid ammonia" is not affected, because
the heat removal in the refrigeration unit is set to the output temperature of 8 °C.
146
Figure 4: Effect of feed temperature deviation throughout the process
Model-based process hazard analysis results
Other significant process variable deviations and their detected consequences are summarized
in Table 2. As shown, the key consequences of process variable deviations in the process of the
presented ammonia synthesis unit are operational problems caused by switching between
different steady states. Investigation of the effect of the mole ratio of one component to another
one was carried out at a constant mass flow of the analyzed stream.
Table 2: HAZOP report example from model-based process hazard analysis of the ammonia
synthesis unit
Deviation from design intent Deviation value 𝐷𝑃 Detected consequences
"lower feed temperature" - 18 % Overall ammonia production
decreased by 99,5 %
"lower feed pressure" - 40 % Overall ammonia production
decreased by 98,2 %
"lower heat removal in
refrigeration" ("lower coolant
flow")
- 11 % Yield of ammonia in the phase
separator decreased by 45 %
"higher feed mole ratio
hydrogen/nitrogen"
+ 54 % Overall ammonia production
decreased by 97,1 %
147
Conclusions
In this work, a software tool combining a chemical unit with model-based HAZOP analysis
was presented. The integration of automated model-based HAZOP study can potentially lead to
the identification of some unexpected deviations and to the reduction of time required for the
process hazard analysis. It can be successfully applied at the beginning stage of the unit design,
for the operation of an already existing unit and also for the training of operators. It was
demonstrated that the simulation of a chemical unit using an appropriate mathematical model is
a suitable tool for safety analysis.
In the presented case study, mathematical modelling of ammonia synthesis in the Aspen
HYSYS v8.4 simulation environment was done which allowed identifying parametric zones,
where shifting between qualitatively different steady states can be expected. Operational problem
of the reaction terminated in a lower steady state was identified. Results of the presented
automated model-based HAZOP tool applied to an ammonia synthesis unit were shown to
exemplify the application. It is important to note that each uncertainty of a model input parameter
can significantly influence the results of a model-based HAZOP study.
Acknowledgments
This work was supported by the Slovak Scientific Agency, Grant No. VEGA 1/0749/15 and
the Slovak Research and Development Agency APP-14-0317.
Reference
Argenti F., Brunazzi E., Landucci G., 2015, Innovative lopa-based methodology for the safety
assessment of chemical plants, Chemical Engineering Transactions, 43, 2383-2388, DOI:
10.3303/CET1543398
Boonthum N., Mulalee U., Srinophakun T., 2014, A systematic formulation for HAZOP
analysis based on structural model, Reliab. Eng. Syst. Saf. 121, 152-163.
Dowell III A.M., 1998, Layer of protection analysis for determining safety integrity level,
ISA Trans. 37, 155-165.
Dunjó J., Fthenakis V., Vílchez J. A., Arnaldos J., 2010, Hazard and operability (HAZOP)
analysis. A literature review, J. Hazard. Mater. 173, 19-32.
Eizenberg S., Shacham M., Brauner N., 2006, Combining HAZOP with dynamic simulation
- applications for safety education, J. Loss Prev. Process Ind. 19, 754-761.
Jeerawongsuntorn C., Sainyamsatit N., Srinophakun T., 2011, Integration of safety
instrumented system with automated HAZOP analysis: An application for continuous biodiesel
production, J. Loss Prev. Process Ind. 24, 412-419.
148
Kang B., Lee B., Kang K., Suh J., Yoon E., 1999, AHA: a knowledge based system for
automatic hazard identification in chemical plant by multi-model approach, Expert Syst. Appl.
16, 183-195.
Khan F.I., Abbasi S.A., 2000, Towards automation of HAZOP with a new tool EXPERTOP,
Environ. Modell. Software 15, 67-77.
Kletz T.A., 1997, HAZOP - past and future, Reliab. Eng. Syst. Saf. 55, 263-266.
Labovská Z., Labovský J., Jelemenský L., Dudáš J., Markoš J., 2014, Model-based hazard
identification in multiphase chemical reactors, J. Loss Prev. Process Ind. 29, 155-162.
Labovský J., Laššák P., Markoš J., Jelemenský L., 2007a, Design, optimization and safety
analysis of a heterogeneous tubular reactor by using the HAZOP methodology, Comput. Aided
Chem. Eng. 24, 1241-1246.
Labovský J., Švandová Z., Markoš J., Jelemenský L., 2007b, Model-based HAZOP study of
a real MTBE plant, J. Loss Prev. Process Ind. 20, 230-237.
Laššák P., Labovský J., Jelemenský L., 2010, Influence of parameter uncertainty on modeling
of industrial ammonia reactor for safety and operability analysis, J. Loss Prev. Process Ind. 23,
280-288.
Molnár A., Markoš J., Jelemenský L., 2005, Some considerations for safety analysis of
chemical reactors, Chem. Eng. Res. Des. 83, 167-176.
Morud J., Skogestad S., 1998, Analysis of instability in an industrial ammonia reactor, AIChE
J. 44, 888-895.
Seligmann B.J., Németh E., Hangos K.M., Cameron I.T., 2012, A blended hazard
identification methodology to support process diagnosis, J. Loss Prev. Process Ind. 25, 746-759.
Srinivasan R., Dimitriadis V.D., Shah N., Venkatasubramanian V., 1997, Integrating
knowledge-based and mathematical programming approaches for process safety verification,
Comput. Chem. Eng. 21, S905-S910.
Švandová Z., Jelemenský L., Markoš J., Molnár A., 2005a, Steady states analysis and dynamic
simulation as a complement in the HAZOP study of chemical reactors, Process Saf. Environ.
Prot. 83, 463-471.
Venkatasubramanian V., Vaidhyanathan R., 1994, A knowledge-based framework for
automating HAZOP analysis, AIChE J. 40, 496-505.
149
Príloha D
150
Predslov k prílohe D
Príloha D dizertačnej práce je konferenčný príspevok z „5th International Conference on
Chemical Technology“ s názvom Multilevel data analysis in computer aided hazard
identification, ktorý bol osobne prezentovaný formou prednášky. V príspevku je predstavené
grafické užívateľské rozhranie Modulu analýzy simulačných dát. Súčasťou príspevku sú ukážky
výstupov rôznych analýz aplikovaných na dvoch prípadových štúdiách.
Z dôvodu transformácie A4 formátu do B5 formátu je prílohou prepis článku. Pre plnú verziu
článku v publikovanej verzii a s formátovaním daného zborníka je čitateľovi odporučené stiahnuť
zborník online – http://www.icct.cz/predchozi-konference/2017 (zborník je voľne dostupný).
151
Multilevel data analysis in computer aided hazard
identification
Janošovský J., Danko M., Labovský J., Jelemenský Ľ.
Institute of Chemical and Environmental Engineering, Slovak University of Technology,
Radlinského 9, 812 37 Bratislava, Slovakia
Abstract
Hazard identification techniques in chemical industry are constantly being improved by
advanced computer simulations of chemical plants. Output from computer simulations is a large
set of simulation data containing relevant information about individual streams and units involved
in the simulated plant such as flow, temperature, pressure, composition, etc. This data is
frequently used in process intensification activities but can be also exploited for process safety
improvement. Software structure and simulation data analysis methods appropriate for computer
aided hazard identification are discussed in this paper. The HAZOP (HAZard and OPerability
study) methodology was adopted to generate process variables deviations and to evaluate their
consequences via multilevel analysis comprising optimised numerical procedures as well as
several graphical interpretations. Software features are previewed in application to two case
studies. Presented approach allows investigating complex chemical processes from the safety
engineering point of view in more depth and provides more effective analysis of complex fault
propagation paths.
Introduction
The constant growth of chemical industry led to the increase of manufacturing processes
complexity to achieve higher product yields and purity at lower costs. Therefore, appropriate
process safety analysis has become one of the most important aspects in plant design and
operation. With the development of CAPE/PSE (computer aided production engineering/process
systems engineering) tools, the demand for computer aided hazard identification has also
increased. Several hazard identification techniques are well established in industrial companies’
policy such as What-If analysis, Checklist, FMEA (Failure modes and analysis) and HAZOP
(HAZard and OPerability study)1,2. Structural and systematic approach of these techniques
qualify them as potential candidates for computer aided hazard identification3,4. In our work,
152
HAZOP principles were chosen to be implemented in software solution because of its robustness
and wide application in chemical industry5.
HAZOP methodology is based on generation of process variable deviations from design intent
and analysis of their consequences. Process variable deviations are created by the combination of
standardised guide words (No, More, Less, As Well As, Part Of, Reverse, and Other Than) and
appropriate process variables (temperature, pressure, flow, level …)6. Although conventional
HAZOP is considered as the most comprehensive hazard identification procedure, various
drawbacks of this method have been noticed by experienced practitioners, e.g. uncommon
hazards overlooking, significant time-consuming character, insufficient design intent definition,
considerations of redundant deviations not leading to scenarios of concern, etc.7 Some
conventional HAZOP drawbacks can be reduced or fully eliminated by implementing simulation-
based approach.
In this paper, a smart software system utilizing HAZOP principles and mathematical
modelling of common chemical processes is introduced. The presented software was tested in
combination with the simulation platform represented by Aspen HYSYS – a commercial process
simulator widely used in chemical industry, particularly in oil and gas industry. The HAZOP
methodology served as a tool for the generation of simulation inputs (HAZOP deviations).
Severity of the simulated process states after HAZOP deviation occurrence (HAZOP
consequences) was determined by multilevel simulation data analysis comprising optimised
numerical procedures. Examples of software graphical user interface (GUI) are also provided.
Case studies
Mathematical models of examined processes prepared in the corresponding simulation
platform are necessary for computer aided hazard identification using the proposed software
solution. For the demonstration of software application variability, two manufacturing processes
employing different reactor types were analysed. Mathematical models of an ammonia synthesis
reactor (Figure 1) and a glycerol nitration process (Figure 2) built in Aspen HYSYS were selected
as case studies. Detailed overview of their model parameters and design intent conditions
considered for HAZOP can be found in our previous works8,9. Implemented reaction kinetic
models in both case studies were verified by experimental results and by comparison with data
from industrial operation10,11.
153
Figure 1. Model of ammonia synthesis reactor built in Aspen HYSYS environment
Figure 2. Model of glycerol nitration process built in Aspen HYSYS environment
Software application – simulation phase
After the reliability and validity of mathematical models were confirmed, the proposed
software tool could initiate the HAZOP analysis. At first, connection with the corresponding
Aspen HYSYS case file was established and information about individual HAZOP nodes was
accessed. The user can browse through three different types of HAZOP nodes: material streams,
energy streams and unit operations (Figure 3). As shown, the software tool found six unit
operations (Figure 3a) suitable for HAZOP analysis of the ammonia synthesis reactor (unit V-
100 represented phase separator which text label in Aspen HYSYS flowsheet (Figure 1) was
hidden) and seven material streams (Figure 3b) suitable for HAZOP analysis of the glycerol
nitration process.
After successful access to Aspen HYSYS data, generation of HAZOP deviations was allowed.
The user can apply logic guide words to any of the permitted process variables to create a list of
HAZOP deviations. Example of the HAZOP deviation list for the ammonia synthesis reactor is
provided in Figure 4 where HAZOP deviations “higher/lower pressure of fresh feed” and
“higher/lower content of nitrogen in fresh feed” were created and stored. Currently, only the
application of quantitative guide words is implemented. The correct use of qualitative guide
154
words in computer aided approach is very limited because of their imprecise definition and
therefore practically infinite possibilities of their interpretation. In the next step, the user can
assign a value range to selected HAZOP deviations. When the final HAZOP deviation list is
completed, the proposed software tool proceeds into the process simulation phase. Stored
HAZOP deviations are selected one-by-one and inserted to Aspen HYSYS. Once the simulated
process correctly converged to a new state, configuration of the Aspen HYSYS simulation case
file, i.e. HAZOP consequence, is assigned to the corresponding HAZOP deviation and stored for
severity determination via multilevel simulation data analysis.
Figure 3. Example of HAZOP nodes’ parameters display in GUI of the proposed software
tool for ammonia synthesis reactor (a) and glycerol nitration process (b)
Figure 4. Example of HAZOP deviation list in GUI of the proposed software tool for ammonia
synthesis reactor
155
Software application – data analysis
Evaluation of HAZOP consequences’ severity is performed in a simulation data analysis
module of the proposed software tool. It employs advanced numerical algorithms for the
automated HAZOP analysis in optional combination with the analysis and monitoring of process-
or unit-specific safety restrictions provided by the user. These two approaches are implemented
for partial automation of the investigation procedure. When the investigation procedure is
completed, the identified hazards and operability problems are assigned to a HAZOP
consequence and stored.
In GUI of the presented software tool, the user can browse through several types of analysis.
An example of such analyses is depicted in Figures 5 and 6. Figure 5 represents the hazard
identification procedure for a process exhibiting nonlinear behaviour with steady state
multiplicity and Figure 6 represents the hazard identification procedure for a process exhibiting
nonlinear behaviour without steady state multiplicity.
The first type of analysis monitors the effect of the HAZOP deviation value on one parameter
of one HAZOP node. This analysis consists of three supplementary methods – analysis of
absolute parameter change from the design intent, analysis of relative parameter change from the
design intent and parametric sensitivity analysis. Parametric sensitivity analysis allows capturing
a sudden change of the process parameter that indicates e.g. the presence of steady state
multiplicity in examined system. Figure 5a and Figure 6a show the difference in the parametric
sensitivity analysis outputs for a system with and without steady state multiplicity. The second
type of analysis monitors the effect of one HAZOP deviation value on selected parameters of one
HAZOP node. The default mode of this type of analysis for graphical interpretation depicts
relative change of selected parameters from the design intent (Figure 5b and Figure 6b). This
analysis allows a more detailed overview of the overall response of one HAZOP node for
particular HAZOP deviation and thus reduces the possibility of hazardous parameter change
overlooking. The third possible analysis is focused on monitoring a change of one parameter of
the selected HAZOP nodes for one HAZOP deviation value (Figure 5c and Figure 6c). This
method provides in-depth analysis of the deviation propagation path through the examined
system.
Identified hazardous events and significant operability problems are formulated in a
simplified HAZOP-like report that can serve as a preliminary analysis and supporting material
for human expert HAZOP teams to detect complicated deviation propagation paths in modern
complex production systems and thus to reduce time requirements of hazard identification in
modern industrial manufactures.
156
Figure 5 – Example of multilevel simulation data analysis in GUI of the proposed software
tool for ammonia synthesis reactor (a – parametric sensitivity analysis, b – relative change of
selected parameters for one HAZOP node, c – relative change of one parameter for selected
HAZOP nodes)
157
Figure 6 – Example of multilevel simulation data analysis in GUI of the proposed software
tool for glycerol nitration process (a – parametric sensitivity analysis, b – relative change of
selected parameters for one HAZOP node, c – relative change of one parameter for selected
HAZOP nodes)
158
Conclusion
Application of a smart software tool for automated model-based hazard identification in two
case studies was presented. Aspen HYSYS was used as the simulation engine and HAZOP was
used as the hazard identification method because of their frequent and successful application in
chemical industry. The proposed simulation-based approach enables identification of process
hazards and operability problems considering the HAZOP deviation size and represents an
upgrade to the conventional HAZOP study where usually only the existence of a deviation is
considered. A set of presented graphical interpretations of deviation propagation in systems with
and without the presence of multiple steady states phenomena demonstrated the variability and
robustness of the hazard identification procedure performed by the proposed software tool. The
developed tool can be easily adapted for other chemical plants using the general modelling
environment of Aspen HYSYS. Future research will be focused on the development of HAZOP
consequences ranking system for more effective hazard assessment and on the proposal of a new
simulation engine optimised for the purposes of hazard identification.
Acknowledgment
This work was supported by the Slovak Scientific Agency, Grant No. VEGA 1/0749/15 and
the Slovak Research and Development Agency APP-14-0317.
Literature
1. Mannan S.: Lees’ Loss Prevention in the Process Industries: Hazard Identification,
Assessment and Control. Elsevier Science, Oxford 2012.
2. Occupational Safety and Health Administration: Process Safety Management (OSHA
3132). U.S. Department of Labor, Washington 2000.
3. Dunjó J., Fthenakis V., Vílchez J. A., Arnaldos J.: J. Hazard. Mater. 173, 21 (2010).
4. Khan F., Rathnayaka S., Ahmed S.: Process Saf. Environ. Prot. 98, 116 (2015).
5. Crawley F., Tyler B.: HAZOP: Guide to Best Practice, 3rd edition. Elsevier Science,
Oxford 2015.
6. Kletz T. A.: Reliab. Eng. Syst. Saf. 55, 263 (1997).
7. Baybutt, P.: J. Loss Prev. Process Ind. 33, 52 (2015).
8. Janošovský J., Labovský J., Jelemenský Ľ.: Acta Chim. Slovaca 8, 5 (2015).
9. Janošovský J., Danko M., Labovský J., Jelemenský Ľ.: Process Saf. Environ. Prot. 107,
12 (2017).
10. Morud J., Skogestad S.: AIChE J. 44, 889 (1998).
11. Lu K., Luo K., Yeh T., Lin P.: Process Saf. Environ. Prot. 86, 37 (2008).
159
Príloha E
160
Predslov k prílohe E
Príloha E dizertačnej práce je konferenčný príspevok z „5th International Conference on
Chemical Technology“ s názvom Inherently safer design of a novel industrial scale reactor
for alkylpyridine derivatives production, ktorý bol prezentovaný formou posteru. V príspevku
je predstavený scale-up laboratórnej výroby 3-metylpyridín-N-oxidu a následne je na zostavenom
matematickom modeli vykonaná identifikácia nebezpečenstva determinujúca bezpečné
prevádzkové podmienky. Súčasťou analýzy je i prevernie spoľahlivosti zvoleného modelu
a dopadu nepresností jeho parametrov na výsledky identifikácie nebezpečenstva.
Z dôvodu transformácie A4 formátu do B5 formátu je prílohou prepis článku. Pre plnú verziu
článku v publikovanej verzii a s formátovaním daného zborníka je čitateľovi odporučené stiahnuť
zborník online – http://www.icct.cz/predchozi-konference/2017 (zborník je voľne dostupný).
161
Inherently safer design of a novel industrial scale
reactor for alkylpyridine derivatives production
Janošovský J., Kačmárová A., Danko M., Labovský J., Jelemenský Ľ.
Institute of Chemical and Environmental Engineering, Slovak University of Technology,
Radlinského 9, 812 37 Bratislava, Slovakia
Abstract
Alkylpyridines and their derivatives are chemical compounds widely used in pharmaceutical
industry and agriculture. In recent years, alkylpyridine-N-oxides have received attention due to
their increased reactivity provided by the N-oxide group. In our paper, design of an industrial
scale continuously stirred tank reactor for production of 3-methylpyridine-N-oxide with the focus
on process safety was discussed. 3-methylpyridine was converted into 3-methylpyridine-N-oxide
by homogeneously catalysed reaction in the presence of hydrogen peroxide as the oxidizing agent
and phosphotungstic acid as the catalyst. Reactor dimensions were proposed based on a scale-up
of a laboratory unit. Sensitivity and uncertainty analyses of selected key process parameters were
performed to determine optimal operating point within the safety constraints. The proposed
continuous process presents a suitable inherently safer alternative to conventional semi-batch
production.
Introduction
With the recent development in process intensification activities, detection of possible
hazardous events and operability problems has become more difficult. One of the basic concepts
of process intensification is the transition towards continuous productions. Continuous reactor is
a preferred alternative to batch or semi-batch reactor not only from the economic point of view
(size reduction), but also because of its inherently safer character1,2. Inherent safety principles
have been applied to the N-oxidation process and design of continuous production of 3-
methylpyridine-N-oxide (3-MPNOX) in a continuous stirred-tank reactor (CSTR) as an
alternative to the conventional semi-batch process is presented. 3-MPNOX belongs to
alkylpyridine derivatives that are frequently used in pharmaceutical industry due to their
increased reactivity provided by the N-oxide group.
This paper compiles necessary activities for safe reactor operation and optimization of the
reaction conditions towards higher product yield utilizing computer simulations. Key operating
162
parameters and construction dimensions of CSTR were proposed based on a scale-up of a
laboratory unit and consequently optimized based on sensitivity analyses and a complex process
hazard identification procedure. Mathematical model for process simulation was built in the
MATLAB® environment. Impact of model parameters’ uncertainties on the optimization results
and proposed operating points of reactor was also studied.
Case study
In pharmaceutical industry, 3-MPNOX (C6H7NO) is produced by the N-oxidation of 3-
methylpyridine (C6H7N, 3-MP) in the presence of phosphotungstic acid (H3PW12O40) as a metal
catalyst and aqueous hydrogen peroxide (H2O2) solution as an oxidizing agent (Equation 1)3. N-
oxidation is carried out at temperatures close to the boiling point of the reaction mixture in an
open semi-batch reactor to allow discharge of oxygen generated by competitive decomposition
of hydrogen peroxide (Equation 2)4. Recent research proposed transition from a semi-batch to a
continuous reactor at elevated pressure (200 – 300 kPa) and temperature (110 – 125 °C) to achieve
more efficient N-oxidation with inherently safer operation5. A scheme of the proposed
manufacturing process is depicted in Figure 1. For the proposed reactor configuration, hydrogen
peroxide decomposition reaction is significantly reduced and can be neglected5,6. Although N-
oxidation is a complex reaction system, reaction rate for the given range of pressures and
temperatures can be calculated from Equation 3 representing simplified reaction kinetic model
where C represents the molar concentration of the corresponding component. Values of kinetic
parameters used in this case study are summarized in Table I3,5. Constant reaction enthalpy of N-
oxidation of – 160×103 J.mol-1 was considered. Key operating parameters were taken from a
laboratory unit model with the reaction mixture volume of 1 L6. After the appropriate scale-up
(Tables II and III), further process intensification and hazard identification were performed.
Mathematical modeling of products’ separation and purification steps is not discussed in this
paper.
𝐶6𝐻7𝑁 +𝐻2𝑂2𝐻3𝑃𝑊12𝑂40→ 𝐶6𝐻7𝑁𝑂 + 𝐻2𝑂 (1)
2𝐻2𝑂2𝐻3𝑃𝑊12𝑂40→ 2𝐻2𝑂 + 𝑂2 (2)
rV =k1aKbC3−MPC𝐻2𝑂2Ccatalyst
1+KbC𝐻2𝑂2+ k1aC3−MPC𝐻2𝑂2 (3)
163
Figure 1. Simplified process scheme of 3-MPNOX production
Table I
Reaction kinetic data
Arrhenius equation Ai
[L.mol-1.s-1]
𝑬𝟏𝒊𝑹,∆𝑯𝒃𝑹
[K]
𝐤𝟏𝐚 = 𝐀𝟏𝐚𝒆𝒙𝒑 (−𝐄𝟏𝐚𝑹𝑻) 3.23×103 3 952
𝐤𝟏𝒃 = 𝐀𝟏𝒃𝒆𝒙𝒑 (−𝐄𝟏𝒃𝑹𝑻) 1.66×1012 12 989
𝐊𝐁 = 𝐀𝑩𝒆𝒙𝒑 (−∆𝑯𝒃𝑹𝑻
) 8.12×1010 7 927
Table II
Reactor inlet and outlet streams after the scale-up of a laboratory unit
Process variable Feed 1 Feed 2 Products
Mass flow [kg.h-1] 219.9 151.3 371.2
Temperature [°C] 50 50 118.9
Ma
ss c
om
po
siti
on
[%]
3-MP 100 - 1.6
H2O2 - 55.9 1.6
H2O - 44.1 29.1
3-MPNOX - - 67.7
164
Table III
Selected parameters of CSTR after the scale-up of a laboratory unit
Reaction
mixture
volume [L]
Molar
ratio of
H2O2 : 3-
MP [-]
Agitator
speed
[rpm]
Cooling water
inlet
temperature
[°C]
Cooling water
mass flow [103
kg.h-1]
Overall heat
transfer
coefficient
[W.K-1.m-1]
1 000 1.05 180 25 12.3 255
Process intensification, hazard identification and results analysis
The goal of process intensification is to maximize the production rate of 3-MPNOX. The
effect of feed temperature, molar ratio of H2O2 : 3-MP, catalyst concentration and cooling
medium inlet temperature was studied. Process variable „feed temperature“ represents the
temperature of streams Feed 1 and Feed 2 leaving the heat exchanger (e.g. operating setpoint for
both feed streams). Figure 2 represents one of the obtained results from two-parametric
optimization of reaction conditions where the production rate of 3-MPNOX as a function of feed
temperature and molar ratio of H2O2 : 3-MP is depicted. As it is shown, increase of feed
temperature and decrease of molar ratio of H2O2 : 3-MP leads to gradual increase of 3-MPNOX
production rate.
Figure 2. Effect of feed temperature and molar ratio of H2O2 : 3-MP on 3-MPNOX production
rate
165
However, process safety limitations have to be considered. As previously mentioned,
operating regime of the reactor was determined in the temperature range from 110 °C to 125 °C.
If the reactor temperature exceeds 125 °C, vaporization of the reaction mixture can occur, which
leads to possible over-pressurization of the reactor. Below 110 °C, secondary reaction of H2O2
decomposition is triggered and thermal runaway occurs. Therefore, safe operating regime has to
be monitored. Possible application of these safety constraints is depicted in Figure 3, where the
temperature in reactor was analyzed. Red zone represents hazardous operating regime and
yellow-to-green zone represents the safe one.
Figure 3. Effect of feed temperature and molar ratio of H2O2 : 3-MP on the temperature in the
reactor after safety constrictions’ application (red zone – hazardous operating regime)
As it can be seen in Figure 3, only a limited number of simulated steady states of CSTR can
be considered safe. Numerical algorithm for finding maximum production rate of 3-MPNOX (in
the matrix visualized in Figure 2) considering temperature limitations (Figure 3) was developed.
User-dependent parameter in the searching procedure was maximum allowed operating
temperature in the reactor. For the parameter uncertainty analysis, six different operating points
are proposed (Table IV). For every operating point, optimal molar ratio of H2O2 : 3-MP was found
in the region of ca. 0.91.
166
Table IV
Proposed operating points after safety constrictions’ application
Process variable Operating points
A B C D E F
Maximum allowed operating
temperature in the reactor [°C] 125 124 123 122 121 120
Feed temperature [°C] 55.0 50.9 46.1 42.4 37.6 33.9
Production rate of 3-MPNOX [kg.h-1] 267.8 267.6 267.3 267.1 266.8 266.5
Parameter uncertainty analysis
Most model parameters involved in the prediction of reactor behavior are uncertain. The
influence of uncertainties in the reaction kinetic parameters (Table I) and reaction enthalpy on
the proposed reactor operating points was analyzed. Reactor behavior was found to be most
sensitive to changes in the reaction enthalpy. Therefore, reaction enthalpy uncertainty was further
examined. The original value of the reaction enthalpy in this case study was – 160×103 J.mol-1.
However, the reaction enthalpy value for 3-MP N-oxidation varies in literature significantly (from
ca. – 120×103 to – 190×103 J.mol-1)6. For the purposes of parameter uncertainty analysis, the
range of reaction enthalpy from – 10 % to + 10 % was studied. The position of the proposed
operating points (Table IV) for various values of the reaction enthalpy is depicted in Figure 4.
Figure 4. Location of the proposed operating points as a function of relative change of reaction
enthalpy value from the original value of – 160×103 J.mol-1
167
As it can be seen in Figure 4, none of the proposed operating points is located in the safety
operating regime for every uncertainty of the reaction enthalpy in the studied value range.
However, if the reaction enthalpy was decreased only by 5 %, all proposed operating points ae
still satisfactory. In case of a reaction enthalpy decrease by 10 %, operating point F was shifted
to the hazardous operating regime because of the reactor temperature decreased below 110 °C
and the consequent decomposition of hydrogen peroxide leading to a runaway would take place.
An increase of the reaction enthalpy had a more significant impact on the position of the proposed
operating points. In case of a reaction enthalpy increase by only 5 %, all but one (F) operating
points were shifted to the hazardous operating regime because of the temperature in the reactor
exceeded the upper safety constraint of 125 °C, which leads to possible over-pressurization of the
reactor. If the reaction enthalpy was increased by 10 %, every proposed operating point was in
the hazardous operating regime.
Conclusion
Simulation-based approach for process intensification and hazard identification combination
was proposed. As a case study, the production process of 3-methylpyridine-N-oxide was selected.
First, a mathematical model suitable for scale-up and reaction conditions optimization of the
CSTR for 3-methylpyridine-N-oxide production was developed in the MATLAB® modelling
environment. In the next step, reaction conditions were optimized towards maximizing the
production rate of 3-methylpyridine-N-oxide with process safety constraints’ implementation.
Six different reactor operating points with the production rate increased by ca. 6 % were proposed
based on process simulation and multi-parametric optimization. Consequent model parameter
uncertainty analysis was performed. For the studied range of reaction enthalpy relative change by
± 5 % from the original value, only one from the proposed operating points was found to be
satisfactory for safe operation. If the range of reaction enthalpy relative change was increased
(relative change by ± 10 % from the original value), none of the proposed operating points was
satisfactory for safe operation.
This study has shown that an appropriate safety analysis is always required prior to the
implementation of an intensified process. The need for model parameter uncertainty analysis in
the simulation-based process intensification and hazard identification was underlined. In our
future work, implementation of the presented procedures into smart software solution for
supporting hazard identification techniques will be studied. Such software tool can be used to
design inherently safer processes and also to train operators and process engineers in existing
industrial plants.
168
Acknowledgment
This work was supported by the Slovak Scientific Agency, Grant No. VEGA 1/0749/15 and
the Slovak Research and Development Agency APP-14-0317 and the AXA Endowment Trust at
the Pontis Foundation.
Literature
1. Kletz T. A.: Chem. Ind. 9124, 287 (1978).
2. Mannan S.: Lees’ Loss Prevention in the Process Industries: Hazard Identification,
Assessment and Control. Elsevier Science, Oxford 2012.
3. Sempere J., Nomen R., Rodriguez J. L., Papadaki M.: Chem. Eng. Process. 37, 33
(1998).
4. Pineda-Solano A., Saenz L. R., Carreto V., Papadaki M., Mannan S.: J. Loss Prev.
Process Ind. 25, 797 (2012).
5. Pineda-Solano A., Saenz-Noval L., Nayak S., Waldram S., Papadaki M., Mannan S.:
Process Saf. Environ. Prot. 90, 404 (2012).
6. Cui X., Mannan S., Wilhite B. A.: Chem. Eng. Sci. 137, 487 (2015).
