POWER April 2023 - 37

INSTRUMENTATION & CONTROL
mat, and the overall calibration plan and
results can be captured and preserved
in a fully auditable format, meeting the
compliance requirements of both the
Occupational Safety and Health Administration
(OSHA) 1910 standard and the
International Organization for Standardization
(ISO) 9000 series standard. APM
calibration management interfaces
with EAM systems, allowing for seamless
work-process integration between
maintenance, operations, engineering,
and instrumentation groups.
Neural Networks and Modeling
Controls
During the energy transition, focusing
on greenhouse gas emissions is even
more critical to get the best performance
out of your asset. As a power
plant operator, there are only so many
control screens and systems that can
be watched at once.
One way to help with the split decisions
necessary to operate the plant is
to have neural networks and predictive
modeling aiding in fine tuning aspects
of heat rate and pollution control systems.
With a neural network and specific
customized modeling in a closed-loop
fashion, operators can focus on other operating
issues, while software is optimizing
the controls with specific outcomes
like heat rate, NOx
and CO2
. This helps
the operator move from operating in a
comfort zone to the best zone (Figure 2).
A digital twin can provide many insights
using data, models, and analytics
to your asset. For the plant manager,
having digital modeling or digital twins
of your plant operations can ensure the
optimal performance of the plant using
the technology available to aid your employees'
operations. This is also helpful
for making sure not only is the plant in
environmental compliance, but at a compliance
rate that optimizes heat rate and
pollution limits. For those working at an
enterprise level for plant performance
and controls, being able to compare
plant operations across plants, utilize
best practices, and understand poor performers
is one way to understand where
resources may be needed.
Overview of Instruments and How
to Improve Heat Rate and Operator
Differences
To exemplify the concepts above, we
will look at Autonomous Tuning, a neural
network model digital twin of the combustion
process in a gas turbine. In this
product, this digital twin is used as a
April 2023 | POWER
Case Study: GE Aeroderivative
LM6000PD Peaking Unit
The neural network model digital twin-
based
control
approach
introduced
2. GE Digital's neural network optimizers
include autonomous tuning and BoilerOpt,
which is a boiler combustion software solution.
Courtesy: GE Digital
method to maintain an up-to-date model
of the combustor as operating conditions
change. The digital twin is supported by
real-time data collected from three sources
of instrumentation that already exist
in most power plants: the unit controller,
the continuous emissions monitoring
system (CEMS), and the fuel properties
monitor. The digital twin stores real-time
data from the data sources and uses it
to maintain a machine learning model of
the recent operation of the combustor.
Two support modules enable the
continuous update of the digital twin's
machine learning model. The first is an exploration
module that sends adjustments
to the unit controller to explore the space
of operation of the combustor, and collect
acoustics and emissions data at a wide
range of operating conditions. The second
is a model training module that uses the
data collected with the exploration module
to replace the previous machine learning
model with a newer one.
In the architecture, a gas turbine digital
twin is a software component that
is synchronized with the physical object
it represents and can support numerous
applications. On the top layer of
the architecture, we have the applications
supported by the digital twin. The
digital twin-enabled application is an
optimizer that uses a machine learning
model of the combustion process in a
gas turbine to find optimal adjustments
to the turbine's base schedule that will
allow it to operate with reduced acoustics
and emissions while improving heat
rate, and without the need to manually
remap the unit.
The digital twin uses data extracted
from various devices and instrumentation
in the power plant to maintain an
up-to-date machine learning model of
combustion, which supports an application
that calculates optimal performance
adjustments. Unlike manual mapping
performed during an outage, exploration
to refresh the digital twin is done when
the turbine is generating power.
www.powermag.com
above was evaluated at a peaking combined
cycle powered plant. The plant is
located in an area where the fuel provider
may change the natural gas composition
daily, switching between fuels
with different lower heating value and
specific gravity.
Such a complex scenario had caused
issues with high emissions and acoustics
over the year prior to the autonomous
tuning implementation and had
required the invention of remote tuners
and field service mappers. Remote
tuners were often called to avoid stage
down for high acoustics/blow out and to
address problems with NOx
(at baseload
operations) and CO (at low loads) emissions
that were at the upper end of the
allowed values. There were also issues
with operability risks and high emissions
in addition to the fall and spring mapping
events that were originally scheduled.
The digital twin-enabled optimizer
demonstrated its capability to adapt to
changes in ambient temperature and
fuel properties. The gas turbine operated
for a space of one year without the
need for tuning services or remapping
events. The optimizer was able to drive
the combustor to acceptable conditions
without the need for human intervention.
Digital twins can enhance decision
making in power generation by leveraging
real-time sensors to maintain an up
to-date model of the operation of power
generation equipment, and allowing engineers
to track the evolution of its operating
conditions.
Helpful Tools Are Available Now
Using digital twins with instrumentation
and modeling of assets is how companies
are going to be better at predicting
failures, achieving best performance,
and meeting pollution targets. Utilizing
software for calibration management
and advanced pattern recognition helps
plant managers predict asset failures
prior to the failure occurring. In addition,
utilizing the closed-loop controls
systems, such as autonomous tuning,
is an effective way to help with inconsistencies
in operations and asset performance
in regards to efficiency and
pollution controls. ■
-Jenny Bulach is an Energy Industrial
Principal with GE Digital, and
Martha Saker is Principal Product
Manager with GE Digital.
37
http://www.powermag.com
POWER April 2023

Table of Contents for the Digital Edition of POWER April 2023
