Pressure relief analysis (PRA) is a critical task at the
engineering design stage of a grassroots oil refinery project. During that phase, the
relief loads must be determined so that the process units and
plant flare systems can be designed.
The key steps of PRA include the determination of individual
relief loads and the evaluation/mitigation of each process
units overall relief load. Individual relief normally
refers to a single piece of equipment or a set of
interconnected equipment systems, such as a distillation column protected with a
pressure relief valve (hereafter referred to as a
A conventional method based on recommended practices and
standards, such as American Petroleum Institute Standard 521
(API 521),1 is normally used for relief load
calculation by designers and licensors. However, the
conventional method has been widely proved to over-estimate the
relief load, leading to the over-design of the flare system.
The over-design of the flare system will not only result in
unnecessary capital investment, but also lead to design and construction difficulties at very
large plantsfor example, a refinery with a capacity of several
tens of million tons per year (MMtpy).
Many recent reports claim that dynamic simulation is more
accurate than the conventional method in predicting relief
load.2, 3 The required relief load, as calculated
with dynamic simulation, is always far less compared to the
relief load calculated with the conventional method.
Furthermore, in the latest edition of API 521, dynamic
simulation is a recommended method.
Although dynamic simulation would be the best way to predict
relief load, building the model for a complete process unit is
time-consuming and labor-intensive. Furthermore, this work
requires detailed equipment and control system information that
would not be available during the early design phase.
An approach combining the conventional method with dynamic
simulation is proposed here. By minimizing the inherent
drawbacks of both the conventional method and dynamic
simulation, this approach can optimize relief load
determination for the entire process unit with minimum modeling
Relief load optimization approach
The relief load determination procedure for a process unit
includes several steps, as shown in Fig. 1.
Note: During engineering design, an iterative
procedure likely will be required due to process modification,
control or safety system reconsideration, etc.; this procedure
appears to be sequential.
1. Approach for relief load
Step 1: Individual load calculation. The
conventional method is the simple, fast way to conservatively
determine the required relief load and, until now, it has been
standard industry practice.
In the first step, the conventional method is applied
(Fig. 1). Before the relief load can be
calculated, applicable relief cases for a specific protected
system must be determined, along with assumptions for these
cases based on API 521. These cases and assumptions are used as
the basis for later dynamic simulation, since even dynamic
simulation should be compliant with the API standard.
The calculation for the distillation column is one of
complicated applications. Three approaches have been used in
the industry, including flash drum, gross overhead vapor and
unbalanced heat load methods. However, the flash drum method
can only be used as a rough estimation at an early stage of
design. The latter two methods are more realistic, although
underestimation of relief load is a common occurrence in the
The last method, unbalanced heat load, is the most
complicated and rigorous method among the three, and it is
widely accepted as the industry standard. A number of authors
have discussed this method in detail,3, 4 and so it
is not repeated here; although, a comparison between the second
and third methods is shown in Table 1.
The gross overhead vapor method gives a smaller number of
figures; therefore, it is not reliable in terms of conservative
consideration. Here, the unbalanced heat load method is taken
as the standard conventional method for the column
Step 2: Ranking of loads. When individual
relief loads for all cases are worked out, these data are
sorted by their values in descending order.
As mentioned, dynamic modeling requires considerable
engineering time and effort, and it is unwise to apply dynamic
simulation for all cases (with the exception of the critical
ones with the largest loads). The purpose of this sorting is to
pick out the critical cases for which further dynamic
simulation is needed. Note: Relief to
different headers must be treated separately.
An example of a 4.2-MMtpy vacuum gasoil (VGO) hydrotreater
is given in Table 2. Individual loads of
high-pressure and low-pressure heaters are ranked and listed
respectively; the largest two or three cases are candidates for
dynamic simulation. It is important to keep in mind that the
weight relief flowrate is not necessarily the largest relief
load, especially for those relief loads with small molecular
Step 3: Matching criteria. In Step 2,
critical cases and corresponding protected systems are
selected. However, not all protected systems can be modeled
properly due to simulation limitations, and, for some cases, it
is not worthwhile to model them. As a result, further screening
of candidates for dynamic simulation is required. In this step,
a judgement will be made vs. a set of criteria. These criteria
typically include, but are not limited to, the following:
- The protected system can be modeled, and the cases can be
- The protected system is a column
- The protected system is a reactor loop.
Fig. 2 shows a comparison of relief loads
estimated by both the conventional method and dynamic
simulation for a 3.7-MMtpy grassroots hydrocracking unit. As
can be seen, with the same assumptions as the conventional
method, dynamic simulation predicts a much smaller peak relief
load for most column cases, thereby improving the relief load
estimation.2, 3 Similar phenomena is observed for
the reactor loop due to its complexity.
2. Comparison of dynamic simulation
with the conventional method.
However, for the drum (including the separator, the flash drum,
the surge drum, etc.) and the compressor, dynamic simulation
cannot make obvious improvements. The reason is that the valve
and the drum are simple pieces of equipment and do not leave
enough room for more rigorous modeling against manual
calculations. Also, the compressors overpressure case is
normally a blocked outlet, in which almost all
inlet vapor must be relieved.
For these reasons, only protected systems of the column
and/or reactor loop require dynamic study, and only separated
modeling of these systems is required. In this way, much
engineering time can be saved.
Step 4: Dynamic modeling. As mentioned in
Step 3, the dynamic models are built for protected systems with
the criteria being matched. Relief cases are executed based on
the model, and the peak relief loads are documented.
A key point is that the same assumptions as the conventional
method are applied during case studies based on a dynamic
model. A typical example is overhead pressure control of the
column. When an overpressure case happens, the increasing
pressure will push the control valve to open wider, and this
will reduce the required relief. However, this conventional
instrumentation response should not be assumed when sizing
individual process equipment pressure relief, according to API
Therefore, in this case study, the pressure controller is
set on manual operation, and the control valve is kept in its
last position. Another example is a fire case. Not only is the
same heat input model as the conventional method used in the
dynamic simulation, but the same assumptions are also applied
as the system is isolated and shut down in the occurrence of a
Step 5: Summation of individual loads.
Previous steps focused on a single protected system. From these
steps, an overall relief load for a process unit is evaluated
and altered for general failure cases (GFCs).
GFCs, typically including general power failures (GPFs),
general water failures (GWFs) and general instrument air
failures (GIAFs), usually are not allowed. The corresponding
safety systems, such as the uninterrupted power supply (UPS),
dual water supply and dual air supply, must be designed to
prevent these failures. However, as far as relief system design
is concerned, the extreme cases must still be considered.
Generally speaking, the summation of all individual relief
source loads in a GFC should be calculated, yielding a rough
and conservative overall load for that case. An example of a
4.2-MMtpy VGO hydrotreater is shown in Table
3, where two protected systems relieve to a
low-pressure header in a GPF case. Note: The
dynamic result in Step 4 should be used instead of the
conventional method, wherever applicable.
Step 6: Depressuring load calculation. The
peak load for depressuring is calculated using the conventional
method. The timing of depressuring and its relevance to general
failures should be carefully evaluated. The engineering
judgement should be made and evaluated if depressuring occurs
simultaneously in a GFC.
Step 7: Matching criteria. One of the
advantages of dynamic simulation is that it considers timing
and interconnection of processes. Normally, multiple protected
systems relieve in GFCs, although not simultaneously. This
scenario provides a good opportunity to mitigate the overall
relief load with dynamic simulation.
In practice, one of the typical situations for which dynamic
simulation can be implemented is a column series with streams
and/or heat interconnections. However, with a simple analysis
on the summation table at Step 5 (see Table
3), if the majority of the relief load (i.e., 80% or
higher) is found to be contributed by a single column (e.g.,
the product fractionator in Table 3), then
dynamic simulation is not encouraged because no considerable
improvement can be expected.
In the same GFC, relief via both the relief valve and the
depressuring valve may happen, but they do not peak
simultaneously. This is another typical application of dynamic
simulation, and a simple summation of those relief loads
normally can be mitigated.
Step 8: Dynamic modeling. Dynamic models
are built for the systems selected in Step 7, and GFCs are
executed. To make the design conservative, the sum of the peak
loads of each protected system should be taken as a design
reference instead of the dynamic peak summation load. This is
further illustrated in the later case study of a delayed coking unit.
Note: The assumptions made for multi-source
simulation would differ from those for individual relief
analysis. The conservative assumptions for the former
simulation are not necessarily the same as in the latter.
Therefore, in this step, careful evaluation of the assumption
is required to ensure that the total relief load is
conservative. It is possible that some models in Step 4 are
used as part of the model in this analysis, but the assumptions
used in other case studies likely will be different.
Step 9: New relief loads in GFCs. For the
relief occurring in GFCs but not included in the analysis at
Step 8, individual peak loads should be used to calculate the
overall load from Step 8. The summation result is the optimized
overall load in the GFC.
Step 10: Optimized loads. The optimized
individual relief load in Step 4 can be used to size the relief
valve of this protected system. For flare header sizing of this
process unit, the maximum individual load, the overall load for
the GFC and the depressuring load are collected and sorted, and
among them the largest one is taken as the design case. One or
more of these loads will be used for the plantwide relief load
Relief load optimization for a delayed coking unit with a
capacity of 3 MMtpy is performed. The relief of the coker drum
is handled separately, and, in this study, only the
fractionation section (as shown in Fig. 3) is
3. Process flow of the delayed coking
Recall the approach procedure in Fig. 1.
Individual relief loads are calculated in Step 1 using the
conventional method, and the three largest loads are listed in
Table 4. As can be seen, the largest loads
come from the coker fractionator in this case, which matches
the criteria in Step 3. Then, a dynamic model is built for the
fractionators, and its relief cases are executed. The new
relief loads are reported in Table 4 for
As expected, the relief loads are considerably mitigated by
dynamic simulation for the first two cases, which are only
around 60% of the relief loads obtained with the conventional
method. No relief occurs for the single power failure case, but
the column pressure increases; however, its peak value of 0.23
megapascal gauge (MPaG) is lower than the relief valve set
pressure of 0.35 MPaG. At this point, Step 4 is finished.
Step 4 is followed by a summation of individual loads at the
GFCs. In this study, the GPF and GWF cases are examined. The
relieving equipment in the GPF includes the coker fractionator,
the stripper feed drum and the debutanizer, while only the
coker fractionator will relieve in the GWF case. Therefore, as
per the criteria in Step 7, the GPF case would be further
analyzed with dynamic simulation. A summary of the results is
listed in Table 5.
As shown in the table, a single coker fractionator
contributes 80% of the total relief load in the GPF case. For
this reason, it is not a practical candidate for dynamic
studies for multiple protected systems; however, it is still
implemented in this study to illustrate Step 8.
As shown in Fig. 4, four protected systems
will relieve in the GPF case, and the dynamic summation (black
curve) peaks at 275,000 kg/hr. However, to make a conservative
estimate, the summation of peak loads of 355,000 kg/hr for four
individual sources is taken as the design reference. A large
improvement is not seen compared with 376,000 kg/hr at Step 5,
due to the reasons mentioned above. This summarizes Step
4. Dynamic relief curve for multiple
protected systems in the GPF case.
Note: An additional relieving protected
systema compressor interstage drumwas found in the
dynamic simulation. It splits the relief flow from other
sources and does not increase the overall load; however, this
phenomenon can only be observed and evaluated through dynamic
Finally, the optimized individual relief loads are collected
and summarized. The results of the conventional method are
replaced by the dynamic simulation results, where applicable.
These data provide sizing references for individual relief valves. The required orifice size of
a relief valve can always be reduced through relief load
mitigation. With the coker fractionator, for example, the GPF
is the controlling case, and the required relief valve size is
reduced from 1,010 cm2 to 623 cm2 (based
on the API relief valve sizing method). It can be determined
that 355,000 kg/hr is the largest relief load for this delayed
coking unit at the GPF case, which
is only 63% of the load without optimization (565,000 kg/hr, as
shown in Table 5), and this value is used to
set the flare header size of the unit.
It is also apparent that modeling the entire flowsheet at
Step 8 is unnecessary, since only a slight improvement can be
obtained compared to the individual load summations. Therefore,
Step 8 is bypassed in real engineering design, leading to
considerable savings of modeling time and effort.
The relief load estimation derived by combining the
conventional method with dynamic simulation was optimized as
expected, with minimum dynamic modeling efforts. This
estimation was successfully applied to a relief load case study
of a large-scale delayed coking unit. Although a refinery process is studied here,
the approach is also applicable to other oil, gas and petrochemical processes.
The authors thank Xiaobo Liu and Xiuwen Zhao from China
Petroleum Engineering & Construction Corp.s East China
Design Branch; Ximei Lv and Zengfu Zhang from SimTech Beijing
Ltd.; and Professor Enxi Lu from South China University of Technology for their contributions
to this article.
1 American Petroleum Institute, Guide for
pressure relieving and depressuring systems, API Standard
521, 5th Ed., January 2007.
2 Depew, C. and J. Dessing, Dynamic simulation
improves column relief load estimates, Hydrocarbon Processing, December
3 Nazami, P. L., Distillation column relief
loadsPart 1, 2, Hydrocarbon Processing, April
4 Rahimi Mofrad, S., Tower pressure relief
calculation, Hydrocarbon Processing, September
Chong Liang Xie is the chief
engineer for China Petroleum Engineering & Construction Corp.s
(CPECCs) East China Design Branch. He has 27
years of experience in refinery design, with
specialized expertise in the delayed coking process and in
general configuration design. Mr. Xie holds a BS
degree from Northeast Petroleum University in
Zhi Gang Wang is the director for
the general design division of CPECCs East
China Design Branch. He has 20 years of experience in
refinery design and is an
expert in general configuration design and in the
fluid catalytic cracking process. Mr. Wang holds a BS
degree from China University of Petroleum in Beijing,
Yun Feng Qin is the technical
manager for SimTech Beijing Ltd. He has more than 10
years of experience in process simulation. Before
joining SimTech, Mr. Qin worked with
Invensys/SimSci-Esscor China for eight years as a
technical support specialist and consultant. He holds
an MS degree in chemical engineering from South China
University of Technology in Guangdong,
China, and a BS degree in chemical engineering from
the Beijing Institute of Light Industry in Beijing,
China. Mr. Qin can be reached at email@example.com.