Reliability of Dynamic Simulation

to reproduce plant dynamics

Repsol – Inprocess
Manel Serra (Inprocess), JoseMaria Ferrer (Inprocess), Jose Garcia
Vega (Repsol Tarragona Refinery), Francisco Cifuentes (Repsol
Refining APC) , Marta Yugo and Maria Luisa Suarez (Repsol
Technology Center)

Boston 6-8 May 2013

 Repsol Presentation
• Integrated company: upstream, downstream, petrochemicals, gas
• Repsol downstream activities regionally based in
Europe and South-America.
• Repsol has 6 refineries, 5 in Spain and 1 in Peru
• It is the refining leader in Iberia and the third LPG company in the world.
• Spanish refineries process 0.9 million bbl/day

Coruña
Bilbao

Tarragona
Puertollano

Cartagena
 Agenda

• Simulation: What is, Why and How good

• Case Study: Double C3Splitter
• Ideas for Future
 What is Process Simulation
From time to time someone tells me:
“ I don´t believe in process simulation”

Well… that´s like saying:

“ I don´t believe in the Bernoulli equation”
Simulation is not a question of believing or not believing

Process simulation is only a macro-

compilation of physics, chemistry and
thermodynamics laws smartly coded
in an interactive computer application.
Just a megaprocesscalculator.
1600 1700 1800 1900 2000
Chemistry, Mathematics, Physics
1614: Napier Logarithms 1738: Bernoulli´s Law 1801: Dalton´s Law Partial P. 1900: Planck´s Raditon Law
1637: Descartes Cartesian geo 1760: Lambert´s Law 1802: Henri´s Gas Law 1908: Grüneisen´s Thermal L.
1662: Boyle´s Gas Law 1768: Euler´s Method 1808: Gay-Lussac´s Law 1913: Heisenberg principle
1665: Calculus (Leibniz) 1785: Coulomb´s Law 1811: Avogadro´s Gas Law 1923: Pauli´s Exclusion prin.
1669: Newton´s Method 1785: Laplace´s transform 1822: Fourier´s Heat Law 1925: Fermi-Dirac distribution
1680: Algebraic logic Leibniz 1787: Charles´s Gas Law 1823: F.T. Calculus (Cauchy) 1949: EO Redlich-Kwong
1687: Newton´s Motion and 1791: Richter´s reaction Law 1829: Graham´s Effusion Law 1972: EO Soave R-K
Cooling 1831: Faraday´s Electrolysis 1976: EO Peng-Robinson
1840: Hess´s Enthalpy Law 1999: EO Elliot-Suresh-Donoh.
1840: Poiseuille´s Flow Law
1850: Clausius´s Law Thermo. 1930: Mechanical calculator
1851: Stoke´s Viscosity Law 1934: Differential Analyzer
This science has been there long time ago, 1852: Beer´s Absortion Law 1939: Turing decrypter
but we are the first generation of people who 1854: Boolean Algebra 1st Generation
1855: Fick´s Diffusion Laws 1946: ENIAC
has in our hands software tools and desktop 1864: Kopp´s Heat Cap. Law 1952: IBM 701
computers capable to simulate dynamically 1866: Maxwell´s Gas Viscosity 2nd Gen.: transistor
1869: Mendeleyev´s Periodic 1959: IBM 1401
entire process units. 1871: Coppet´s Freezing Point 3rd Gen.: integrated circuit
1871: Boltzmann´s Distribut. L 1964: IBM System/360
1873: EO Van der Waals 4th Gen.: Microprocessor
Most of their applications are still in the 1882: Raoult´s Vapor Pressure 1971: Intel 4004
early stages 1885: van´t Hoff´s Osmotic Pr. 1977: VAX-11/780
1893: Sutherland´s Gas Visco. 1978: Intel 8086
1980: Sinclair ZX80
1801: Punched Cards 1982: Intel 80286, 1985: Intel
1822:Mechanical Computer 80386, 1989: Intel 80486
1200: Abacus
(Babbage) 1993. Intel Pentium
1621: Slide Rule
1879: Cash Register (Ritty) 2006. Intel Core line
1673: Leibniz’s Step Reckoner
1700 2010. Intel Core i3,i5,i7
Computers
 Why Dynamic Simulation
Consolidated
Exploring
4.- Design control layout 6.- Develop virtual sensors
1. Equipment sizing and
process layout verification: • Scenarios analysis
• Online Analyzers backup
• Compression systems • Perturbation rejection
• Fault diagnostic
• Pipeline networks • Control loops selection
• Look-ahead sensors

2.- Flare Load calculation Dynamic Model

and PSV sizing 7.- DCS checkout
• Design/revamp flare • DCS FAT with virtual plant
networks
• Operating procedure test

3.- Emergency System 5.- Prototyping MPC

verification and HAZOP 8.- Operator Training System(OTS)
studies support • Obtain MPC models • Operator Training
• HIPPS studies • Study non-linearities • Emergency scenarios
• Cause & Effect matrixes • Test/Tune MPC controller • Knowledge base system
 How good are the models
Well… plants are built based in steady-state models (they should
be good enough)
But when moving to Dynamics, how good are they?
If the plant is being built there is no way to know it. You have to
trust in the tool and the experience of the modeler

If the plant is available there are three methods:

1. Compare responses of single moves

2. Compare DMCplus models (plant vs. model)
3. Feed historical data into the model (presented here)
 Method 1: Single moves
Shown in a debutanizer in AspenTech ACO UGM 2005 Barcelona

Real Plant HYSYS Dynamics

Reference: www.aspentech.com/publication_files/HP0906_Gonzalez_PDF.pdf
 Method 2: DMCplus models
Shown in a C3 Splitter in AspenTech UGM 2008 Berlin (APC track)
Blue: Real Plant
Red: Simulation
STEP-TEST

Dynamic
Model

DMCplus
Model

Other References:
1. www.aspentech.com/publication_files/Hydrocarbon_Engineering_Nov_2004.pdf
2. www.aspentech.com/workarea/downloadasset.aspx?id=6442451960
 Double C3 splitter polymer grade
1st column: 189 trays, lateral extraction chemical grade
2nd column: 205 trays, reboiler is condenser of 1st column
Challenging to control: very long settling times, heat interaction, external
disturbances and intrinsic non-linearity
 HYSYS Dynamics model
A HYSYS Dynamics model integrated with a DMCplus controller was developed
in order to analyze unit interactions and dynamics, change basic regulatory
controllers, generate HYSYS based DMC model and train engineers on their use.
 Model Building steps

Steady-State Mode Dynamic Mode

INITIAL MODEL DETAILED CALIBRATED DYNAMIC DYNAMIC

MODEL
From Process Eng. SS MODEL SS MODEL MODEL
VALIDATED

2. Intro Data 3. Calibrate 4. Switch & 5. Dynamic

15% 25% stabilize Validation and adjust
15% 25%

ENGINEERING PROCESS

DESIGN DATA PLANT DATA

1. Collect Data
20%

Percentages are efforts required for the model building

 Dynamic modeling
The model was developed with HYSYS 7.2.
Some features used in the model were:
- Level taps for column bottom levels
- Algorithms of PIDs (TDC3000 Eq C)
- Automation of step-test in DMCplus block
- Automatic calculation of residence times and
warning if integrator step-size is excessive
 Self regulated top pressure
The liquid level in the shell depends
on the differential pressure between
top and reflux tank.

Heat transfer coefficient (U) of the

2nd column condenser fully depends
on the liquid level in the shell side.

U Condensing Zone: 400 – 1000 Btu/h·ft2·ºF

U Subcooling Zone: 10 – 30 Btu/h·ft2·ºF

Changes in the Cooling Water temperature (day/night)

affects to the condenser duty and hence to the column
top pressure and condenser shell liquid level.
 CW Condenser in HYSYS
The Shell&Tube exchanger
Column Top Temperature
of HYSYS Dynamics doesn´t
consider the effect of shell
liquid variations in the heat
transfer coefficient.

Black: Real plant

Red: HYSYS with fix UA factor
Blue: HYSYS with variable UA factor
Therefore a calculated
UA Factor
variable UA factor has been
introduced in the specified
UA of the exchanger. It is a
correlation based in
pressures and design UAs.
 CW Condensers
It was historically believed that these 6 CW Condensers worked at full
capacity all the time with most of the tubes exposed to the hydrocarbons gas.

The HYSYS Dynamics

model with a variable UA
factor was fitting better
with plant data, revealing
that condensers work
partially inundated.
This was effectively
verified by the 2-3 Deg C
difference between the
low shell zone (subcooled)
and the high shell zone
(equilibrium).
 Validation Method 3:
For 5-days validation period, all the events that occurred in the real plant are
synchronically (1 min) introduced into the dynamic model (DMCplus actions,
measured disturbances, operators actions) in order to compare the variables
calculated by the dynamic model with those obtained from the real plant.
INPUT DATA :
Reflux1 (MV1) Excel Macro
Side-draw (MV2) OUTPUT DATA :
Bottom Flow (MV3) All compositions
Reflux2 (MV4) Temperatures
Feed Flow (FF1) Pressures
%C3 Feed (FF2) Product flows
CW Temp. (FF3) etc
Feed Temp.
Steam Temp.
LC´s SP
This type of validation is only useful if the main disturbances in the real plant
are measured, as in the event of there being strong disturbances which go
unmeasured they cannot be introduced into the simulation model, with the
result that the responses may well be different
 Some input variables
The DMCplus actions and disturbances were
MV4: Reflux2 (210-235 m3/h)
introducing significant changes to the unit

FF3: CW y Feed Temp. (10-30 °C)

FF1: Feed Flow (40-49 m3/h) MV3: Bottom Flow (41-46 m3/h) MV1: Reflux1 (500-540 m3/h)
 Validation (Method 3)
1st Column Bottom Quality % (C3= in C3)
Black: Real Plant
Red: Simulation
 Validation (Method 3)
1st Column Bottom Level (%)
Black: Real Plant
Red: Simulation
 Validation (Method 3)
2nd Column Top Quality % (C3 in C3=). Scale: 0 to 0.7%
Black: Real Plant
Red: Simulation
 Validation (Method 3)
2nd Column Differential Pressure (mbar)
Black: Real Plant
Red: Simulation
 Model usage issues
• The use of dynamic models by Control and Operation
engineers is not an standard practice
• The model was running a 3-4 times realtime (Intel
Core i5 – 680 at 3.6 GHz), still needing more than a
day to do a 5-days run.
• It is still complex to handle process data (from real
plant and dynamic model), joint use of HYSYS with
DMCplus, needs time to run simulation and
understand results.
 Ideas for Future: Virtual Sensor
• Steady-state online models need a reconciliation step in order
to close Heat & Material balances of imbalanced real plants.
• Dynamic online models would not need this reconciliation
step, but a proper input variable selection and some self
adapted parameters (fouling, efficiencies, etc). Not an obvious
task.
• Dynamic online models could provide virtual sensors for
compositions, using them as backups of online analyzers
which frequently require costly maintenance.
• With online dynamic models, the number and location of the
instruments can be revisited, helping to reduce the
instrumentation CAPEX.
 Ideas for Future: Fault Detection
• What happen if an online dynamic model suddenly diverge
from some plant data?
REAL PLANT INSTRUMENTATION PLANT REAL TIME DATABASE

INPUT VARIABLES
OPC data
DYNAMIC MODEL ONLINE

The reason can be:

1) Wrong modeling approach
2) Instrument Fail
3) Equipment Fail/Constraint
Fortunately physical
laws don´t lie !
 Ideas for Future: Look-ahead
Who does say this?:
“I've just picked up a fault in the AE-35 unit.
It's going to go a hundred percent failure
within 72 hours.”
Past Future

• Online dynamic models can

run faster than realtime
providing predictions of DPI prediction
critical variables could anticipate
flooding
• Realtime look-ahead trends
can be combined with plant
trending application.
 Ideas for Future: APC
• Look-ahead dynamic model could calculate rigorous
CVs predictions, which could replace DMC predictions
• It could be also used as a test bed for DMC tuning
REAL PLANT DMC1 controller PCWS Interface

CVs

MVs

INPUT VARIABLES
DMC2 controller PCWS Interface
DYNAMIC MODEL ONLINE
CVs

MVs
 Conclusions
• When the thermodynamic package represent well the
components and the main disturbances into the unit
are measured, HYSYS Dynamics models can reproduce
plant dynamics with acceptable precision and
validation method 3 can be used.
• Exploiting the value of such models is still an issue
• Specific education on simulation for non-process
engineers is required (Dynamic Simulation for Control
Engineers course)
• Some HYSYS Dynamics improvements would be
desirable