Beruflich Dokumente
Kultur Dokumente
Olaf Abel Ethylene Apc
Olaf Abel Ethylene Apc
in the Chemical
Process Industry
Bram Jansen and Olaf Abel
BASF Antwerpen N.V.
STI Automation Services
Workshop Honorary Doctorate Degree
Prof. Dr. Lennart Ljung
Leuven, 12./13.10.2004
Agenda
33.361 Mio
EBIT:
2.658 Mio
Cashflow:
4.878 Mio
3.666 Mio
EBIT:
383 Mio
Cashflow:
468 Mio
Applications
Plastics and
Fibers
(46,0%)
Chemicals and
other products
(35,3%)
ethylene, propylene,
benzene, PIB, amines.
crop-protecting, pharmaceutical
products, additives for fuels and
lubricants, ...
Fertilizers and
Inorganics
(9,7%)
fertilizers
Performance
Products
(8,9%)
Automation Services
Central competence center for Operations
Enterprise Resource Planning,
Business IT, SAP/R3
Advanced Process Control,
Real-time Optimization
Process Information
Management Systems
Process Control
Measure,
Communicate
Operate
ERP
APC, RTO
PIMS, LIMS
Automation Services
DCS, PLC
Operations
Agenda
Achieving optimality
The interesting
quantities are often
difficult to measure.
The influence of the
DoF on the OF is
not obvious.
The used problem
formulation is
plant-specific and
time-dependent.
product qualities
safety
Possible requirements:
Dynamic
Nonlinear behaviour
Resolution in time and space
Extrapolation capabilities
Property distributions
First principles:
Based on balance and
phenomenological equations
Process knowledge required
Comparatively expensive
Data driven:
Identified using process data
Comparatively inexpensive
Agenda
Data collection
server
PC
router
PC
DCS
PC
collector
Identification
Models
Steady state
Dynamic
Neural Networks
State Space
Least Squares
Backpropagation
Numerical optimization
Least Squares
Subspace identification
Prediction error methods
Numerical optimization
SAS JMP
IPCOS Presto
Mathworks MATLAB System Identification Toolbox
Tailored applications
Experiences
Integration of
knowledge
Noise
Feasibility
Disturbances
Requirements
static / dynamic
Maintenance
Preliminary
investigation
model
structure
selection and
parameter
estimation
Preprocessing
raw data
Fine tuning
and
confidence
interval
Validation
Analysis
Interpolation
Sensitivity
analysis
Outlier detection
Online
application
Agenda
Reaction
Granulation
laboratory
analysis
QI
Extraction
Drying
15 h
18 h
Polymer
Monomer
QI
Relative viscosity is measured twice per day with high precision (0.01)
(laboratory / offline) analysis time is 4h
Potential in avoiding off-spec production:
fast detection of new product specification during grade change
fast reaction on disturbances influencing relative viscosity
reduction in number of (expensive) offline analysis
continuously
and easily
measurable
variables
= f(T,p,m)
model
prediction
of relative
viscosity
UNILAB
(Siemens)
PHD
(Honeywell)
JMP
(SAS)
PRESTO
(IPCOS)
EXCEL
(MS)
Multiple Linear
Regression (MLR)
13 - 17 input parameters
(reaction and granulation section)
laboratory
analysis
soft sensor
18 months
Relative viscosity
2,3
2,2
labo analysis
2,1
2
1
1441
2881
19.2.2003
4321
5761
7201
22.2.2003
8641
10081
11521
25.2.2003
12961
14401
15841
28.2.2003
Relative viscosity
is provided continuously
with high precision!
Case 2: Steamcracker
Capacity:
C2: 800.000 t/a
C3: 500.000 t/a
C4: 300.000 t/a
Benzene, TX
Mass fraction
furnace decoke
C3H6
H2
Cross over temperature
C2H4
900 C
850 C
Residence time
setpoint
(from optimizer)
controlled variable
(predicted)
measured
manipulated
variable
time
control horizon
prediction horizon
Multivariable
Rigorous incorporation of constraints
(Limited) optimization capabilities
Process data
collected by
executing step tests
PRBNS input
Mainly FIR models
(linear application)
Identification based
on LS algorithms
Process Part
MV
FF
CV
34
Ethane furnace
10
11
36
Cold side
36
20
82
Depropanizer
14
11
Propylene fractionator
11
29
Feedmaximizer
143
137
448
Results: C2 splitter
(separation ethane/ethylene)
Feed to C2 splitter
Ethane pollution in
top ethylene product
Ethylene loss in
bottom ethane
Propane in top
propylene product
(1. and 2. column)
Combined
feed to both columns
Propylene loss in
bottom propane
(1. and 2. column)
Agenda
Conclusions