Cloud Frontiers: A Deep Dive into Serverless Spatial Data and FME
238 iit conf 238
1. Sensor Fault Diagnosis for Wind
Turbine Generators Using Kalman
Filter
Guided by:
Dr. R . Saravana Kumar
Professor
School of Electrical Engineering(SELECT)
VIT University.
Tej Enosh. M
M.Tech
Power Electronics and Drives
VIT University.
2. Outline
Introduction
Doubly Fed Induction Generator operation(DFIG)
Modeling of DFIG
Kalman Filter & Filter Bank
Generalized observer Scheme & Dedicated Observer Scheme
DFIG State Estimation with Kalman Filter
Fault detection using Dedicated Observer Scheme In DFIG
Modeling of PMSG
Fault Detection in PMSG
3. Objective
• To create a model of DFIG
• To Identify the Current Sensor Fault in DFIG with Dedicated
observer scheme using kalman Filter
• To create a state space model for PMSG
• To Identify the Current Sensor Fault in PMSG with Dedicated
observer scheme using kalman Filter
• To Identify the Current Sensor Fault in PMSG with Augmented
kalman Filter
4. Introduction
• Wind energy - the fastest-growing source of energy in the
world
• The doubly feed induction generator (DFIG) is one of the most
used drive in wind energy because of its
low cost, simplicity of maintenance, reliability
• When a fault occurs, it must be detected as soon as possible
• The data validation is important in this processes
• The control system operates with the information of the
system provided by sensors --- can be go through faults.
5. Cont..
•When a fault occurs, it must be detected as soon as possible,
even where all observed signals remain in their allowable limits.
•The fault must then be located and its cause identified
•This aspect becomes more and more investigated because of
the construction of high capacity offshore wind parks.
6. Problem Identified
• The control system operates with the information of the
system provided by sensors, which can be subjected to faults
• For The isolation of the fault the two following fault scenarios
will be used
i) multiple but non simultaneous faults scenario
ii) simultaneous faults scenario.
• The state observer for fault detection and isolation
• Filter bank used to estimate the dynamical behaviors of the
system in order to detect then to isolate the fault.
• Previous Method For Study of Current sensor Fault and
Voltage sensor Fault is Luenberger observers method
• It has proposed an observer scheme base on Kalman filter to
diagnosticate the current sensor fault of a DFIG because of its
discrete property
7. Methodology
Operating principle of a wind turbine using doubly fed
induction generator
Modeling of the doubly fed induction generator
The Kalman filter bank based on Generalized Observer
Scheme
The Kalman filter bank based on Dedicated Observer Scheme
Validation of simulated DFIG & PMSM for current sensor FDI
8. Kalman Filter
• The Kalman filter uses the dynamical model, the known inputs
to that system as well as the measurement (which given by
sensors) to estimate the state of the system.
• Widely use in automatic filtering as a mathematical technique
to extract a signal from noisy measurements.
xk+1 = Axk + Buk + wk
zk = Hxk + vk
A,B,H are matrices of approximated dimension
• p(w) ∼ N(0,Q) Q --process covariance noise
• p(v) ∼ N(0,R) R-- measurement covariance noise
w- Process noise and v- measurement noises
9. • The implementation of Kalman filter could be divided in two
steps.
• Prediction step and Correction step.
• Prediction Step:
• Correction Step:
• The diagnostic scheme with Kalman filter is capable to detect
the fault but it is unable to locate the fault.
• To resolve this problem, a filter bank will be used.
10. Filter bank for the FDI problem
• To Design state observer for fault detection and isolation is a
well known problem.
• Filter bank used to estimate the dynamical behaviors of the
system in order to detect then to isolate the fault.
• The first kind of filter bank is Dedicated Observer Scheme
(DOS).
• The second one, Generalized Observer Scheme (GOS).
• Each filter bank is composed by a number of observers, which
are supplied with all of the input and different subsets of
output of the system.
• A Decision unit diagnosticate whether or not faults are
presented in the sensors and which one is faulty by comparing
the estimated outputs with the measured ones
11. Generalized Observer Scheme and
Dedicated Observer Scheme
•Generalized Observer Scheme – Can Detect Single Sensor Fault
•Dedicated Observer Scheme (DOS) – Can Detect a Simultaneous Faults
GOS
The structure of a GOS
for a MIMO system
DOS
In this scheme each observer
is driven by a different single
output.
12. Generalized Observer
Scheme
• Structure of GOS for MIMO
System
• Supervised System with 4
outputs
• ith observer is driven by the
input u and all of the output
except yi
• By this way residual vector
rG,i depends on all but the ith
fault
13. Dedicated Observer Scheme
• In this observer scheme
each observer is driven by a
different single output.
• Hence ith observer is only
sensitive to the failure of yi
• Then the residual rD,i
represents the failure of the
ith sensor.
Advantage: It allows to detect
and isolate simultaneous
faults.
16. Modeling Of Double Fed Induction
Generator
• In this work, we consider that the DFIG operates at a fixedspeed
• Crotor convertor should be considered as control signals.
• The generated power is determined by the currents in the
windings of stator and rotor; these currents are to be
measured.
• The stator voltages are the voltages of the grid as known
external inputs.
• The model of DFIG was transformed in dq reference frame.
• The d-axis is chosen to coincide with stator phase r-axis at
t = 0 and
• The q-axis leads the d-axis by 90 degree in the direction of
rotation.
19. FDI of the Current Sensor Faults
•For The isolation of the fault the two following fault scenarios
will be used
•i) multiple but non simultaneous faults scenario
•ii) simultaneous faults scenario.
Fault detection using Kalman filter
•Residual rK obtained from the Kalman filter with no sensor’s
failure.
•The sensor’s faults are detected.
• Fault detection and isolation using Generalized Observer
Scheme
• Fault detection and isolation using Dedicated Observer
Scheme Model in the Loop validation
27. Overview
• Variable Speed operation of Modern wind turbine enables
• Optimization of the performance
• Reduces the mechanical loading
• Delivers various options for active power plant control
• Mathematical Modeling of PMSG
• Kalman Filter for State estimation in PMSG
• Fault detection Using Kalman Filter
• Augmented State Kalman Filter for PMSG
28. Estimation, Fault Diagnosis
Architecture
mi Ԑ Z
PMSG System
PMSG System
Weights
& initial state
information
Estimator0
Estimator1
Estimator2
EstimatorN
Kalman Estimator bank
State
Estimation
& residual
generation
37. Residual of Augmented state PMSG Without fault
Augmented Model
Augmented Model
2
3
1.5
2
1
0.5
1
residual-r
residual-r
0
-0.5
0
-1
-1
-1.5
-2
-2
-2.5
-3
0
20
40
60
80
100
Id
120
140
160
180
200
-3
0
20
40
60
80
100
Iq
120
140
160
180
200
38. Conclusion
• In this project, problem of current sensor Fault Detection in DFIG
and PMSM of wind turbine was treated.
• Detection and the isolation of multiple sensor faults was addressed
using the Kalman filter bank in a Dedicated observer scheme(DOS).
• All the multiple and simultaneous faults is detected and located
with the observer scheme.
• There is no miss detection.
• The employed DOS based FDI processes has shown its capacity to
detect and to isolate simultaneous faults
39. References
• H.Chafouk, G.Hoblos, N.Langlois, S.L. Gonidec, and J.Ragot, “Soft computing
algorithm to data validation aerospace systems using parity space
approach”, Journal of Aerospace Engineering, vol 20, no .3, pp. 165-171,
July 2007.
• R.Isermann, Fault-Diagnosis Systems: An Introduction from Fault Detection
to Fault Tolerance, Springer, 2005
• Bolognani, S.; Oboe, R.; Zigliotto, M., "Sensorless full-digital PMSM drive
with EKF estimation of speed and rotor position," Industrial Electronics, IEEE
Transactions on , vol.46, no.1, pp.184,191, Feb 1999
• D.H.Trinch and H.Chafouk, “Current sensor fdi by generalized observer
scheme for a generator in wind turbine”, in International Conference on
Communications, Computing and Control Applications(CCCA11),
Hammamet, Tunisia, March 2011
• O.Anaya-Lara, N.Jenkins, J.Ekanayake P.Cartwright, and M.Hughes, Wind
Energy Generation-Modelling and Control. John wiley sons, Ltd, 2009
The Simulink’s
model of the DFIG, as well as of the supervise system
(filter bank, decision unit,...) were translated in C code
using Matlab Real-Time Workshop then downloaded to the
test bench. The result of the FDI processus is display on
LEDs of the dSPACE DS4002 card.