You are on page 1of 5

Should you rely on your simulation results?

Avoiding some of the pitfalls of process simulation

ATUL CHOUDHARI
Tata Consulting Engineers

P
rocess simulation is a useful synthesis using simulation software More choice, more confusion?
and powerful tool to model yields a heat and mass balance that As more options become availa-
chemical process flowsheets of is consistent with thermodynamics. ble, it is expected and assumed that
varying complexities. Modern day This heat and mass balance is a firm the user is knowledgeable enough
simulators are built with a compre- basis for all downstream process to select appropriate methods and
hensive, pure component databank, design activities including prepara- employ correct choices to solve
an exhaustive library of thermody- tion of equipment and instrument the flowsheet. Since simulation
namic systems, physical property specifications. software uses advanced high-end
estimation methods, initial estimate Mathematical simulation models computer hardware for such sophis-
generators, and inbuilt algorithms offer a drag-drop type of graphi- ticated and advanced simulation
for every unit operation with a user- cal interface for setting up process programs, there is no programming
friendly graphical interface. These flowsheet configurations. Graphical limit with respect to presenting
simulators can solve and optimise user interfaces have matured to multiple choices to the user. In fact,
virtually any flowsheet synthesis the extent that they make it easy to they are increasingly interpreted,
problem. However, despite the sim- build large and complex models; falsely, as ‘more choice, more capa-
ulator’s sophisticated and rigorous they also provide a step-wise menu bility’ products. However, so many
modelling techniques, process sim- and online help to enable users to choices surely pose a challenge for
ulations at times fail to represent define input parameter specifica- process simulation engineers to
real life plant data. In most cases, tions easily. check the applicability and suita-
the user shows blind faith in the Simulators serve a wide variety bility of every choice to model a
inbuilt configurations and default of unit operations in refining and given process, verify the advantage
selection of methods in simulators, include mathematical solver algo- or disadvantage of the selection,
which can introduce erroneous rithms specific to every unit opera- and determine its effectiveness over
results for specific systems. Since tion model. For example, distillation the operating range. Thus, to uti-
simulators can generate multiple columns can be solved by selecting lise the capability of the simulation
and, at times, conflicting solutions any of the suitable prebuilt algo- software, adequate knowledge and
for the same set of external input rithms like Inside Out or chemdist. specific process experience become
data parameters, this raises doubts In short, virtually all unit operations a prerequisite. However, this basic
about their effectiveness and reli- are well supported and can be easily and fundamental issue could affect
ability. There are a number of rea- defined using a typical simulation the quality of simulation output
sons for simulation failure. This suite. These mathematical models results.
article discusses various issues that are supported with internal data- With increased availability of sim-
may help users to derive mean- bases covering physical and ther- ulation models, it has become easier
ingful results from simulation and modynamic property databanks. In for users to study various design
thus enhance the reliability of their a bid to enhance the capability of cases. Normally, these design cases
simulations. the software, these simulation pack- include modelling the same flow-
ages present multiple choices for the sheet for various case studies, for
Elements of process simulators user to configure and then to solve instance the effect of processing
With advances in computer tech- the flowsheet. A few of the impor- different feedstocks, or checking
nology and the availability of mod- tant user choices include selection of performance under different oper-
ern tools, commercial steady state thermodynamic methods (for calcu- ating conditions. In today’s world
simulation software packages have lating pure component and mixture of high computing speeds, compu-
become an integral part of process physical properties) and/or specific tation time is no longer a constraint
design practices. Process simulation algorithms for solving individual while selecting or adding as many
is utilised in all stages of process unit operations, deciding calculation check cases as required. The user
plants, from concept and feasibil- sequences, methods for generating can check performance for multiple
ity, to basic design, to detail design, initial estimates, methods for defin- cases, conduct what-if analysis, per-
even extending to commissioning ing a non-library component, and form sensitivity studies, and opti-
and revamps. Process flowsheet so on. mise process designs.

www.digitalrefining.com/article/1002561 PTQ Q4 2020 47


lems can be dealt with accurately
Pure
and reliably. Unfortunately, this is
component
Thermodynamic
property
not a reality.
database Physical database Accuracy of simulation output
property
database
is critical, irrespective of the rep-
utation of the simulation package
Circulation Flowsheet deployed, or the speed to obtain
Predefined unit operation mathematical models
algorithms tolerances a solution. Unless the simulation
model accurately describes the
Initial Multivariable interaction of different components
Inbuilt calculation sequences
estimates controllers at varying temperatures or pres-
sures using reliable methods, sim-
ulation results will not represent
Data Stream data, PFDs,
sheets heat and heat release reality. Simulators are just auto-
mass balances curves mated mathematical model solvers
based on inputs provided by users.
Therefore, simulation input param-
Figure 1 General architecture of a typical commercial steady state simulator eters require skilful scrutiny. Also,
the results need to be carefully ana-
Using a simulation tool, one can Blind faith by inexperienced engineers lysed based on fundamental princi-
significantly increase the profita- Knowing how to configure a flow- ples and the specific objectives of a
bility of a process by optimising sheet in a simulation environment simulation.
design and operating parameters. and familiarity operating the sim-
Thus, simulators provide a reliable ulator does not necessarily confirm Differences in simulation and design
platform by solving any process the availability of skills required to It is now widespread practice to
flowsheet using an inbuilt mod- effectively solve and analyse flow- employ process simulators to solve
elling approach. However, a few sheet synthesis results. Here is the design cases. Although a simulator
problems still exist with respect main hitch. Although simulator can be configured to provide solu-
to multiple solutions that can be programs provide early warnings tions to a design problem, there is a
generated for the same flowsheet for missing input data or inade- difference in approach and method
by different users using the same quate input data, there may be no using simulators to arrive at a solu-
simulator and using the same set warning if the user selects an inap- tion. In order to obtain a uniform
of input data. Multiple solutions propriate calculation method. This solution, any design problem is
are possible as multiple choices are is especially true when blind faith solved in a step-wise manner utilis-
available to users while configur- is exhibited by users on the appli- ing a fixed and defined method or
ing the flowsheet. As an example cability of various choices such as procedure for every step. However,
from experience, a typical glycol selection of a thermodynamic sys- when it comes to solving the same
dehydration tower for the same tem. As Figure 1 shows, modern day design problem using a simulator,
input and output specifications will simulators carry a wide variety of an iterative or case study type of
require eight theoretical stages if an unit operation libraries, expanded approach has to be adopted. Before
equation of state based thermody- thermodynamic data libraries, com- dealing with a simulation’s pitfalls,
namic model is selected against 12 prehensive pure component and it is important to understand the
theoretical stages if a liquid activity binary interaction parameter data- differences between a design prob-
coefficient based thermodynamic bases, initial estimate generators, lem and a simulation problem.
model is selected. In both cases, the and so on. Looking at these capa- As Figure 2 shows, a design prob-
simulator will converge but will bilities, it seems that most chemical lem can be defined with a known
yield two different hardware config- plant modelling problems or pro- or given set of input parameters;
urations for the same mass balance. cess design and optimisation prob- the objective of a design problem
is to obtain design variables for an
already known set of performance
parameters. For example, a sim-
Set of input Design Set of performance
plified problem statement can be:
parameters variables parameters
“Design a heat exchanger to cool
water from 60°C to 50°C.” In this
Known Unknown Known
Example Example Yield simple problem the set of input
Temperature Equipment sizes Flow parameters is known. These are
Pressure Unit operation configuration Temperature
Flow Pressure the component (water), its physical
Composition Phase fraction properties, and initial conditions
(60°C). The desired output per-
formance parameter (50°C) is also
Figure 2 Definition of a design problem known. With input and output per-

48 PTQ Q4 2020 www.digitalrefining.com/article/1002561


formance parameters known, the
objective is to find design variables
(heat exchanger geometry) to satisfy Set of input Design Set of output
the desired output parameters. This parameters variables parameters

design problem is then systemati-


cally solved in a stepwise manner so Known Known Unknown
as to obtain the outlet temperature.
As Figure 3 shows, a simulation
problem can be defined with a Figure 3 Definition of a simulation problem
known or given set of input param-
eters. The objective of a simulation of design methods, it is possible to specified. Even for the same set of
problem is to obtain output (or per- generate multiple solutions for the input data, various simulators will
formance) parameters for prede- same set of input parameters. predict somewhat different results
fined design variables. In the above Arriving at the most appropriate, depending on the source of thermo-
example, the simulator will use a most suitable or most optimised physical or other data in its inter-
set of available input parameters. solution can then be a dilemma. nal databases. The accuracy and
The user needs to define a design This issue can become very subjec- correctness of output data depends
variable (the exchanger geometry) tive. Therefore, it is necessary to on various modelling techniques.
so that the simulator can check the verify the simulation results with Chemical engineering’s empiri-
suitability of the exchanger geom- some baseline data. Preferably, the cal correlations pose a challenge
etry to obtain the desired output simulation results are compared in representing the true nature of
(50°C) parameter. In every subse- with actual plant operating data many processes and applications.
quent run, the exchanger geometry and/or proven baseline design data Therefore, it is imperative that any
can be readjusted by the user until supplied by technology suppliers. simulation model must be cali-
the desired output is obtained. Matching the data for the operating brated based on representation of
This means that when a simu- case enriches the designer’s expe- true laboratory or plant operating
lator is to be used for designing rience and understanding to per- data.
an exchanger, a user should first form similar designs effectively. In Initially, a simulation model
assume a suitable design geome- the absence of verification against must be checked to reflect true and
try and then check if the assumed such baseline reference data, the real-life conditions for comparison,
geometry satisfies the output simulation results cannot be fully either from the plant or from the
requirements. The user thus needs trusted. Their accuracy remain laboratory. Simulation results for
to follow an iterative or ’trial and unverified. Simulation of grass- the base case must be compared
error’ approach to solve the prob- roots design cases can be based on to verify and validate their close
lem statement. Unlike the design careful analysis of laboratory data resemblance to actual plant or lab-
case, a good approximation for the or from simulation experience for oratory data. Once it has been ver-
most suitable exchanger geometry similar operating units/operations ified that the simulation results
as a starting point is essential in or pilot scale/laboratory scale set- match closely with benchmark or
solving a design problem using a ups. When any of the reference real-life data, further studies such
simulator. The solution and number data is somehow not available for as process performance predictions
of cycles for convergence depend on new designs, then the user must be with separate sets of operating con-
the user’s ability to assume a config- expert in modelling the problem ditions, different capacities, differ-
uration close to a reasonable design. with experience in selecting the ent feedstock conditions, and so
Analysis of the convergence history most suitable methods to represent on can be simulated. Even in such
of each cycle provides a direction the system. However, ignoring the cases, the extent of extrapolation of
for the user to decide which design validation exercise and opting to go the operating envelope may matter.
parameters in subsequent runs need ahead with the simulation results This works for building simu-
to be adjusted to obtain the desired based on the user’s gut feeling lation models of operating facili-
output. Lack of proper understand- may result in unreliable simulation ties. However, it is not a fool proof
ing will lead to solution failure and results. method; gathering error-free, con-
divergence may be observed instead sistent data from an operating
of convergence. Concept of mirroring plant and laboratory data (which
An operating plant provides realis- is discrete data) is not easy. Even
Result validation tic and true data for various oper- with the right data and with the
The user can select a particular cal- ating variables that need to be validation exercise completed, fun-
culation method from a variety of reflected in a mathematical simu- damental mistakes in selecting ther-
options offered by the simulator. lation model. This is the concept of modynamic methods (for instance)
Simulators also provide multi- plant mirroring. cannot be eliminated. In any case,
ple choices of design variables for As previous discussion points out, this may not help in new designs
configuring the given problem. process simulation results can vary other than that the user’s under-
Depending on the user’s selection for every user based on the inputs standing is enhanced. Thus, for

www.digitalrefining.com/article/1002561 PTQ Q4 2020 49


by derating the calculated pressure
Input process drop values by 10%. However, for
parameters Process another case involving close boil-
simulation
model
Output data
ing point component separations,
Assumed
Known
hardware a deviation in vapour pressure of
plant data
configuration Lab or 10% between simulated and plant
plant data
data will be unacceptable. There
Initial estimate Model inputs, Compare are no fixed rules for an acceptable
generators like thermo data
degree of deviation. It depends on
the application, hence a user must
No cautiously compare results while
mirroring the process plant.
Mirroring Output within
Yes
complete acceptable tolerance Pitfalls in process simulation
Process simulation enables exper-
imentation with predictions in the
Figure 4 Simplified procedural overview for plant mirroring form of what-if scenarios for a real
process plant. The success of a sim-
effective representation/mirroring port properties of the mixer outlet ulation model depends on design
one must have: stream, the simulation model can techniques applied while modelling
• Good understanding of the unit/ be accepted with respect to its main the problem. It is common belief
operations to be simulated objective of predicting the heat of that simulators can provide sophis-
• Error-free data from various mixing. ticated and accurate results in the
sources (DCS, equipment specifi- The user should be aware of shortest time. However, the truth
cations, lab analysis) which is also the limitations and capabilities of is that simulations can fail badly.
consistent with respect to time a selected method or a selected They can fail in all the ways that
• Correct selection of options in equation. For example, a simula- theories can fail, and in all the ways
simulation – thermo, component tion model is set up to calculate that experiments can fail as well.
data banks, solving options, and so pressure drop in a cross-country An understanding of the many
on pipeline. Initially, in the absence of ways in which simulations can fail
• Correct interpretation of devia- any data for comparison, the user offers better clarity on the reasons
tions/ results may provide some design mar- for failure and, in a way, helps the
Figure 4 shows a procedural over- gins based on sound engineering user to be cautious. Some pitfalls
view of plant mirroring when using practices and may decide to trust of simulation can be categorised as
a simulation model. This exercise the simulation results. On com- follows:
in verifying simulation results with missioning, the user notes that the
available plant or laboratory data to actual pressure drop calculated by Data entry
reflect the real world is commonly the simulator turned out to be 10% This is very common and the sim-
termed mirroring the plant. higher. In an assignment involv- plest source of simulation error. A
Many engineers struggle to fix ing exactly the same service con- decimal shift error has serious con-
an acceptable degree of accuracy ditions, should the user reject the sequences. For example, a decimal
when comparing simulated data previous simulation model because level error in specifying the wash
with plant data. While it is desira- of its large deviation from reality? water flow rate injected on a col-
ble to match the data as closely as It is important to know the extent umn overhead stream, say inflat-
possible, there are no fixed rules for of deviation and possible reasons ing it, will result in multiple errors
accepting or rejecting the simulation for such deviation, and thus deter- in hardware design including an
data. The results must be analysed mine the applicability of a given increase in condenser heat duty, an
and compared considering the sim- method. Many times, the proper- increase in cooling media consump-
ulation objective of the application. ties of an actual fluid may be dif- tion, oversized pipes, an oversized
For example, if the objective of the ferent from the design fluid, or the reflux drum with boot for water
simulation is to calculate the heat roughness factor used in designs separation, and so on.
of mixing of, say, caustic dilution in could be different from the actual.
water, then temperature data at the It really does not matter, even Incorrect units of measurement
mixer outlet is compared with infor- though there are deviations. As Although simulators have an auto-
mation from the operating plant. long as the user is aware of the matic feature for converting units of
As long as this temperature com- limitations of the selected design measurement, there is potential for
parison matches with an acceptable options, the simulation results can wrong selection from the graphic
degree of accuracy, the model can be accepted with work-around interface menu. As an example,
be accepted. In this example, while adjustments. In this example of entering the feed stream data in
comparing the data, even if a large pressure drop calculation, the user molar flows instead of mass flows
deviation is noticed in the trans- can accept the simulation model will lead to erroneous results.

50 PTQ Q4 2020 www.digitalrefining.com/article/1002561


Data reconciliation modynamic methods for various approaches, when the enthalpy
Operating plant data obtained from components/situations. The user model itself is only ±20% accurate.
a plant’s DCS system are rarely also needs to be familiar with the It is important to apply cause and
constant. Depending on the rela- specific methods/guidance avail- effect type of error analysis skills.
tive mode of operation and posi- able from the simulation software The user should have a fair under-
tion of other control variables, the guide. Selection of a suitable ther- standing of the impact of one varia-
results and trend line for the same modynamic model and physical ble on the entire flow sheet. Process
controller vary with respect to time. property model is crucial to the suc- engineering knowledge cannot be
There are typically no unique sin- cess of any simulation problem. replaced by a process simulator.
gle fixed-point data for any given Missing, inaccurate or inade-
variable. For example, a simple one quate physical and thermodynamic Conclusion
feed, two product stream binary properties compromise accuracy A process simulator is one of the
distillation column with flow and of results. Most inbuilt proper- most powerful tools for modelling
composition measurement on feed ties have a predefined applicabil- a process flow sheet. It is highly
and both products lines will not ity range. The user needs to verify flexible when applied to the design
perfectly satisfy a feed flow = prod- and confirm that the inbuilt data- and analysis of many simple or
ucts flow mass balance rule. Bigger bases associated with the selected complex systems. At the same
errors may crop up as feed and thermodynamic methods cover time, its results can be misleading
product analysis timing may not be the operating range required for a and prone to generating multiple,
consistent with each other or other given process. Estimation of miss- conflicting results if proper model-
operating data. Therefore, data rec- ing parameters using various tech- ling techniques are not understood
onciliation must be done carefully niques (the UNIFAC method, or and followed. The fundamental
for obtaining satisfactory repre- data regression, for instance) must understanding of thermodynam-
sentative data which is consistent be done carefully. Estimated data ics, engineering principles and unit
and satisfies heat and material bal- must be validated before its use in operations, cause and effect type
ances. Even with correct modelling simulation. of error analysis skills, the ability
techniques and appropriate choice to understand variability and con-
of equations to represent the sys- Too much faith in input data sistency gaps in input data, the
tem, data reconciliation if not done The quality of input data clearly degree of sensitivity of performance
properly results in horrendous influences the quality of output parameters to the variable param-
results. results. Any data used without eters, and so on, play a vital role in
proper analysis for any disconti- successful simulation. Validation
Ignoring warning messages nuity in material and heat balances of a simulation model by mirroring
A fully converged complex flow or plant data used without recon- actual operational performance in a
sheet using rigorous procedures is ciliation may provide unexpected plant/unit or pilot/laboratory scale
no guarantee of dependable results. results. Negligence in mirroring of equipment can provide confidence
Modern simulators can display the plant or laboratory data will result of reliable results. It is crucial that
results for a selected unit operation in an inaccurate simulation model. simulation results are not blindly
or selected stream with a click of the accepted. The success of a simula-
mouse. The detailed simulation text Configuration of simulation defaults tion depends on the user’s ability
report runs to several pages; most Every simulation model has default to interpret input and output data.
of the time, such a detailed report is configuration values for many The onus of accuracy, reliability
not generated while converging the parameters. These include ini- and relevance never lies with the
simulation model. A detailed text tial estimation generation meth- simulator.
report contains warnings and mes- ods, selection of a minimum tear
Atul B Choudhari is Technology Principal
sages which if ignored can lead to an stream algorithm to decide a calcu-
with Tata Consulting Engineers Ltd, Navi
erroneous simulation. For example, lation sequence, flow sheet toler- Mumbai, India. His 27 years’ experience
the report may highlight a warn- ance values, water decant options, includes flowsheet simulations, process
ing about unavailability of binary and methods for estimating miss- optimisation, basic and detailed engineering
interaction parameters for certain ing data. These defaults need to be for petrochemicals, refinery and hydrocarbon
pairs. Such pairs are defaulted to reviewed and accepted consciously processes. He holds a B.E. Ch.E. from
‘ideal behaviour’ in the absence of and may need to be modified as and Marathwada University, Aurangabad.
any other user supplied data. If a when necessary.
detailed report is not generated and
all warnings are not analysed cor- Apparent accuracy LINKS
rectly, simulation failures can occur. Engineers sometimes feel obliged
to use simulation tools to solve More articles from the following
Physical and thermodynamic optimisation problems. There is categories:
property model no point in investing effort in opti- Instrumentation, Automation and
The user needs to have adequate mising a process based on the heat Process Control
Process Modelling and Simulation
knowledge of the suitability of ther- exchangers’ small temperature

www.digitalrefining.com/article/1002561 PTQ Q4 2020 51

You might also like