Presentation of Jacques Niederberger for the "Workshop Virtual Sugarcane Biorefinery"
Apresentação de Jacques Niederberger realizada no "Workshop Virtual Sugarcane Biorefinery "
Date / Data : Aug 13 - 14th 2009/
13 e 14 de agosto de 2009
Place / Local: ABTLus, Campinas, Brazil
Event Website / Website do evento: http://www.bioetanol.org.br/workshop4
Yaroslav Rozhankivskyy: Три складові і три передумови максимальної продуктивн...
Modelling, Simulation and Optimization of Refining Processes
1. Modelling, Simulation and Optimization of
Refining Processes
Jacques Niederberger, M.Sc.
PETROBRAS Research & Development Center (CENPES)
August/2009
4. PETROBRAS
AN INTEGRATED ENERGY COMPANY
Total Investments: 15 Refineries
US$ 29 billion in 2008 Installed Capacity: 2.125 million bpd
Natural Gas Production:
Employees: 74,204 420 thousand boe per day
Net Operating Revenues
US$ 127 billion (2008)
Proved Reserves : Oil Production:
15.1 billion barrels of oil 1,980 thousands barrels per
and gas equivalent (boe) day (bpd) of oil and LPG
Natural Gas Sales:
Gas stations: 6,485
65 million m3/d
Thermoeletric Energy
Plants : 10
Installed Capacity : 1,912 MW Dec 2004
7. R&D EXPENDITURES
2.000
1.750
1.500
1.250
R$ MM
1.000
750
500
250
0
2000 2001 2002 2003 2004 2005 2006 2007 2008 CENPES
Ano 137 Laboratories
EXAMPLES OF MAIN CHALLENGES 14 TECHNOLOGY PROGRAMS
Ultra deep water production technology
Production in the Pre-salt sequence
Lower environmental impact products
Better output products Optimization
Pre-salt &
Zero discharge / zero emissions processes Reliability
8. TECHNOLOGICAL INTEGRATION
R&D
CENTER
Types:
Types:
Contracts and agreements with Universities Joint Industry Projects
and Research Centers Cooperating Research
Strategic Alliances
National networks of excellence - about Technology Interchange
different oil & gas themes
Over 120 Brazilian Institutions Over 70 International Institutions
11. EXPERIMENTAL DATA
Complete assay contains:
Distillation curve
Specific Gravity curve
Light end contents
Viscosity
Sulphur, nitrogen and metals contents
Other properties
12. TRADITIONAL
CHARACTERIZATION
PROCEDURE
True Boiling Point Curve - TBP
• Product withdraws at constant volume or at
constant temperature
• Near ideal fractionation
• Long time demanded, high cost
13. TRADITIONAL
CHARACTERIZATION
PROCEDURE
Crude Oil TBP
temperature, C
o
% vaporized
14. TRADITIONAL
CHARACTERIZATION
PROCEDURE
Crude Oil TBP
temperature, C
o
% vaporized
15. TRADITIONAL
CHARACTERIZATION
PROCEDURE
Distillation curve,
Specific gravity Pseudo-components
Characterization
Method
Pseudo-component: fake component, oil fraction.
Crude oil and its derivatives are hydrocarbons mixtures,
well described by cubic equations of state (SRK, PR)
The characterization method provides pseudo-
component properties: Tc, Pc, w, PM, d60, Teb, etc.
16. IMPROVED
CHARACTERIZATION
Instead of pseudocomponents, real
molecules.
• Group of molecules typically present in a
determined fraction
• Bulk properties: distillation curve and
specific gravity
• Mixture composition obtained through an
optimization method
19. EFFECTS OF THE
CHARACTERIZATION
METHOD
Processes involving chemical reactions:
Heavy Feedstock → Gases + Light
Distillates + Medium Distillates +
unconverted
or
Heavy Feedstock + H2 → Organic Gases
+ H2S + NH3 + Light Distillates + Medium
Distillates + unconverted
20. EFFECTS OF THE
CHARACTERIZATION
METHOD
How to model chemical reactions ?
Kinetics x Thermodynamics
Kinetics: reaction order, kinetic
parameters
Thermodynamics: Gibbs free energy
21. EFFECTS OF THE
CHARACTERIZATION
METHOD
Either Kinetics or Thermodynamics
require pure component data.
Pseudo-component approach:
not good!
Compositional approach:
no big deal!
22. EFFECTS OF THE
CHARACTERIZATION
METHOD
If we characterize using molecules:
23. EFFECTS OF THE
CHARACTERIZATION
METHOD
•How to build phenomenological
models of conversion processes
dealing with pseudocomponents ?
•Relating the overall conversion and
product profile to bulk properties of
the feedstock and process
conditions.
24. REFINING PROCESSES
MODELLING
•We model phase equilibrium and
separation process with the traditional
tools provided by Thermodynamics
•And for the conversion processes we
build semi-empirical models
26. REFINING PROCESSES
MODELLING
For instance, in the FCC process:
Gasoil → Combustible gas + LPG +
Naphta + LCO + DO + coke
•Overall conversion depends on:
feedstock properties
catalyst properties
hardware geometry
process conditions
27. REFINING PROCESSES
MODELLING
•Product profile depends on:
feedstock properties
catalyst properties
hardware geometry
process conditions
•Product properties depend on:
...
28. REFINING PROCESSES
MODELLING
How do we address any other effect
not directly taken into account by the
semi-empirical model ?
Introducing adjustable tuning
parameters in the model.
Process data is necessary for fitting
the parameters.
29. REFINING PROCESSES
MODELLING
Quality of the model predictions
equals the quality of process and
feedstock data
31. REFINING PROCESSES
OPTIMIZATION
What does optimization mens ?
Generally speaking, any improvement
in a process with a few degrees of
freedom may be called optimization.
From our point of view, optimization is
finding THE best solution, in a system
with one ore more degrees of freedom.
32. SCOPE X TIME SCALE
Task Scope Time horizon
Planning operations and All the eleven Petrobras’ 5 to 20 years
The scope of the optimization problem and
invesments for the next refineries
the time horizon varies in the same
years
direction.
Designing a new plant One or more units of a 5 years
refinery
Planning the production One single refinery Monthly, weekly
of a sigle industrial plant
Optimizing operating Crude distillation + FCC Every 1 or 2 hours
conditions of one or converter + FCC
more units of a single fractionation section of a
plant refinery
33. SCOPE X MODEL
COMPLEXITY
The largerTask scope, the simpler must be
the Model type
Planning operations and investments for the Linear models (linear
the model.
next years programming)
Planning the production of an entire refinery Linear models (linear
programming)
Designing a new unit Rigorous mixed integer-non-
linear models (MINLP)
Optimizing operating conditions of one or Rigorous non-linear models
more units of a single plant
34. OPTIMIZATION &
PROCESS DESIGN
Design Synthesis
Initial estimates
Decision variables
Analysis
Mass & energy
balances
Optimization
Equipment sizing and
cost estimates Parametric Structural
Optimization Optimization
Economic
Evaluation
Final Design
35. OPERATING CONDITIONS
OPTIMIZATION - OFF LINE
PROCESS DATA
DATA RECONCILIATION RECONCILED PROCESS
DATA
MODEL TUNING & GROSS ERRORS
OPTIMIZATION DETECTION
UNIT
MAINTENANCE
PROCESS ENGINEER
CONTROL SYSTEM
OPERATOR
37. OPERATING CONDITIONS
OPTIMIZATION - RTO
Many plants don’t have a much stable
operation.
Optimal conditions for one
determined run may not be the best
for another run.
If optimization is off-line, we need to
re-optimize for every different run.
38. OPERATING CONDITIONS
OPTIMIZATION - RTO
Imagine if we had an optimization
machine that could read process data
at real time, tune automatically the
process model, run automatically the
optimization problem and send
automatically the optimal conditions
for the digital control system …
That would be Real Time Optimization -
RTO.
39. RTO STRUCTURE
Hibernation
Steady State Detection
No
Stationary ?
Yes
Model tuning
Optimization
No
Solution
obtained?
Yes
New setpoints for the control
system
40. RTO benefits
Real Time Optimization
PETROBRAS experience: RTO implemented on
Distillation and FCC Units using Equation Oriented and
Sequential Modular approaches
41. RTO benefits
Real Time Optimization
FCC Example: Operational modifications (Reaction
temperature, Feed temperature and Main Fractionator
top reflux) due to RTO
42. RTO Challenges
RTO runs only when the unit is Steady
but what is Steady State?
commercial applications use a kind of statistical
approach (mean, std dev, Student and F-test) along
with some heuristics (“tuning factor”) on a set of the
most representative variables (temperatures and flow
rates linked to the unit heat and mass balance)
do we really have to wait Steady-State?
it can take 1-2 hours between runs
if a disturbance enters the unit in between no RTO
run maybe for a long period
Change the “tuning factor” or improve APC / Regulatory
control
43. RTO Challenges
Real Time Optimization
How to deal with the “unknown” feed composition (especially
in Distillation)?
Online analyzers NMR or NIR?
Lab analysis frequence? Methods?
Feed Reconciliation as long as you have
confidence on the model, use it as an analyzer
Redistribute the amount of the pseudocomponents in
order to match some information from the unit
(operations and product quality)
It is an optimization problem maybe the most difficult
one (more than the profit optimization)
44. RTO Challenges
Real Time Optimization
Non convergence tracking: it is a hard task, sometimes, to
find out the origin of the failure, especially, when it is not
associated with instrumentations or well-known process
problems
Initialization techniques
Scaling: heuristic rules X numerical analysis of the system
Integrating multiple process unities: how to deal with the
increasing problem size to get the most of integrated unities
optimization and its flexibilities?
How to deal with non convergence?
45. RTO Challenges
Real Time Optimization
Entire plant rigorous RTO – feasible, but still not possible
Multi-scale Optimization: integration and information
exchange between different optimization levels is an issue
that demands more attention
Dynamic RTO: it is still an open issue
Computational efforts?
Numerical issues?
How to implement it on industrial applications?