The appropriate use of advanced process control
(APC)specifically, multivariable predictive control
(MPC)has been well established in the hydrocarbon
processing industry over multiple decades, and it is widely
considered an essential contributor to production maximization
on liquefied natural
gas (LNG) trains. If correctly applied, APC software
delivers more efficient operation of existing hardware assets
and essentially provides a cruise control for the
control room operator.
The Woodside-operated Karratha Gas Plant (KGP) has been
progressive in the application of APC across all major process
units, generating sustained benefits. Although the site is a
mature APC user, there is a continual focus on innovation and
design evolution to further improve APC benefits.
This article describes the implementation of APC on an LNG
liquefaction train. Several generic APC project aspects are
investigated, such as the use of a dynamic simulator and
automated step testing to aid development. Also, details of the
projects significant operability and economic
benefitsincluding a 4,000% return on investmentare
discussed with commentary on whether this success has been
sustained beyond the honeymoon period.
Woodside engaged Apex Optimisation to assist with a revamp
of the existing APC on LNG train 4 (LNG4) and the
implementation of a new APC on LNG train 5 (LNG5). The project
was a collaborative effort, with both parties heavily involved
in the design, implementation, commissioning and post-audit of
the new APCs. The implementation kicked off in March 2010 after
a functional design specification phase. The revamped LNG4 APC
and the new LNG5 APC were commissioned in May 2010 and
September 2010, respectively. A successful site acceptance test
signaled handover to site support engineers in October
Challenges to development.
The execution of the project was challenging due to a range
The design evolution significantly pushed the
previous projects boundaries. Additional compressor
power-management handles were included, the site electrical
power-generation spinning reserve and fuel gas system capacity
limits were added (these global constraints are relevant to
both trains), and a more sophisticated approach to optimizer
functionality was adopted. Hence, the scope of the modeling and
custom functionality required was substantially different from
that of the previous LNG4 APC application.
The new applications are relatively large, with
each having over 20 manipulated variables (MVs) managing more
than 60 controlled variables (CVs) and some complex
interactions (i.e., relatively high model density).
Parts of the process are highly nonlinear in
their behavior, and this can limit the applicability of linear
APC technologies. Improved performance was needed during lower
production conditions (e.g., turndown or hot summer
temperatures), and this required some innovative use of
transforms, gain scheduling and automatic logic to manage
variable usage. Dynamic simulation was leveraged to develop the
gain scheduling relationships.
As the existing LNG4 APC had been unused for
over a year, there was limited operator expertise with APC on
the LNG4/LNG5 distributed control system (DCS) panel. This
situation required careful management of the reintroduction of
APC and operator training.
The LNG5 train was relatively young, with a
limited operating history. Furthermore, its operation was very
different from that of LNG4, despite the equipment design being
essentially identical. Mechanical changes to the LNG5 train
during the execution phase of the APC project significantly
changed the train operation and reset the LNG5 APC design
needs. The project engineers had to remain flexible to adapt to
the changing basis while maintaining the project schedule.
Interfacing to some of the compressor packages
required an exotic approach. In particular, one key compressor
handle was hosted on a separate DCS network on the other side
of the control room. This context required careful software
design and operator training to ensure that the final mechanism
was robust and intuitive to both DCS operators.
Automatic step testing was adopted in order to
reduce the duration of the step-testing phase; this had not
been previously attempted onsite.
An aggressive schedule was required to
commission two large applications within seven months, which
kept the intensity high throughout the duration of the project.
Fig. 1 shows a schematic illustrating the process design for
the two liquefaction trains.
1. Process design for LNG4 and
These challenges were overcome through teamwork among the
participants. Close operator involvement was critical to
project success, as this fostered ownership of the project and
ensured that each process control improvement implemented was
intuitive for the operators and appropriate for the widest
range of process conditions.
One of the major APC benefits delivered is improved
consistency in how the process is managed. To realize this
benefit via sustained APC usage, consistency in how the APC is
operated is paramount. Therefore, thorough operator training is
essential to the project process. Fig. 2 shows Woodside DCS
operators at work in the control room.
2. Woodside DCS operators at
USE OF DYNAMIC SIMULATORS TO ASSIST MODEL
In recent years, the use of a dynamic simulator (i.e., an
operator training simulator, or OTS) has been promoted by
advocates as a more efficient way of developing APC. The
ability to speed up real time, avoid real-life plant reliability and load disturbance
impacts, reduce engineering support requirements, and
potentially complete the APC development well before the plant
is commissioned makes the OTS very appealing to cost- and
schedule-focused customers. These factors prompted Woodside to
investigate the use of an existing OTS to assist with the
conceptual design and initial (seed) model for the
automated step-test phase.
While the OTS is typically fit for the purpose of
investigating an APC optimization strategy and controller
structure, is it appropriate for APC model development? One can
build an OTS to varying levels of fidelity (with cost
implications), and the main objectives are
Enabling thorough DCS and emergency shutdown
system checkout and verification before construction
Providing useful operator training on the process
with the target system interface
Providing a useful what if? tool for
engineering analysis of process changes.
Ascertaining OTS fidelity.
To achieve these objectives, the OTS requires a level of
fidelity that is well practiced and accepted by OTS developers.
However, a standard OTS may not have the fidelity required for
complete APC model development; what is required is a function
of both the APC modeling needs (the APC design) and the nature
of the process included in the APC scope. Even if it is
identified as an OTS objective up front, the distant APC topic
may struggle to justify a costly increase in the OTS fidelity
among more traditional construction project needs.
The question then becomes, How can it be known if the
OTS has the required fidelity? This question is not an
easy one to answer unless an operating plant can be used as a
datum, or unless the process is extremely well understood from
a modeling perspective and the required fidelity exists.
In our LNG liquefaction APC example, the OTS system was
developed alongside the construction project, with
traditional objectives in mind and well before APC was
considered. The development of the OTS was given heavy focus
(including post-commissioning improvements to OTS accuracy in
selected areas), with high acceptance of the simulators
value. When using the OTS for the APC model development, we
found that the thermodynamics-related models were reasonably
accurate at base-case production rates. However, there were
discrepancies around many of the ∆P-related models
(especially those associated with complex devices such as
hydraulic turbines with multiple flow elements) and
turndown-related models (such as those associated with flow
regime changes experienced inside the spiral-wound cryogenic
heat exchanger). Given the exotic nature of the cases where
accuracy was lacking and the relative importance of these items
to the traditional OTS objectives, this is not a surprising
outcome from a traditional OTS used outside of its original
The value of the OTS in our LNG APC case was essentially
limited to the actions listed below:
Formulating the optimization strategy and
Being able to interrogate turndown cases, which are
relevant for hot-weather operation, without suffering
production losses on the plant or needing to contemplate a
second step test in more difficult summer conditionsthus,
providing valuable data on relative gain changes, which was
used in the gain scheduling logic
Providing useful, initial models for the automatic
stepping tool. As the new APC design was different in both DCS
control basis and scope, the previous model could not meet this
need in all areas.
Benefits of simulation.
A dynamic simulator of typical fidelity (OTS or desktop
engineering tool) can be useful in verifying an APC design
concept in terms of control and optimization strategies. This
need is more relevant for complex processes where the pre-APC
operation does not exploit all the available degrees of freedom
and some methodology needs to be developed. The APC model
accuracy required for accurate model development and full APC
benefits would be much higher than that required for strategy
A complete OTS-based APC model was developed as part of the
functional design phase to support the automated step test.
After the final model was verified post-commissioning, a
comparison was performed to assess the accuracy of the
OTS-based model. The results in several key areas are presented
in Table 1.
In summary, the knowledge gained from using an OTS for APC
model developments (as distinct to APC design and optimization
strategy) reinforces the following guidelines:
Understand the relevant accuracy of the OTS well.
There are obvious implications for developing APC on young or
difficult OTS processes prior to plant commissioning. In some
instances, the OTS has relevant accuracy inherently (e.g., the
C3 splitter example, where the distillation models are the key
aspect2). In other areas, the important APC needs
are not necessarily aligned with key OTS objectives.
Understand the value of using the OTS in APC
development; i.e., is it prohibitive to step test on the real
plant for operational or economic reasons?
Do not underestimate the value of working on the
real plant and interacting with operators for developing an
operations understanding (as distinct to a process
understanding) and cultivating APC understanding.
Always be prepared for some model error when
commissioning the APC on the real plant, and allocate
sufficient time to resolve any problems.
USE OF AUTOMATED STEP-TEST TECHNIQUES
Automated step-test techniques have been promoted in recent
years as a way of providing a rich data set in a short period
of time, thereby reducing project cost. Also, simultaneous
testing of multiple MVs could improve the accuracy of the gain
ratios that are important to the performance of the
This LNG liquefaction APC project was the first incidence in
which the site had used this technique as the primary step-test
approach, after successful testing on the liquid petroleum gas
(LPG) fractionation unit suggested it would be a time-saving
option. Despite the best endeavors of the project team, the LNG
train experience was somewhat different, with the net result
being neutral relative to a traditional, manual step test.
The reality was that this particular LNG
liquefaction process was not well-suited to this technique, for
the two reasons listed below:
1. The daily variation due to ambient temperature
swings is six times the maximum MV step size allowed for the
test. The automated tool works purely on process feedback,
whereas anyone operating the plant knows what moves have to be
made before the sun comes up. The manual test is superior in
this case, as the tester can plan moves using all information
available, not just APC variables. Thus, when using the tool as
intended, the moves required to control the process swamped the
random steps required for model identification.
2. Also, the extent of the load disturbances
encountered during a normal day demands both the need for
minimal optimizer action and the inclusion of extra steps in
addition to the automated steps.
For other processes where this is not the case, and manual
step-test costs are greater, this approach may offer a tangible
reduction in the step-test duration.
Test automation results.
Based on our cumulative experience with a range of automated
step-test techniques, our conclusions from test automation are
set out below:
Using the available APC model as a true model
identification seed model (as opposed to simply a
model used by the APC to manage the process during the test)
may considerably speed up the model development process. A
further enhancement would be the ability to assign confidence
to sub-models to assist the initial model identification.
With some processes, it is not viable to switch
off the optimizer action for long periods, much less for the
duration of a step test. In our LNG example, the superseded DCS
controls provided a high level of optimization that had to be
matched during the step test. The automated test must
accommodate this need with some sort of mild optimization.
It may be useful to automatically change step
direction if a full step size is not feasible due to potential
limit violations. If partial moves are applied, additional
steps may be required to achieve the same data quality.
As there can be a need to make extra moves on a
real plant, it may be desirable to include all moves made
during the step testnot just those made by the automated
toolin the model identification approach, as a means of
reducing the total test duration.
Real-time model identification can be very
useful, but one should not rely only on automated model
identification to signal that testing is complete. In one
instance, this approach produced some false negatives, which
would have prolonged the test further if additional
identification was not undertaken using traditional
approachesi.e., manual data grooming, careful slicing,
and finite impulse response (FIR) generation over multiple
times to steady state (TSS).
Engineers should not be required to work more
intensely than a manual step test in order to manage the
automated testing. Keeping in mind that the traditional
approach offers some additional value:
1. Time for detailed discussions
with operators at the panel is very effective from both a
public relations and training perspective.
2. Time to observe the plant
behavior and experience the challenge for the APC
provides useful insight into how the APC should act and sets
helpful expectations for the model identification.
Unfortunately, this valuable experience is generally negated by
automated testing tools, which step multiple MVs simultaneously
as the CV responses can no longer be seen by the eye.
3. Time to consider DCS control
servo response and make repairs early can greatly improve the
It is widely regarded that most efficiency tools added to a
well-proven methodology are no replacement for sound
engineering judgment. Generalizations about efficiency
improvements will be tested by the more challenging APC
projects. One needs to have confidence in significant
efficiency gains to warrant deviation from the trusted
methodology, especially when the payback on these projects is
CUSTOMIZATION OF APC APPLICATIONS
Woodside has nearly 15 years of experience with APC
applications in the relatively demanding environment of an
integrated production facility. The context is demanding in the
sense that personnel turnover is high at the remote site, and
the costs of poor performance are severe. Accordingly, effort
is required to maintain appropriate skill levels at the
This experience has proven the value of appropriate APC
customization to improve availability and robustness. Indeed,
the inability of the previous APC application to accommodate
the full range of operations was one of the main reasons for
its demise. A few examples of how the generic APC software was
augmented are discussed below.
Gain scheduling for turndown.
Analysis of previous APC performance and OTS scenarios
confirmed significant gain changes at reduced production rates.
These changes demanded custom logic to manage gain scheduling,
according to production rate ranges using discrete gain
multipliers. (Continuous formula-based gain scheduling was not
preferred due to the risk of producing ill-conditioned
matrices.) The logic also provided some automatic shedding of
specific MVs and CVs during turndown to accommodate the unique
Model adaptation for hydraulic turbines.
The power extraction from the hydraulic turbines is akin to
climbing to the summit of a hill, with constraints applying a
ceiling on how high one can climb. The model gains are very
much a function of the status of the surrounding DCS controls,
and if the alternative flow path opens up (the Joule Thompson
[JT] valve), the wicket gate is moved in the opposite direction
to maximize power extraction (i.e., one is on the opposite side
of the hill and needs to walk in the other direction to climb
In the past, this scenario had constituted a challenge for
the APC that was avoided by instructing the operators to ensure
that the JT valve was shut before giving the wicket gate
control to the APC. However, it was still possible to suffer
load disturbances, which bounced the process onto the opposite
side of the gain inflexion point. The results were not
With the addition of simple logic to flip the gain sign and
drop/activate specific constraints, the new APC has improved
robustness by allowing the operators to give the hydraulic
turbine control to the APC, regardless of the DCS control
state. The APC will honor the correct constraints with
appropriate wicket gate moves, and will walk the process over
to the correct side of the hill when feasible.
Product price-driven optimization.
Another feature of the new APC design is the ability to
specify product prices and use them to dictate the subtleties
of the optimization toward either maximizing LNG production or
LPG extraction. This arrangement is different from simply
specifying maximum LNG or maximum LPG, as each of the relevant
MVs has differing effects on the yield of each product. It is
useful to provide some shades of gray in terms of
the optimization options.
Aside from a purely economics-driven optimization, the APC
has maximum LNG and maximum LPG modes to assist logistics needs
without sacrificing valuable production (e.g., tank-top
scenarios that affect only one product).
The overall results of the project were exceptional, given
the challenges faced. Results included:
Excellent operator acceptance of all the
developments implemented during the project (i.e., DCS control
improvements, instrument repairs and APC commissioning), with
APC uptimes consistently greater than 97%. Operator feedback
shows that the new APC makes objectives easier to achieve.
A tangible contribution to improved reliability as a result of the APC
maintaining the process within constraints on a
minute-by-minute basis. In particular, the APC manages some
difficult operating envelope constraints associated with the
large axial compressors employed in the liquefaction process.
Prior to the APC, manual management of this relatively tight
feasible space, coupled with the production changes driven by
diurnal swings, left the DCS operators under continual
The production increase achieved with the same
process equipment represents a decrease in specific energy
consumption and a relative reduction in carbon footprint for
this important clean energy-producing process.
The project was completed on schedule and within
budget, despite an evolving design datum being prevalent
throughout the execution.
The APC benefits delivered a significant boost
to the bottom line for North West Shelf Joint Venture Partners,
with a 3%5% increase in LNG4/LNG5 production (depending
upon ambient conditions) and a 4.7% increase in LPG production
verified. This production increase delivered an overall project
payback of less than two weeks, or a return on investment of
At the 2011 Process And Control Engineering
(PACE) Zenith Awards, the project won the Oil & Gas
category and the Project of the Year Award ahead of 50
The LNG production benefits are best illustrated by the
reduction in compressor power giveaway, which is an inherent
characteristic of the process design. That is, production is
either limited by the helper motor power on the mixed
refrigerant (MR) compressor or the propane (C3)
compressor. The amount of spare compressor power not applied to
the process represents a production loss. Fig. 3 shows power
consumption of the primary compressors before the APC.
3. Power consumption of the primary
Following the commissioning of the new APC, the higher average
power consumption was a significant contributor to the
increased production capacity. Fig. 4 shows power consumption
of the primary compressors after APC commissioning.
4. Power consumption of the primary
compressors post-APC commissioning.
It is important to note that the project benefits have been
sustained one year later, with no deterioration in performance
or in operator satisfaction detected. Fig. 5 shows a comparison
of production vs. technical maximum capacity. This project demonstrates how the
appropriate use of APC technology can provide a tangible
and sustained improvement in plant profitability and
operability in a cost-effective manner. HP
5. Comparison of production vs.
1 Stephenson, G. and L. Wang, Dynamic
simulation of liquefied natural
gas processes, Hydrocarbon Processing, July
2 Alsop, N. and J. M. Ferrer, Avoiding plant
tests with dynamic simulation, Hydrocarbon
Processing, June 2008.
Taylor is a principal consultant with Apex
Optimisation, based in Australia. His responsibilities
include all aspects of APC application design,
implementation and maintenance. In his 20 years
of experience, he has contributed to over 100 APC
applications. Previously, he was employed as a consultant
with Honeywell in South Africa and the UK and with Mobil
in Australia. Mr. Taylor holds a BE degree in engineering
science from the University of Auckland and is a
chartered professional member of Engineers
Jamaludin is a senior process control engineer
at Woodside Energy Ltd. and has 12 years of experience in
the LNG industry. He was previously employed by Petronas
in Malaysia. Mr. Jamaludin has published numerous papers
for technical journals and international industrial
conferences. He contributed to the development of the
first LNG train automatic cool-down
advanced controller, and has led the design and
implementation of multiple APC applications. He holds a
BS degree in chemical engineering from the University of