Computational Tools FSU Poroseva 2008

2008 ESRDC Team Meeting, 20-21 May, Austin, TX
Computational Tools
for Early Stage
Ship Design
Center for Advanced
Power Systems
Florida State University
2000 Levy Avenue, Tallahassee, FL 32310
Outline
• Challenges & Goals
• Facilities & Capabilities
- CAPS Development
- High-End Tools
2008 ESRDC Team Meeting, 20-21 May, Austin, TX
Challenges & Goals
Challenges
-
Increased complexity and integration of all-electric ship subsystems
Uncertainty in requirements to ship system architecture
Uncertainty in component/subsystem characteristics, behavior, interaction
New concepts and technologies
Modeling and Simulation at early stage of ship design is vital
Goals
• Develop and validate
- New models for components/subsystems behavior and interaction
- New approaches & algorithms to enhance performance of computational tools
- New ship system architectures
• Real-time simulation
2008 ESRDC Team Meeting, 20-21 May, Austin, TX
Facilities &Capabilities
• CAPS Models, Methodologies & Algorithms
• High-Performance Real Time Digital Simulator
• Virtual Test Bed
• Hardware-in-the-Loop
• PC-based Software
MATLAB/Simulink, PSCAD, PSIM etc.
2008 ESRDC Team Meeting, 20-21 May, Austin, TX
CAPS Development
• Real-Time Particle Swarm Optimization
for PMSM Parameter Identification
• Neural Network Controller Design
• Parametric Sensitivity & Uncertainty Analysis
• Survivability Analysis for Power Systems
• Structural Analysis for Automated
Fault Detection & Isolation
2008 ESRDC Team Meeting, 20-21 May, Austin, TX
Real-Time Particle Swarm Optimization
for PMSM Parameter Identification
Wenxin Liu, Li Liu, and David A. Cartes
Motivation
• Previous PSO applications were offline solutions due to time requirements
for evaluating candidate solutions
• Online implementation of PSO will result in more efficient and accurate
parameter identification
Objectives
• Develop approaches/algorithms to conduct faster-than-real-time simulations
• Implement PSO in real time using a hardware controller
• Investigate its performance for parameter identification of PMSM
2008 ESRDC Team Meeting, 20-21 May, Austin, TX
Real-Time Particle Swarm Optimization
for PMSM Parameter Identification
Achievements & Considerations
Currents - Ia,Ib, Ic
• Developed a method to conduct fasterthan-real-time PSO-based simulations
• Implemented the PSO algorithm in
Simulink using Simulink modules &
Matlab Embedded Functions
• Successfully identified two parameters
in a PMSM model
5
0
-5
0
100
200
300
400
500
600
Index of samples
700
800
900
Comparison between measured
& simulated data
Future Research
• Consider other approaches such as the method of direct integration
(dimension adaptive collocation) in collaboration with Dr. Hover (MIT)
• Use properly simplified models to further speed up simulations
• Extend the approach to other online identification, optimization,
and control problems
2008 ESRDC Team Meeting, 20-21 May, Austin, TX
1000
Neural Network Controller Design for 3-Ф
PWM AC/DC Voltage Source Converters
Wenxin Liu, Li Liu, and David A. Cartes
Motivation
• Most controllers in power electronics are designed based on simplified
linear models, which limit their performances to certain configuration
and operating range
• Existing nonlinear controller designs usually have trouble handling
parameter impreciseness and parameter drifting
Objectives
• Design a novel intelligent controller based on a nonlinear model
• Approximate parameters of the system using neural network to
obtain a robust system
• Achieve both unity power factor and regulated output DC voltage
2008 ESRDC Team Meeting, 20-21 May, Austin, TX
Neural Network Controller Design for 3-Ф
PWM AC/DC Voltage Source Converters
Approach
k2
_ u
q
iq (e2 )
• NN based MIMO Control to
regulate Id & Iq indirectly to
realize control objectives
• PI control to speed up the
convergence of zero dynamics
and generate reference signal
for the NN control
vo
V
(.)
Vr2
Achievements
• Introduced a novel NN-based
adaptive nonlinear controller design
• Tested the control algorithm using
Simulink and PSIM
• Tested the algorithm using dSPACE,
RTDS-based Hardware-In-The-Loop
2
id
_
+
1
k p  ki
s
Zero dynamic controller
+
_
I d*
_
iq (e2 )
measurement (id , iq , vo )
X
1
log sig (.)
+
2
2

W2 1
s
W2
+
Output Layer
Input Layer
_
+
1
1
Neural Network Controller
W1 1
s
W1
+ ud
_
e1
k1
Structure of the adaptive NN controller
Future Research
• Design path-following type of control
to control [iq, vo]  [0 ,Vo*] directly
• Design a new algorithm to overcome
the unstable zero dynamics and
stability analysis problems
• Consider other control problems
2008 ESRDC Team Meeting, 20-21 May, Austin, TX
Parametric Sensitivity & Uncertainty Analysis
J. Langston, A. Martin, M. Steurer, S. Poroseva / J. Taylor, F. Hover (MIT)
Motivation: need to quantify uncertainty in results of simulation due to
-
Environmental (random) variables (e.g. load)
Sensitivity of simulation results to artificial parameters (e.g. time-step size)
Model (unknown) parameters (confidence bounds) (e.g. machine data)
Objectives
• From small number of evaluations of computationally expensive,
physics-based model, develop empirical surrogate models describing
system behavior as a function of model parameters
• Apply sensitivity and uncertainty analysis to computationally
inexpensive surrogate models
2008 ESRDC Team Meeting, 20-21 May, Austin, TX
Parametric Sensitivity & Uncertainty Analysis
Surrogate Models
• Polynomial models
• Gaussian Process models
• Additive models
Sampling Approaches
• Classical experimental designs
• Orthogonal arrays
• Prediction variance based designs
• Quadrature integration techniques
Achievements
• Constructed surrogate models
involving from 6 to 27 parameters
• Performed sensitivity & uncertainty
analysis for various models,assessed
propagation of effects of a pulse load
charging event
Future Research
• Uncertainty in surrogate models
• Uncertainty in distributions of
parameters
2008 ESRDC Team Meeting, 20-21 May, Austin, TX
Survivability Analysis for Power Systems
S. V. Poroseva, S. L. Woodruff/N. Lay, M. Y. Hussaini (SCS, FSU)
Survivability is the system ability to accomplish mission in spite of multiple
faults caused by adverse conditions (combat damage, software failure etc.)
Motivation
Survivability of Integrated Power System is vital for ship survivability
Integrated Power System
Power Loss
Control, Propulsion, Combat, Service Loads
Mission failure
Personnel loss
Ship destruction
Objectives
• Mathematical framework to assess system survivability
• Numerical algorithms to calculate survivability of large power systems
• New system architectures of enhanced survivability
2008 ESRDC Team Meeting, 20-21 May, Austin, TX
Survivability Analysis for Power Systems
Current focus: Topological survivability, which is due to the system topology a number of generators, their connections with one another and loads
Achievements
•
•
•
•
•
•
Developed the probabilistic description of topological survivability
Assessed survivability of topologies including 2 - 4 generators
Developed a graph-based algorithm
Compared design strategies (redundancy, link partition & position)
Suggested a new topology based on bio-prototype (patent pending)
Conducted dynamic simulation for a new generator bus of 2 generators
Future research
•
•
•
•
Bio-prototype
Susceptibility
Larger-system algorithms
Dynamic simulation for full system
Fault detection & isolation
2008 ESRDC Team Meeting, 20-21 May, Austin, TX
Web
Structural Analysis for Automated
Fault Detection & Isolation
D. Düştegör, S. V. Poroseva, S. L. Woodruff /M. Y. Hussaini (SCS, FSU)
Motivation: automated wide-area FDI methodology is required to address
current Navy demands of reduced manpower, system survivability, reliability,
availability, effective and efficient protection and control
Objectives
• Without a detailed power system model (in early stage of ship design):
assess a given system topology with respect to
- Fault Detectability
- Fault Isolability
- Extra sensor placement
• With a detailed analytical model
- Residual generator
2008 ESRDC Team Meeting, 20-21 May, Austin, TX
Structural Analysis for Automated
Fault Detection & Isolation
Approach / Methodology
• Structural model: only the relation between
variables and equations are investigated
• Canonical decomposition: yields the
“structurally” monitorable part of the system
• Matching: investigates how to eliminate
state variables and generate residuals
• Residual signature: shows which faults are
detectable and isolable from each other
Bipartite-graph based model
Efficient graph-based
algorithms
Sensor placement guideline
Preliminary Results
• Developed methodology & graph-based algorithm
• Applied to simple topologies (2-4 generators)
• Determined minimum number of sensors necessary for full fault isolability
Future Work
• Application to real-size power system topologies
• Dynamic simulation
2008 ESRDC Team Meeting, 20-21 May, Austin, TX
High-End Tools
• CAPS Models, Methodologies & Algorithms
• High-Performance Real Time Digital Simulator
• Virtual Test Bed
• Hardware-in-the-Loop
• PC-Based Software
MATLAB/Simulink, PSCAD, PSIM etc.
2008 ESRDC Team Meeting, 20-21 May, Austin, TX
5 MW RTDS-PHIL Facility at CAPS
5 MW AC-DC-AC
PEBB-based Converter
“Amplifier”
Reproduces simulated
voltage waveforms
4.16 kVAC, 1.15 kVDC nominal
+20%/-100%
40-65 (400) Hz
Bandwidth up to 1.2 kHz
Equipment delivered 10/01/2007
Commissioning started 10/22/2007
5 MW VVS – 3-Line Diagram
Grid
connection
4.16 kV
DC load
connection
1.15 kV
Design drawing
by ABB
AC load connection 4.16 kV 06/09/2005
PHIL Experiments with a Superconducting
Fault Current Limiter (FCL)
+
-
VFCL
5 MVA Converter
RTDS
Simulation
LSRC
RSRC
• First user
RLOAD
RHDWR
application of
ISRC
5 MW VVS
ISRC
VSRC
VFCL
IHDWR
• Medium voltage
FCLs may be
ISRC = IHDWRLimitedFCL
Comparison of Prospective and Measured
Fault Cu
applied to ship
2
systems
Prospective current Prospective Curre
Limited Current
• FCL is a non- 1.5
FCL voltage
FCL Voltage
linear device
Limited current
posing some
1
challenges to
0.5
PHIL
1.8 kV FCL
• Peak power was 0
1.4 MW
• Current tracking-0.5
within 10% of
-1
reference
Fault
a
b
Current (kA)/Voltage (kV)
c
-0.05
0
0.05
Measured FCL voltage
(s)
andtime
current
0.1
0.15
Cryostat
Future: High Speed Machinery HIL Facility
Machine and system simulations in RTDS
• Secured funding to
establish
experimental
facilities for medium
(3,600 RMP) and
high-speed (22,500
RPM) rotating
machinery
• Allows for testing
high speed
generators, motors,
or gas turbines
40-400 Hz
0…4.16 kV
5 MW /
6.25MVA
2 –stage gear box
proposed under
DURIP
Recommended for
funding by ONR
Future HIL R&D at CAPS
• Improving HIL interface algorythms for Non-linear loads
– Accomodate large changes of apparent impedances
– Provide robustness against noise in fedback signals and
unpredicted load behavior
– Improve transient response of 5 MW VVS
• Devolping „virtual“ motor capability using RTDS and
various amplifier converters
– Will alow testing of motor drives w/o the need to install a real load
machine
• Implementing of high-rpm machinery test capability
– Applies a recently pateted method for HiFi torque control on load
side of gear box
• Characterizing of CAPS test bed for HiFi modeling
– Facilitates future user projects through transparent model sharing
• Geographically distributed simulations
– Collaboration with MSU (RTDS) and University of Alberta,
Canada (OPAL-RT)