Dynamic Simulation for APC projects

A case study on a Reformate Splitter

with side draw

Dr Sebastien OSTA – TOTAL

Jose Maria FERRER – Inprocess
 Introduction

• Steady-state simulation is used traditionally for engineering design,

process analysis and troubleshooting, performance monitoring and
real-time optimization

• Dynamic simulation is used traditionally for process control studies,

operability studies, safety and HAZOP studies and operator training
simulators

• Dynamic simulation could possibly be used to assist Advanced

Process Control engineers with speeding up the deployment of some
APC projects, as well as enhancing the quality of the linear models
embedded within the multivariable predictive control applications

• This presentation shows the current status and the preliminary results
of a dynamic simulation project applied to the Reformate Splitter at
TOTAL La Mede refinery

2
 Agenda

• Vocabulary and Objectives

• Building the Dynamic Simulation

• Exploring the Steady-State Simulation

• Exploring the Dynamic Simulation

• Results

3
 Vocabulary

• APC stands for Advanced Process Control – In this case APC refers to
Multi-Variable Predictive Control, with the use of linear models
• An APC model is the dynamic function representing the effect of the
change of an independent variable (called here MV, i.e. Manipulated
Variable) to a dependent variable (called here CV, i.e. Controlled
Variable)
• As an example, increasing a column tray temperature by 1°C causes
the overhead product flow to increase by an extra 1.0464 T/h (i.e. “the
model gain”), reaching steady-state after approximately 150 minutes

4
 Project Objectives

• Contract a 3rd party, the company Inprocess (specialized in simulation)

to build a Dynamic Simulation of TOTAL La Mede Reformer
Fractionation Column
• Validate the Dynamic Simulation using online data and check the
prediction of benzene concentration in the 3 product streams
• Run step-testing within the Dynamic Simulation, exploring a wide
range of operating domains, e.g.:
• High/Low Reformer severity
• High/Low Benzene concentration in bottom’s
• High/Low throughput
• Build linear APC models from simulated step-tests data
• Define a strategy to account for non-linearities in the process, e.g.:
• Swap between several linear models depending upon process
conditions
• Use a single APC model with gain adaptation depending upon
process conditions
• Re-commission the APC controller and check the results
5
 BUILDING THE
DYNAMIC SIMULATION

6
Reformer Fractionator

7
 Plant Data

• TOTAL supplied Inprocess with all process data, including:

• PFD’s, P&ID’s
• Vessels, exchangers, air coolers
• Pumps, column, piping
• Valves and instruments
• Process description and test-runs
• PID controllers and tuning
• Process and lab data
• DCS calculations and inferentials equations
• Inprocess’ project methodology is to condense all necessary
information in a few Excel spreadsheets
• Component list (101 chemical compounds)
• Equipment data
• Tag list, including PID tuning

8
 Steady-State
Simulation
• Inprocess first builds the steady-state simulation in Aspen HYSYS
• Tuned on an agreed test-run (here 2 simulations were made available, for
high and low severity conditions on the reformer)
• This allows to initialize later the dynamic simulation
• This also allows to run case studies (explained later)

Mixer to recalculate feed

composition from products GC

Uses HYSYS standard distillation

column object

9
 Validation of the SS
Simulation
• The main validation criteria for the steady-state simulation is the
comparison of the simulated and actual temperature profiles for the
test-run data. Obtaining a similar temperature profile ensures that
simulated composition along the column will match with plant data

Temperature profiles (Y axis) for

the low severity case, as a
function of tray number (X axis)

 The simulated temperature

profile matches well the plant
data

10
 Dynamic Simulation

• Inprocess reproduces the plant as-is during the test-run

• With all equipment characteristics
• With all valves, control loops and specific PID algorithms (here Foxboro)
• With inferential calculations (inferential = virtual quality estimator)

11
 Validation of the Dyn
Simulation
• The main validation criteria for the dynamic simulation are:
• The comparison of the simulated and actual temperature and pressure
profiles for the test-run data
• The reason for potential differences between the simulated steady-state
and dynamic profiles comes mainly from the difference in algorithms used
to solve the problem, as well as extra parameters in the dynamic
simulation such as the elevation of equipments

12
EXPLORING THE STEADY-
STATE SIMULATION

13
Understanding the
Process

Side draw

Feed

14
 Question #1

1. What are the steady-state gains, obtained from the steady-state

simulation, between the APC MV’s and benzene concentration in
overhead (Bz top), side draw (Bz Medium) and bottom (Bz bottom)
products ? How do these gains compare with the gains found from
actual step-tests data ?

The dotted lines represent the

APC gain of the tray#45
temperature to Bz top (+1.29)
and to Bz bottom (-0.67),
computed from actual step-tests
data

 Strong non-linear effect of

temperature to benzene in the
side draw, with change in sign

Evolution of the gain of the tray#45 temperature to benzene

concentration (%LiqVol), as this temperature increases from 76 to 86°C
15
 Question #2
2. How do MV’s starting point influence the steady-state gains ? In other
words, are the previous gain functions valid across the entire range of
process conditions ?

With the same ovhd pressure, the

same reboiler duty, the same feed
temperature and flow, the side draw
flow is increased from 6.2 T/h (base
case) to 9.2 T/h (new condition)

 The gain function of tray#45

temperature to benzene in side draw
(Bz Medium) changes significantly,
and more importantly, the curve shifts
along the temperature axis

Evolution of the gain of the tray#45 temperature to benzene concentration (%LiqVol), as

the temperature increases from 76 to 86°C, with influence of side draw flow
16
 Question #2 – Con’t

Another representation of the gain of the tray#45 temperature to the

benzene concentration in the side draw, as the temperature varies
from 71 to 83°C, and the side draw flow varies from 5 to 12 T/h
17
 EXPLORING THE
DYNAMIC SIMULATION

18
 Running the Dyn
Simulation

Change of pressure PC5011

setpoint from 0.564 to 0.664 bg

19
 Question #3

3. How do the gains computed from step-tests data generated by the

dynamic simulation, compare with the gains obtained from actual step-
tests performed on the plant ?

This simulation runs ~ 9 times

faster than real time on a standard
laptop

 Step-tests took approximately 4

hours
 But we didn’t explore (yet) the
entire envelope of process
conditions

Automated step-tests in the dynamic simulation

20
 Question #3 – Con’t

Step-tests took approximately 6

days in 2013, when the plant was
running mostly at high severity

4 additional days in 2015 for

revamping the APC models, with
the plant running low severity,
which is now the usual mode of
operation

Actual step-tests on the plant 21

 Question #3 – Con’t
Light Ref FC Reflux FC %Bz Light %Bz Medium %Bz Heavy Reboiler Temp

Ovhd
PC

Tray
TC

Med
Ref FC

Steam
FC

Feed
TC

Feed
FC

Comparison of APC models obtained from

Actual plant data (2013) Dynamic simulation data Actual plant data (2015)
 Models are quite close, for both the dynamic shape and the steady-state gain
 Differences observed for %Bz Medium are expected due to the non-linear behavior
 Some differences also in the dynamic response of the reboiler
22
PROJECT RESULTS

23
 Results

• Confirmed benefits of using Simulation for APC purpose

• Running quickly multiple case studies, with varying operating parameters
on the plant
• Gaining deep understanding of the process
• Finding APC-type steady-state gains between MV’s and CV’s, with the
possibility to highlight non-linear behaviors
• Determining Pressure Compensated Temperature parameters for
improved basic control at the plant
• Generating high quality data for building inferentials
• Helping with designing APC controller structure
• Revealed benefits from using Dynamic Simulation for APC purpose
• Performing step-tests like on the real plant but significantly faster
• Building an APC model that can be used as a ‘seed model’ for constrained
automatic step-testing
• Immediate benefit for TOTAL La Mede was to improve the inferential
for benzene concentration in the bottom product
• More benefits are to come once all available data is fully exploited

24
 Thank You

Any Question ?

25
BACK-UP SLIDES

26
 Q #1 – MV Gains

Gain of side draw FC5008 Gain of steam flow FC5005

APC gain to Bz top 0, to Bz medium APC gain to Bz top -0.17, to Bz medium
variable and to Bz bottom -0.24 +0.43 and to Bz bottom -0.14

Gain of pressure PC5011 Gain of feed temp TC5000

APC gain to Bz top +0.18, to Bz medium APC gain to Bz top 0, to Bz medium
-0.33 and to Bz bottom +0.11 +0.36 and to Bz bottom 0
27
 Q #2 – Influence of
other MV’s

Influence of pressure PC5011 to gains Influence of steam flow FC5005 to gains from
from tray temp TC5031 tray temp TC5031
1.164 bg vs 0.564 (base case) 8900 KW instead of 10300 KW (base case)
Equivalent to a change in steam flow of 2 T/h

28
 Q #2 – Influence of
other MV’s – Con’t

Influence of feed temp TC5000 to gains Influence of feed flow FC5000 to gains from
from tray temp TC5031 tray temp TC5031
90.7°C vs 100.7°C (base case) 100.5 T/h instead of 85.5 T/h (base case)

29