Sie sind auf Seite 1von 9

139f

A CREATIVE SOLUTION TO A SPLITTER WITH MULTIPLE CONSTRAINTS USING APC


Anthony Congiundi
Senior Advanced Process Control Engineer
Equistar Chemicals LP, a LyondellBasell company
Kurt Rickard
Consulting Advanced Process Control Engineer
Equistar Chemicals LP, a LyondellBasell company
Prepared for Presentation at the 2013 Spring National Meeting
San Antonio, TX, April 29 - May 2, 2013
AIChE and EPC shall not be responsible for statements or opinions contained in papers or
printed in its publications

A CREATIVE SOLUTION TO A SPLITTER WITH MULTIPLE CONSTRAINTS USING APC


Anthony Congiundi
Senior Advanced Process Control Engineer
Equistar Chemicals LP, a LyondellBasell company
Kurt Rickard
Consulting Advanced Process Control Engineer
Equistar Chemicals LP, a LyondellBasell company

Abstract: A splitter overhead condenser system in one of our Olefin plants has unique heat
integration with the feed propane vaporization system and cooling water system. Similar to
other olefins plants, the reboiler has heat integration with the quench system. The complex
heat integration results in several challenging constraints that must be managed. In addition,
the active constraint set changes often depending on ambient and unit conditions. These
challenging constraints were not well handled in the existing Advanced Process Control (APC).
To address this issue, the APC was revamped using a creative design to robustly handle these
constraints. In addition, several advanced techniques and unique solutions were used
throughout the design, modeling, and implementation of this revamp. A unique aspect of this
revamp is that the APC uses a very long time-to-steady-state model (30 hours) while using a
steady-state model for the product composition. Most APC models are significantly shorter
(about 2 hours) with ramp models for the product composition.

Background
The C3 Splitter in one of our Olefins plants is capable of producing both chemical grade
and polymer grade propylene. Throughout the plants history, AspenTech DMCPlus
controllers have been implemented in most of the units between 1999 and 2007. Several new
DMC applications and revamps have occurred since 2007. In 2011, the C3 Splitter was slated
for revamp.
Figure 1 below shows an overview of the C3 Splitter process flow and Distributed
Control System (DCS) controls. The product DCS control configuration uses an internal reflux
to total reflux ratio to control product draw. The internal reflux calculation is derived using a
material balance (reflux side draw) and energy balance (the additional reflux caused from
the sub-cooled reflux). When not in DMC control, an analyzer controller will drive the ratio.
The quench water to the reboiler is controlled using a duty controller which is reset by a
lower tray temperature when not in DMC control. The duty controller is cascaded to a quench
water flow controller.
The tower pressure is controlled by manipulating the outlet valve of the cooling water
condenser, which indirectly controls the level in the shell side of the condenser. The tower
pressure is also impacted by a parallel condenser that provides vaporization of fresh propane
feed (Vap A).
Since Vap A is integrated with the propane vaporization system, the other parts of the
propane vaporization system can impact the C3 Splitter. The propane system is made up of
three parallel propane vaporizers. Fresh propane feed is split between the three vaporizers.
Vap B and C are on level control and are typically maximized to provide maximum vaporization.
These vaporizers provide subcooling to other process streams that are downstream of cooling
water condensers. Propane feed to Vap A, as mentioned earlier as providing condensing duty
to the C3 Splitter overhead, controls the pressure to the propane header. As a result, Vap A is
the swing propane vaporizer and rejects any disturbances in the propane feed system. These
disturbances are passed onto the C3 Splitter.
Over time, the propane vaporizers can become fouled. As fouling increases, the
propane vaporization regularly becomes a constraint for the C3 Splitter. When more duty is
required to vaporize propane, the level in Vap A will rise above the top of the tube bundle.
Frequently, the tube bundles in Vap B and C are already completely covered by propane,
resulting in a limit to the propane vaporization system. One method to provide additional
vaporization is to increase the C3 Splitter tower pressure, which provides more heat in Vap A.
Reasons for Revamp
The C3 Splitter DMC controller was slated for revamp because it did not account for the
propane vaporization constraint and it was missing key models that led to poor pressure
control. The previous controller had 3 manipulated variables (MVs): pressure, internal-to-total
reflux ratio, and duty. The pressure and duty MVs were disabled and controlled in the DCS
because of poor performance. Pressure control was poor because no models existed from
pressure SP to Pressure PV, quench water valve, pressure valve, Vap A level and product
analyzer. Several feed forward variables were also missing, such as ambient temperature,

propane vaporization system variables and furnace feed. The duty control did not maximize
up to the valve constraint properly. The analyzer control was adequate, but could be
improved.
New DMC Controller Design
A new creative design was developed that corrected the previous designs issues, properly
maximizes heat duty and minimizes pressure against changing tower constraints (including the
propane vaporization constraints) while maximizing product impurity. The objectives of the
new DMC include the following:

Maximize propane in product close to specification target.


Maximize reboiler duty to constraints (minimizes propylene losses out of the tower
bottom)
Maximize overhead pressure output to constraints (minimizes tower pressure against
constraints)

The controlled variables, manipulated variables and feedforward variables are shown in
Figure 1. The new design is capable of handling several changing constraint sets. The DMC
controller writes directly to the internal-to-total reflux controller (FFIC) to maximize the
propane in the product to its upper limit. The controller maximizes the duty up to the valve or
tower pressure upper limits to minimize propylene loss. The controller minimizes pressure
using the pressure controller output until it hits the upper limit of the propane vaporizer level.
Minimizing pressure also minimizes propylene losses by increasing the temperature differential
between the tower and quench water. The controller uses several feedforwards to improve
predictions and minimize relying on feedback.
The revamped DMC controller is capable of controlling against changing constraint sets.
The constraint sets the C3 Splitter DMC will push against are as follows:
1. High propane in product, high output on quench water flow valve, high level in Vap A
2. High propane in product, high tower pressure, high output on tower pressure valve
3. High propane in product, high output on quench water flow valve, high output on tower
pressure valve
In all constraint sets, the propane in the product is maintained at the upper limit.
Constraint set 1 is the constraint set that the tower experiences most of the year. Since the
propane vaporizers are currently fouled, the tower pressure must be increased to vaporize
more propane in Vap A. The higher tower pressure causes less temperature difference and
heat duty in the reboiler, which causes additional propylene losses.
Constraint set 2 is experienced mostly in the hotter summer months during the day when
the tower is at a condensing limit. When this occurs, the tower pressure is maintained at the
upper limit by reducing reboiler duty, and the pressure control valve remains wide open to
achieve the maximum condensing duty. At night, the constraint set commonly switches back
to constraint set 1 and the tower pressure is reduced to the propane vaporization constraint
and the duty is added back to the valve constraint.

Constraint set 3 is experienced when the tower is not heavily loaded and offers the most
efficient operation of the C3 Splitter. The tower operates at the lowest possible pressure,
maximum duty, and lowest propylene losses while maintaining the product close to the
specification limit. The duty is maximized up to the valve limit, and the tower pressure is
minimized to the condensing limit while not violating the propane vaporization constraint.
Heat Integration
There are multiple ways in which heat integration of this process causes disturbances to
the DMC controller that are not normally encountered when applying this control technology to
a super-fractionator. Beyond the integration loops shown in Figure 1, another feedback path
exists from the tower through the cooling water system and back to other points in this tower.
In Figure 1, the quench water loop integrates with the reboiler, and the propane
vaporization system integrates with the C3 Splitter condenser. The direct effect of this
integration is due to the resulting flow changes from the tower bottoms to one of the
condensers. In addition, a more subtle connection exists through the cooling water that
connects the propane vaporization system, quench water system and C3 Splitter. This subtle
connection creates a feedback loop that adds a layer of complexity to this controller.
Interaction between the quench loop and the cooling water starts from controller
induced changes in the reboiler duty, which changes the duty available in the quench water
loop. This in turn impacts the cooling water temperature via a cooling water cooler in the
quench water loop. The cooling water temperature then impacts the C3 Splitter condensers,
which then can cause further controller induced changes in the reboiler duty to keep pressure
well controlled. The nature of this connection sets up a positive feedback loop that is well
managed by the active control of this tower and the quench system.
The cooling water temperature also impacts the propane vaporization system through
Vap B and Vap C. As mentioned earlier, these vaporizers are downstream of cooling water
condensers on other process streams. These cooling water condensers provide certain degree
of subcooling depending on the cooling water temperatures. When the cooling water
temperature drops, further subcooling lowers the temperature entering Vap B and C causing
less duty to be available for propane vaporization. When at a maximum duty capacity in Vap
A, Vap A will fill up and the controller will increase column pressure to provide additional duty
to the propane vaporization system (Constraint set 1 above). The connection through the
quench water system can then result in an increase in cooling water temperature creating a
feedback loop that converges.
Advanced Techniques / Solutions
Throughout the design, modeling, and implementation of the C3 Splitter revamp,
several advanced techniques and unique solutions were used. One unique aspect of this
revised controller is the very long time-to-steady-state model (30 hours) which allows the use
of a steady-state model for the product composition, whereas, many APC models for
distillation columns are significantly shorter (about 2 hours) with ramp models for the product
composition. This long model provides for excellent control of the product composition. To

achieve good control of the fast dynamic variables with the long model, 64 future moves and
small move suppressions (0.1 to 0.2) were required.
A nonlinear transform is applied to the internal-to-total reflux ratio MV (see Figure 1) to
achieve a good-quality, linear model to the product composition. The internal-to-total reflux
ratio is transformed using a piece-wise, linear transform using the function 1/(1-FFIC) which is
equivalent to a reflux-to-distillate ratio. Reflux-to-distillate ratio exhibits linear dynamic
response with respect to propane in the product. Figure 2 shows the case results from the
untransformed ratio and transformed ratio to the product composition. Note, for the
transformed case, the different time-to-steady-state models converge to the same steadystate gain. This transform was used in lieu of revising the DCS control configuration and
retraining operations personnel, thereby achieving improved control while saving project
execution time.
Another advanced technique in this DMC controller is the use of prediction errors to
extract the independent portion of two slightly dependent variables. This process is affected
by both the ambient temperature and cooling water temperature, so both should be included
as feedforward variables. However, these two temperatures are not totally independent of
each other. In order to extract the independent portion of the cooling water temperature, a
cooling water prediction error was developed by modeling the ambient temperature to the
cooling water temperature. This model is then included as a controlled variable (CV) in the
final DMC controller to provide a prediction of cooling water temperature. A prediction error is
calculated on-line as the difference between the actual cooling water temperature and this
prediction of cooling water temperature. This prediction error is used as a feedforward signal
in place of the actual cooling water temperature.
Prediction errors are also used to detect unmeasured disturbances for two other
measurements, which reduces the disturbances on important CVs. Without these prediction
errors as feedforward variables, the controller would rely solely on feedback to compensate for
these disturbances. A duty prediction error was developed to account for unmeasured
disturbances in the quench loop and account for saturation in the quench water supply valve.
This prediction error was developed by modeling the duty setpoint and quench water inlet
temperature against the duty process variable. Any unmeasured disturbances in the quench
loop would result in a prediction error, which is fed into the model as a feedforward and
improves the predictions of important CVs. With more accurate predictions, the controller can
adjust its move plan to reject these disturbances. A similar prediction error was developed for
the sum of the fresh propane and propane recycle. This prediction error accounts for
unmeasured disturbances in the propane vaporization system related to the tube-side of the
propane vaporizers, Vap B and Vap C in Figure 1.
In order to properly maximize the duty up to the valve limit, a technique known as sag
control is used. This technique is used to allow smooth dynamic control at the true valve limit.
This is achieved by creating an effective output that could increase linearly beyond 100%
based on the error between the SP and PV of the MV (flow controller) variable. Without the
use of this technique, the steady-state targets calculated by the controller would flip between
a duty to achieve the valve limit and a wound up value. Prior to implementing this technique,
the performance of the control was degraded significantly. Adding the effective output solved
the issue and allowed the DMC controller to successfully control the duty at the true valve limit.

Another unique solution used to achieve improved control of Vap A level (see Figure 1)
is using the column pressure controller output as an MV instead of the SP of the DCS loop.
The DCS pressure controller does not control pressure tightly enough to provide good control
over Vap A level. The DMC controller is better able to control the pressure because it is aware
of all variables impacting the pressure. In addition, setting the output allows direct and fast
control of Vap A level. Using the Pressure SP as an MV in the DMC controller would add
unnecessary lag and degrade the quality of the models of CVs that depend on column
pressure.
The revamped DMC controller needed to have the ability to control both chemical grade
propylene (CGP) and polymer grade propylene (PGP). To accomplish this, two CVs were
created for CGP and PGP respectively. CCF calculations, as shown in Figure 3, were used to
dynamically switch on the appropriate CV for CGP or PGP and switch off the other CV. When
using this technique, an interesting effect was noticed. The controllers dynamic move plan for
the active CV was slowed down by the dynamic equal concern errors (ECEs) of the inactive CV.
Even though the inactive CV was turned OFF, the dynamic ECEs are scaled based on all of the
variables. This can dramatically slow down the dynamic move plan if the scaling of the
variables is largely different. To solve this issue, CCF calculations were added, as shown in
Figure 3, to zero the gain multipliers associated with the inactive CV.
In order to obtain the models necessary for both CGP and PGP, two step tests were
performed. As discussed in Huang and Riggs (2002), this can require a large amount of data
for models with long time-to-steady-states. To obtain this data, advanced testing techniques
were used. A manual step test was performed for CGP first for two weeks to get initial seed
models. These models were short time-to-steady-state models and used a ramp model for the
product analyzer. Then a longer 1 month step test was performed using SmartStep
without around-the-clock coverage. The feedforward variables were manually stepped by the
operators. Another manual step test was performed for PGP for one week to adapt the CGP
models. Then a longer two month step test was performed using SmartStep, again without
around-the-clock coverage.
Results
The DMC controller has been performing quite well controlling the product composition,
maximizing duty, and minimizing tower pressure while keeping the propane vaporization
constraint under control. The tower now operates more efficiently at the lowest possible
pressure and maximum duty while maintaining the product close to the specification limit,
resulting in lower propylene losses. The controller has provided significant benefit to the
process unit by resulting in a 1.5% increase in propylene production.
Acknowledgements
Specific thanks to Dennis Williams, Consulting
Montes, Principal APC Specialist, LyondellBasell for
commissioning.
In addition, the authors thank
LyondellBasell, and Robert Hawkins for their expert

APC Engineer, LyondellBasell and Gilbert


their contribution during step test and
James Hackney, Senior APC Advisor,
advice. The authors thank the team of

operators, I&E, Technical, GES APC and many others that contributed to the C3 splitter
DMCPlus implementation.

TI
CW

LI

PIC
Legend

Ambient
TI

Vap A

Controlled Variable
(Constraint)
Manipulated
Variable

FIC

LIC

Feedforward
Variable

AIC

Feed

FFIC

FIC
Product
Propylene

UIC

TIC

LIC
TI

FIC
LIC
Vap B

Quench
Water

PIC

Fresh
Propane
LIC
FIC
AI
Vap C

Figure 1. C3 Splitter Overview with DCS controls and DMC variables

Cracking

Untransformed Ratio

Transformed Ratio

Figure 2. Untransformed and transformed ratio to the product composition case

Figure 3. CCF calculations used to switch on and off CGP and PGP CVs
References
1. Huang, Haitao and James B. Riggs (2002), Including Levels in MPC to Improve Distillation
Control, Ind. Eng. Chem. Res., Vol. 41, No. 16, pp. 4048-4053.

Das könnte Ihnen auch gefallen