You are on page 1of 11

Implementation of Unit Wide Advanced Process Control

In Ethylene Plant
Gabe Haarsma
Advanced Process Control Specialist
Shell Chemical LP.
Rajendra Mutha
Senior Consultant
Shell Global Solutions (US) Inc.


Abstract: This paper gives a general overview of the Advanced Process Control (APC) solution
in the Shell Chemical Ethylene plant in Deer Park TX. APC has been implemented on both the
hot side (pyrolysis furnaces and pyrolysis fractionation column) and the cold side
(compressors, core heat exchangers and fractionation train) of the plant. Implementation of
the APC is performed with the SMOCPro Advanced Process Control package from Shell Global
Solutions. Particular features, like intermediate variables, are highlighted. The APC is
experiencing very high uptime numbers and remains on during furnace transitions. The paper
also shows various economic benefits that are achieved with the APC solutions.

Introduction
Since the early 1980s, Shell uses its own APC technology called SMOC (Shell
Multivariable Optimizing Controller). From the beginning Shell has selected state space
modeling for the controller design and uses parametric models for process identification. Two
very useful features that follow from this approach are the possibility to use intermediate
variables and the use of cascade correction for controller model design.
The concept of intermediate variables allows the engineer to include more process
knowledge in the controller design and to build more robust and transparent controllers. This
is also known as the grey box model approach (as opposed to the black box approach).
The concept of cascade correction allows the engineer to explicitly model the behavior of
base-layer PID controller into the MPC design. This allows for a much more relaxed APC
solution, when the base-layer PID controller is capable of rejecting the disturbance.
Originally SMOC controllers used intermediate variables and cascade correction for
relatively small controllers (unit size, e.g. a column or a single cracking furnace). Since the
release of SMOCPro it is possible to use the grey box approach not only for controllers on a
unit level, but also to build large controllers covering a whole plant. For ethylene plants, this
means that a plant wide controller can be built using intermediate variables to link the unit
sub-controllers.

Figure 1 Difference between Black Box Model and Grey Box Model

In the Black Box model there are inputs & outputs and direct models (transfer
functions) between the inputs & outputs. There are no intermediate variables and there is no
specific knowledge about the process configuration contained in the model. In the Grey Box
model the inputs & outputs are linked via intermediate variables, which represent measured
variables (such as intermediate flows or temperatures). In this way the model represents the
actual process much better and allows a more robust feedback mechanism, by using the
process measurements for these intermediate variables. It is also more intuitive for engineers
and operators to build a multivariable controller this way.
The next section describes the benefits of using the grey box modeling approach, using
intermediate variables and cascade correction in real-life examples from a recent ethylene
plant APC project. The paper is concluded with describing the various benefits of the APC
project.

Brief technology description
The main APC technology used for this project is SMOCPro for the multivariate advanced
process controller. Some advanced SMOCPro features include:
Advanced modeling capabilities including intermediate variables, cascade correction
of manipulated variables and explicit options to model the magnitude and dynamics
of unmeasured disturbances.
The concept of sub-controllers, which allows the control designer to break a unit-
wide controller down into smaller sections that ease the tuning of the controller as
well as reducing the computing time of the unit-wide controller.
Robust (bi)linear steady state optimization, which allows the control designer to
define (bi)linear economic functions. The steady state optimization has build in
robustification, in order to avoid LP chatter.
On-line updateable model gains, allowing the control designer to employ gain
scheduling and nonlinear control techniques.

Robust Quality Estimators used for the project are built with RQEPro. Good process
models form the basis for robust and profitable performance of any model-based advanced
control scheme. To build the dynamic process models in the APC project, Advanced
Identification and Data Analysis is performed with AIDAPro. To sustain the application after
project completion it is important to define performance indicators. Target sheets are
developed to continuously show stakeholders that the system is performing well. MDPro is
Shells process control monitoring and diagnostics tool, which can used both online and offline.


Intermediate variables Furnace Example
The given ethylene plant has a large number of furnaces. These furnaces are of several
different process designs and can process different feedstocks. Therefore there are a large
number of possible controller models. With the intermediate variable approach we can simplify
this problem, while still covering all possible combinations.


Figure 2 Cracking furnace Coil Outlet Temperature (COT) control

From a process design and operation point of view, the furnaces can be divided into a
furnace feed side (tube side) and a furnace firing side (fire-box). The control is linked via the
Coil Outlet Temperature (COT) controller, which acts on the fuel gas supply pressure
controller. Whenever feed flow, steam-to-oil-ratio or COT controller adjustments are
happening on the process side, this always can be translated for the firebox side as a change
in fuel gas pressure. The fuel gas supply pressure can therefore used as an intermediate
variable.
The furnace model matrix can be divided in two independent matrices; one for the feed
side and one for firebox side. The fuel gas pressure is used in both matrices. In the tube side
matrix it is a model output and on the firebox side it is a model input. The two matrices are
linked, using the fuel gas pressure as intermediate variable.
The intermediate variable approach offers the following advantages:
1) It saves significant time during model identification. In principle, full plant tests
are required on one furnace type and one feed type only. If different feed types are
used (for example one furnace on naphtha and another on butane) then this will
affect the feed side models, but not the firebox models.
2) Less transfer functions are required and therefore a smaller model is achieved. In
the example of figure 3 we have 17 transfer functions (9 on the tube side plus 8 on
the firebox side). If we would have used a black box model, then 30 transfer
functions would be required (4 inputs * 6 outputs = 24 transfer functions).
3) The controller has better unmeasured disturbance rejection. If there are
fluctuations in the fuel gas pressure (for example due to variation in BTU value),
then this unmeasured disturbance is passed on to the firebox constraint (feed
forward). In a black box model the controller can only react after the change in
fuel gas pressure has actually affected the firebox constraints (feedback).
COT
P

Figure 3 - Furnace model, using fuel gas pressure as intermediate variable


Cascade Correction Furnace Example
In general APC writes its controller output solution to the setpoints of base-layer
controllers. Typically these are PID controllers residing in the DCS. If the closed loop response
time of the base-layer controller is sufficiently fast (around 1-3 iterations of the APC), this
poses no problem for the APC. Many flow loops fall into this category.
If however the closed loop response time of the base-layer loop is much slower (>5
iterations), and sufficient disturbances affect the base layer controller, problems arise for an
APC that is based on an input-output model. Many furnace outlet temperature controllers and
distillation tray temperature controllers fall into this category. APC solutions with an input-
output model only, have no notion that the base-layer controller is working to reject the
disturbance. When the disturbance is picked up by a furnace severity indication or a distillation
column analyzer, an APC with input-output model only will make moves to the base-layer
(temperature) controller that were not needed.
It is possible to break open the base-layer controller and write directly to the valve, but
this has significant drawbacks:
PID controllers in the DCS, typically execute every second or faster. In order to
reject disturbances in a similar manner, the APC execution frequency has to be
increased.
During plant step testing, more loops are in manual, which complicates the step
testing procedure. Since several disturbances are not rejected anymore, the quality
(signal to noise ratio) of the step test declines.

Because SMOC is based on a state-space model, it offers a third approach. It is possible
to completely model the behavior of the base-layer PID controller into the APC disturbance
model. There is no need to break open the base-layer controller and SMOC has proper
knowledge about the base-layer behavior.

X-over Temp
Severity
Indication
Fuel gas pres.
SP
Fuel Gas
pres. OP
Firebox O2 BTU/HR
Feed Rate SP X X X
Steam/Feed Ratio SP X X X
Coil Outlet Temp SP X X X X X X
Draft pres. SP X X X X
X
Feed model Firebox model
Figure 4A - Furnace model, normal input-output action model

Figure 4B - Furnace model, disturbance model

It can be seen in figure 4B that when an unmeasured disturbance hits the base-layer
COT PID controller, SMOC has knowledge about how the disturbance gets rejected and what
the effect is on other process variables. When the severity analyzer gets a higher reading, due
to the temperature disturbance, SMOC will not make additional changes to the COT setpoint,
since the base layer PID is capable of rejecting the disturbance.
Cascade correction offers significant advantages due to faster and more accurate
unmeasured disturbance rejection. Unnecessary moves are not generated when the base-layer
PID is capable of rejecting the unmeasured disturbance. This results in a more relaxed
controller behavior.

Plant Level Gray box modeling
The previous examples showed the benefits of using the intermediate variables and
cascade correction approaches on a unit level (i.e. furnace). A similar approach, using
intermediate variables is also very powerful for a plant-wide controller. On this project Shell
has also implemented a plant-wide controller with one dynamic model for the entire ethylene
plant. Figure 5 shows a schematic overview of this plant wide controller.



COT PV
C
O
T

S
P
Severity
Indication
Fuel Gas
press. SP
Firebox O2 COT PV
C
O
T

S
P
Severity
Indication
Fuel Gas
press. SP
Firebox O2
C
O
T

P
V

(
e
r
r
o
r
)
COT PV
Severity
Indication
Fuel Gas
press. SP
Firebox O2
C
O
T

P
V

(
e
r
r
o
r
)
COT PV
Severity
Indication
Fuel Gas
press. SP
Firebox O2

Figure 4 Overview of plant-wide model, using intermediate variables

The various sub-models are linked, using intermediate variables. The feed to each
column is an intermediate variable with dynamic models from upstream sections of the plant.
Similarly the feed to the core heat exchangers (DMFT) is modeled from the furnace sections of
the plant. The SMOC state space model predicts the intermediate flows. The real flows (from
plant measurements) are used to provide feedback to the SMOC controller, providing robust
dynamic control.
The Propylene Refrigerant Compressor (PRC) and Ethylene Refrigerant Compressor
(ERC) have many interactions with the DMFT and various columns. Interactions come from
heat integration and provided cooling to condense overheads of columns. These interactions
are not drawn in the schematic, but are important to model. This modeling effort, which can
significant, is important in order to maximize energy recovery in the core heat exchangers and
unload the refrigeration system. This will produce significant energy savings.
The use of intermediate variables provides a very robust and practical Advanced
Process Control Solution for ethylene plants with many advantages over conventional APC
solutions, such as:
Furnace
Type 1
Furnace
Type 2
Furnace
Type 3
Pyro
Frac
Process Gas
Compressor
LPS HPS
DeMeth
Feed Train
De-
Methanizer
De-
Ethanizer
C2 Splitter
De-
Propanizer
De-
Butanizer
ESD
Fuel Oil
CLGO
C5 Gasoline
Propylene
C5 Gasoline
C4
Ethylene
H
2
/ CH
4
Ethane Recycle Feed
ERC PRC
Propylene and Ethylene
Refrigeration Systems
= Intermediate Variable
Furnace
Type 1
Furnace
Type 2
Furnace
Type 3
Pyro
Frac
Process Gas
Compressor
LPS HPS
DeMeth
Feed Train
De-
Methanizer
De-
Ethanizer
C2 Splitter
De-
Propanizer
De-
Butanizer
ESD
Fuel Oil
CLGO
C5 Gasoline
Propylene
C5 Gasoline
C4
Ethylene
H
2
/ CH
4
Ethane Recycle Feed
ERC PRC
Propylene and Ethylene
Refrigeration Systems
= Intermediate Variable
Using the intermediate flows and product flows as intermediate variables in the
SMOCPro controller provides continuous feedback and consolidation of the internal
SMOCPro model. This gives very robust plant-wide control.
Production can be optimized with a dynamic controller. Furnace feed rates and
cracking severities can be balanced to provide maximum production. This is an
advantage over the more conventional feed pusher solutions where feed is
maximized, but severity is not adjusted.

Benefits of Advanced Process Control project
Typically reported benefits from the implementation of APC on an ethylene plants vary
from 1 to 10 million dollars per year, normally resulting in project pay-back times of 1 year or
less. The benefits are being generated from the moment the project activities start. In some
cases the project already paid for itself before it is even finished. The potential benefits of
course depend on the plant capacity, its feed slate, the constraints and local economics.
The cold side controller has been commissioned as first, about 1.5 years ago. A post
implementation review has been conducted to capture economic benefits and well as other
non-tangible benefits.
A first criterion for any successful APC project is good operator acceptance and high
uptimes. Uptime has been very high since controller commissioning. See table 1 for monthly
data. This is a key indicator that operator acceptance is high. It is also clear from the uptime
numbers that the controller remains on during furnace transitions. SMOC is keeping the olefins
plant on target during furnace transitions and remains on control.

Table 1 Monthly uptime numbers of the various sub-controllers.
PGC LPS HPS DMFT DC1 DC2 C2S ESD DC3 DC4 PRC ERC
99.5 97.4 98.9 99.2 97.7 97.6 97.5 98.9 96.0 97.3 98.7 99.4
99.9 99.9 99.9 99.6 98.0 99.8 98.8 99.9 99.8 99.5 99.7 99.9
98.6 99.0 99.0 96.8 98.6 98.6 92.5 98.8 97.4 92.1 94.2 98.6
98.3 97.3 97.8 98.0 97.8 96.3 97.2 98.3 97.5 93.4 96.1 98.3
88.5 97.7 99.0 77.2 99.7 78.7 99.1 99.1 99.2 99.7 65.3 84.1
90.6 97.6 99.9 96.6 93.4 96.1 95.4 98.9 95.7 94.0 96.3 91.1
96.7 96.9 97.6 97.3 97.4 97.7 96.2 98.8 97.6 95.2 95.7 95.2
88.5 97.7 97.7 97.7 97.6 97.7 96.1 96.8 97.3 91.6 97.7 97.7
99.8 100.0 100.0 100.0 100.0 99.8 100.0 99.6 99.7 99.0 100.0 100.0
95.1 95.5 95.1 95.2 94.1 95.5 94.1 95.7 95.2 95.5 95.2 94.9
100.0 93.1 99.9 100.0 100.0 100.0 100.0 98.7 83.6 100.0 89.6 100.0
99.8 99.4 100.0 99.9 99.8 99.8 100.0 99.3 64.9 99.9 100.0 100.0
99.8 99.9 99.8 99.9 92.0 97.9 99.9 99.7 85.3 85.4 99.7 99.8
99.7 99.7 99.7 99.7 99.2 99.1 99.7 99.3 91.1 92.8 99.7 99.6
98.6 99.7 99.7 99.7 98.8 99.5 96.9 99.3 87.0 98.3 99.3 99.2
99.9 99.8 99.9 99.7 99.8 99.6 97.9 99.7 87.0 86.9 99.8 99.3
99.9 99.9 100.0 99.7 100.0 100.0 100.0 100.0 98.2 0.0 100.0 99.7

The new APC loops were designed to maximize economic benefit to the plant. Most
loops maximize economic benefit by minimizing energy consumption. For others, maximum
benefit is achieved by maximizing product recovery and or yield. For each loop, economic
calculations were performed using rigorous plant steady state model process models to define
the loop economics objectives. Table 2 summarizes economic benefits of these new controls
in terms of reduced energy consumption and increased product recovery.


Table 2 Economic benefits of cold side APC

Achieving
$ /yr
PGC Suction pressure

PRC discharge pressure 648,000
PRC suction pressure 368,000
ERC suction pressure 570,000

Ethane header pressure 401,000

Ethylene in HPTG 95,000
DC1 control tray temperature 116,000

Controlling propylene in DC2 Overhead
Raising ethane in DC2 Bottoms 93,000

Raising ethane in ethylene product 93,000
Lowering ethylene in recycle ethane

Raising C4's in PDC3 and SDC3 Overhead 330,000
Raising C3's in Crude BD

Raising C5's in Crude BD 575,000

Lowering HPS tray temperature 504,000

Controlling C4's in Pygas (LPS and DC4 both)

Sum of incentive/benefit 3,793,000

Significant savings are generated in the Propylene Refrigerant Compressor (PRC). A lot
of the benefits are generated by reducing the load on the refrigerant systems due to better
heat integration in the core set of heat exchangers or by reducing the refrigerant load on
condensers on overheat sections of distillation columns. These savings are only possible if the
APC has good understanding about the various interactions come from heat integration and
provided refrigerant cooling to condense overheads of columns. Ultimately, these savings in
refrigerant use result in reduction of steam to the turbines of the refrigerant compressors.

Figure 5 PRC Steam consumptions per LB C3 and lighter products

Beyond the dollars and cents there are also significant other benefits from an APC
project. Although these might not help you with the project justification from an economic
standpoint of view, it is important to capture them. Figure 6 shows a significant reduction in
quality excursions for the main products of the olefins plant from pre APC (2005) to
commissioning of the APC (mid 2006) to post APC (2007).

Figure 6 Product quality excursions for Ethylene and Propylene
C3 and lighter products produced
P
R
C

1
2
5
0

p
s
i

S
t
e
a
m
PRC Utiliz vs C3 and Lighter Products Post APC PRC Util vs C3 and Lighter
C3 and lighter products produced
P
R
C

1
2
5
0

p
s
i

S
t
e
a
m
PRC Utiliz vs C3 and Lighter Products Post APC PRC Util vs C3 and Lighter
Ethylene Propylene
2005
2006
2007
Ethylene Propylene
2005
2006
2007
Although it is not possible to take full credit for the reduction in product quality
excursions as other productivity, reliability and quality initiatives have been rolled out in the
unit, a partial credit is certainly due. Perhaps the following board operator feedback for the
APC project sums it all up:


I'm very satisfied with the SMOC controllers. Like most everything, it takes a little time to gain
confidence in a new system, but this system proved itself quickly. One main test I wanted to
evaluate was a big ethane furnace swap. When a swing like that comes through the unit it took
a lot of operator time and attention to manage the event. The SMOC controllers now handle the
furnace changes with normal monitoring.
I can now trust the SMOC controllers to make adjustments where needed, monitor the
adjustments, and concentrate on staying within my targets. Another benefit is that I think the
reliability has increased. Because I can focus on staying within targets and allow the controllers
to make the minor adjustments the entire unit runs in a more steady state and steady means
more reliable. In these times of increased pressure to run more environmentally friendly, safely
and more productively, SMOC has proved to be a mighty tool.

You might also like