Process and fluid dynamics simulation

Process simulation

Fabrizio Bezzo
[email protected]
 049 8275468
www.dii.unipd.it
Outline
 General definitions
 Chemical components and property models
 Simulation of unit operations
 Flowsheeting modelling
 Using process simulators

2 Bezzo 2024
References
 Towler, G., R. Sinnott (2022), Chemical Engineering Design: Principles, Practice
and Economics of Plant and Process Design (3rd ed.), Butterworth-Heinemann,
Oxford (U.K.), Chapter 4  Available online
 Equivalent book: Towler, G., R. Sinnott (2020), Chemical Engineering Design –
Coulson and Richardson’s Chemical Engineering Series (6th ed.), Butterworth-
Heinemann, Oxford (U.K.), Chapter 4
 Turton, R., J.A. Shaeiwitz, D. Bhattacharyya (2018). Analysis, Synthesis, and
Design of Chemical Processes (5th ed.). Pearson, Upper Saddle River (U.S.A.),
Chapters 13, 16
 Seider, W.D. et al. (2016). Product and Process Design Principles. Synthesis,
Analysis and Evaluation (4th ed.). Wiley, New York (U.S.A.), Chapters 7, 8
 Bertucco, A., E. Barbera, G.B. Guarise (2023). Separation Unit Operation and
Process Simulation. CLEUP, Padova (Italy), Chapters 1, 2
 Adams II,T.A. (2022). Learn Aspen Plus in 24 hours (2nd ed.).
McGraw Hill, New York (U.S.A.)
 Westerberg, A.W., H.P. Hutchison, R.L. Motard, P. Winter (1979). Process
flowsheeting. Cambridge University Press, Cambridge (U.K.)

3 Bezzo 2024
Process simulation

General definitions
What is a flowsheet
A flowsheet is a graphic representation of a process by
means of units that are connected

(PATH, Siemens, 2019)

5 Bezzo 2024
Elements on a flowsheet
 Sources: they bring endless material, energy or information into the
flowsheet
 Sinks: they are an endless outlet for material, energy or information
 Units: units typically change something, e.g. change the temperature
or composition due to reaction or separation

Units
Sink

Sources
6 Bezzo 2024
Elements on a flowsheet
 Sources and sinks are simplifications of reality
 No need to worry about where they are coming from
(sources) or how they are going to be consumed (sinks)
 Example:
Sink of pie:
Source of eggs:
Customers that consume
A basket with infinite eggs
all the pies produced

7 Bezzo 2024
Elements on a flowsheet
 Connections: transfer material, energy or information
 Material and energy flows are called streams. The minimum variables
representing a stream are temperature, pressure and composition
 Port: Start or end point for a connection
Port

Connection

8 Bezzo 2024
Elements on a flowsheet
 Dialogue boxes: windows that communicate information
to the user and prompts them for a response

Dialogue
box

9 Bezzo 2024
Building a flowsheet - unit by unit
 Add elements (sources, sinks, units) and enter data for
each element

10 Bezzo 2024
Building a flowsheet-unit by unit
 Add elements (sources, sinks, units) and change the settings

11 Bezzo 2024
Building a flowsheet-unit by unit
 Connect one unit with sources and sink; make sure you
can simulate the flowsheet

Sink

12 Bezzo 2024
Building a flowsheet-unit by unit
 Extend to another unit until the whole process is finished

Sink

13 Bezzo 2024
Building a flowsheet-unit by unit
 Extend to another unit until the whole process is finished

14 Bezzo 2024
A typical flowsheet in chemical engineering

15 Bezzo 2024
Process simulators
What do they do?
 “Flowsheeting (= process simulation) may be defined as the use
of computer aids to perform (steady-state) heat and mass
balances, sizing and costing calculations for a chemical
process” (Westerberg et al, 1979)
 Process simulation is a fundamental tool to process design.
The process design effort can be divided into three basic steps
 Synthesis: the step where structure is chosen (i.e. the equipment to
be used and its interconnection) and variables completing the DOF
are set
 Analysis: the step were we solve heat and mass balances, size the
equipment, evaluate its cost
 Optimisation: the step where we perform parameter and
structural optimisation (e.g. by modifying some operating
temperature and pressure levels, or even by altering equipment or its
interconnection)

16 Bezzo 2024
Process simulators
What do they do?
 Process simulation software typically does not solve the
momentum balance (which requires ad hoc numerics and
methods  computational fluid dynamics)
 However, simplified balances, e.g. to estimate pressure drops, or
rigorous calculations in simplified geometries, e.g. a pipe, may
sometimes be available
 Note the recent developments make it more difficult to draw
a clear separation line between process simulation software
and general purpose modelling tools such us MatLab or
Python (the gPROMS process simulation environment is a
typical example)

17 Bezzo 2024
 Structure of a process simulator

(Turton et al., 2018)

18 Bezzo 2024
Process flowsheeting
Degrees of Freedom
 If we have n independent equations in m (m > n) variables, then the
number of degrees of freedom (DOF) d is:

d= m − n
 Thus m−n of the variables must receive their values from other
means (e.g. additional equations) before we can solve the problem
 What is the minimum number of variables required to specify fully a
stream? Duhem’s theorem states that whatever the number of
phases, of components or of chemical reactions, the equilibrium state of a
closed system, for which we know the initial masses of each component, is
completely determined by two independent variables.
 It can be proved (Westerberg et al, 1979) that the same holds for a stream.
In other words, a stream is fully characterised once we know the
molar flowrate of each component and 2 additional variables
(typically temperature and pressure).

19 Bezzo 2024
Process flowsheeting
Degrees of Freedom
 Let us consider a flowsheet and assess the number of degrees
of freedom we have:
 Let us consider the following block flow diagram assuming it
represents a simplified ammonia production process where four
species (c = 4) are accounted for (NH3, N2, H2, Ar)
s7 s6
splitter

B4
s5

s1 s2 s3 s4
mixer reactor separator

B1 B2 B3

 To characterise each stream in terms of mass and energy balances,

we need c + 2 variables. Thus, in this example there are 6×7 = 42
stream variables to be determined.
20 Bezzo 2024
Process flowsheeting
Degrees of Freedom
 How many equations can we provide?
 For each unit operation we can write 4 mass and 1 energy balances,
i.e. 20 balances altogether ( d = 42−20 = 22)
 Since momentum balances are typically not solved in process
simulators, we can assume that all stream pressure levels are given.
This provides 7 values for unknown variables ( d = 22−7 = 15)
 Note that typically we do not set the stream pressures, but rather set the
operating pressure (or pressure drops) in a unit: the streams exiting the
unit are assigned the same pressure as the unit itself
 The mixer and the reactor do not provide new variables
 In the reactor additional variables may need to be assigned such as the
equilibrium constant or the conversion or the extents of reaction, and so
on. The variables to be assigned depend on reactor model being used (e.g.
CSTR, conversion reactor, equilibrium reactor, and so on).

21 Bezzo 2024
Process flowsheeting
Degrees of Freedom
 If the separation unit is a distillation column, the two outlet
streams (i.e. since pressure levels are set, 10 unknown variables) are
fully specified once 4 variables are assigned (e.g. two product
specifications, reflux ratio, and feed stage). Apart from the five mass
and energy balances (already considered), it means that the equations
of the distillation models provide the missing relationship to
determine all unknowns ( d = 15−5 = 10)
 If the separation unit is a flash unit, 2 variables are assigned (usually
temperature and pressure, but note that pressure has already been
accounted for), whereas 4 additional equations, one per component, are
provided by phase equilibria constraints (again 5 DOF are specified)
 One additional equation is provided in the splitter by providing the
split fraction, which determine 4 additional equations (one per each
component); we also constrain the temperatures of streams s6 and s7
to be the same and equal to the temperature of s5; note that one of
these equation is equivalent to the energy balance, thus only one
additional equation is added ( d = 10−5 = 5)
 The remaining 5 DOF are set by specifying the c + 1 = 5 feed
variables (pressure was set already)
22 Bezzo 2024
Process flowsheeting
Degrees of Freedom (DOF)
 Note that in general we cannot arbitrarily choose the
combination of variables to specify the DOF, and that may not
be a trivial task
 However, the great advantage of process simulation software is
that the simulator “tells us” which variables need
specifying so that the number of equations is equal to the
number of unknown variables and we can solve the model
 Sometimes alternative options are available (e.g. in a flash unit, we
can select temperature and pressure or temperature and vapour
fraction of the mixture)
 The choice of the variables to specify may have an impact on
convergence

23 Bezzo 2024
 Process simulation

Chemical components and property models

Physical property models
General comments
 Thermodynamic and physical properties are the back bone of
most chemical engineering flowsheets (and models)
 To solve a flowsheet the software needs to calculate things like
vapor-liquid and chemical equilibrium equations, enthalpies,
densities etc.
 Selecting the right property and thermodynamic
models is crucial to get the correct answer
 Select components, thermodynamic model and database

25 Bezzo 2024
Pure components
 Process simulation programs contains a large data bank of
pure component compounds
 NOTE: The fact that a pure component is listed in a
simulator data bank does not guarantee that any of the
properties given for that component are based on
measured data. If the properties of a compound are critical to
process performance, then the scientific literature should be
consulted
 A key decision is choosing the right number of
components
 If too few components are used, the model will not correctly predict
the performance of reactors and separation equipment
 If too many components are used, the model can become difficult to
converge, particularly if there are multiple recycles in the design

26 Bezzo 2024
Pure components
Guidelines for choosing them (Towler and Sinnott, 2022)

1. Always include any component that has a specified limit in any of the
products if that component is present in any of the feeds or could be formed in
the process
 Critical to determining whether the separations are meeting product specifications
2. Always include any component that has a specified limit in any of the feeds
 They can be a source of by-products or can act as catalyst or enzyme inhibitors. They must
be tracked to ensure that they do not accumulate in the process
3. Always include components that are expected to be formed in side reactions
or consecutive reactions
 Important to understand where these components will accumulate or leave the process
4. Always include any compounds that are expected to be present and are known
to have significant health, safety, or environmental concerns
 They must be tracked to make sure that they do not reach unsafe levels in any stream
5. Usually include any compound that might be present at a mass or mole fraction
greater than 2% in any process stream
6. Do not include isomers unless the process specifically requires distinction
between isomers
 Considering all of the possible isomers of organic compounds becomes combinatorically
explosive at high carbon numbers

27 Bezzo 2024
Pure components
Other components
 Pseudocomponents
 Pseudocomponents are components created to match the boiling curves of petroleum
mixtures. Crude oil, fuels such as gasoline, kerosene and diesel consist of many different
hydrocarbon compounds. (e.g., for diesel, crude oil and heavy fuel oils, the number of
possible compounds can be from 104 to >106)
 A large number of compounds with boiling points in a given range are “lumped” together
and represented by a single pseudocomponent with a boiling point in the middle of that
range. A set of ten to thirty pseudocomponents can then be fitted to any petroleum assay
and used to model that oil
 Solid and salts
 Most chemical and pharmaceutical processes involve some degree of solids handling
(crystals, insoluble salts, cells, coal and ash particles, catalysts…)
 The way they are handled depends on the process simulator (often a distinction is made
among i) materials with measurable properties that can participate in any of the phase or
reaction equilibria; ii) materials that can participate only in reaction equilibrium; and iii)
materials that are not involved in either phase or reaction equilibrium and behave like
inerts)
 User components
 Components not available in the software database can be created by the user, typically
based on the molecular structure. Minimum required information is the molecular weight
and often the normal boiling point (additional data can be added by the user)

28 Bezzo 2024
Physical property models
General comments
 In general any property model comprise:
 One or more mathematical equations that allow to calculate the property of
interest as a function of the variables of the process stream or unit (such as
temperature, pressure and composition). These equations contain a number of
parameters, which are defined as physical variables whose value is however
related to (and fixed for) the specific component or mixture
 The values of the model parameters for the specific component or mixture
 A range of applicability of the model (in terms of ranges of relevant process
variables). A model should never be used outside these ranges
 Information about the accuracy in the calculation of the property of interest.
This is important to evaluate how much the process calculation is affected by the
property model inaccuracy
 It is always a good practice to benchmark the physical properties predicted
using a process simulation program against measured data.
 There are many good literature sources of data for properties of individual
compounds, but much less data are available for mixtures
 Be careful when inputting external data: the reference state should be
consistent with that used by the simulator: this is especially important with
enthalpy data, particularly in reactive systems

29 Bezzo 2024
Physical property models
Sources for property data
 International Critical Tables (1933): probably the most comprehensive compilation of physical properties and
is available in most reference libraries
 Touloukian, Y. S. (1970). Thermophysical properties of matter, TPRC data services. Plenum Press. 1977.
 The Engineering Sciences Data Unit (ESDU, www.ihsesdu.com) was set up to provide validated data for
engineering design, developed under the guidance and approval of engineers from industry, the universities,
and research laboratories
 The NIST (National Institute of Standards and Technology) Source Database comprises over 4 million
experimental data points. It includes data for over 24000 pure components and 30000 binary mixtures
 the DECHEMA vapor-liquid and liquid-liquid data collection (DECHEMA, 1977): experimental phase equilibrium
data have been published for several thousand binary and many multicomponent systems.Virtually all the
published experimental data have been collected together in the DECHEMA databank
 The Journal of Chemical Engineering Data specializes in publishing physical property data for use in chemical
engineering design
 the Physical Property Data Service (PPDS) and the The Design Institute for Physical Properties (DIPPR) databases
 PPDS was originally developed in the United Kingdom by the Institution of Chemical Engineers and the
National Physical Laboratory. It is now available as a Microsoft Windows version from NEL, a division of the
TUV Suddeutschland Group (www.tuvnel.com/content/ppds.aspx)
 The DIPPR projects are aimed at providing evaluated process design data for the design of chemical processes
and equipment (www.aiche.org/TechnicalSocieties/DIPPR/index.aspx)
 Many sources of engineering information are subscription services. The American Chemical Society’s
Chemical Abstracts Service is the best source for chemical properties and reaction kinetics data. Chemical
abstracts can be searched online through the SciFinder subscription service (www.cas.org)

30 Bezzo 2024
Physical property models
Prediction of physical properties
 The process simulation programs contain subroutines that predict the physical
properties of pure compounds and mixtures as functions of temperature, pressure,
and composition.
 Techniques are available for the prediction of most physical properties, with
sufficient accuracy for use in process and equipment design
 However, the accuracy of the predictions should always be assessed by comparing the
model output with data from experiments, pilot plants, or operating units.
 Specialized texts on physical property estimation should be consulted, such as those by
Reid et al. (1987), Poling et al. (2000), Bretsznajder (1971), Sterbacek et al. (1979), and
AIChE (1983, 1985), and the data should be confirmed experimentally.
 The two most common approaches used in predicting properties are group
contribution methods and the use of reduced properties
 Group contribution techniques are based on the concept that a particular physical
property of a compound can be considered to be made up of contributions from the
constituent atoms, groups, and bonds, with the contributions being determined from
experimental data
 Reduced property models (also known as method of corresponding states models) predict
properties based on knowledge of the critical conditions of a compound. They are useful if
values for the critical properties are available or can be estimated with sufficient accuracy

31 Bezzo 2024
Group contribution method
Example

Example 4.1 (Towler & Sinnott, 2022)

Using Chueh and Swanson’s method, estimate
the specific heat capacity of ethyl bromide at
20°C.

Solution
Ethyl bromide CH3CH2Br

Molecular weight is 109, thus the specific heat

capacity is 104.92/109 = 0.96 kJ/kg°C

The experimental value is 0.90 kJ/kg°C

32 Bezzo 2024
Pure components
Setting components in a process simulator: hands-on

Consider species acetophenone (C8H8O). Include it as a component in a process

simulator.

Use both Aspen Plus and PRO/II.

33 Bezzo 2024
Pure components
User components: hands-on

Consider species 4-isobutylacetophenone (C12H16O):

Use both Aspen Plus and PRO/II to include it in the list of components for simulation.
From the literature, the normal boiling point of C12H16O is equal to 277 °C.

Do you obtain a similar result?

Can you use experimental data directly?

34 Bezzo 2024
Physical property models
Phase equilibrium models
 Everything from energy balances to volumetric flowrates to separation in
equilibrium-stage units depends on accurate thermodynamic data
 Note that only few data are required to estimate parameters in simple
thermodynamic models
 For cubic equations, critical temperature and pressure of pure components are
sufficient to estimate model parameters
 However, blindly using thermodynamic models without experimental data will
miscalculate the entire flowsheet!
 A safe choice of a thermodynamic model requires knowledge of the
system, of the thermodynamic model, of the calculation options
of the simulator, and the margin of error
 If the thermodynamic option used by the simulator is a mystery, the meaning of
the results obtained by the process simulator will be equally mysterious
 In real-life practice, the choice and tuning of the thermodynamic model is
often the most demanding and critical step, and may add up to 30-40% of
the overall time needed to set up and converge the model

35 Bezzo 2024
Physical property models
Phase equilibrium models
 Whenever possible, phase equilibrium data should be used to
regress the parameters in the model, and the deviations between
the model predictions and the experimental data should be studied
 There are two types of fugacity models: equations of state (EoS),
used in the ϕ −ϕ approach for equilibrium calculations, and activity
coefficient models, used in the γ −ϕ approach
 An EoS is an algebraic equation for the pressure of a mixture as a
function of composition, temperature, and volume
 Activity coefficient models provide algebraic equations for the activity
coefficients (γi) as a function of composition and temperature
 For theoretical explanations about main thermodynamic models,
refer to relevant courses attended in the past
 A practical user-oriented approach will be considered here
 From now on, it will be assumed that EoS are cubic EoS, but different
equations, too, were proposed for specific systems

36 Bezzo 2024
Thermodynamic models
Equations of state
 Little known fact: all cubic EoS are comparable in terms of
calculation accuracy, if their parameters are estimated using a
suitable procedure (Bertucco and Fermeglia (2023), Fluid Phase Equilibria, 566,
113678)
 Because of “tradition” and the way they are implemented in
commercial simulators, only two EoS are used for most
applications:
 The SRK (Soave-Redlich-Kwong) EoS (Soave, 1972)
 The PR (Peng-Robinson) EoS (Peng and Robinson, 1976)
 Normally they use three pure-component parameters, and one
binary-interaction parameter (BIP) per binary pair

37 Bezzo 2024
Thermodynamic models
Equations of state
 Starting from the well-known van der Waals EoS (with ac and bc estimated at
critical conditions):
RT a
= P − 2c
v − bc v
Soave’s intuition was to add a temperature-dependent third parameter α:
RT α (T )ac
=P −
v − bc v2
which can be obtained by solving the isofugacity equation for a given T once the
(experimental) pure vapour pressure Pi* is known
ϕi*,V ( Pi* , T ,α ) = ϕi*, L ( Pi* , T ,α )
 A similar approach was proposed by Peng and Robinson with the goal of improving
the molar volume prediction (still not satisfactory) and performance around the
critical point.
RT α (T )ac
= P −
v − bc v ( v + bc ) + bc ( v − bc )

38 Bezzo 2024
Thermodynamic models
Equations of state
 For non-polar components Soave proposed the following function for α:

α (T ) =
1 − m 1 − T Tc
m=C1 + C2ω + C3ω 2
where ω is the acentric factor, and C1, C2, and C3 are constants, whose
values were estimated over extensive vapour pressure data
 For polar components, α(T) requires a regression on experimental data
 Since the iso-fugacity condition cannot be solved for components in super-
critical conditions, suitable functions were proposed to extrapolate alpha values
fitted below the critical temperature (a special equation is needed for hydrogen
(Mathias, 1983))
 The problem of evaluating α has no solution when vapour pressure data cannot
be obtained. As a consequence, EoS cannot be used for polymers or
electrolytes
 Note that the capability of EoS to represent VLE does not imply that they
can be used to estimate molar volumes, enthalpies, etc. accurately
 An additional parameter was introduced to represent molar volume (the so-
called Peneloux translation (Peneloux and Rauzy, 1982); this does not affect the EoS
capability to represent VLE
39 Bezzo 2024
Thermodynamic models
Equations of state
 The extension to fluid mixtures is based on the assumption that a a
mixture may be described as a pure fluid, whose properties are a “blend” of
the components in the mixture (one-fluid theory). This implies that
vapour and liquid properties can be calculated using a standard EoS (e.g.,
RKS or PR)
 The only difference is that parameters ac and bc are substituted with mixture
parameters am and bm, which are defined according to a mixing rule. The
classical mixing rule is as follows:
NC NC
=am ∑∑ x x i j ac ,i ac , j (1 − kij )
=i 1 =j 1
NC
bm = ∑ xibc ,i
i =1

where NC is the number of components, and kij are the BIPs.

 Other mixing rules were proposed. In particular, it was found that the classical
mixing rule is unsatisfactory for polar components. One solution is given by the
mixing rule proposed by Huron and Vidal (1979), which should be used
whenever an EoS is applied in the presence of strongly non-ideal mixtures.

40 Bezzo 2024
Thermodynamic models
Activity coefficient models
 Activity coefficient models are based on the γ −ϕ approach to
represent phase equilibrium:
 1 P L 
ˆ
ϕi yi P = Pi xiγ iϕi exp  ∫ * i 
V * *
v d P
 RT Pi
 Note that values for the pure vapour pressures at the temperature of the
system are needed. For components with a critical temperature lower than the
system temperature, activity coefficients cannot be defined and Henry constants
are used. Such components need to be defined as a “Henry component” in
process simulators.
 Available models aims at representing the γ term
 As no activity coefficient models available include a pressure dependence, the γ −
ϕ approach is limited to the pressure range between 0 and 30-40 bar
 If no BIPs are available, an activity coefficient model will give similar results as
those for an ideal solution
 BIPs are typically highly correlated: very different BIPs can lead to similar
fits of experimental data
 Parameters regressed from VLE data are often unreliable when used for LLE
predictions (and vice versa)

41 Bezzo 2024
Thermodynamic models
Activity coefficient models
 Among the available local composition models, the Non-Random Two-
Liquid (NRTL – Renon and Prausnitz, 1968) and UNIversal
QUAsiChemical (UNIQUAC – Renon and Abrams, 1967) are the best
ones for process simulation purposes. Key assumptions are (Wilson, 1964):
 interactions among molecules are expressed in terms of binary parameters only
 the temperature dependency of these parameters is made explicit
 NRTL is theoretically simpler, but requires three adjustable parameters
per component pair, whereas UNIQUAC has sounder theoretical
foundations and requires two adjustable BIPs only
 There are no reliable rules to choose an activity coefficient model a priori

Example: NRTL equations

The 3 parameters are the two BIPs τ12 and τ21,
and the non-randomness parameter αij (it is
assumed αij = αji).
Also note that in general τij is a function of
temperature.

42 Bezzo 2024
 Thermodynamic models
Example: setting BIPs to 0

VLE for Ethanol/Water at 1 atm:

solid curves are for CHEMCAD
databank BIPs for NRTL; dotted
curves are for NRTL BIPs set to zero

(Turton et al., 2018)

43 Bezzo 2024
Thermodynamic models
Predictive models
 The most famous of fully predictive models is the UNIQUAC Functional-
group Activity Coefficients model (UNIFAC, Fredenslund et al, 1977)
 UNIFAC was proposed to calculate the activity coefficient γi of a
component i in a liquid mixture starting from the activity coefficients of
functional groups.
 It is denoted as fully predictive because it only requires the knowledge of the
chemical structures of all components in the mixture to be applied (plus the
mixture composition)
 UNIFAC is based on group contributions, where each functional group in the
molecules contributes to the calculation of γi, which depends on the group
activity coefficients Γk . They are evaluated applying the UNIQUAC models to
groups (group BIPs are based on fitting a large VLE database)
 PRACTICAL NOTE: when determining the number of groups in a molecule, always
start with the largest group
 UNIFAC main advantage is its applicability in the absence of data. However:
 it is impossible to check the calculation accuracy of UNIFAC (there are no
data)
 a posteriori validations tests demonstrated that UNIFAC may introduce
large and unpredictable errors

44 Bezzo 2024
Thermodynamic models
Special systems
Electrolyte systems
 Several important chemical engineering applications involve electrolytes
 Gas treatment processes (e.g. acid gas removal processes with alkaline-salt
solutions)
 Wastewater treatment processes (e.g. stripping processes involving ammonia)
 Sour water systems: the term sour water is used for water containing carbon
dioxide, hydrogen sulfide, and ammonia encountered in refinery operations.
 Electrochemical processes (e.g. fuel cells, batteries, electrolysis, corrosion
modelling)
 Separation processes (e.g. extractive distillation, seawater desalination, and
solution crystallization)
 When water and salts are present in a mixture, the salts can dissociate into
ions in aqueous solution.
 The phase equilibrium model must account for dissociation and the presence of
long-range interactions between charges on ions
 Special electrolyte models and databases have been developed for
electrolyte systems. These models are available in the commercial process
simulation programs (they may require an additional fee)

45 Bezzo 2024
Thermodynamic models
Special systems
Equilibria with a solid phase
 Solid-liquid equilibrium (SLE) is important in several process operations
(e.g., crystallisation) and metallurgical processes. Some process simulators
have empirical or semi-empirical approaches for calculating SLE.
 In terms of activity coefficients, the isofugacity condition can be expressed
as:
 T H iL − H iS 
x γ = z γ exp  ∫
i i
L
i i
S
2
d T 
 Tmi RT 

where Tmi is the melting temperature of pure species i.

For calculating γiL a suitable activity coefficient model is used. For
calculating γiS a correlation such as the Margules or van’t Hoff equations is
often used.
 There are also systems where solid-vapour equilibrium (SVE) is important
(e.g., lyophilisation processes or the formation of gas hydrates). A γ −ϕ
approach is typically adopted with γiS computed as above.

46 Bezzo 2024
Selecting a thermodynamic model
General guidelines (I)
 EoS usage is typically recommended for simple systems (non-polar, small
molecules) and high pressure (including super-critical) conditions
 Activity coefficient model are a reasonable choice for polar molecules
and whenever liquid-phase interactions are non-ideal. They should not be
used at high pressure (>10 bar) or when some components are in super-
critical conditions
 The UNIFAC method is not recommended for process simulation
applications. The specific limitations of the method are:
 Pressure not greater than a few bar (say, limit to 5 bar)
 Temperature below 150 °C
 No non-condensable components or electrolytes
 Components must not contain more than 10 functional groups
 Remember that the availability of a VLE model does not imply that
parameters (especially BIPs) are also available for a specific binary system

47 Bezzo 2024
Selecting a thermodynamic model
General guidelines (II)
 EoS  Activity coefficient models
 Good for vapor phase  Good for liquid phase modeling
modeling and liquids of low only
polarity  Many binary parameters required
 Fewer binary parameters  Binary parameters are highly
required temperature dependent
 Parameters extrapolated  Inconsistent in critical region
reasonably with temperature
 Can represent highly non-ideal
 Consistent in critical region liquids
 Typically limited in ability to
represent non-ideal liquids

48 Bezzo 2024
Selecting a thermodynamic model
General guidelines

Are there any POLAR

components in the system?
No Yes

Are the operating conditions near

Use EOS Model Yes
the CRITICAL region of the mixture?

No
Are there LIGHT gases or
SUPERCRITICAL components
in your system?
No Yes

Use Activity Use Activity Coefficient

Coefficient model Model with Henry’s Law

(PATH, Siemens, 2019)

49 Bezzo 2024
Selecting a thermodynamic model
General guidelines

50 Bezzo 2024
 Selecting a thermodynamic model
General guidelines

B-W-R: Benedict-Webb-Rubin equation

C-S: Chao-Seader equation
G-S: Grayson-Streed equation
P-R: Peng-Robinson equation
R-K: Redlich-Kwong equation
R-K-S: Redlich-Kwong-Soave equation (Towler and Sinnott, 2022)

51 Bezzo 2024
Selecting a thermodynamic model
General guidelines

(Jaubert and Privat,

Thermodynamic models for
chemical processes, 2021)

52 Bezzo 2024
Selecting a thermodynamic model
General guidelines

choose property choose

method components

check parameters

do they NO
get parameters
exist?

YES

NO are results YES start building the

reliable? flowsheet

53 Bezzo 2024
Selecting a thermodynamic model
10 golden rules (Fermeglia, 2021)
1. Check vapor pressure
2. Check pure and mixture densities – for aqueous mixtures excess
volumes are important
3. Check pure and mixture enthalpy and heat capacities
4. Check transport properties (for heat exchangers and trays)
5. Check surface tension if you design trays for distillation columns
6. Azeotropes: check if they are present
7. Check trace components behavior versus infinite dilution activity
coefficients
8. Talk to people, interact with chemists (new processes), talk to
process simulators vendors
9. Beware of estimation methods for screening alternatives
10. Check the simulation results versus the reality, talk to the plant
personnel, consider the reality (air leaks,…)
54 Bezzo 2024
Thermodynamic models
Final recommendations
 Always verify that all BIPs for all binary pairs are available in the
process simulator
 If some parameters are not available, look for phase equilibrium data (e.g., from
databanks, articles in the scientific literature, patents, …) to estimate missing
BIPs
 For binary pairs with no data, the UNIFAC estimation option of the
simulator can be used (be careful!)
 For any section of the process flowsheet it is suggested to select VLE binary data
at pressure and temperature close to the process operating conditions, and to
choose the thermodynamic model that is most suitable to represent those
process conditions (it is possible to use different models in different process
sections)
 A key issue that must be considered for accurate thermodynamic modelling
is the calculation of the pure-component vapour pressures
 The wrong estimation of vapour pressures may lead to significant errors in the
description of binary vapour–liquid equilibria: these errors cannot be
compensated for by optimally selecting the EoS BIPs
 Remember that few data are better than no data at all
 For example, data concerning azeotropic compositions are extremely helpful

55 Bezzo 2024
Thermodynamic models
Final recommendations
 Never assume that the availability of BIPs in the databank of a process
simulator is an indication that the model is of acceptable accuracy
 These parameters are merely the “best” for some specific objective function or
set of data (but they may be the best ones at T and P conditions being considered)
 Test different (suitable) models to simulate a unit operation (e.g. a flash
unit): if the calculation gives substantially different stream flows or
compositions, the designer should seek more experimental data to
determine which model is “best”
 Simulation differences are a way to quantify model uncertainty
 Often no single thermodynamic model can be used reliably to predict the
fugacities of such a wide rage of components as those that can be found in
real processes.
 Simulators allow using hybrid models: these may include Henry’s models for
non-condensable gases together with activity coefficient models or the usage of
EoS to describe the vapour phase while activity coefficient models are used to
represent the liquid phase
 Keep in mind that the description of VLE at high pressures is not trivial
and special care is needed for choosing and calibrating the thermodynamic
model
56 Bezzo 2024
Thermodynamic models
Compare thermodynamic models: hands-on
Use different thermodynamic models (EoS and activity coefficient models) to
compare the following binary systems (Barolo & Guarise, Esercizi di impianti chimici, 2008):
non poral- ideal behavior EoS

• benzene – toluene
one polar and other not
azeotrope point
non ideal
• ethanol – benzene activity
coefficient model
polarity
H bonds ----------> non ideal
water hydrocarbon

• water – i-butanol
activity
coefficient model

57 Bezzo 2024
Thermodynamic models
Compare thermodynamic models: hands-on (Turton et al, 2018)

Use the PR and SRK EoS to calculate the methane vapour molar flowrate
from a flash at the conditions set in the table.

Variable Value Compare the results for BIPs from the process
Temperature 225 K simulator databank and with BIPs set to zero.
Pressure 60.78 bar
Feed flowrates
CO2 6 kmol/h
H2S 24 kmol/h
CH4 66 kmol/h
C2H6 3 kmol/h
C3H8 1 kmol/h

58 Bezzo 2024
Thermodynamic models
Compare thermodynamic models: hands-on (Fermeglia, 2021)

A chemical plant has a settling tank that they use to decant off the water from
a mixture of water and chlorobenzene; the inlet stream to the tank also
contains some carbon-dioxide and nitrogen; the tank and feed are at ambient
conditions (25 °C, 1 atm) and have the following flow rates:
• Water: 230 kg/hr
• Chlorobenzene: 2000 kg/hr
• CO2: 340 kg/hr
• N2: 20 kg/hr

59 Bezzo 2024
Thermodynamic models
Validation and BIP regression: hands-on
The accuracy of the model parameter values is essential as it provides the
precise prediction of all thermodynamic properties.
The objectives of this exercise are:
a) Evaluate the VLE of a binary combination using the default thermodynamic
model of the process simulator;
b) Regress the binary interaction parameters of a thermodynamic model;
c) Predict the VLE of a binary-mixture at a desired operating condition.

Let us consider a binary mixture made of water and isopropanol.

Let us assume that the process being investigated operates in the pressure
range between 0.3 and 5 bar.

60 Bezzo 2024
Thermodynamic models
As a support to process design choices: hands-on (I)

Let us consider a mixture of methanol (CH3OH) and chloroform (CHCl3).

Produce T-xy VLE diagrams at pressures: 1.01325 bar, 5 bar, and 10 bar.

Now imagine that you wan to separate a stream containing 70%mol methanol
and 30%mol chloroform: what would you do?

(The example is taken from Adams II, Learn Aspen Plus in 24 hours, Tutorial 2)

61 Bezzo 2024
 Thermodynamic models
As a support to process design choices: hands-on (II)

(Adams II, Learn Aspen Plus in 24 hours, 2022)

62 Bezzo 2024
Process simulation

Simulation of unit operations

Unit operations
Library
 All commercial process simulation software has a library
of predefined unit operations
 Reactors
 Heat exchangers
 Distillation and separation units
 Hydraulics (pumps, compressors, valves…)
 Solid handling
 User models
 …

64 Bezzo 2024
Unit models
Reactors
 Conversion reactors
 A conversion reactor (or stoichiometric reactor) requires a
reaction stoichiometry and an extent of reaction (usually specified as
an extent of conversion of a limiting reagent)
 No reaction kinetics information is needed, so it can be used when
the kinetics are unknown or when the reaction is known to proceed
to full conversion.
 Conversion reactors can handle multiple reactions, but care is
needed in specifying the order in which they are solved if they use
the same limiting reagent
 Aspen Plus: RStoic ; Pro/II: Conversion Reactor

65 Bezzo 2024
Unit models
Reactors
 Equilibrium reactors
 An equilibrium reactor finds the equilibrium product distribution
for a specified set of stoichiometric reactions. Phase equilibrium is
also solved
 The engineer can enter the outlet temperature and pressure and let
the reactor model calculate the duty needed to reach that condition,
or else enter a heat duty and let the model predict the outlet
conditions from an energy balance
 An equilibrium reactor only solves the equations specified, so it is
useful in situations where one or more reactions equilibrate rapidly
while other reactions proceed much more slowly.
 A typical example is the steam reforming of methane to hydrogen. In this
process, the water-gas-shift reaction between water and carbon
monoxide reaches equilibrium rapidly at temperatures above 450 °C,
while methane conversion requires catalysis even at temperatures above
800 °C.
 Aspen Plus: REquil ; Pro/II: Equilibrium Reactor

66 Bezzo 2024
Unit models
Reactors
 Gibbs reactors
 The Gibbs reactor solves the full reaction (and optionally phase
equilibrium) of all species in the component list by minimization of the
Gibbs free energy, subject to the constraints of the feed mass balance
and either isothermal or adiabatic operation
 A Gibbs reactor can be specified with restrictions such as a temperature approach
to equilibrium or a fixed conversion of one species
 The Gibbs reactor is very useful when modelling a system that is known
to come to equilibrium, in particular, high-temperature processes
involving simple molecules. It is less useful when complex molecules
are present, as these usually have high Gibbs energy of formation;
consequently, very low concentrations of these species are predicted
 The designer must specify the components carefully when using a Gibbs reactor in
the model. If a component that is actually formed is not listed in the component
set, then the Gibbs reactor results will be meaningless. Furthermore, if some of
the species have high Gibbs free energy, their concentrations may not be properly
predicted by the model.
 Aspen Plus: RGibbs ; Pro/II: GIBBS Reactor
67 Bezzo 2024
Unit models
Reactors
 Continuous stirred tank reactors
 The CSTR is a model of the conventional well-mixed reactor. It can be
used when a model of the reaction kinetics is available and the reactor is
believed to be well mixed
 By specifying forward and reverse reactions, the CSTR model can model
equilibrium and rate-based reactions simultaneously.
 The main drawback of using the CSTR model is that a detailed
understanding of kinetics is necessary if byproducts are to be predicted
properly
 Aspen Plus: RCSTR ; Pro/II: CST Reactor
 Plug-flow reactors
 A PFR models the conventional plug-flow behaviour, assuming radial
mixing, but no axial dispersion. The reaction kinetics must be specified
and the model has the same limitations as the CSTR model
 Most of the simulators allow heat input or removal from a plug-flow
reactor
 Aspen Plus: RPlug ; Pro/II: Plug Flow Reactor

68 Bezzo 2024
Unit models
Reactors
 Yield shift reactors
 The yield shift reactor allows the designer to specify a yield pattern.
Yield shift reactors can be used when there is no model of the kinetics,
but some laboratory or pilot plant data are available, from which a yield
correlation can be established
 Yield shift reactors are particularly useful when modelling streams that
contain pseudocomponents, solids with a particle size distribution or
processes that form small amounts of many byproducts. These can all be
described easily in yield correlations, but can be difficult to model with
the other reactor types
 The main difficulty in using the yield shift reactor is in establishing the
yield correlation. If a single point, for example from a patent, is all that
is available, then entering the yield distribution is straightforward. If, on
the other hand, the purpose is to optimize the reactor conditions, then a
substantial set of data must be collected to build a response surface
model that accurately predicts yields over a wide enough range of
conditions.
 Aspen Plus: RYield

69 Bezzo 2024
Unit models
Reactors
 Batch reactors
 Batch reactors cannot be represented in steady-state process
simulation models, but process simulators (e.g. Aspen Plus) provide a way
to model batch reactors in continuous processes. A cyclic unit operation
is integrated into a continuous process by means of holding tanks.
 As illustrated in the figure, the continuous feed always feed into Tank 1 (never
stopping); Tank 1 is drained periodically, at the start of each batch reactor run; the
batch reactor runs in batches, typically with a fixed run time; when the batch
finishes, the products are drained to Tank 2; the contents of Tank 2 are
continuously drained to downstream process
 Batch reactors
 Aspen Plus: RBatch

70 Bezzo 2024
Unit models
Summary of reactor models in Aspen Plus

Model Stoichiometry Kinetics Rigorous Feed

RStoich Yes No No Any
RYield No No No Any
REquil Yes No No Any
RGibbs No No No Any
RBatch Yes Yes Yes 1
RCSTR Yes Yes Yes Any
RPlug Yes Yes Yes 1

71 Bezzo 2024
Unit models
Reactors
 Industrial reactors are usually more complex than the simple simulator
library models
 Real reactors usually involve multiple phases and have strong mass transfer,
heat transfer and mixing effects. The residence time distributions of real
reactors can be determined by tracer studies, and seldom match the simple
CSTR or PFR models
 Sometimes a combination of library models can be used to model the
reaction system.
 For example, a conversion reactor can be used to establish the conversion of
main feeds, followed by an equilibrium reactor that establishes an equilibrium
distribution among specified products. Similarly, reactors with complex mixing
patterns can be modelled as networks of CSTR and PFR models
 When using a combination of library models to simulate a reactor, it is a good
idea to group these models in a sub-flowsheet. The sub-flowsheet can be given a
suitable label such as “reactor” that indicates that all the unit operations it
contains represent a single piece of real equipment
 Detailed models of commercial reactors are usually written as user models

72 Bezzo 2024
Unit models
Reactors: hands-on (I)
Let us consider the batch reaction of Allyl Alcohol (AA) and Acetone (ACE) to
produce n-Propyl-Propionate (nPP), as shown in the following:

Ea
−
The reaction kinetics is described by the following power-law: − rAA =
ke RT
C AAC ACE
where k = 1.5 × 109 m3 kmol-1 min-1 and Ea = 6 × 107 J kmol-1, and CAA and CACE are
the molar concentration of AA and ACE, respectively.

Task 1: Simulate the process to produce nPP from 200 g/s of AA and 280 g/s of ACE
using an integrated homogeneous liquid-phase batch reactor system. The feed is at 30
°C and 1 bar, and the reactor should have a cooling system that maintains a constant
temperature of 30 °C. Assume no pressure drop and use NRTL model. Each batch
cycle is considered complete when the reaction has achieved 98% conversion of the
limiting reagent.
(The example is taken from Adams II, Learn Aspen Plus in 24 hours, Tutorial 7)
73 Bezzo 2024
Unit models
Reactors: hands-on (II)
Task 2: Assume that a lab technician took 5 g of AA and 7 g of ACE in separate bakers
and placed each baker in a bath of warm water with a heating control system to
maintain the temperature of 30 °C. He left the bakers in the baths until the content
also reached 30 °C. Once that was finished, he poured the contents of the AA baker
into the ACE one, and started the reaction. He collected sample of liquid every 2
minutes and determined the composition of each sample, recording the following data:
Sample time (min) Mole Fraction of AA Mole Fraction of nPP
2 0.2268 0.3513
4 0.1359 0.5320
6 0.0941 0.5724
8 0.0588 0.6632
10 0.0424 0.6517
12 0.0292 0.7017
Use the Regression functionality to find the pre-exponential coefficient k which best
fits the simulation model used with RBatch to the experimental data.

Task 3: Determine the volume of a liquid plug flow reactor to achieve 98.5%
conversion of Allyl Alcohol for the same inlet conditions as in Part 1. Keep the
diameter of the reactor at 40 cm and operate isothermally at 32 °C.
74 Bezzo 2024
Unit models
Reactors: tutorial

Use stoichiometric reactors and Gibbs reactors in Aspen Plus and PRO/II for the
following reaction:

(The example is taken from Adams II, Learn Aspen Plus in 24 hours, Tutorial 7)
75 Bezzo 2024
 Unit models
Reactors: a focus on stoichiometric coefficients

Let us consider the following reaction: H2S + 3/2O2 → SO2 + H2O

From the literature (An and Jung, Case Studies in Thermal Engineering, 21, 100694), we have:
Ea
−
ke pH 2 S pO1.52
− rH 2 S = RT

with k = 14 mol/(cm3∙s∙atm2.5) and Ea = 11000 cal/mol.

Let us implement the kinetics in both Pro/II and Aspen Plus, considering a CSTR with a
volume of 1783 L and operating at 850 °C and 1 atm. Consider the following inlet
flowrates for H2S, O2 and H2O, respectively: 200, 300, and 100 kmol/hr

Now write the kinetics as: 2H2S + 3O2 → 2SO2 + 2H2O

Does anything change?

76 Bezzo 2024
Unit models
Heat exchangers
 All of the commercial simulators include models for heaters, coolers,
heat exchangers, fired heaters, and air coolers
 The only inputs that are usually required on the process side are the
estimated pressure drop and either the outlet temperature or the duty
 A good initial estimate of pressure drop is 0.3 to 0.7 bar
 Heater, cooler, and heat exchanger models allow the design engineer to enter
estimates of film transfer coefficients and hence calculate the exchanger area
 Note that process-to-process heat exchangers (typically used for heat
integration based on, e.g., pinch analysis) introduce a number of recycle loops
making convergence difficult: their analysis via sub-problems simulation is often a
wiser choice
 Another problem that is often encountered when simulating heat exchangers is
temperature cross. A temperature cross occurs when the cold stream outlet
temperature is hotter than the hot stream outlet temperature. When this
occurs, many types of shell and tube heat exchangers give a very poor
approximation of counter-current flow, and consequently have low F-factors and
require large surface areas. In some process simulators, models will indicate if
the F factor is low, and the designer should split the exchanger into several shells
in series so that temperature cross is avoided.

77 Bezzo 2024
Unit models
Heat exchangers
 Aspen Plus and Pro/II provides several models for heat
exchangers. Only the most commonly used will be reviewed.
 In Aspen Plus the main models are Heater and HeatX.
 Heater is a sort of “half-heat exchanger”: it models only of the two
fluids in the heat exchange process (typically the process one). It is
convenient for fast calculations during conceptual design and to
estimate heating/cooling duties
 HeatX is a complete heat exchanger models and allows for a higher
input/output detail of information
 This model will be able to detect if there is a violation of the second law
of thermodynamics leading to a temperature crossover
 In Pro/II, the equivalent models are Simple HX and
Rigorous HX
 Additional models for air coolers or fired heaters are available, too.

78 Bezzo 2024
Unit models
Heat exchangers
 One interesting feature in the Aspen Plus Heater block is
that it can be used to model complete heat exchangers by
increasing numerical robustness, particularly in complex
flowsheets
 This is done by exploiting a heat stream (available in Aspen Plus)
 Discuss the following example (Adams II, Learn Aspen in 24 hours, Tutorial 4, 2022)

79 Bezzo 2024
Unit models
Heat exchangers: hands-on

80 Bezzo 2024
Unit models
Distillation and separation units
 The commercial process simulators contain a range of
distillation models with different degrees of sophistication.
 The design engineer must choose a model that is suitable for the
purpose, depending on the problem type, the extent of design
information available, and the level of detail required in the solution.
 In some cases, it may make sense to build different versions of the
flowsheet, using different levels of detail in the distillation models, so
that the simpler model can be used to initialize a more detailed
model.

81 Bezzo 2024
Unit models
Distillation and separation units
 The simplest distillation models to set up are the shortcut models based
on Fenske-Underwood-Gilliland or Winn-Underwood-Gilliland methods to
determine i) the minimum reflux and number of stages, or ii) the required
reflux given a number of trays, or iii) the required number of trays for a
given reflux ratio
 The shortcut models can also estimate the condenser and reboiler duties and
determine the optimum feed tray
 Minimum information needed to specify a shortcut distillation model is:
 The component recoveries of the light and heavy key components
 The condenser and reboiler pressures
 Whether the column has a total or partial condenser
 In some cases, the designer can specify the purities of the light and heavy
key components in the bottoms and distillate, respectively.
 Care is needed when using purity as a specification, as it is easy to specify
purities or combinations of purity and recovery that are infeasible
 Shortcut distillation models are robust, and they are often used to initialize
a rigorous simulation
 Main drawback of shortcut models is that they assume constant relative
volatility
82 Bezzo 2024
Unit models
Short-cut distillation models: hands-on

Use the short-cut distillation models in Aspen e Pro/II to characterize a

distillation unit separating a stream whose flowrate is 100 kmol/h and the
molar composition is 50% water and 50% methanol at 25 °C and 1.4 bar.
The goal is to separate the feed into 99% methanol and 99% water.

(The example is taken from Adams II, Learn Aspen Plus in 24 hours, Tutorial 8, with tips from Tutorial 5)

83 Bezzo 2024
Unit models
Distillation and separation units
 Rigorous models carry out full stage-by-stage mass and energy balances
 Rigorous models allow many more column configurations, including use of side
streams, intermediate condensers and reboilers, multiple feeds and side strippers
and rectifiers
 Rigorous models can be much harder to converge
 The two main types of rigorous distillation models are equilibrium-stage
models and rate-based models
 Equilibrium-stage models assume either full VLE on each stage or else an
approach to equilibrium based on a stage efficiency entered by the designer.
When using an equilibrium-stage model for column sizing, the stage efficiencies
must be entered
 Rate-based models do not assume phase equilibrium, and instead solve the
interphase mass transfer and heat transfer equations. Rate-based models are
more realistic than the idealized equilibrium-stage models, but because it can be
difficult to predict the interfacial area and mass transfer coefficients, rate-based
models are less widely used in practice
 Rigorous distillation models can be used to model absorber columns,
stripper columns, refluxed absorbers, three-phase systems such as
extractive distillation columns, …

84 Bezzo 2024
Unit models
Distillation and separation units
 Other multistage vapor-liquid separations such as absorption and stripping
can be modeled using variations of the rigorous distillation models, as can
multistage liquid-liquid extraction
 Single-stage liquid-liquid or vapor-liquid separation can be modeled as a
flash vessel, but some caution is needed. The simulation programs assume
perfect separation in a flash unless the designer specifies otherwise
 If there is entrainment of droplets or bubbles, the outlet compositions of a real
flash vessel will be different from those predicted by the simulation
 Most of the simulation programs allow the designer to specify a fraction of each
phase that is entrained with the other phases
 The following modules are available:
 Aspen Plus: Flash2 and Flash3 (flash units), DSTWU (short-cut column),
RadFrac (equilibrium-stage column or rate-stage column), MultiFrac (3-
phase columns absorber/stripper combinations), PetroFrac (petroleum
refining)
 Pro/II: Flash (flash unit), Shortcut (short-cut column), Distillation
(equilibrium-stage or rate-stage columns)

85 Bezzo 2024
Unit models
Distillation and separation units
 Convergence of distillation column models is probably the most common
problem in process simulation. Possible causes of convergence problems
are as follows:
 Infeasible specifications: care must be taken when using purity as a
specification (e.g., it is not possible to make a 99% pure distillate if the feed
contains 2% of a component that boils lighter than the light key)
 Poor initialisation: if the simulation model is set up with fewer than the
minimum number of trays or less than the minimum reflux, it will not converge.
Shortcut models can be used to provide an initial design that is feasible, which
can then be used as a starting point for a rigorous model
 Poor initial estimates: The faster distillation column solution algorithms run
much better when provided with a good initial estimate of tray temperatures
 A good practice is to start by using simple hand calculations to check
that the column specifications are feasible  check mass balances!
 A shortcut model can then be used to determine the minimum reflux and the
number of trays and feed tray location. The rigorous model should then be
initialized with the reboiler and condenser temperatures found from the
shortcut simulation, a linear temperature profile, and pressure gradient that
makes an approximate allowance for the pressure drop per tray (about 5 cm of
liquid is typical).

86 Bezzo 2024
Unit models
Distillation and separation units
 Different algorithms are available in commercial simulation packages. They
vary in terms of rapidity and robustness.
 The inside-out algorithm is the most commonly used in distillation
problems. It decomposes the stage equations into two nested iteration
loops
 If it fails, other methods can be tried (e.g., in PRO/II: CHEMDIST can be a good
choice for systems that form two liquid phases or azeotropes, or systems that
have free water (thermodynamic models based on liquid activity coefficients are needed); SURE
is often the method with better convergence, but it is slow – in Aspen Plus: there
is a convergence menu where several parameters can be changed: increasing the
number of maximum number of iterations or turning on the numerical Jacobian
matrix calculation may help)
 Note the some design specifications are “easier” than other ones.
 For instance, it is good practice to initially run the rigorous model with easy
specifications, such as reflux ratio and either distillate rate or bottoms rate.
Typically, these will guarantee convergence and generate a realistic column
temperature profile that can then be stored as the estimate for future runs. The
model specifications can then be changed to the desired design specifications
(purity or recovery specifications).
 Sometimes, adding trays may help convergence

87 Bezzo 2024
Unit models
Distillation: hands-on

Use specifications and results obtained from the short-cut distillation model to
set up a rigorous distillation simulation both in Aspen Plus and Pro/II.

(The example is taken from Adams II, Learn Aspen Plus in 24 hours, Tutorial 8, with tips from Tutorial 5)

88 Bezzo 2024
Unit models
Mixing and splitting units
 All process simulation software provides simplified units for
mixing and splitting streams
 Mixers solve the following balance (where fij is the flowrate of
component i in stream j): f i,1
fi,2 fi,out
f i ,out = ∑ f ij
fi,S
j

 Splitting units can be divided into (assuming two product

stream):
fi,1
 Component-splitters: fi,in
fi,2
f i ,1= ξi f i ,in f i ,2= (1 − ξi ) fi ,in
 Stream-dividers (splitters), where all streams have the same
composition:
f i ,1= α f i ,in f i ,2= (1 − α ) fi ,in

89 Bezzo 2024
Unit models
Fluid pumping
 All process simulators contains models for pumps, compressors,
turbines, and valves
 For pumps and valves, the minimum requirement is the discharge pressure (or
the pressure increase/drop)
 For compressors and turbines, apart from pressure specifications, the mode of
compression/expansion (adiabatic, isentropic, polytropic) is needed, too
 Several process simulation programs contain models for pipe segments,
tees, and elbows, which can be used to make an initial estimate of system
pressure drop for the purposes of sizing pumps and compressors.
 If a process hydraulic model is built, care must be taken to specify pressure drop
properly in the unit operation models. Rules of thumb are adequate for initial
estimates, but in a hydraulic model these should be replaced with rigorous
pressure drop calculations.
 Care is needed when modeling compressible gas flows, flows of vapor-liquid
mixtures, slurry flows, and flows of non-Newtonian liquids. Some simulators use
different pipe models for compressible flow.
 In most of these cases, the simulation model should be replaced by a
computational fluid dynamics (CFD) model of the important parts of the plant.

90 Bezzo 2024
Unit models
Solids handling
 The commercial simulation programs were originally developed
mainly for petrochemical applications, and none of them has a
complete set of solids-handling operations.
 Although models for filters, crystallizers, decanters, and cyclones are
present in many simulators, user models may have to be created for
different operations
 Because solids are handled in many commodity chemical processes
as well as pharmaceuticals, polymers, and biological processes, the
simulation software vendors are under pressure from their
customers to enhance the capability of the programs for modeling
solids operations (e.g. gPROMS FormulatedProducts).

91 Bezzo 2024
Unit models
User models
 Spreadsheet models
 Models that require no internal iteration are easily coded as spreadsheets. Most
of the simulators offer some degree of spreadsheet capability, ranging from
simple calculation blocks to full Microsoft Excel functionality
 Sometimes you may simply want to get data from Aspen Plus to Excel
 For example, Aspen Plus can exploit spreadsheet capability using Microsoft Excel,
which can be specified as a calculator block. For more sophisticated spreadsheet
models, Aspen Plus allows the user to link a spreadsheet to a simulation via a
user model. Similar functionalities can be found in Pro/II.
 You can use Excel to change things inside Aspen Plus, and therefore you can write your
own custom model into Excel or you can even write a program in Visual Basic inside
Excel that runs Aspen Plus, making changes as you go.
 User subroutines
 Models that require internal convergence are best written as subroutines rather
than spreadsheets, as more efficient solution algorithms can be used. Most user
subroutines are written in FORTRAN or Visual Basic or Python, though some of
the simulators allow other programming languages to be used (e.g. gPROMS is
coded to allow the user create their models using the simulator programming
language).
 Detailed instructions on how to write user models to interface with commercial
simulation programs can be found in the simulator manuals.
92 Bezzo 2024
Unit models
Spreadsheet models: hands-on

This tutorial shows how to link Aspen Plus and Excel

(The example is taken from Adams II, Learn Aspen Plus in 24 hours, Tutorial 9)

93 Bezzo 2024
Process simulation

Flowsheeting modelling
Computational blocks
Design and calculator blocks
 Other than the unit operation blocks, process simulators have two types of
computational blocks that need to be solved (if implemented)
 The first type (sometimes called design block) satisfies the design specifications
and works like a feedback controller: it satisfies the design specification by
manipulating an input variable to a block or some specification of a block within
some given bounds
 Aspen Plus: Design spec; PRO/II: Controller
 The second type (sometimes called calculator block) calculates the value of
some parameter or variable or input condition based on user-defined
calculations. These blocks may be compared to a feedforward controller
 Aspen Plus: Calculator; PRO/II: Calculator
 Note that differently from control systems, computational blocks do not
consider the system dynamics, and just represent the steady-state
condition reached by the control system
 Note that design/calculator blocks may determine relationships between
units that are similar to the ones caused by recycles

95 Bezzo 2024
Design and calculator block
Calculator example: hands-on (I)

Let us consider the methane reforming reaction:

CH4 + H2O  CO + 3H2

and define the following flowsheet (equilibrium reactor, SRK EoS):

Isothermal, 32 bar, 925 °C
Methane (32 bar, 75 °C, 200 kmol/hr)
Reactor Wet syngas
Steam (saturated, 32 bar)

We will use the calculator block to overwrite a value, and then to make a
helpful calculation.

(The example is taken from Adams II, Learn Aspen Plus in 24 hours, Tutorial 9)

96 Bezzo 2024
Design and calculator block
Calculator example: hands-on (II)

Step 2:

Step 3:

97 Bezzo 2024
Process flowsheeting
Degrees of Freedom
 DOF analysis
 While building a flowsheet step by step, we need to make sure the
DOF is satisfied
 This can be achieved by entering data in the dialogue boxes:
commercial software issues an error message if DOF is not satisfied
 Note that DOF depends on the process configuration: in some
cases the degree of freedom assignment may change
 E.g.: forming recycles

?
? ?

Unit3
Unit3

Unit1 Unit2 Unit1 Unit2

98 Bezzo 2024
 Process flowsheeting
Degrees of Freedom (DOF)
 Computational blocks may change DOF
 Usually, we specify source 1 and 2, simulate the rest of the process
 If we know the production, or we have a target of the production (e.g. Product1),
we can use a design block to back-calculate source (Source2): so the software
will adjust the value in the source (Source2) to reach a certain desired value in
the product
 Of course you need to make sure that what you adjust really influences what
you try to change
 Note that most simulators will still require to satisfy the DOF without the
design block, which intervenes “on top”. But be careful: wrong DOF assignments
in the design block can lead to a simulation crash

Adjust
Block ?
Product1 Design block changes the
Source2 DOF by specifying product
1 instead of source 2
Unit3

? Source1 Unit1 Unit2

Product2
99 Bezzo 2024
Process flowsheeting
Solution algorithms
 There are two types of solution algorithms:
 In the sequential-modular (SM) approach, the equations describing equipment
units are grouped together and solved in modules, i.e. the process is solved
equipment piece by equipment piece. The executive program sets up the
flowsheet sequence, identifies the recycle loops, and controls the unit operation
calculations, while interacting with the unit operations library, physical property
data bank, and the other subroutines
 Note that individual equipment blocks may require iterative solution algorithms

 In an equation-oriented (EO) simulator the model equations for all the unit
modules and connecting streams are collected together and solved
simultaneously. The executive program solves the equations using data from the
unit operations library and the physical property data bank and calling on the file
of thermodynamics subroutines
 The SM approach is by far the most widely used algorithm for steady-state
simulation software (implemented in Aspen Plus, HYSYS – steady-state, Pro/II,
UniSim). The only fully EO commercial simulator is gPROMS (although EO
solvers and/or functions exist in other commercial software, too)
 Dynamic process simulation software (e.g. gPROMS, HYSYS – dynamic, DYNSIM…)
adopt an EO approach
100 Bezzo 2024
Process flowsheeting
Solution algorithms
 SM simulators  EO simulators
 PROs  PROs
 Easier initialisation  Dynamic simulation
 Moderate computer power  Higher robustness for
needed complex problems
 Easier troubleshooting  Easier to set up and solve
(convergence failure is easier optimisation problems
to identify)  Great flexibility
 CONs  CONs
 Convergence can be difficult  Initialisation can be an issue
due to iterative methods (often SM is used to provide
 Dynamic simulation are not initial guess)
possible  Troubleshooting can be
 Less flexibility (optimisation, complicated (no smoking guns)
inclusion of additional  Significant computer power
functions) needed

101 Bezzo 2024

Process flowsheeting
Commercial software
 Most common commercial process simulation software:
 ASPEN PLUS (Aspen Technology)
 Steady-state (SM standard + EO as an option)
 https://www.aspentech.com/en/products/engineering/aspen-plus
 ASPEN HYSYS (Aspen Technology)
 Steady-state (SM standard + EO as an option) and dynamic (EO)
 https://www.aspentech.com/en/products/engineering/aspen-hysys
 PRO/II Simulation and Dynamic Simulation (AVEVA)
 Steady-state (SM) and dynamic (EO)
 https://www.aveva.com/en/products/pro-ii-simulation
 UniSIM Design (Honeywell Forge)
 Steady-state (SM standard + EO as an option) and dynamic (EO)
 https://www.honeywellforge.ai/us/en/products/industrials/unisim-design
 gPROMS (Siemens PSE)
 Steady-state (EO) and dynamic (EO)
 https://www.psenterprise.com/products/gproms

102 Bezzo 2024

Sequential-modular approach
The recycle problem
 Most chemical engineering flowsheets have recycles
 Reactions and separations are rarely perfect, which means we may want to
recover some of our reactants or products
 Minimise waste and raw materials
 Creates better mixing
 Can act as a buffer (e.g. pick up excess heat from reaction)
 Recycles
 create integration of the blocks that are part of the loop
 typically have a significant effect on the solution
 make it harder to solve the flowsheet, particularly in a SM simulator

In a SM simulator, units are solved one-by-one. However, to get solved,

Unit 1 requires information concerning the recycle stream (e.g. flowrate,
composition, temperature), which is an output of Unit 3, which depends
Recycle
on the output from Unit 1, and so on…

Sources Unit1 Unit2 Unit3 Sink

103 Bezzo 2024

Sequential-modular approach
How to handle recycles
 For a SM simulation program to be able to solve a flowsheet with a recycle,
the design engineer needs to provide an initial estimate of a stream
somewhere in the recycle loop.
 This is known as a tear stream, as the loop is “torn” at that point. The
program can then solve and update the tear stream values with a new
estimate. The procedure is repeated until the difference between values at
each iteration becomes less than a specified tolerance, at which point the
flowsheet is said to be converged to a solution.

104 Bezzo 2024

Sequential-modular approach
Solution methods
 The solution sequence in a SM simulator is defined by the following steps:
1. Partitioning: the flowsheet is divided into independent subsystems (partitions)
2. Precedence ordering: the order in which the partitions are solved sequentially
is determined
3. Tearing: tear streams (or tear variables) are determined
4. Sequencing: precedence ordering of the blocks in each partition is determined,
and the system containing one or multiple tear streams is solved

Case study
Acid gas removal (AGR)
using chilled methanol
(Turton et al., 2018)

105 Bezzo 2024

Sequential-modular approach
Steps 1 and 2
 Partitioning determines the smallest group of unit operation blocks that
can be solved independently (strong component)
 If there are no recycles (or design/calculator blocks), each partition contains only
one unit operation block
 In the general case, flowsheet are represented as directed graphs (DG). For
example, the AGR process can be represented as in figure 1
 It can be observed that there are 3 partitions: {B1}, {B2, B3, B4, B5, B6}, and {B7},
with one subsystem involving a recycle loop. It is quite intuitive that the
precedence ordering is the one illustrated in figure 2: first {B1}, then {B2, B3, B4,
B5, B6}, and finally {B7}

Figure 1 Figure 2

106 Bezzo 2024

Sequential-modular approach
Steps 1 and 2
 For complex systems, efficient algorithm are required for both partitioning
and precedence ordering
 Partitioning has a drastic effect on computational time
 Note that partitioning may change
in a “hidden” way because of user’s
interventions
 For instance, first (a) imagine
introducing a calculator block setting
the makeup solvent flowrate F3 as
follows
P1 + S7
F3 =
F1
 Then (b), the makeup is computed as:
P2 + S7
F3 =
F1
 There is no change in the results
(P1=P2), but the partitioning change,
and the second case is
computationally more demanding

107 Bezzo 2024

Sequential-modular approach
Step 3
 Partitions may contain single or multiple units. When a partition contains
multiple units, there is one or multiple recycles: in a SM approach units
need to be solved sequentially
 Tear stream are identified automatically (sometimes badly), and an algorithm
must be applied to update the torn stream information at each iteration, until
convergence is reached
 User intervention may be necessary to speed up convergence or improve
robustness  a clear understanding of how the algorithm works is needed
 The first question is: what stream(s) should be tear stream(s)? In general this
problem has multiple solutions and is formulated as an optimisation problem,
based on several criteria, typically (in Aspen Plus default criteria are 2. and 3.):
1. Minimisation of number of tear streams
2. Minimisation of total number of stream variables (unknowns to be converged)
3. Minimisation of a cost function assessing the difficulty in guessing stream variables
4. Minimisation of the number of times loops are torn
 For complex criterion, where the initial guess of tear streams is very important
(i.e. criterion 3.), user intervention may be necessary

108 Bezzo 2024

Sequential-modular approach
Step 3
 For example, if criterion 1. is applied to the flowsheet below

one may notice that not all three recycles need tearing and that it is
sufficient to tear either stream M3 or stream M4
 Now the question whether to tear M3 or M4, which may depend on the
number of variables to be converged (criterion 2.) or the ease in getting an
initial guess (criterion 3.)
 E.g. if we assume that U3 is a mixer and U4 is a heat exchanger, then
temperature in M4 is likely to be known, while an initial guess for M3 is probably
not available. According to criterion 3., M4 should be chosen as tear stream.

109 Bezzo 2024

Sequential-modular approach
Step 4
 Once recycles are removed, the precedence ordering can be implemented
as in Step 2.
 The key question at step 4. is how to update the initial guess at each
iteration.
 Let’s consider the following flowsheet. Once an initial guess for M4′ is given, the
solver can obtain a solution for M4.

 Let’s assume that the vectors of stream variables (e.g. species flowrates,
temperature, pressure) for M4 and M4′ is denoted by y and x. Then, the problem
is to find the solution to equation: x − y = 0
 During a given iteration k, values xk are guesses, whereas yk is obtained by solving
all blocks in the sequence; yk is then used in calculating the next guess for x, i.e.
xk+1. A number of algorithms are available to solve this problem. We will discuss
the main ones.
110 Bezzo 2024
Sequential-modular approach
Step 4
 Direct substitution
 In this method, an initial estimate, xk, is used to calculate a new value of the
parameter, yk. The estimate is then updated using the calculated value:
x k +1 = y k
 This method is simple to code but is computationally inefficient and convergence
is not guaranteed.

111 Bezzo 2024

Sequential-modular approach
Step 4
 Bounded Wegstein
 Efficiency and convergence can be improved using relaxation methods by using a
vector of relaxation (also called acceleration) parameters αi:
1
xik += α i xik + (1 − α i ) yik
 If αi < 0, the method helps to accelerate convergence; if 0 < αi < 1, the method
helps convergence stabilisation by damping.
 The bounded Wegstein method is the default method in most of the simulation
programs. It is a linear extrapolation of successive substitution.The Wegstein
method initially starts out with a direct substitution step:
x1 = y (0)
 The relaxation parameter αi can then be calculated:
si yik − yik −1
=αi = with α i
si − 1 xik − xik −1
 The method may lead to instability if is not bounded. Usually, αmin ≤ α < 1, and
αmin is about -5. The bounded Wegstein method is usually fast and robust. If
convergence is slow, αmin can be reduced If convergence oscillates, then consider
damping the convergence by setting bounds such that 0 < α < 1.

112 Bezzo 2024

Sequential-modular approach
Step 4
 Newton’s and quasi-Newton’s methods
 The Newton method uses an estimate of the gradient at each step to calculate
the next iteration
x − (J ) (x − yk )
−1
x k +1
= k k k

where J ij = ∂ ( xi − yi ) ∂ x j is the jacobian matrix

 In quasi-Newton methods such as the Broyden method evaluation of the
Jacobian is avoided by approximation functions. This approach reduces the
number of calculations per iteration, although the number of iterations may be
increased.
 Newton and quasi-Newton methods are used for more difficult convergence
problems, for example, when there are many recycle streams or many recycles
that include operations that must be converged at each iteration, such as
distillation columns.
 The Newton and quasi-Newton methods are also often used when there are
many recycles and design/calculator blocks.
 The Newton method should not normally be used unless the other methods
have failed, as it is more computationally intensive and can be slower to converge
for simple problems

113 Bezzo 2024

Numerics
Sequential-modular vs. Equation-oriented
 EO solvers are based on Newton-method or quasi-Newton-Method
 A typical numerical challenge is given by the sparsity of the Jacobian matrix 
block-decomposition algorithms are applied

 Note that there is a key difference in SM and EO simulators.

 In EO simulators, convergence is defined in absolute terms, and, given a vectors
of some tolerances ε, we need that for all variables in the model:
x k +1 − x k < ε

 In SM simulators, there is a sort of “double convergence”: on the one side, each

block unit will have to reach convergence according to specific algorithms used
to solve its equations; on the other side, convergence needs achieving for the
loops involving recycles and design/calculator blocks. For variables involved in
the loop, again we need:
x k +1 − x k < ε

 Tolerance ε can be modified by the user. In commercial software the default

tolerance , is typically 10-4

114 Bezzo 2024

Process simulation

Using process simulators

Convergence issues
Practical advice
 Becoming a process simulation expert requires deep knowledge
on the process, on fundamentals (e.g. thermodynamics), on
modelling, on numerics… as well as time and patience
 Some advice to help solving convergence problems is given in the
following slides… but remember that a converged model does
necessarily mean a correct model: always remember the GIGO
principle  Garbage In Garbage Out!

116 Bezzo 2024

If simulation fails
Debugging your flowsheet
 If the problem is in a large flowsheet, cut the flowsheet down and
introduce local sources and sinks to isolate where problem occurs
 Once we have fixed it, we go back to full flowsheet and check if this has
resolved the problem

117 Bezzo 2024

If simulation fails
Debugging your flowsheet

Replace the rest of

the flowsheet with a
source and a sink

118 Bezzo 2024

If simulation fails
Analyse the simulation behaviour
 Find a situation that works and change the values to desired
values/inputs/specs in a smooth way
 Look at key variables (temperatures, pressures, flowrates, compositions)
and how they change while moving along this path
 If unsuccessful, use simpler version of units (e.g. a component splitter
instead of a distillation column) to see if the problem is due to complexity
of unit or with the flowsheet
 Models can be simplified by using faster and more robust unit operation models
(e.g. substituting shortcut column models for rigorous distillation models)
 Models can also be simplified by reducing the number of components in the
model. Reducing the number of components often leads to a good estimate of
the bulk flows and stream enthalpies, which can be useful if there are
interactions between the mass and energy balances
 ADVANCED: Design/calculator blocks and tear variables can be solved setting
up nested iterations or be solved simultaneously. Furthermore, tear streams (or
design/calculator blocks) can be nested inside or outside a convergence loop
 Although any simulator will make default choices, the user can act on this and results
can be dramatically different

119 Bezzo 2024

If simulation fails
Some common problems
 Pressure drops may sometimes lead to unfeasible solution (during
iterations it may decrease until 0, and then the simulator will declare
convergence failure): pressure changing devices (pumps, compressors,
valves) can solve the problem by resetting pressure at each iteration
 Convergence problems are often related to recycle streams. The
convergence of recycle calculations is better if a good initial estimate of the
tear stream is provided
 If the tear stream is chosen carefully, it may be easy to generate a good initial
estimate or it may be worthwhile to develop a simplified simulation model to
achieve a first estimate of tear stream composition, flow rate, and conditions
(temperature and pressure).
 Sometimes, it may be useful to “open” a recycle, thus generating an inlet and
outlet streams, then to iterate manually until similar values are obtained for inlet
and outlet, and finally “close” the loop so that the solver starts from a reasonably
good solution

120 Bezzo 2024

Unit models
Spreadsheet models: hands-on
Open file ExampleTearStream in Aspen Plus, run simulation, and observe
that there are warnings:
The following Unit Operation blocks were
completed with warnings:
HX1 COL1
Check the blocks and notice that for the heat exchanger:
* WARNING
AT LEAST ONE FEED HAS ZERO FLOW
BLOCK BYPASSED
The following messages were issued during Report Generation:
* WARNING
AT LEAST ONE FEED HAS ZERO FLOW
BLOCK BYPASSED
And for the column:
* WARNING
FEED(S) CONTAINS ONLY ONE COMPONENT - BLOCK BYPASSED
Now try to give an initial guess to the recyle stream R1 (T = 200 F; P = 15 psia;
water flowrate = 75 kmol/hr). Now it works! Note that the input you’ve given to
R1 is just a guess, which will be overwritten.
(The example is taken from: https://www.youtube.com/watch?v=VvSTBWl76PI)
121 Bezzo 2024
If simulation fails
Step by step
 When first simulating a process, input only the data required to perform
the material and energy balances for the process
 This entails building up the model starting from a simplified version and
successively adding detail while re-converging at each step. As more complexity
is added, the values from the previous run are used to initialize the next version.
This is a slow but effective method.
 Remember to save the intermediate versions every so often, in case later
problems are encountered.
 Complex unit operations or detailed representation of unit operations may
have to simulated independently as stand alone models, and only simplified
units, possibly deriving from rigorous models, will be included in the
process flowsheet
 When multiple recycles are present, it is sometimes more effective to solve
the model in an EO mode rather than in a SM mode. If the simulation
program allows an EO approach, this can be attempted.

122 Bezzo 2024

If simulation fails
Additional advice
 If a flowsheet is not converged, or if the process simulation software
runs and gives a statement “converged with errors,” the first steps
that an experienced user would usually take are:
 Make sure that the specifications are feasible
 Perform a manual mass balance around distillation columns to determine whether
the specifications are feasible
 In distillation columns, the use of recovery specifications is usually more robust,
but care is still needed to make sure that the reflux ratio and number of trays are
greater than the minimum required
 a component may accumulate in recycle loops because of the separation
specifications that have been set: adding a purge stream usually solves this problem
 Try increasing the number of iterations
 Try a different convergence algorithm
 Try to find a better initial estimate
 Try a different tear stream

123 Bezzo 2024

If simulation fails
Look for information sources
 Check documentation of the simulation software
 Videos on YouTube or software vendors website
 Ask local experts (company’s experts, professors)
 Contact support
 Software vendors usually provide different levels of supports to their
customers

124 Bezzo 2024

If simulation fails
An example from Aspen Plus user’s guide

125 Bezzo 2024

If simulation fails
An example from Aspen Plus user’s guide

126 Bezzo 2024

Process simulation tools
Sensitivity analyses
 In a sensitivity study, one or more input variables are modified to generate results
for some output variables. Typical applications comprise:
 Assessing the effect of design variables (i.e. decision variables) on key outputs (e.g.,
process productivity, product specifications, critical flowrates…)
 Identifying variables for process optimisation
 Identifying initial guesses or ranges of variations for manipulated variables
 Assessing the impact of uncertainty (parameters or inputs) on resulys
 Small incremental steps increases the chance to obtain a solution at each step
 EO simulators are often more effective and robust for sensitivity analyses
 Note that the approach one-factor-at-a-time may be misleading since it assumes
that changes will have an additive effect
 A Design-of-Experiments (DoE) approach is needed when assessing more than one input
variables

Product A
mole fraction

Product B

conversion

127 Bezzo 2024

Process simulation tools
Design specifications
 Design blocks can be used to set up design specifications. The typical instruction
is as follows:
 Adjust variable x to value y by manipulating variable z (e.g. adjust the flowrate of a limiting
reactant in order to target productivity or adjust reflux ratio in order to target purity in
the distillate)
 Design specifications are typically managed through the Secant numerical method (if the
model is linear, solution is found at the third iteration  2 point to find the linear correlation, a 3rd
point to get the solution)
 Design blocks can be used to capture all kinds of design constraints and
specifications. They are particularly useful for setting feed ratios and controlling
purge rates and recycle ratios to achieve target compositions.
 Care is needed to ensure that they are used sparingly; otherwise, too many recycles
of information can be introduced and convergence becomes difficult
 Design blocks behave much like recycles, and it is usually a good idea to generate a
converged simulation to act as a good initial estimate before adding design blocks
(sensitivity analysis may identifying the range within which variable z is allowed to vary)
 When a design specification is changed, it should be done “gradually”
 The designer changes the relevant parameters in small steps to reach the new conditions while re-
converging at each step.
 The results of each step then provide a good initial estimate for the next step, and convergence
problems are avoided.
 This is a good piece of advice anytime a converged simulation is modifies (e.g. an input is changed)

128 Bezzo 2024

Sensitivity analys and design specification
Example: hands-on

Let us consider the following cycle:

Then, carry out the following tasks:

1. Create a model in Aspen Plus (assume that 200 MW are needed to take the
reactor effluent from 420 °C to 150 °C)
2. Use Design specifications to estimate the steam flowrate needed to achieve a
steam temperature of 360 °C
3. Use a Sensitivity analysis to vary the outlet pressure in the first turbine and
determine how the power produced changes with the intermediate pressure

(The example is taken from Adams II, Learn Aspen Plus in 24 hours, Tutorial 3)
129 Bezzo 2024
Sensitivity analys and design specification
Example: Guided problem 2 (I)
The process considered consists in a reactor for the conversion of toluene and hydrogen into
benzene, followed by a heat exchanger that cools the reaction products, and subsequently by a
gas-liquid separator.
The vapor fraction is sent to further treatments, while the liquid fraction, which contains the
benzene produced and the non-reacted toluene, goes to a distillation column for the recovery of
benzene and of toluene, which is recycled to the reactor.

Base case
Feed 1: 25°C e 25 bar, total flow-rate: 10000 kg/h (pure toluene).
Feed 2: 25°C e 25 bar; total flow-rate: 820 kg/h (95.02% of hydrogen e 4.98% of methane, molar).
The reactor is operated at a temperature of 600°C and a pressure of 25 bar;
The reaction is the following:

C7H8 + H2  C6H6 + CH4 (conversion of toluene equal to 0.75).

After the reactor, a heat exchanger cools the reaction products to a temperature of 38°C.
The following separator allows to recover a gaseous stream containing only methane and
hydrogen.
The distillation column is made of 20 stages (the inlet stream is fed at stage 10), and operates so
that the residue flow-rate is equal to 28% (by weight) of the feed flow-rate, with a mass reflux
ratio of 0.2.

130 Bezzo 2024

Sensitivity analys and design specification
Example: Guided problem 2 (II)
Sensitivity analysis
Evaluate the behavior of the molar fraction of benzene in the distillate stream as a function of the
reflux ratio in the column, and plot the results.

Design specification
Determine the value of reflux ratio necessary to obtain a distillate having a molar composition of
benzene equal to 99.6%.

131 Bezzo 2024

Process simulation tools
Optimisation
 All commercial process simulation programs all have the ability to solve
optimisation problems that can be posed as nonlinear programming
(NLP) problems
 EO simulators have a clear advantage in solving optimisation problems
 Very few process simulators (e.g. ASPEN PLUS, gPROMS) allows the
designer to carry out discrete optimisation using integer variables.
 Thus, in most cases, it is therefore not possible to optimise integer parameters
such as feed tray location while simultaneously optimizing continuous variables
or to perform superstructure optimization.
 Optimization of a large process simulation model is intrinsically difficult,
particularly if there are multiple recycles
 The formulation of the objective function may not be an easy task in
process simulation software
 Note that the objective function for industrial design is typically a measure of
economic performance

132 Bezzo 2024

Process simulation tools
Optimisation
 Before attempting any optimisation:
 Start by converging a simulation of the flowsheet
 Carry out a sensitivity analysis to determine which variables are most impactful
on the objective function. These are the variables that should be used as decision
variables. It is also important to determine reasonable ranges for these variables
and set upper and lower bound constraints. If the ranges set are too narrow,
then the optimum may not be found. If they are too wide, then convergence may
be difficult.
 While carrying out the sensitivity analysis, see if the optimum is broad or sharp.
If there are only small changes in the objective function, further optimization may
not be justified
 Another approach that is often used is to carry out optimization using
simplified models to fix the process structure and determine the
approximate values of key decision variables.
 A final NLP optimization can then be carried out using a rigorous model.

133 Bezzo 2024

Optimisation
Example: hands-on
The concentration of dichloromethane in an aqueous stream needs to be reduced from
0.3% (mol/mol) to 0.015% (w/w), by stripping with steam available at 4 atm at dew
temperature. The separation is obtained with two subsequent flash units. Determine the
amount of steam to be supplied to the first and the second stage respectively, optimizing
the total steam consumption.

Inputs: FEED
Feed flowrate = 100000 kg/h;
T = 90 °C; STEAM

Pressure = 1 atm;
H2O= 0.997 (mol/mol)
CH2Cl2 = 0.003 (mol/mol) STEAM
Steam: Pressure = 4 atm; vapor fraction = 1

The flash units are adiabatic and operate at atmospheric pressure.

The use of NRTL as thermodynamic model is recommended.

134 Bezzo 2024

Additional tools
Using utilities
 Several process simulators provides tols for choosing utilities
and to reduce their costs in a process (we will consider the
Utility feature in Aspen Plus)
 Typical information needed comprise the type of utilities
available (e.g. fired heat, cooling water, high- or low-pressure
steam…) and their operating conditions

135 Bezzo 2024

Using utilities
Example: hands on
Let us consider the following process, where we are separating a mixture of acetone,
methanol and n-butanol.

Assign utilities to the process units.

(The example is taken from Adams II, Learn Aspen Plus in 24 hours, Tutorial 6.1)
136 Bezzo 2024
Additional tools
Capital cost estimation
 Some process simulators incorporate additional software for capital cost
estimation
 Possibly the most widely used tool is the Aspen Capital Cost Estimator (in
PRO/II there is a Simulation’s Economics unit)
 Capital cost modelling is an important part of process engineering and
plant design
 Aspen Capital Cost Estimator is a highly complex software with a huge
amount of information inside
 The purpose is to estimate the cost of common chemical process equipment
 You can select units, currency, country (i.e. cost data are taken for that country’s
plants and applications), key economic parameters (tax rate, interest rate, labour
cost…)
 You need to add individual pieces of equipment and then run economic analyses
for the whole plant
 Capital cost estimates can be directly integrated with Aspen Plus
 You can export an Aspen Plus flowsheet into Aspen Capital Cost Estimator
 Alternatively, you can have capital costs predicted in Aspen Plus

137 Bezzo 2024

Additional tools
Energy integration
 Several process simulators also include tools for plant energy
integration, more specifically to design heat exchanger
networks (HENs) for energy minimization
 The goal is to improve energy recovery among process streams, and
thus reduce heating and cooling supply from utilities
 Different methods may aim at minimising utility usage or at
minimising total cost (capital cost + utility usage)
 Typically the software allows the user to implement energy
integration approaches such as pinch analysis (manual intervention
from the user is needed) or to apply optimisation techniques that
still somewhat refer to pinch analysis concepts
 The Aspen Plus tool is called Aspen Energy Analyzer

138 Bezzo 2024

Conclusions
 Remember that you are responsible for the results you
obtain, and not the process simulation software
 Convergence does not mean correctness
 Engineering knowledge and expertise is needed to interpret
and critically assess the results

139 Bezzo 2024