Application of Digital Twin Technology

Application of Digital Twin Technology
on Simulation and Optimization

of Prime Movers in Energy Systems
Vili Panov and Samuel Cruz-Manzo
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2 Modeling and Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.1 Model Zooming Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3 Model-based Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.1 Gas Turbine Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.2 Performance Monitoring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.3 Performance Tracking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.4 Diagnostics (Detection and Isolation) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.5 Prognostics (Prediction and Life Estimation) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.6 Compressor Fouling Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4 Gas Turbine Faults and Degradation Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.1 Faults and Degradation Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
4.2 Compressor Turbine Degradation Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
5 Integrated Control and Health Monitoring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
5.1 Accommodation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
5.2 Life Extension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
5.3 Model-Based Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
5.4 Soft Sensor Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
6 Implementation and Deployment of Digital Twins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
6.1 Performance Tracking Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
V. Panov ()
School of Engineering, University of Lincoln, Lincoln, UK
Siemens Energy Industrial Turbomachinery, Siemens Energy, Lincoln, UK
e-mail: vpanov@lincoln.ac.uk; vili.panov@siemens-energy.com
S. Cruz-Manzo
School of Engineering, University of Lincoln, Lincoln, UK
e-mail: scruzmanzo@lincoln.ac.uk
© Springer Nature Switzerland AG 2022 1

M. Fathi et al. (eds.), Handbook of Smart Energy Systems,
https://doi.org/10.1007/978-3-030-72322-4_153-1
2 V. Panov and S. Cruz-Manzo
Abstract
The advent of digital twin technology enabled next steps in the evolution of
the current state-of-the-art optimization solutions for prime movers in energy
systems. Conventional offerings for operational optimization of prime movers
were predominantly based on non-real-time and off-line solutions. Emerging
digital twin technologies facilitated the next generation of simulation and opti-
mization techniques, which in return enabled advanced utilization of engineering
assets. These solutions exploit online functionalities which are distributed across
the whole IoT chain consisting of Embedded, Edge, and Cloud computational
platforms. In this contribution, we present various simulation and optimization
techniques applicable to gas turbine systems. Gas turbines play a vital role in
energy production as prime movers and they are extensively utilized as the
workhorse within the power generation sector.
Keywords
Digital twin · Simulation · Model-based optimization · Monitoring ·

Diagnostics · Prognostics · Tracking · Control · Gas turbines
1 Introduction
The proliferation of the Digital Twin concept in industrial environments has led
to the development of several variants of this technology applicable to almost the
entire product life cycle (Zhong et al. 2017; Rasheed et al. 2019). Traditionally
we can recognize three main types of Digital Twins, namely: Product, Production,
and Performance Digital Twin. Product Digital Twin is used to support the efficient
design of new products by providing virtual-physical connection which enables
performance analysis of complex multi-physics domains. Production Digital Twin
supports an optimized manufacturing process and enables more efficient production
planning. Combining data generated from Product and Production Digital Twins, a
business can devise efficient production methodologies. Performance Digital Twins
capitalize on data obtained from products in operation and make this analysis
available to activities such as asset management and operational optimization
(Zaccaria et al. 2018).
The various engine health-monitoring systems of today provide a basic level
of monitoring and asset management. Their capabilities are relatively limited, and
usually, they are a collection of separated, unrelated technologies. The information
that they provide is used mostly to initiate maintenance actions, but not for
online decision-making in real time. While these traditional control and diagnostics
techniques are reliable, they are not optimal, and new advanced techniques provide
the promise to meet the challenging requirements of increased reliability, improved
efficiency, and extended operational life. Using a Real-Time Executable Digital
Twins (RT xDT) to meet the challenging control and diagnostics requirements
Application of Digital Twin Technology on Simulation and Optimization of. . . 3
- Performance optimization
Physical Asset - Trip Prevention Digital Twin
Configuration
Asset Performance Digital Twin

Digital Twin RMS
Utilization - Tracking Analysis
- Diagnostics
Conditions - Prognostics
Based
Maintenance
Recommendations relayed back
to the customer
Customer Operations Centre Review

Alternative scenarios played out
Fig. 1 Digital Twin platform for integrated model-based control and prognostics and health
management
has established itself as the most viable approach. This emerging Digital Twin
technology can provide a unified framework for advanced model-based control and
diagnostics functionalities (Fig. 1).
The Digital Twin functionalities can be distributed across different computational
platforms such as Embedded, Edge, and Cloud platforms, depending on operator
needs related to real-time operation and requirements associated with the processing
power of available platforms. The Digital Twin, virtual and physical systems, are
closely integrated through a concept of multiple network levels: Engine network,
Plant network, and Fleet network, to provide physical systems with new capabilities
(Fig. 2). In this way, system capabilities are expandable with non-real-time func-
tionalities which are encapsulated within software solutions deployable on various
IoT platforms, which can support hosting of agents dedicated to fleet and asset data
analytics (Panov and Cruz-Manzo 2020).
2 Modeling and Simulation
Generally speaking, mathematical models are a description of a system or process

using mathematical concepts, and they are an essential part of simulation and design
of engineering systems. Typically, mathematical models are categorized based on
their structure. Therefore, classification of mathematical models would distinguish
between Static vs Dynamic, Linear vs Nonlinear, Physic-based vs Data-driven, and
Continues vs Discrete modeling (Fig. 3).
The data driven models, for example, artificial neural network models, give
relationship between input and output variables that can be obtained on the basis
Fig. 2 Integration of digital twin platforms at multiple network levels
Fig. 3 Classification of mathematical models
of available real data without the need of gas turbine system knowledge. This can
be seen as an advantage over physics-based models, because this approach has a
theoretical possibility to exclude model inaccuracy from the diagnostic process.
A disadvantage of data-driven approach is the reduced availability of data that
can be used to form a representative fault classification because of the occasional

occurrence of faults and on the other hand high cost of real fault simulation on a test
bed.
The physics-based models, for example, thermodynamic models, require detailed
knowledge of the gas turbine and generally present more or less complex software.
Physical models usually consist of a set of different component models. The
behavior of single components is described by physical equations or by using
component characteristics, which are obtained using rig tests or various prediction
tools. The main drawback of physics-based modeling approach is that model
inaccuracy can cause elevated errors in the estimation of fault parameters.
Engine models used in gas turbine diagnostics fulfill two main purposes.
Firstly, they are used to determine performance baseline in order to calculate
differences between measurements and such a baseline. Secondly, they are used for
obtaining fault signatures, which represent different engine faults and degradation
mechanisms.
We can also recognize two different modeling approaches in gas turbine diag-
nostics. The goal of a system identification technique is to determine gas turbine
fault parameters that minimize difference between measured engine variables and
model-based variables usually obtained by physics-based models. The second
approach is based on the pattern recognition technique and mostly uses data-driven
models. Once when the fault pattern is available, a data-driven recognition tool is
usually trained without detailed knowledge of the gas turbine system, and used for
diagnostic purposes.
Gas turbine thermodynamic models are predominantly physics-based models,
and typically they are extensively used for the simulation of steady-state and tran-
sient behavior of gas turbine engines. Over the years, steady-state thermodynamic
models established themselves as a main tool in gas turbine gas path analysis.
Mathematically, these are nonlinear steady-state models , represented by the system
of algebraic equations reflecting mass, heat, and energy balance for all components
operating under stationary conditions. Steady-state thermodynamic model computes
vector y of gas path monitored variables, as a function of steady-state variables x,
control variables u, and operating conditions v:
y = g (x, u, v) (1)
To describe various types of gas turbine deterioration and faults such as fouling,
tip rubs, erosion, etc., this set of equations could be augmented with an additional
state vector h containing health or fault parameters characterizing different engine
degradation modes:
y = g (x, h, u, v) (2)
For a given steady-state operating point, the above model can be reduced to:
y = g(h) (3)
The simplified model can be obtained by linearization of nonlinear dependency

between the gas path y and health h parameters determined for a fixed steady-state
operating condition:
δy = H δh (4)
This linear steady-state model connects a vector of δh small changes of the health
or fault parameters with a vector δy of the corresponding relative deviations of the
monitored variables by influence matrix H.
Although most of the diagnostics methods are developed at a steady state, the
current trend demonstrates increasing interest in the diagnostics during transient
operations (Loboda et al. 2006). Dynamic physics-based models are used to describe
the behavior of engines under transient conditions. A nonlinear dynamic model is
described with the following generalized system of equations:
ẋ(t) = fx (x(t), u(t), v(t), t) (5)
y(t) = g (x(t), u(t), v(t), t) (6)
where x and u stand for the state and the input variables, respectively, and t is the
time. In steady-state conditions, the left-hand side in the above equation (Eq. 5)
is zero, and it degenerates into an algebraic equation or system of equations.
The second equation (Eq. 6) is the output equation that describes system output
variables y.
Many existing model-based methods are designed with the assumption that the
system exhibits linear behavior in the neighbor of a steady-state operating point, and
therefore, linearization-based methods are used. By linearizing the engine model
(Eqs. 5 and 6) around a nominal operating condition, that is, x0 and u0 , the following
linear dynamic model is obtained:
ẋ(t) = A(t)x(t) + B(t)u(t) (7)
y(t) = C(t)x(t) + D(t)u(t) (8)
where A(t), C(t) and B(t), D(t) are state and input distribution matrices, respectively,
for the linear state and output equations.
The above equations (Eqs. 7 and 8) represent a linear time-varying model, and
in this general form, system matrices are functions of time. Further simplifications
in the gas turbine model could be introduced by assuming that the system parame
ters are constant in time. In that case, the system model simplifies to linear time-
invariant model, which can be described with the following set of equations:
ẋ(t) = Ax(t) + Bu(t) (9)

y(t) = Cx(t) + Du(t) (10)
where elements of system matrices A, B, C, and D can be treated as constants.

Approaches that involve linearizing engine dynamics around steady-state condi-
tions and subsequent blending of parameters and controllers for selected operating
points can lead to a rather complicated procedure. In addition, when the effects of
various faults are taken into account, modeling complexity could be significantly
increased, which makes the design of model-based techniques and their real-
time implementation very difficult. Therefore, simulation models used for control
and diagnostic purposes can benefit significantly from modeling methods that are
directly based on intrinsic nonlinearities of the engine dynamics.
The accurate modeling of engine behavior is a complex task, because of the
large number of effects that have to be taken into account (Fig. 4). For example,
if compressor and combustor stability effects are considered by the simulation
tool, high-frequency gas volume dynamics should be included in the model. These
high dynamic effects are in the range of 1 kHz and are usually neglected by the
majority of dynamic models. On the other hand, heat transfer effects and shaft inertia
associated with the low-frequency dynamics are crucial for the dynamic behavior of
the engine and control system, and are often considered by gas turbine transient
models. When gas dynamics is not included in modeling architecture, the range of
model sampling frequency reduces typically to 10 Hz.
Fig. 4 Dynamic phenomena

Predominantly, dynamic models have a component-oriented architecture where

an engine model is composed of modules representing individual components
(Schobeiri et al. 1994; Visser and Broomhead 2000; Camporeale et al. 2006; Panov
2009). The different component models are usually connected via conservation laws
for mass, momentum, and energy. The component models include conservation of
mechanical energy for engine shafts, heat-soaking effects for metal parts (blades,
discs, and casing), and conservation of thermodynamic energy within different gas
volume in the engine. This modeling approach generally leads to a set of nonlinear
differential equations, which can be solved by appropriate numerical methods such
as integration algorithms.
The detailed dynamic model of gas turbine engine can be expressed with a system
of nonlinear differential equations in state space:
ẋ = fx (x, u, v) (11)
ym = gm (x, u, v) (12)
yn = gn (x, u, v) (13)
where x is the state coordinate vector, u is the control vector, and v is the vector
of operating conditions. Vector ym contains measurable observable parameters and
vector yn non-measurable parameters.
2.1 Model Zooming Example
Using examples of 0-D and 1-D compressors models, this section presents the
concept of modeling zooming for gas turbine engines. Multistage axial compressors
are used in industrial gas turbines and comprise a series of stages. The pressure of
the air is increased across the different stages of the axial compressor. Each stage
comprises a set of rotor blades followed by a set of stator blades. The rotor blades
increase the kinetic energy of the air, and the stator blades transform the kinetic
energy of air into static pressure. Correct diffusion of air across the different stages
of the axial compressor can be achieved by the variation of the angles of the stator
guide vanes (VSGV) (Cruz-Manzo et al. 2019).
A 0-D compressor model or lumped axial compressor model considers perfor-
mance maps for flow and efficiency as a function of the pressure ratio and rotational
speed. The angle of the VSGVs to change the geometry of the compressor and
change the overall compressor maps can also be considered in the 0-D compressor
modeling architecture. The inputs of a 0-D compressor model are the rotational
speed and the pressure ratio. These 0-D models are typically used for stability
optimization of steady-state and transient operation of compressors.
The simulation results shown in Fig. 5 represent the overall performance of an
axial compressor during load acceptance in a twin-shaft gas turbine. The results
Fig. 5 Representation of
compressor transient response
on compressor performance
map for 13–40% load and
26–53% load with bleed air.
(Figure published in
Cruz-Manzo et al. 2020, page
10. Figure published under
license of Creative Commons
Attribution 4.0 License (CC
BY))
shown in Fig. 5 were obtained by considering a 0-D compressor model within a 0-D
gas turbine model. The steady-state and transient responses of the compressor can
be represented in the compressor map as shown in Fig. 5. The compressor discharge
pressure and flow increase with increasing the load of the gas turbine. The influence
of compressor air bleed on compressor running line is also typically considered by
0-D compressor models.
To obtain a better understanding of stagewise performance of multi-axial com-
pressors, a more detailed modeling is required, and that is usually provided by 1-D
compressor models. A 1-D compressor model simulates the increase in pressure
and temperature across the different stages of the axial compressor. The increase in
temperature and pressure across the different compressor stages can be calculated
from trigonometric relations of velocity triangles and thermodynamic principles.
The 1-D compressor model requires stage-wise performance maps which can be
estimated from a standard streamline curvature through flow analysis code that
computes the aerodynamic parameters across the compressor stages. The inputs
of the 1-D compressor model are the flow, efficiency, and rotational speed. The
variation of VSGV angles, to change the compressor geometry and stagewise
performance maps, is also typically considered in the 1-D compressor model.
Figure 6 shows the simulation of the transient response of the axial compressor
represented in stage performance maps. Typically, stage performance maps are
represented with stage load coefficient in function of flow coefficient . The
simulation results represent the performance of the second and eleventh stage of
an axial compressor. The stagewise simulation results shown in Fig. 6 correspond
to the simulation of the gas turbine load acceptance maneuver shown in previous
Fig. 5. Simulations performed by 1-D models can provide more insight into the
stability of individual stages and therefore their cumulative contribution to the
overall performance of compressor components. The provided example showcase
that the transient compressor response, when the engine load is increased from 13%
to 40% of the maximum load, is moving more toward the stability limits of the stage
characteristic maps compared to the transient response for the case at 26–53% load.
Fig. 6 Representation of stage performance map. The running line (RL) at different rotational
speeds is represented. (Figure published in Cruz-Manzo et al. 2020, page 16. Figure published
under license of Creative Commons Attribution 4.0 License (CC BY)
3 Model-based Techniques
Prime movers’ operational requirements continue to become more and more

demanding in response to the need for extended component life, increased reli-
ability, and improved overall efficiency of assets. To support the requirements
of emerging smart energy systems, novel model-based control and diagnostics
concepts have been introduced.
Model-based techniques exploit engine models to estimate gas turbine internal
conditions, enabling in that way implementation of advanced control and diagnostic
methods. Model-based information is the foundation of many diagnostics and
control strategies, ranging from simple threshold to sophisticated pattern recognition
methods.
State-of-art degradation monitoring, fault detection, and isolation tools vary

widely in their complexity and applications and are primarily built upon both model-
based and sensor-based analyses. Model-based monitoring techniques exploit
engine models to estimate asset internal conditions, enabling in that way imple-
mentation of various diagnostic methods.
Among other intended uses for such a model, is also to enable online tracking of
engine performance changes and engine parameter synthesis for fault detection and
accommodation. Model-based tracking methodology offers a means to compensate
engine to engine variations, and furthermore aligns the model to a particular engine
being monitored to ensure accurate performance tracking while engine performance
deteriorates with time.
Traditionally, a gas turbine control system transforms real engine limits, into
limits that are based on measured engine variables. As a result of that, engines
operate with increased safety margins and thus with non-optimal performance. To
overcome this problem model-based control concepts have been proposed. Model-
based control approach exploits engine models to estimate control feedback signals,
enabling the implementation of novel control methods.
Model-based diagnostics employs engine models tuned to match the observed
engine state in the same manner as a model-based control. The residual deviations
between predicted and sensed parameters are modeled, again usually as variations
in component losses and flow capacity, and the best match is used to identify likely
component degradation modes and faults.
The use of model-based techniques to diagnose and adaptively manage degrada-
tion of engine component characteristics is crucial for the operational effectiveness
of gas turbines. Model-based prognostic approach, therefore, is increasingly favored
due to its capability to provide a more accurate estimation of remaining useful life.
This approach is naturally extensible to the physics of failure, and hence it has been
considered to be a promising enhancement of the conventional prognostic tech-
nology. The potential payoffs of prognosis-based asset management are increased
component life entitlement, improved time in service, better asset utilization, and
lowered engine maintenance costs.
This section gives an overview of current and evolving model-based techniques
and discusses the benefits of these concepts in operational management of the gas
turbine assets.
3.1 Gas Turbine Systems
The gas turbine systems are typically represented with component-based modeling
tools. The component models include conservation of mechanical energy for engine
shafts, heat-soaking effects for metal parts, and conservation of thermodynamic
energy within different gas volumes in the engine. The detailed dynamics model
of a gas turbine engine can be expressed with a system of nonlinear differential
equations in state space:
ẋ = fx (x, u, v) (14)
y = gy (x, u, v) (15)
where x is the state coordinate vector, u is the control vector, v is the vector of
operating conditions, and vector y contains measurable ym and non-measurable
parameters yn (Fig. 7).
The dynamic gas turbine models are usually devised as physics-based models,
and they have module-oriented architecture, where each module represents an
individual component. Different engine configurations could be created using
generic simulation tools, and subsequently, these models could be used for non-
real-time and real-time simulations.
To be able to carry out dynamic simulation of the whole system, a gas turbine
model is complemented with a control system module that contains various control
loops and corresponding set points (Fig. 8). Typically, a gas turbine control system
besides set-point controllers contains also transient and limit protection controllers.
3.2 Performance Monitoring
Performance monitoring relies on engine measurements to infer the evolution of

degradation variables. Measurement z deviations from expected, that is, nominal
Fig. 7 Generic system

inputs, outputs, and states
Fig. 8 Gas turbine system

Fig. 9 Performance monitoring
measurements y represented by the nominal model, reveal problems in gas turbine

components (Fig. 9).
As a gas turbine engine undergoes internal changes, these changes may be
manifested in performance degradation. To account for this degradation, original
state and output equation could be augmented with an additional vector h containing
health parameters:
ẋ = fx (x, h, u, v) (16)
ym = gm (x, h, u, v) (17)
yn = gn (x, h, u, v) (18)
where ym and yn are output vectors of measured and non-measured variables.

The vector h contains health parameters that indicate the engine health condi-
tions. Health parameters are usually represented by efficiencies and flow capacities
of the engine components. As they deviate from their normal health conditions, the
performance delivered by each component degrades and this can be recognized as a
performance shift of “new and clean” component characteristics (Razak 2007).
3.3 Performance Tracking
Since the gas turbine model represents a “nominal” engine, it must be adapted
or tuned to the performance of the real engine as it deviates from the nominal
Fig. 10 Performance tracking
baseline with time. To mitigate this problem, tuning of the engine model can be
performed so that model aligns to the actual engine being monitored using a model-
based tracking approach (Fig. 10). The idea behind this approach is to minimize
the deviations or modeling errors of the engine model that runs in parallel to a gas
turbine, by correcting parameters of the “nominal” engine’s behavior. This approach
effectively removes the uncertainty introduced with engine-to-engine variability,
different disturbances, unknown initial conditions, and modeling simplifications.
The tracking methods are usually based on the errors between the measured
engine variables and the corresponding simulated variables, where generated errors
are used to correct engine component characteristics. The measured variables z are
compared with simulated variables ym , and then the corresponding simulation error
vector yerr is multiplied by a matrix K, where the resulting vector is used to correct
the model state variables x and health parameters h. Therefore, a dynamic model
described with the system of equations Eqs. 16, 17, and 18 expanded with tracking
filter takes the following form:

x̂˙ = fx x̂, ĥ, u, v + K z − ŷm (19)

ŷm = gm x̂, ĥ, u, v (20)

ŷn = gn x̂, ĥ, u, v (21)
where function K represents the gain matrix, which determines model correction and
sensitivity to measurement and modeling process noise. Vectors x̂ and ĥ represent
the estimates of the state variables and health parameters, and vectors ŷm and
ŷn represent the estimates of the measurable and non-measurable model outputs,
respectively. Gain matrix K can be designed using linear quadratic theory to form
the Kalman filter gain matrix (Dewallef and Leonard 2003; Tagashira et al. 2009).
3.4 Diagnostics (Detection and Isolation)
Fault detection and isolation play a critical role in enhancing engine reliability and
reducing the operating cost of gas turbine engines. Engine component degradation
and faults may occur in various degrees of severity and at various locations, and
numerous scenarios are possible. We can distinguish three general classes of engine
faults, namely, sensor, actuator, and gas turbine component faults.
The detection process evaluates the residuals between measurements and esti-
mates, and subsequently continues with monitoring if a fault has occurred. The
approach commonly used for model-based diagnostics (Fig. 11) is composed of
two steps (Jaw and Wang 2006). The first step consists of the generation of residual
signals from measurements and their nominal values:
(t) = z(t) − ym (t) (22)
followed by calculation of normalized relative deviation Δz for measured variables

from their nominal values:
(t)
Δz(t) = (23)
σym (t)
where standard deviation σ accounts for measurement uncertainty, and it is cal-

culated based on the assumption of Gaussian distribution for measurement noises.
The second step contains a comparison of residuals with thresholds to make fault
detection decisions:
z(t) > γ (24)
Most gas turbine diagnostics tools are based on engine steady-state measurements,
because during the engine life most of the time the engine will operate at steady-
state conditions. However, significant diagnostic content can be found in transient
operation of the engine, and hence transient gas turbine performance deterioration
and diagnosis were analyzed by many researchers (Merrington 1988, 1993; Meher-
Homji and Bhargava 1992; Bird and Schwartz 1994).
In order to diagnose gas turbine faults under transient conditions, the above
equation for normalized relative deviation (Eq. 25) could be transformed into the
following form (Li 2003):
Fig. 11 Diagnostics
Fig. 12 Isolation
t
t0 [z(t) − ym (t)] dt
z(t) = t (25)
σ t0 ym (t)dt
Once a fault is successfully detected, the next step is to isolate the particular
fault from other potential faults (Fig. 12). The goal of fault isolation is to identify
which degradation state variable, that is, health parameter has exceeded the alarm
threshold:
ĥ(t) > ε (26)

Fig. 13 Prediction and life estimation
3.5 Prognostics (Prediction and Life Estimation)
Prognostics holds a central place in prognosis-based asset management, and it

has the potential to contribute toward reducing the operating cost of running gas
turbines. Condition-based component lifting and inspection, instead of operating to
fixed intervals, is based on engine health, where component state\level of damage
and asset profitability dictate when an optimal inspection should take place (Cerri
et al. 2008).
Prognosis is the ability to predict the remaining useful life of a component based
on the engine performance (Fig. 13). The main task of prognostics is to predict how
performance degradation will deteriorate to an unacceptable level, which can be
summarized as follows:
z (t + d) − ym (t + d)
Δz (t + d) = >κ (27)
σym (t + d)
ĥ (t + d) > μ (28)
where κ is the acceptable performance limit for useful service/maintenance, μ is

the corresponding state variable, that is, health parameter degradation limit, and d is
time to failure or remaining useful life of the component.
It is common practice to base damage calculation on a fleet average loading cycle
mix. Using a model-based prognostic technique, it is possible to perform these
calculations for each engine, taking into account engine to engine variation and
specific engine applications. Damage assessment could be done for selected critical
components and could account for different failure mechanisms. This proactive
approach can be used to perform engine-specific schedule maintenance and logistic

activities.
Broadly speaking, we can recognize several types of maintenance and logistic
activities. Reactive maintenance is an event-based concept where system repair is
performed when problems arise. The periodic inspection and maintenance activities
are time based, and they are part of preventative maintenance. The maintenance
based on estimated usage of the system is usually referred to as scheduled
maintenance. Predictive maintenance is based on condition or prognostic evidence
for maintenance activities.
Following the current practice of preventative and scheduled maintenance,
engine components are retired with useful remaining life. Therefore, by introduc-
ing prognosis-based management, components can realize their life entitlement,
because in this case the component life prognostics is based on actual engine
performance. This approach enables a shift from the current practice, that is, fixed
predetermined maintenance intervals, and leads toward lower engine operational
costs (Hindle et al. 2006).
3.6 Compressor Fouling Example
In this section, an example of performance deterioration detection for compressor

fouling degradation model is presented and discussed. Fouling in a compressor
reduces the performance of a gas turbine and is attributed to airborne particles (dust,
mineral particles, etc.) attached to the blades. The surface roughness of the blades
is increased by fouling particles. The angle of air diffusing through the compressor
blade is also changed by fouling particles. During compressor fouling conditions a
reduction in airflow and efficiency is expected (Tarabrin et al. 2008; Maleki et al.
2018).
Figure 14 shows measured data from a twin-shaft gas turbine during compressor
fouling conditions. The twin-shaft gas turbine was operated as a mechanical drive
in low-power condition and as a power generator at high-power condition. The gas
turbine was continuously operated at a steady state during low and high-power
conditions and from a clean compressor condition to a highly fouled condition.
Figure 14a shows the measured compressor discharge pressure when a gas turbine
is operated as a power generator and from a clean condition at t = 0 h to a fouling
condition at t = 1152 h. Figure 14b shows the measured compressor discharge
pressure when the gas turbine is operated as a mechanical driver and from a clean
condition at t = 0 h to a fouling condition at t = 1944 h.
The performance of a gas turbine at compressor fouling conditions can be
simulated with a gas turbine model containing compressor characteristic maps.
Figure 14a and b also show the simulated compressor discharge pressure generated
by a gas turbine model comprising the compressor characteristic that takes into
account compressor health parameters.
The airflow and efficiency of the compressor were not measured during the
operation of the twin-shaft gas turbine. These parameters can provide relevant
information about faults and degradation mechanisms of the compressor; hence
the airflow and efficiency can be considered as “health parameters.” A gas turbine
model can calculate health parameters and can assist the development of prognostic
algorithms to predict the evolution of the gas turbine performance at compressor
fouling conditions during a longer period. Based on the simulation results shown
in Fig. 14, a reduction in airflow and efficiency of the compressor during the gas
turbine operation at fouling conditions was estimated by the gas turbine model as
observed in Fig. 15. Figure 15a and b demonstrate that a more intense level of
Fig. 14 Measured and simulated compressor discharge pressure during fouling conditions: (a)
power generator; (b) mechanical driver. (Figure published in Cruz-Manzo et al. 2018a, page 11.
Figure published under license of Creative Commons Attribution 4.0 License (CC BY))
Fig. 15 Estimation of dynamic reduction of health parameters during fouling conditions (a)
compressor discharge airflow; (b) compressor efficiency. (Figure published in Cruz-Manzo et al.
2018a, page 16. Figure published under license of Creative Commons Attribution 4.0 License
(CC BY))
fouling was present when the gas turbine was operated as a power generator under
high-power condition.
4 Gas Turbine Faults and Degradation Modes
The physical faults in a gas turbine engine include problems such as erosion,
corrosion, fouling, foreign object damage (FOD), worn seals, burned or bowed
blades, etc., (Fig. 16). These physical faults can occur individually or in combination
Fig. 16 Gas turbine faults and degradation modes
and cause changes in the performance of the different gas turbine components.
These changes in the performance of the gas turbine components result in changes
in the gas path measurement parameters, which are typically used for diagnostic
purposes (Li 2002).
Typically, any abnormality in performance is preceded with a precursor event,
which is the earliest detectable sign of performance deterioration. Usually, abnor-
mality is defined as an unknown or not seen before performance deviation of
operating condition which deviates from normal or expected operating conditions.
On the other hand, known or repeatable abnormal deviations from nominal
operating conditions are generally recognized as faults. These faults could be
induced by usage degradation (natural deterioration), or they can be caused by
external factors that initiated abrupt or accelerated rate of performance degradation.
Therefore, the whole purpose of the diagnostic process is to timely capture
these various degradation modes and faults which could lead to failure where the
performance of a product has degraded to the point so that can no longer provide
intended functionality, or even to total breakdown which can cause a catastrophic
scenario.
Engine performance is represented by a set of so-called health parameters.
These health parameters deviate from initially healthy baseline values as the engine
components degrade. Estimation of health parameters from engine data is often
referred to as gas path analysis (Urban 1972, 1974). The use of weighted-least-
square estimation (Doel 1992, 1993) and Kalman filters (Kobayashi and Simon
2003; Kobayashi et al. 2005) are widely used for gas path analysis (GPA). More
recently various techniques such as Neural Networks (Ogaji et al. 2003), Bayesian
Belief Networks (Lee et al. 2010), Genetic Algorithms (Sampath et al. 2003),
Fig. 17 Gas turbine performance deviation
Polynomial Functions (Cerri et al. 2011), and different hybrid methods (Volponi
2005; Volponi et al. 2007) have been explored for use in performance fault diagnosis
and tracking.
Generally speaking, we can recognize two main reasons for engine performance
deviation: engine-to-engine variations and engine deterioration (Fig. 17). The
engine-to-engine variation accounts for manufacturing and assembly variation
found in new engines, and it is based on engine parameter measurements of a sample
of the engine population. This engine-to-engine variation adds typically a normally
distributed random value to the nominal value of each measured engine parameter.
On the other hand, over the lifetime of operation, engine components will also
undergo some amount of deterioration. This engine degradation may be gradual or
abrupt. Some of the known deterioration mechanisms are seal and secondary flow
leaks, clearance increases, erosion, and fouling. While engine health deterioration
is a normal aging process that occurs in all engines as a result of usage, abrupt
abnormal events such as foreign object damage are not predictable as they happen
unexpectedly.
4.1 Faults and Degradation Classification
Model-based diagnostics employ engine models tuned to match the observed engine
state in the same manner as model-based performance tracking (Panov 2013). The
residual deviations between predicted and estimated health parameters are modeled,
Fig. 18 Gas turbine faults classes
again usually as variations in component losses and flow capacity, and the best
match is used to identify likely component degradation modes and faults.
A number of causes can result in gas turbine performance deterioration. The goal
of diagnostics is to attempt to detect one or more of these causes that are responsible
for the deterioration of engine performance. Usually, this detection process is based
on monitoring of so-called “health indices.”
Health indices are means of determining the deteriorated component character-
istics. They represent the percentage change in component characteristics usually
due to component faults or gradual degradation. Typically, two health indices can
be defined for any component and they correspond to the capacity and efficiency
index:
Γξ −Γh ΔΓ
IΓ = 100 × = 100 × h (29)
Γh Γ
ηξ − ηh Δη
Iη = 100 × = 100 × h (30)
ηh η
Fault detection and isolation play a critical role in enhancing engine reliability
and reducing the operating cost of gas turbine engines. Engine component degrada-
tion and faults may occur in various degrees of severity and at various locations, and
numerous scenarios are possible (Fig. 18). We can distinguish three general classes
of engine faults, namely, sensor, actuator, and gas turbine component faults.
Fig. 19 Gas turbine fault classification
In reality, there is a very wide range of different engine faults. Classification

of faults is usually based on different criteria, and generally, they can be divided
into single and multiple faults. It would be ideal to address all these faults (sensor,
actuator, and component faults) under one unified diagnostic framework, and several
researchers have investigated the development of such a diagnostic framework
(Dewallef and Leonard 2003; Volponi et al. 2003; Surrender and Ganguli 2004).
In practice, for the analysis of the engine degradation, engine faults have to be
divided into a limited number of fault classes (Fig. 19). Typically, it is considered
that every fault class corresponds to one engine component. For example, compres-
sor fouling and turbine erosion are considered as two different component fault
classes. Compressor fouling fault class is detectable with decreased compressor
efficiency and flow capacity, and erosion of turbine components can be detected
as an increase in turbine flow capacity and decrease in turbine efficiency.
To isolate the diagnostic information, a classifier is added to the model-based
detection process, and numerous techniques have been applied in the past as a
classification engine (Li 2002). The detection process consists of a preliminary
variable selection and computation of variable residuals. Subsequently, further
features such as health parameters are calculated by using previously determined
variable residuals. And finally, after the feature selection process, the determined
features are subjected to classifications, where the classification engine isolates the
final fault state (Panov 2015).
4.2 Compressor Turbine Degradation Example
In the following example, the gas turbine degradation mode that corresponds to a
compressor turbine component is presented. Compressor turbine erosion mode, has
been diagnosed and isolated by comparison of typical turbine health vector against
a bank of fault signatures for different gas-path components.
The compressor-turbine or gas generator turbine in a twin-shaft industrial gas
turbine drives the compressor through a mechanical shaft. Cooling on the trailing
edge of the turbine blades is usually difficult to achieve. This can result in loss of
trailing edge material in the turbine blade. An increase in the flow capacity and a
decrease in efficiency in the compressor-turbine can be a consequence of a hot-end
blade damage incident (Cruz-Manzo et al. 2018b). Figure 20 shows the measured
power output of a twin-shaft gas turbine during a hot-end blade damage incident in
the compressor-turbine. Four hot-end blade damage incidents were identified in the
set of measured data, but the first and last incidents are shown in Fig. 20. Incident
1 corresponds to the earliest compressor-turbine damage condition and Incident
4 is the latest compressor-turbine damage condition as the compressor-turbine is
degraded with time. A reduction in the measured power output from an initial
steady-state condition to the steady-state condition resulting from a hot-end damage
condition is observed in Fig. 20. Incident 4 presents a small % variation (Delta) of
the power output which can be attributed to degradation in the compressor-turbine
material.
A thermodynamic model of a twin-shaft gas turbine can simulate the physical
parameters such as pressure, temperature, and flow rate of the gases entering and
leaving the different stations of the gas turbine. During a hot-end blade damage
incident, a change in the flow capacity and efficiency from the compressor-turbine
is expected. Therefore, by the implementation of non-dimensional coefficients that
Fig. 20 Normalized measured power output during Incidents (hot-end blade damage) 1 and 4.
Incident 4 represents a higher degradation in the compressor-turbine blade. (Figure published
in Cruz-Manzo et al. 2018a, page 19. Figure published under license of Creative Commons
Attribution 4.0 License (CC BY)
Fig. 21 Measured and simulated compressor discharge pressures during a hot-end damage
condition in the compressor-turbine: (a) Incident 1; (b) Incident 4. (Figure published in Cruz-
Manzo et al. 2018a, page 24. Figure published under license of Creative Commons Attribution 4.0
License (CC BY))
correspond to the flow and efficiency of the compressor-turbine performance of

different stations of the gas turbine during a hot-end blade damage incident can
be simulated.
Figure 21 shows the compressor discharge pressure measured during gas turbine
operation and during the hot-end blade damage incidents in the compressor-turbine.
The measured compressor discharge pressure shown in Fig. 21 corresponds to the
measured power output of the twin-shaft gas turbine previously shown in Fig. 20.
Figure 21 also shows the simulated compressor discharge pressure generated by
the gas turbine model and considers the degraded turbine health parameters. A
reduction in the compressor discharge pressure and a reduction in the pressure in
the gas turbine system during a hot-end blade damage incident can be related to the
loss of trailing edge material in the turbine blade which in turn increases the flow
capacity in the compressor-turbine.
Fig. 22 % Delta for flow capacity and efficiency in compressor-turbine during a hot-end blade
damage incident in the compressor-turbine. Figure published in Cruz-Manzo et al. 2018a, page 27.
Figure published under license of Creative Commons Attribution 4.0 License (CC BY)
Figure 22 shows the variation (% Delta) of the flow capacity and efficiency of
the compressor-turbine. The flow capacity and efficiency (health parameters) are
calculated by the gas turbine simulation model.
Figure 22 also shows the flow capacity and efficiency of the compressor-turbine
calculated during the four hot-end blade damage incidents and during gas turbine
operation. The calculated health parameters shown in Fig. 22 correspond to the
measured data of the gas turbine shown in Figs. 20 and 21. The last incident or
fourth incident presents the smallest absolute value of % Delta flow capacity and %
Delta efficiency. The fourth incident also presents a small value of % Delta power
output compared to the first incident as shown in Fig. 20. The small absolute value
of % Delta flow capacity, efficiency, and power output in Incident 4 can be attributed
to irreversible degradation mechanisms in the compressor-turbine as the gas turbine
continues operating after each hot-end blade damage incident.
5 Integrated Control and Health Monitoring
Adaptive model-based control with integrated model-based diagnostics, isolation,

and accommodation plays a significant role in the operational improvement of gas
turbine engines. These systems can diagnose and adaptively manage the degradation
of engine components while taking into account engine-to-engine variation and
current operating conditions (Fig. 23). The model-based control strategies include
methods for optimization of operational limits such as compressor and combustor
stability, turbine life, and emissions (Panov 2016). In addition, life-extending control
Fig. 23 Integration of model-based control with prognostics and health management system
can contribute toward the reduction of operational costs, taking into consideration
the impact on component life usage in ways that trade minor performance degrada-
tion, for significant component life extension.
Although conventional control strategies have certain advantages due to their
simplicity (easily understood and verified), they are also conservative and therefore
assets operate with increased safety margins using approximate operability limits.
Alternatively, model-based control strategies offer good accommodation of
engine-to-engine variation and ambient and fuel variability. They are also robust,
that is, easily adaptable and expandable, and moreover optimal with direct control
of stability limits. Model-based control strategies, therefore, can offer explicit
accommodation for engine deterioration.
There are a number of technical challenges that must be met in order for model-
based concepts to be effectively used for operational improvement of gas turbine
engines. Development and implementation of online damage models is prerequisite
for robust prognostic and damage prevention strategies. More accurate gas turbine
models that capture the dynamics of interest are required to guide the development
of both, diagnostic and control methods, and on the other hand, adequate dynamic
models are needed to enhance the performance of the engine under adaptive control
while maintaining stability and operability margins.
5.1 Accommodation
The goal of a fault accommodation system is to regain operability and maintain

stability after a fault has been detected and isolated. The output of the diagnostic
system triggers the accommodation, that is, actuator control adjustments, to achieve
stability margins in a faulted engine that are equal to or greater than the pre-fault
values (Fig. 24).
Applied accommodation strategies can vary from simple trimming of control set-
tings and schedules to complex adaptive adjustments of component characteristics
to achieve required stability margins. A more sophisticated fault accommodation
method utilizes analytical redundancy, that is, virtual measurements to compensate
for instrument faults.
A successful implementation of model-based isolation and accommodation
strategy should protect engine stability and operability under steady-state and
transient conditions. In addition to compensation for different operating condi-
tions, successful accommodation methods also provide compensation for different
ambient conditions and fuel compositions. Due to a large number of possible
accommodation parameters, these control methods have to be carefully balanced
to ensure the overall stability of the gas turbine and its subsystems (Rausch et al.
2005).
5.2 Life Extension
Gas turbine engine components operate under cyclic loading conditions and harsh
environments, and hence they are subjected to deterioration. This component dete-
Fig. 24 Fault accommodation

rioration is generally described by damage evolution. Due to elevated temperatures

and operational stresses, the design life of a hot section component is significantly
reduced compared to that of a cold section component. The most common failure
modes for a gas turbine engine include: low cycle fatigue (LCF), high cycle fatigue
(HCF), thermo-mechanical fatigue (TMF), creep, rupture, corrosion, oxidation, and
erosion.
The most important aspect of life-extending control concept is the identification
of the type of damage that is most life-limiting. Thermo-mechanical fatigue,
creep, and rupture are the main candidates for damage control and life extension
on a continuous operation basis. Creep and creep rupture are primary damage
modes for turbo-machinery, and these damage modes are further aggravated by
extreme temperature cycles. By reducing temperature gradients in hot section
components, a significant life extension of critical components could be achieved.
These thermal gradients in hot section components induce thermal stresses, which
eventually can cause component failure. Especially during engine acceleration and
deceleration high thermal gradients could be achieved, and hence a control policy
that reduces engine transients can significantly influence the remaining life of
critical components (Guo et al. 2005).
The control strategies that are re-evaluated to include operating cost and to
estimate the life of the engine can achieve life-extension objectives (Fig. 25). In
this way, maintenance cost could be potentially reduced if the life of engine is
considered as an objective during the design of the engine controls philosophy. The
disadvantage of life extension control concept lies in the fact that performance and
life of engine are in opposition of each other. However, the utilization of intelligent
engine control algorithms can drastically increase the engine life, with a minimum
sacrifice in engine performance (Behbahani et al. 2006). Further development of
damage models for turbine engines, should lead to the implementation of online
damage models in real-time that will allow more robust damage prevention.
5.3 Model-Based Control
Model-based control explicitly uses virtual parameters created by gas turbine

engine models (Fig. 26). Not all of the engine variables can be measured, or they can
be measured only with reduced accuracy. Using real-time online engine models, it is
possible to obtain non-measured engine variables, such as temperatures, pressures,
and mass flow at relevant engine stations. By using those virtual measurements, it is
possible to calculate the dynamic responses of parameters, which are not available,
or data that are affected by high measurement lags or low update frequencies.
Another benefit that integrated real-time engine models can offer is the analytical
redundancy of existing sensors, which has been considered to be more cost-
effective than commonly used hardware redundancy.
Since the component characteristics change significantly during service intervals,
these control systems must sense degradation and damage to multiple components
and adapt to them. Proposed model-based control methods include numerous
Fig. 25 Prognostics and life extension
Fig. 26 Model-based control
strategies such as clearance and compressor stability control, blade flutter, and
combustion instability suppression (Lietzau and Kreiner 2001; Turevskiy et al.
2002). In addition, as previously discussed, life-extending control also has potential
to reduce the impact of thermo-mechanical fatigue on hot components life caused
by transients and short-term over-fueling.
Various control philosophies such as adaptive (Fuller et al. 2006), active
(Gastineau 2001; Garg et al. 2010), and predictive control (Brunell et al. 2004),
have been explored for use in model-based control concepts. Some of the prime
candidates for model-based control strategies that can enhance existing gas
turbine control functions encompass: turbine life extension, compressor stability

management, active combustion dynamics, and predictive emission control.
5.4 Soft Sensor Example
This section illustrates fault diagnostics, isolation, and accommodation (FDIA)

functionality of integrated control and health monitoring systems. In this example
application of “soft sensors” in dual lane control with analytical sensor, redundancy
is demonstrated.
Dual lane redundancy is often introduced into the gas turbine control systems to
improve the reliability of the measurement chains. In a dual-lane configuration, both
lanes simultaneously measure the engine parameters and compare the measurements
to enable sensor health monitoring. One lane is considered to be in the control mode
while the other is in the back-up mode. Once an instrument fault is confirmed, lane
switching is initiated.
In Fig. 27, the use of a real-time engine model (RTM) in condition monitoring
of dual-lane control system and lane-to-lane switching logic is depicted. In this
configuration, double redundancy is achieved with one hardware and one soft
sensor, where the output from the online engine model is used as an analytical
second channel for gas turbine applications. Through the comparison of redundant
channels, the system diagnoses two types of faults in sensors.
When discrepancy between analytical and sensor measurement exceeds pre-
scribed tolerance levels (soft fault – min tolerance/hard fault – max tolerance),
the sensor fault diagnosis logic determines the state of switching logic in dual lane
configuration. In depicted configuration when soft failure of the sensor is indicated,
lane logic will select a lane with conservative control settings, and in case that
sensor hard failure is detected, the faulty sensor signal is replaced by the virtual
measurement.
The dual-lane control system based on a real-time online model must accom-
modate gas turbine operation under the steady-state and transient conditions. The
sensor fault detection and accommodation functionalities are demonstrated with
Fig. 27 Dual lane control with dual sensor redundancy. (Figure published in Panov and Maleki
2017. Figure published under license of Creative Commons Attribution 4.0 International License)
sensor fault injection at steady-state conditions and subsequent adjusted transient

response using a typical load acceptance test case.
A fault injection into compressor delivery pressure measurement is depicted
in Fig. 27a. A fault is injected after the engine achieved steady-state operating
conditions, and time histories of a faulty sensor measurement, nominal measure-
ment, and virtual measurement generated by online real-time engine model are
depicted. The realignment of the soft sensor, due to the ability of the real-time engine
model to reconfigure itself, allows the implemented analytical measurement to be
subsequently used as a redundant measurement channel.
The state of the switches in dual lane logic (Fig. 27) is presented in Fig. 28b.
The state of switches is defined with analytical residual between the soft sensor
and compressor delivery pressure (CDP) measurement. When the analytical residual
exceeds the minimum (min) threshold, the switch 1 is enabled, and subsequently,
switch 2 changes states when the analytical residual exceeds the maximum (max)
threshold. This figure (Fig. 28b) also shows the adaptive minimum and maximum
thresholds during transient conditions.
The FDIA system conducts fault diagnosis and isolation and additionally acti-
vates the reconfiguration logic. Reconfiguration control logic selects the appropriate
controller by means of a switching mechanism. In, Fig. 29 consequence of control
logic reconfiguration is presented. In this figure, responses of gas turbine engine
during load acceptance maneuver for three different sensor states (no fault, soft fault,
and hard fault) and corresponding control logic configurations are demonstrated.
The engine operational parameter (power turbine speed) and control system
variable (fuel demand) are shown in Fig. 29a and b. The engine response to
reconfiguration of the control logic is presented in Fig. 29a. Three different cases
are compared, and two distinctive responses are observed. The first two cases
correspond to the conventional compressor delivery pressure (CDP) limiter and the
second response (Case 3) to the model-based controller with conservative regulator
settings. Accompanying plots of control system response (fuel demand), are given
in Fig. 29b.
a b
160 1.1
140 residual
0.9
min
120
dimensionless
max
0.7
100 switch 1
switch 2
%
80 0.5
compressor sensor
60
virtual sensor 0.3
40
nominal value
0.1
20
0 -0.1
time time
Fig. 28 (a) Compressor delivery pressure measurement, analytical measurement, and nominal
conditions. (b) Analytical residual and switch states in dual-lane control logic
Fig. 29 (a)Power turbine speed profile for block load acceptance. (b) Fuel demand for block load
acceptance – Conventional and model-based CDP limit. (Figure published in Panov and Maleki
(2017). Figure published under license of Creative Commons Attribution 4.0 International License)
6 Implementation and Deployment of Digital Twins
The current practice of implementation of processes that can accommodate devel-

oping Cyber-Physical Systems, offers the possibility of seamless connectivity of
engineering systems with Internet of Things Platforms. A Digital Twin concept,
as a part of the Industry 4.0 strategy, emerged as an answer for these challenges.
Introduction of digitalization strategies enabled integration and deployment of dif-
ferent variants of Digital Twins of the physical assets onto various systems such as
Embedded, Edge, and Cloud Platforms (Panov and Cruz-Manzo 2021). Depending
on intendent use and required processing time of available data (Real-Time, Active,
Reactive, Historical), various processing techniques such as real-time, streaming,
and batch processing (Fig. 30) are utilized by IoT systems.
The real-time data processing is based on the processing of continuous data as a
prerequisite for the implementation of embedded solutions for real-time monitoring
and optimization of assets. The processing of streaming cross-device data in near
real-time is typically used in monitoring, decision support, and generation of reports.
Embedded
systems Edge
systems
Real-Time
Processing
Stream Enterprise
Processing System
Value of lloT Data
Real-Time
Data Cloud
Batch
Agents
Processing
Activa
Data
Big Data
Reactive
Analytics
Data
Historical data
Milliseconds Seconds Minutes Hours Days Months Years
Time
Fig. 30 Value of IoT Data
Furthermore, processing a large volume of historical data is usually performed in

a batch, where processed information is stored and available for subsequent data
exploration, training of models, and reporting (Fig. 31).
Current state-of-art distributed control systems support seamless integration
of control and communication capabilities through concept of multiple network
levels (Engine, Plant, and Fleet). This hierarchical network architecture enables
deployment of heterogenous computing platforms closely integrated with assets
embedded system. The Digital Twin solutions in this way fully exploit deployment
of modular functionalities distributed across the whole IoT chain consisting of
Embedded, Edge, Cloud, and Enterprise platforms (Fig. 32).
The executable Digital Twins utilize the integration of self-configuring virtual
plants with physical assets, to achieve objectives of the flexible and robust operation.
In this way, operational assets are capable of adaptation to external and internal
operating conditions. The hybrid Cyber-Physical System configurations typically
contain a Virtual Plant, generated from models of sensors and actuators, and a
model of a core Physical Plant. The virtual plant is integrated with physical plant
via deployment onto distributed Control System and capability of the whole system
is extended with IoT connectivity. The overlapping nature of Physical, Cyber, and
IoT Domains is presented in Fig. 33.
The Physical Domain contains a real plant, which is connected to an automa-
tion system via multiple sensors and actuators. Human-in-the-loop interaction is
also part of the Physical Domain, where the operator interacts through multiple
modalities with other platforms of the system. The self-configuring Virtual Plant
Fig. 31 Real-time, stream, and batch data processing
Fig. 32 Embedded, edge, cloud and enterprise platforms
as a part of the Cyber Domain enables the Physical Plant to be monitored and
controlled via adaptation to external operational and internal health conditions.
The IoT Domain contains state-of-the art network technologies based on standard
Fig. 33 CPS and IoT

domains
Human Plant Internal States
Self-configuring
Plant
Virtual Plant
Virtual Virtual
Actuator Sensor Actuator
Sensor
Plant DCN Network
Fleet Network
IOT Platform
Fig. 34 Building blocks of CPS and IoT domains
and interoperable communication protocols. The integrated IoT Platforms provide

a range of functionalities in support of Cloud-based applications and services.
The physical trackable data objects (such as sensor measurements), and virtual
smart data objects (such as internal engine states and virtual measurements), are
seamlessly integrated into the information network. The Control System, which
contains logical and communication components, is a subset of all three domains
and serves as an integration platform for CPS and IoT Domains (Fig. 34).
The Real-Time Executable Digital Twins (RT xDT) are developed as a sub-class
of Performance Digital Twins, with the purpose to serve as digital platforms for
integration of Model-Based Control (MBC) and Prognostics and Health Manage-
PERFORMANCE Asset
PERFORMANCE Utilization
Performance
Modelling
Detection PERFORMANCE Management
Auto-tuning of component
Prediction degradation
Isolation Condition
TRACKING
Based
Analytical Redundancy DIAGNOSTICS Life Maintenance
Estimation
Model-Based
Sensor Fault Virtual Tracking PROGNOSTICS
Detection Measurements Diagnostics
Isolation Prognostics
PERFORMANCE
Sensor Fault
Accommodation Optimization of Component life Model-Based
operational limits extension Control
OPTIMIZATION
Fig. 35 Performance digital twin functionalities
ment (PHM) Technologies. From a functionality perspective, the RT xDT typically

focuses on three main areas: Model-Based Control (virtual and soft sensors),
Performance Tracking/Diagnostics/Prognostics, and Condition Based Maintenance
(Fig. 35).
6.1 Performance Tracking Example
The example of performance tracking of a gas turbine engine, by use of real-

time Gas Turbine Performance Digital Twin, is presented in this section. Feature
of performance tracking is based on adjustment of health parameters, such as
efficiencies and flow capacities, of engine gas path components (Panov 2014).
As a consequence of parameters adjustment, Digital Twin is aligned with real
engine and virtual measurements are tuned to represent actual engine parameters,
instead of parameters that correspond to the “nominal” engine model. The following
example demonstrates the capability of Digital Twin to track engine parameters
such as compressor inlet mass flow and turbine entry temperature for different
compressor geometry settings that correspond to Variable Guide Vanes. These
engine parameters typically are not measured with engine sensors, but instead,
they are measured and evaluated using testbed instrumentation and performance
evaluation tools.
Fig. 36 Comparison of real-time model estimation and performance evaluation tool output
Multi-shaft free turbine engines operating at high-pressure ratios often use vari-
able guide vanes (VGV) to ensure satisfactory and safe operation of a gas turbine.
VGV’s are typically used to optimize the performance and operating stability of
the compressors. This example demonstrates tracking of engine parameters during
engine operation outside of “nominal” geometry settings for Variable Guide Vanes.
The Fig. 36 shows how different VGV schedules, influence the compressor and
combustor operating characteristics during transient maneuvers. Figure 36a shows
a comparison of the compressor inlet mass flow measured on the testbed and the
value generated by the real-time model. Profile of typically non-measured engine
parameter, Turbine Entry Temperature (TET), is given in Fig. 36b. This figure shows
a comparison of TET evaluated using testbed measurements and TET generated by
deployed performance tracking tool.
References
A. Behbahani, E.A. Jordan, R. Millar, A feasibility study of life-extending controls for aircraft
turbine engines using a generic air force model, ASME Paper No. GT20006-90717 (2006)
J.W. Bird, H.M. Schwartz, Diagnosis of turbine engine transient performance with model-based
estimation technique, ASME Paper 94-GT-317 1994
B.J. Brunell, D.E. Viassolo, R. Prasanth, Model adaptation and nonlinear model predictive control
of an aircraft engine, ASME Paper No. GT2004-53780 (2004)
S.M. Camporeale, B. Fortunato, M. Mastrovito, A modular code for real time dynamic simulation
of gas turbines in Simulink. J. Eng. Gas Turbines Power 128, 506–517 (2006)
G. Cerri, M. Gazzino, F.A. Lacobone, A. Giovannelli, Optimum planning of electricity production.
ASME Paper GT20008-50956 (2008)
G. Cerri, L. Chennaoui, A. Giovannelli, C. Salvini, Gas path analysis and gas turbine re-mapping.
ASME Paper GT2011-46424 (2011)
S. Cruz-Manzo, V. Panov, Y. Zhang, Gas path fault and degradation modelling in twin-shaft gas
turbines. Machines 6, 43 (2018a)
S. Cruz-Manzo, S. Maleki, V. Panov, F. Agbonzikilo, Y. Zhang, A. Latimer, Analysis of
performance of a twin-shaft gas turbine during hot-end damage in the gas generator turbine,
GT2018-75374, ASME Turbo Expo 2018b, 11–15 June, Oslo, Norway
S. Cruz-Manzo, S. Krishnababu, V. Panov, Y. Zhang, Performance Analysis of a Multi-Stage Axial
Compressor Through a Semi-Empirical Modelling Framework, GPPS-BJ-2019-123, GPPS
Beijing19, 16–18 September 2019, Beijing, China
S. Cruz-Manzo, S. Krishnababu, V. Panov, C. Bingham, Inter-stage dynamic performance of an

axial compressor of a twin-shaft industrial gas turbine. Machines 8, 83 (2020)
P. Dewallef, O. Leonard, On-line performance monitoring and engine diagnostic using robust
Kalman filtering techniques, ASME Paper No. GT2003-38379 2003
D.L. Doel, TEMPER -38379.mance monitoring and engine diagnostic using r, ASME Paper No.
92-GT-315 1992
D.L. Doel, An assessment of weighted-least-squares based gas path analysis, ASME Paper No.
93-GT-119 1993
J.W. Fuller, A. Kumar, R.C. Millar, Adaptive model-based control of aircraft propulsion systems:
Status and outlook for naval aviation applications, ASME Paper No. GT2006-90241 (2006)
S. Garg, K. Schadow, W. Horn, H. Pfoertner, I. Stiharu, Sensor and actuator needs for more
intelligent gas turbine engines. ASME Paper No. GT2010-22685 (2010)
Z.D. Gastineau, Turbine engine performance improvements a proactive approach, ASME Paper
No. 2001-GT-0371 (2001)
T. Guo, L. Chen, L. Jaw, Intelligent life-extending controls for aircraft engines, NASA/TM-2005-
213373 (2005)
E. Hindle, R.V. Stone, C. Brogan, J. Vandike, K. Dale, N. Gibson, A prognostic and diagnostic
approach to engine health management, ASME Paper No. GT2006-90614 2006
L.C. Jaw, W. Wang, Mathematical formulation of model-based methods for diagnostics and
prognostics, ASME Paper No. GT2006-90655 2006
T. Kobayashi, D.L. Simon, Application of a Bank of Kalman filters for aircraft engine fault
diagnostics, NASA/TM-2003-212526 2003
T. Kobayashi, D.L. Simon, J.S. Litt, Application of a constant gain extended Kalman filter for in-
flight estimation of aircraft engine performance parameters, ASME Paper No. GT2005-68494
2005
Y.K. Lee, D.N. Mavris, V.V. Volovoi, M. Yuan, T. Fisher, A fault diagnosis method for industrial
gas turbines using Bayesian data analysis. J. Eng. Gas Turbines Power 132(4), 041602(6 pages)
(2010)
Li, Y. G., 2002, Performance analysis-based gas turbine diagnostics: A review”eJ. Power Energy,
216, 2, pp. 363–377
Y.G. Li, A gas turbine diagnostic approach with transient measurements. J. Power Energy 217(2),
169–177 (2003)
K. Lietzau, A. Kreiner, Model based control concepts for jet engines, ASME Paper No. 2001-GT-
0016 (2001)
I. Loboda, S. Yepifanov, Y. Feldshteyn, A generalized fault classification for gas turbine diagnostics
on steady states and transients, ASME Paper No. GT2006-90723 (2006)
S. Maleki, S. Cruz-Manzo, V. Panov, C. Bingham, Performance analysis and prediction of
compressor fouling condition for a twin-shaft engine, GPPS-2018-035, 2nd Global Power and
Propulsion Forum 2018, 10–12 Jan, 2018, Zurich, Switzerland
C.B. Meher-Homji, R. Bhargava, Condition monitoring and diagnostic aspects of gas turbine
transient response, ASME Paper No. 92-GT-100 1992
G.L. Merington, Fault diagnosis of gas turbine engines from transient data. J. Eng. Gas Turbines
Power 111, 237–243 (1988)
G. L. Merrington, Fault diagnosis in gas turbine using a model-based technique. J. Eng. Gas
Turbines Power 116, 374–380 (1993)
S.O.T. Ogaji, S. Sampath, Y.G. Li, R. Singh, Gas path fault diagnosis of a turbofan engine from
transient data using artificial neural networks, ASME Paper No. GT2003-38423 2003
V. Panov, GasTurboLib – Simulink Library for Gas Turbine Engine Modelling, GT2009-59389,
ASME Turbo Expo 2009, Orlando, USA 2009
V. Panov, Model-based control and diagnostic techniques for operational improvement of gas
turbine engines, ETC2013-0098, 10th European Turbomachinery Conference, Lappeenranta,
Finland (2013)
V. Panov, Auto-tuning of real-time dynamic gas turbine models, GT2014-25606, ASME Turbo
Expo 2014, Dusseldorf, Germany, 2014
V. Panov, Gas turbine performance diagnostics and fault isolation based on multidimensional
complex health vector space, ETC2015-0051, 11th European Turbomachinery Conference,
Madrid, Spain (2015)
V. Panov, Integrated model-based control and health management for industrial gas turbine,
PHM2016-041, 2016 IEEE International Conference on Prognostics and Health Management,
Ottawa, Canada (2016)
V. Panov, S. Cruz-Manzo, Gas Turbine Performance Digital Twin for Real-Time Embedded
Systems, ASME Turbo Expo 2020, 21–25 September 2020, Virtual Conference
V. Panov, S. Cruz-Manzo, Embedded, Edge and Cloud Computing for Gas Turbine Digital Twins,
ICAM 2020, 18–21 May 2021, Moscow, Russia
V. Panov, S. Maleki, Model-based compensation of sensor failure in industrial gas turbine,1st
Global Power and Propulsion Forum 2017, 16–18 Jan 2017, Zurich, Switzerland
A. Rasheed, O. San, T. Kvamsdal, Digital twin: Values, challenges and enablers. arXiv:1910.01719
(2019)
R.T. Rausch, K.F. Goebel, N.H. Eklund, B.J. Brunell, Integrated in-flight fault detection and
accommodation: A model-based study, ASME Paper N0. GT2005-68300 (2005)
A.M.Y. Razak, Industrial gas turbine performance and operability, Woodhead Publishing Ltd.,
Cambridge, England (2007)
S. Sampath, Y.G. Li, S.O.T. Ogaji, R. Singh, Fault diagnosis of a two spool turbo-fan engine using
transient data: A genetic algorithm approach, ASME Paper No. GT2003-38300 2003
T. Schobeiri, M. Abouelkheir, C. Lippke, GETRAN: A generic, modularly structured computer
code for simulation of dynamic behavior of aero and power generation gas turbines engines. J.
Eng. Gas Turbines Power 116, 483–494 (1994)
V. Surrender, R. Ganguli, Adaptive myriad filter for improved gas turbine condition monitoring
using transient data”, ASME Paper No. GT2004-53080 (2004)
T. Tagashira, T. Mizuno, M. Koh, N. Sugiyama, ATF test evaluation of model based control for a
single spool turbojet engine, ASME Paper No. GT2009-59854 2009
A.P. Tarabrin, V.A. Schurovsky, A.I. Bodrov, J.P. Stalder, An analysis of axial compressor fouling
and a blade cleaning method. J. Turbomach. 120, 256–261 (2008)
A. Turevskiy, R. Meisner, R.H. Luppold, R.A. Kern, J.W. Fuller, A model-based controller for
commercial aero gas turbine, ASME Paper No. GT-2002-30041 (2002)
L.A. Urban, Gas path analysis applied to turbine engine conditioning monitoring, AIAA/SAE
Paper 72-1082 1972
L.A. Urban, Parameter selection for multiple fault diagnostic of gas turbine engines, ASME Paper
No. 74-GT-62 1974
W.P.J. Visser, M.J. Broomhead, GSP, A Generic Object-Oriented Gas Turbine Simulation Environ-
mentrbine engines,T-2 2000
A. Volponi, Use of hybrid engine modelling for on-board module performance tracking, ASME
Paper No. GT2005-68169 2005
A.J. Volponi, H. Depold, R. Ganguli, C. Daguang, The use of Kalman filter and neural network
methodologies in gas turbine performance diagnostics: A comparative study. ASME J. Eng. Gas
Turbine Power 125(4), 917–924 (2003)
A. Volponi, T. Brotherton, R. Luppold, uEmpirical tuning of an on-board gas turbine engine model
for real-time module performance estimation035ASME Paper No. GT2007-27535 2007
V. Zaccaria, M. Stenfelt, I. Aslanidou, K.G. Kyprianidis, Fleet Monitoring and Diagnostics Based
on Digital Twin of Aero-Engines, GT2018-76414, ASME Turbo Expo 2018, Oslo, Norway,
2018
R.Y. Zhong, X. Xu, E. Klotz, S.T. Newman, Intelligent manufacturing in the context of industry
4.0: A review. Engineering 3(2017), 616–630 (2017)

Application of Digital Twin Technology

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Application of Digital Twin Technology

Uploaded by

Copyright:

Available Formats

Application of Digital Twin Technology

on Simulation and Optimization

Vili Panov and Samuel Cruz-Manzo

© Springer Nature Switzerland AG 2022 1

Digital twin · Simulation · Model-based optimization · Monitoring ·

Asset Performance Digital Twin

Customer Operations Centre Review

2 Modeling and Simulation

Generally speaking, mathematical models are a description of a system or process

Fig. 2 Integration of digital twin platforms at multiple network levels

Fig. 3 Classification of mathematical models

can be used to form a representative fault classification because of the occasional

The simplified model can be obtained by linearization of nonlinear dependency

ẋ(t) = fx (x(t), u(t), v(t), t) (5)

y(t) = g (x(t), u(t), v(t), t) (6)

ẋ(t) = A(t)x(t) + B(t)u(t) (7)

y(t) = C(t)x(t) + D(t)u(t) (8)

ẋ(t) = Ax(t) + Bu(t) (9)

y(t) = Cx(t) + Du(t) (10)

where elements of system matrices A, B, C, and D can be treated as constants.

Fig. 4 Dynamic phenomena

Predominantly, dynamic models have a component-oriented architecture where

2.1 Model Zooming Example

Prime movers’ operational requirements continue to become more and more

State-of-art degradation monitoring, fault detection, and isolation tools vary

3.1 Gas Turbine Systems

3.2 Performance Monitoring

Performance monitoring relies on engine measurements to infer the evolution of

Fig. 7 Generic system

Fig. 8 Gas turbine system

Fig. 9 Performance monitoring

measurements y represented by the nominal model, reveal problems in gas turbine

where ym and yn are output vectors of measured and non-measured variables.

3.3 Performance Tracking

Fig. 10 Performance tracking

3.4 Diagnostics (Detection and Isolation)

(t) = z(t) − ym (t) (22)

followed by calculation of normalized relative deviation Δz for measured variables

where standard deviation σ accounts for measurement uncertainty, and it is cal-

z(t) > γ (24)

ĥ(t) > ε (26)

Fig. 13 Prediction and life estimation

3.5 Prognostics (Prediction and Life Estimation)

Prognostics holds a central place in prognosis-based asset management, and it

where κ is the acceptable performance limit for useful service/maintenance, μ is

approach can be used to perform engine-specific schedule maintenance and logistic

3.6 Compressor Fouling Example

In this section, an example of performance deterioration detection for compressor

4 Gas Turbine Faults and Degradation Modes

Fig. 16 Gas turbine faults and degradation modes

Fig. 17 Gas turbine performance deviation

4.1 Faults and Degradation Classification

Fig. 18 Gas turbine faults classes

Fig. 19 Gas turbine fault classification

In reality, there is a very wide range of different engine faults. Classification

4.2 Compressor Turbine Degradation Example

correspond to the flow and efficiency of the compressor-turbine performance of

5 Integrated Control and Health Monitoring

Adaptive model-based control with integrated model-based diagnostics, isolation,

The goal of a fault accommodation system is to regain operability and maintain

5.2 Life Extension

Fig. 24 Fault accommodation

rioration is generally described by damage evolution. Due to elevated temperatures

5.3 Model-Based Control

z(t) > γ (24)