AIRCRAFT ELECTRICAL MONITORING & FAULT
DETECTION
Dr. George Kusic
Department of Electrical Engineering University of Pittsburgh Pittsburgh, PA 15261
Keywords: Aircraft Electrical Systems, State Estimation, Fault Detection
Abstract: Fault detection and electrical system monitoring (management) for aircr
aft/spacecraft dc or 400 Hz
electrical systems is presented. Real-time ‘snapshot’ data is collected from current and voltage measurement
transducers on radial or loop aircraft electrical system and introduced into a State Estimator. The State
Estimator ‘smoothes’ the data, detects bad transducers, and calculates the best estimate of the voltage and
phase angle at busses of the network, i.e., the ‘state’ of the network. Experimental results of estimation and
fault detection are presented.
1 INTRODUCTION
Commercial aircraft and NASA spacecraft employ
distributed generation/load points and limited
transducer monitoring. Recent air disasters such as
the Sept. 2, 1998 Swiss Air flight 111 and the July 7,
1996 TWA flight 800 crashes have implicated the
electrical system. The FAA has listed 26 reports of
accidents or serious electrical system incidents since
1983. Some of these electrical system problems
could have been avoided if the aircraft were
equipped with improved monitoring and fault
detection.
Present day aircraft and aerospace vehicles
cont
ain many kilometers of wire throughout the
fuselage, wings, and tail structure. A military aircraft
may contain 20 kilometers, and a commercial
aircraft on the order of 240 km of wire. The wires
run through stringers, bulkheads, engine pods and
compartments, into the cockpit, wheel-wells, behind
panels, etc. Most of the wires are in harnesses, and
the terminals are inaccessible unless panels are
opened. The wires are subject to fraying in
maintenance operations, contacts corrode at high
altitudes, insulation degrades due to time or weather.
All are faults in the wiring.
Electrical system monitoring instruments
p
resented to the pilots have up to this time generally
been analog-type, and very limited in number. The
pilot’s monitoring points are typical of the locations
where voltage measurement contacts and current
transducers should be installed for monitoring and
fault detection. Detection of an abnormality in the
electrical system generally is treated by transfer to a
redundant generator and if it fails, to the auxiliary
generator. The abnormality has to be sufficient to
trip a circuit breaker (hard fault) or trip a limit
device. There are no methods on modern aircraft to
detect small amounts of current
line-to-ground (soft fault) or t
o discern if the trip
signal is valid (bad data). Present-day commercial
aircraft have sufficient monitoring only on the
primary (or high power) circuits. The extensive
secondary circuits have only over-current protection.
Figure 1 is a one-line diagram of a Boeing 737
prim
ary 3-phase electrical system. The port side of
the dual system is energized through the normal
position of transfer relay #1 (24-50-01). This
diagram for the port side has about 17 busses and 15
wires. On this part there is a 115Vac, 400 Hz
network, transformer-rectifiers, inverters, and
rectification for the 28Vdc sub-system. This one-line
diagram is representative of redundant, radial
systems on other aircraft such as the Boeing 777
(Andrade and Tenning, 1992) (Tenning, 1992). The
electrical system is basically a radial topology where
only one source at a time, Generator #1, #2 or the
APU supplies the system. Figure 1 does not show
the AWG size or length of the power transmission
lines, connectors, or wires bundled together in
harnesses.
Consider generator #1 as supplying all power in
no
rmal operation. The generator feeds the network
through breaker 1 to generator bus 1. From generator
bus 1 there are 3 active lines, to main bus 1, to
transfer bus 1, and to the battery charger via the
133
Kusic G. (2004).
AIRCRAFT ELECTRICAL MONITORING & FAULT DETECTION.
In Proceedings of the First International Conference on Informatics in Control, Automation and Robotics, pages 133-141
DOI: 10.5220/0001125501330141
Copyright
c
SciTePress
ground service bus. Thus far, this is a 5 bus, 4 line,
radial topology.
2 STATE ESTIMATION
State Estimation is a well-known power system
monitoring algorithm which is extensively used on
3-phase earth power systems. It is almost a
‘standard’ program resident in large utilities energy
management system computers (Kusic, 1985).
Figure 1: Boeing 737 Primary Electrical System
ICINCO 2004 - ROBOTICS AND AUTOMATION
134
The network simulated by the State Estimation
program consists of single phase pi equivalent
models for balanced 3-phase transmission lines, and
busses where power is injected or extracted from the
network. Each bus of the electrical network has
voltage, current, and injected power measurements
at the bus with additional metering on transmission
lines connected to the bus (node) as shown in Figure
2. Pi equivalent line parameters are used to describe
a distributed line.
V
ii
δ
x
Injections
Bus
i
jY j
C
=
ω
2
x
x
x
line to bus
m
line to bus
l
Bus
k
jY j
C
=
ω
2
x
..
G
jB
Figure 2: Connection of a Bus into a Network
(Measurements shown by X’s)
A dc network has only resistance, but the more
general case is where each 400 Hz ac transmission
line is modeled by the pi - equivalent where
GjB RjL+= +1/( )
ω
with R the line series
resistance, L the line series inductance and
G
j
B
is the derived admittance. The line-to-neutral
capacitance is C and ω represents the power
frequency in radians per second. If a three-phase
power scheme is used, then the pi network is the
balanced flow single line equivalent. For the case of
dc power, B and C are set to zero and there are no
corresponding reactive power flow or phase angles.
The sensor signals are measured quantities from
the real electrical network and indicated by X’s of
figure 2. These are measurements from the network
as performed by current transformers, step-down
potential transformers, inherent to equipment such as
regulators, over-current protection, etc. For power
flow on the line from bus i to bus k, near bus i the
measurement would be positive in value while near
bus k it is negative.
Each X on the network of figure 2 consists of up
to 4 electrical quantities. The injection
measurements at bus i are given in figure 3 where
p.u. stands for per unit as defined on the power base.
The bus injection measurements P
i
and Q
i
can be
either positive or negative and often represent the
power demand of lower voltage circuits or many
lines connected to this point. The real and reactive
power measurements describe the relative phase
angle between the bus voltage and local current
measurements
Figure 3: Bus Injection Measurements
P
i
Real power injected (Either watts or
p.u.)
Q
i
Reactive power injected (Either not-
amperes – reactive, VARS, or p.u.)
I
i
Absolute magnitude of the current
(Either amperes or p.u.)
V
i
Absolute magnitude of the line-to
neutral voltage (Either volts or p.u.)
Observe there is no phase angle measurement
associated with voltage and current. The phase angle
difference between bus i and bus k, which is
necessary to describe line power flow, is not
measured with present-day instrumentation on earth
power systems. However, for compact aircraft
systems,
it may be possible to directly measure
phase angles. Angle measurements would re-
formulate the State Estimator.Injection and line flow
measurements for the complete network are obtained
from the transducers at an instant of time referred to
as a ‘snapshot’. ‘Snapshot’ is State Estimator
terminology for simultaneous measurements
performed by every transducer. If the network has
slowly changing electrical power demand and the
computer time for a scan of all digitally encoded
data is fast enough, the ‘snapshot’ concept is valid.
Experience on the U.S. 60 Hertz power transmission
system has shown that thousands of lines and busses
can be scanned fast enough by the data acquisition
system to represents a ‘snapshot’ valid for the state
estimation computation.
Figure 4: Line Flow Measurements
P
ik
Real power flow from bus i towards bus k
(Watts or p.u.)
Q
ik
Reactive power flow from bus i towards
bus k (Volt-Amperes- Reactive, VARS,
or p.u.)
I
ik
Absolute magnitude of the line current
(Either amperes or p.u.)
V
i
Line to neutral voltage at the bus (Volts or
p.u.)
The measurements of a snapshot are the ‘sensor’
data for the State Estimator and fault detection. The
weighted measurements from the power system such
as a line power flow measurement near the i bus
toward bus k,
SZ jZ
ik
mm
=
+
12
where the
complex j term is for reactive power, are used in the
performance index:
AIRCRAFT ELECTRICAL MONITORING & FAULT DETECTION
135
J = Sum{W
i
i
(
(
h
h
i
i
-
-
Z
Z
m
m
i
i
)
)
2
2
} (1)
In equation 1, the h
i
are calculated using the state
of the power system. For example, real and reactive
power flow on a transmission line connected from
bus i to bus k are calculated as:
(2A)
h P GV GVV BV V
ik i i k i k i k i k1
2
== cos( ) sin( )
δδ δδ
()
(
)
hQ YVBVGVV BVV
ik
iii
k
i
k
i
k
i
k
2
22
== + sin cos
δδ δδ
(2B)
where G, B, and Y are parameters of the transmission
line. Injections at a bus i are essentially a sum of all
the flows on transmission lines connected to the bus.
The J performance index of equation 1 is
minimized by up-dating the state vector x at each
iteration by means of increments calculated in the
Newton-Raphson expression:
1
2
2
t
V
JJ h h
x
1
J
W
x
xxxx
∂∂
δ
∂∂
⎡⎤
∆= =
⎢⎥
⎣⎦
]
(3)
where only the vector
appears in the
Jacobian approximation and second-order effects in
the Jacobian are ignored. The partial derivatives of
h are derived from analytical expressions for line
flows, currents, etc, similar to equations 2A and 2B.
For example, partial derivative of equations 2A and
2B exist with respect to variables
, ,
[
t
Vx
δ
=
i
V
k
V
i
δ
,
and
k
δ
.
Let H =δh/δx be the linearized gradient evaluated
at the final value of the state vector,
. H is used to
compute the covariance matrix for the
measurements:
x
Σ
2
= W
-1
- H [H
t
W H]
-1
H
t
(4)
From this matrix, Σ
i
is the standard deviation of
measurement i and is the square root of the i
th
diagonal. When the standard deviation is used to
normalize the residual value
τ
τ
=
=
|
|
h
h
(
(
X
X
)
)
i
i
-
-
Z
Z
m
m
i
i
|/Σ
i
(5)
for each measurement, it has been proven that the
largest among all normalized residuals is the
most
probable bad data (Broussolle, 1978). Thus voltages,
currents, etc. are compared on the same basis to find
the bad data. This is a salient feature of State
Estimation—to detect bad measurement data from
transducers.
With no faults on the network, the residual
standard deviation values calculated by equation 5
are due to errors in the ‘true’ or physical line
parameters compared to the values used in the
computation. Also the parasitic errors in the
transducers such as accuracy, linearity, and D/A
round-off are part of the residual. A ‘bad’ data
measurement must exceed several multiples of the
residual value to be considered invalid and avoid
false alarms.
Assume that sensor data is valid and that a high
impedance line-to-line fault is present on the
network. The fault impedance (resistive only) is
sufficiently high so as to not trip the protective
circuit breakers at the ends of the line, but it is
sufficient to be detected by accuracy checks on
measurements and cause ‘bad data’ at both ends of
the line. The location of the fault is one the line with
‘bad’ data on both ends. A series fault such as occurs
due to contacts separating, corrosion of the contact,
or wire strands broken, is found by propagating
through the network the differences between
calculated sensor values and measurements.
The time interval to scan all the network
transducers is considered to be very small compared
to the time duration of electrical load/generation
changes on the network. Thus the ‘snapshot’ of data
used in the State Estimator is a steady-state
operating condition. This same time approximation
may be extended to future variable frequency
systems, where the frequency at the start of the
snapshot is sufficiently close to the frequency at the
end of the snapshot, such that a steady-state exists.
These fault detection methods based upon State
Estimation have been demonstrated on 6 terrestrial
power systems with 3-phase transmission lines
between 118 kV and 565 kV. They have also been
applied to the first aircraft electrical system, the
Wright-Patterson Air Force 270 Vdc MADMEL test
bed
(Maldonado et al., 1997) (Kusic, 2000), which is
representative of an F-18 aircraft. The MADMEL
tests indicate the derivation for 3-phase ac systems is
valid for the special case of direct current systems.
For dc systems, there is no reactive power flow, no
line inductance or line charging, and the state
consists of the voltages at the busses.
ICINCO 2004 - ROBOTICS AND AUTOMATION
136
3 A 270 Vdc NETWORK
Main power lines on aircraft are large gage wire, or
for flexibility consist of parallel smaller gage wires,
both of which have on the order of several milliohms
resistance. Figure 5 shows a radial topology 5 bus
(diamonds) network with resistances on the same
order of magnitude of the primary circuit of the 270
Vdc MADMEL electrical test bed. The network of
figure 5 is used to demonstrate State Estimation (for
the dc special case) and fault detection methods.
Because resistances are very low in the network
of figure 5, the voltage drops when current is being
conducted are on the order of tenths of volts, which
is often less than the 0.5% or 1% accuracy of ground
to bus voltage measurement transducers. Such small
voltage drops were accurately measured
(Kusic, 2000)
in the Wright-Patterson Air Force Base MADMEL
test bed for the F-18 by means of a differential
voltage scheme using the 270Vdc as a reference as
shown in figure 6. This MADMEL instrumentation
was duplicated to obtain measurements on the
network of figure 5.
5
1
4
Load
Bank
L3
21.9 m
Load
Bank
L2
26.2m
Load
Bank
L1
3
2
Line #4
9.69 m
1.08 m
REF
I
4
I
2
I
3
I
4
I
1
I
jump
Line #1
Line #2
Line #3
Jumper
108 m
I
k
Ammeter
symbol
Figure 5: Experimental Test Topology (The jumper is for
calibration only.)
A photograph of the line structure and ammeters
in a laboratory experiment setup is shown in figure
7. The resistance of an ammeter is incorporated into
the line resistances specified on figure 6. Line #4
from the 270Vdc reference to bus 1 is constructed of
calibrated dc current shunts, but the equivalent
resistance is also affected by the contacts.
The jumper wire shown in figure 6 is not a
normal part of the radial network. The jumper was
sequentially connected to bus 2, then 3, then 4 to
estimate the resistance of the transmission lines.
Line parameters B, G, Y are determined from the
gradient and Jacobian in a Newton-Raphson
computation:
p
B
G
Y
J
p
J
p
h
p
W
h
p
J
p
t
== =
2
2
1
1
(6)
ork. The jumper is
moved for normal operation.
The loop is opened then the unused
measurements are then used to determine line
parameters. This method applies if the line is in a
loop. For a radial network, a “jumper” must be
added to “calibrate” the netw
re
BCA T1
Line Resistance
Return
Resistance
AIRCRAFT
STRUCTURE
270
Vbus
270 - Vbus
+5
POWER
SUPPLY
REFERENCE
A/D
Figure 6: M surement
Used on the Test Bed of Figure 6
ethod of Differential Voltage Mea
Figure 7: Current Metering on the 270 Vdc Laboratory
twork (Ammeters are in lines #1, #2, and #3 connecteNe d
to resistive load banks L1, L2, and L3 respectively)
AIRCRAFT ELECTRICAL MONITORING & FAULT DETECTION
137
The State Estimator uses a ‘snapshot’ of
transmission line flow measurements and bus
measurements to compute the state of the network.
For the dc network of figure 6 the state is the
voltages at all the busses, V = [V1, V2, V3, …..] and
the ‘snapshot’ co
nsists of voltage and current
ea nts rm n t etw bu
Z
m
V
3mea
#5
is a
t of data with a jumper
nn bet th fer b #2 of
Z
meas
=
69.985 2 .94738 26
6),
e #1 to be
esti
ined with a source
oltage of Vdc = 260, and resulted in the estimated
arameters shown in figure 9.
mission
Line
ted
m sureme perfo ed o he n ork with s
#5 set to 270 Vdc:
eas
= [V
1meas
, V
2meas
,
s
, V
4meas
, …, I
1,2meas
,
I
1,3meas
, I
2,3meas
, I
3,2meas
, ….] (7)
In equation 7 the I
j,k meas
are measured line
currents at bus j towards bus k. If no errors are
present in the data snapshot, the state agrees exactly
with the state computed by a power flow calculation
(Kusic, 1985). An important principle is the State
Estimation method uses one bus as the ‘slack’ or
reference bus on the system, and all other bus
voltages are computed as differences of voltages
between it and the slack bus. For equation 6, bus
t 270 Vdc. The slack bus principle is the reason
the differential scheme of figure 7 was effective.
The complete snapsho
co ected ween e re ence us and bus
the network is as follows:
[V
1meas
, V
2meas
, V
3meas
, V
4meas
, …,
I
1,2meas
, I
1,3meas
, I
1,4meas
, I
1,5meas
, I
jump
. ]
=[2 17, 69 , 9.88851,
269.93169, 4.2, 3.8, 3.95, -11.95, 0.65]
(8)
When the measurement vector of equation 8 was
used in the parameter estimation program (equation
it resulted in the resistance of transmission line
#1 to be: Transmission line #1 = 12.56 milliohms
This was the only transmission line that could be
estimated with the data of equation 8 before
accumulated measurement errors resulted in
impossible parameter estimation for other lines, or
repeatedly returned transmission lin
mated. Observe that transmission line #1 was
originally taken as 9.69 milliohms.
Using the same steps that resulted in the
measurement vector equation 8, two other jumper
connections were used to create loops for parameter
estimation. These were obta
v
p
Trans Initial
Resistance
Estima
Resistance
#1 9.69
milliohms
12.56
milliohms
#2 26.2
milliohms
24.85
milliohms
#3 21.9
milliohms
-----
#4 1.08
milliohms
----
Jumper 108 milliohms 100.93
milliohm
s
Figure 8: Summary of Transmission Line Estimates for the
e 6
I
1
to I
4
,
Z
m
10)
om the snapshot of equation 10. All normalized
as = 27000 0.0030
US SU ***********
ERR.
LOAD_L3 0.9510 0.0000
REF BUS 0.9519 0.0000
Experimental Network of Figur
4 BASE CASE
The transmission line parameters of the
experimental network have been estimated (figure 8)
in the previous section. The network is considered
un-faulted. A ‘snapshot’ is taken of the radial
network operating at nominal loads, without
jumpers, in order to obtain residual values of bad
data and residual errors for fault detection. Residual
errors are due to inherent inaccuracies in
measurements, transmission line parameters, and
temperature changes. A ‘snapshot’ is taken on the
radial network with nominal loads on at busses #2 to
#4. The measured bus voltages and currents
with the reference bus #5 operated at 257Vdc are:
eas
= [256.97510, 256.92351, 256.87323,
256.75924, 4.87, 3.85, 10.0, -18.6] (
The State Estimator results for this operating
condition using the estimated transmission line
parameters are presented in figure 10. Figure 10
shows calculated values of P,Q,I,V as estimated
fr
residual errors, equation 5, are less than 5 X 10
-5
Number of measurements + parameters = 44
Pb e Vbase = 270 stest =
******* B MMARY********
meas NORM.
1 BUS 1 0.9518 0.0000
2 LOAD_L1 0.9516 0.0000
3 LOAD_L2 0.9514 0.0000
4
5
ICINCO 2004 - ROBOTICS AND AUTOMATION
138
* OW SUMMARY**************** LINE FL **
UR
BUS #1 0
US #1 EF BUS
UR
LOAD_ 0
OAD_L1 BUS #1
R
LOAD_ 0
OAD_L2 BUS #1
UR
LOAD_
OAD_L3 BUS #1
V
UR
R
REF BUS BUS # 000 0.0000
EF BUS BUS #1 V,pu 0.9519 0.0000
F
er than 4.1 X10 .
Wh
in the
harnesses plus connectors plus transfer relays are all
included in the line resistance (e.g., figure 8).
5
1 and taking a snapshot of
er the addi al nce. The
Z
meas
=
5m
]
e base case, but operating with the fault. The State
0.
270 stest = 0.0030
************ B
OLTA u
ERR.
OAD_L3 0.9845 0.0000
EF BUS 0.9852 0.0000
Bus 1 V= 0.952 p.u.
FROM TO
MEAS E NORM ERROR
BUS #1 LOAD_L1 P,pu 0.0464 0.0000
BUS #1 LOAD_L1 I,pu 0.0487 0.0000
BUS #1 LOAD_L1 Q,pu 0.0000 0.0000
BUS #1 LOAD_L1 V,pu 0.9518 0.0000
BUS #1 LOAD_L2 P,pu 0.0366 0.0000
BUS #1 LOAD_L2 I,pu 0.0385 0.0000
BUS #1 LOAD_L2 Q,pu 0.0000 0.0000
BUS #1 LOAD_L2 V,pu 0.9518 0.0000
BUS #1 LOAD_L3 P,pu 0.0952 0.0000
BUS #1 LOAD_L3 I,pu 0.1000 0.0000
BUS #1 LOAD_L3 Q,pu 0.0000 0.0000
BUS #1 LOAD_L3 V,pu 0.9518 0.0000
BUS #1 REF BUS P,pu -0.1770 0.0000
REF BUS I,pu -0.1860 0.000
B R Q,pu -0.0000 0.0000
BUS #1 REF BUS V,pu 0.9518 0.0000
Bus 2 LOAD_L1 V= 0.952 p.u.
MEAS E NORM ERROR
LOAD_L1 BUS #1 P,pu -0.0463 0.0000
L1 BUS #1 I,pu -0.0487 0.000
L Q,pu -0.0000 0.0000
LOAD_L1 BUS #1 V,pu 0.9516 0.0000
Bus 3 LOAD_L2 V= 0.951 p.u.
MEASU E NORM ERROR
LOAD_L2 BUS #1 P,pu -0.0366 0.0000
L2 BUS #1 I,pu -0.0385 0.000
L Q,pu -0.0000 0.0000
LOAD_L2 BUS #1 V,pu 0.9514 0.0000
Bus 4 LOAD_L3 V= 0.951 p.u.
MEAS E NORM ERROR
LOAD_L3 BUS #1 P,pu -0.0951 0.0000
L3 BUS #1 I,pu -0.1000 0.0000
L Q,pu -0.0000 0.0000
LOAD_L3 BUS #1 V,pu 0.9510 0.0000
BUS 5 REF BUS = 0.952 p.u.
MEAS E NORM ERROR
REF BUS BUS #1 P,pu 0.1770 0.0000
EF BUS BUS #1 I,pu 0.1860 0.0000
1 Q,pu 0.0
R
igure 9: State Estimator Results for Experimental Test
Bed (No Faults)
In figure 9 note the voltage base is 270 Vdc, so
the reference bus is operating at 257/270 =
.951851851 per unit. The base current is 100
Amperes, such that current in line #1 is 4.87
Amperes or .0487 per unit. The real power is
com
puted as measured voltage times measured
current. Observe that the reactive power is carried
through the computation, but is always zero.
An examination of the normalized residual errors
in figure 9 (not shown here), indicates the largest
value is 4.1 X10
-5
for the voltage mismatch on line
#2 due to current flow from bus #3 to bus #1. The
criterion for bad data is set to stest = 3X10
-3
which is
many orders of magnitude high
-5
en a normalized residual is above stest in value,
it is considered to be ‘bad data’.
The base case represents the aircraft electrical
system ‘as built’, or in other words, the original un-
faulted condition. The resistance of the wires
SERIES FAULT
DETECTION
Consider that the aircraft has aged over a period of
years and either corrosive effects or handling and
bending of connecters or lines has increased the
resistance of a line. The as-built values of the
transmission lines are the estimated parameters of
figure 8. However, transmission line #1 has an
increase of 0.5 milliohms resistance increase due to
a series fault. This fault was experimentally set up
by inserting a 0.5 milliohm calibrated resistance into
transmission line #
op ation with tion resista
snapshot f dat s:
o a i
[V
1meas
, V
2meas
, V
3meas
, V
4meas
, …,
I
1,2meas
, I
1,3meas
, I
1,4meas
, I
1, eas
=[265.978, 265.972, 265.887, 265.832,
7.05, 3.92, 8.05, -18.8] (11)
This snapshot is a different operating condition from
th
Estimator results with this snapshot are in figure 1
Number of measurements + parameters, = 44
Pbase = 27000 Vbase =
US SUMMARY ************
BUS V GE (p )
meas NORM.
BUS 1 0.9851 0.0000
LOAD_L1 0.9848 0.0000
LOAD_L2 0.9847 0.0000
L
R
AIRCRAFT ELECTRICAL MONITORING & FAULT DETECTION
139
***********LINE FLOW SUMMARY ********
FROM TO
5 p
UR
US #1 EF BUS
R
0.0000
OAD_L1 US #1
.9
R
.0000
OAD_L2 BUS #1
3 0.
UR
0 0.0000
OAD_L3 US #1
V
R
REF BUS BUS #1 852 0.0000
Figure 10: State Estimator Results for 0.5 milliohm Series
LOAD #1 increases to
4.8
est weighted error
always
(pU)
BUS 1 BUS #1 V= 0.98 .u.
MEAS E NORM ERROR
BUS #1 LOAD_L1 P,pu 0.0694 0.0000
BUS #1 LOAD_L1 I,pu 0.0705 0.0000
BUS #1 LOAD_L1 Q,pu 0.0000 0.0000
BUS #1 LOAD_L1 V,pu 0.9851 0.0000
BUS #1 LOAD_L2 P,pu 0.0386 0.0000
BUS #1 LOAD_L2 I,pu 0.0392 0.0000
BUS #1 LOAD_L2 Q,pu 0.0000 0.0000
BUS #1 LOAD_L2 V,pu 0.9851 0.0000
BUS #1 LOAD_L3 P,pu 0.0793 0.0000
BUS #1 LOAD_L3 I,pu 0.0805 0.0000
BUS #1 LOAD_L3 Q,pu 0.0000 0.0000
BUS #1 LOAD_L3 V,pu 0.9851 0.0000
BUS #1 REF BUS P,pu -0.1852 0.0001
BUS #1 REF BUS I,pu -0.1880 0.0001
BUS #1 REF BUS Q,pu -0.0000 0.0000
B R V,pu 0.9851 0.0000
BUS 2 LOAD_L1 V= 0.985 p.u.
MEASU E NORM ERROR
LOAD_L1 BUS #1 P,pu -0.0694 0.0000
LOAD_L1 BUS #1 I,pu -0.0705 0.0000
LOAD_L1 BUS #1 Q,pu -0.0000
L B V,pu 0.9844 0.0004
BUS 3 LOAD_L2 V= 0 85 p.u.
MEASU E NORM ERROR
LOAD_L2 BUS #1 P,pu -0.0386 0.0000
LOAD_L2 BUS #1 I,pu -0.0392 0.0000
LOAD_L2 BUS #1 Q,pu -0.0000 0
L V,pu 0.9848 0.0000
BUS 4 LOAD_L V= 984 p.u.
MEAS E NORM ERROR
LOAD_L3 BUS #1 P,pu -0.0793 0.0000
LOAD_L3 BUS #1 I,pu -0.0805 0.0000
LOAD_L3 BUS #1 Q,pu -0.000
L B V,pu 0.9846 0.0001
BUS 5 REF BUS = 0.985 p.u.
MEASU E NORM ERROR
REF BUS BUS #1 P,pu 0.1852 0.0001
REF BUS BUS #1 I,pu 0.1880 0.0001
REF BUS BUS #1 Q,pu 0.0000 0.0000
V,pu 0.9
Fault in Line #1
The normalized voltage error in transmission #1
at LOAD #1 has a value 4 X 10
-4
so line #1 is
identified as the ‘faulted’ transmission line because
it has a much higher residual than the criterion of 4.1
X10
-5
of normal operation. When the fault
magnitude is increased to 5.0 milliohms, the
normalized voltage error at
X10
-4
, so the detection is relatively insensitive to
the magnitude of the fault.
The line where the fault is located, so long as
normalized error is above the minimum level, or a
threshold set from the base case (no fault condition).
To treat the case of multiple normalized errors above
the minimum level, the errors are weighted and
propagated through the network to find the largest
weighted error. This larg
cor
10
and
3.3 X10
respectively. It is necessary to eliminate
the ‘bad data’ before series fault detection.
ctions. When the normalized
res
t (dc system only) are above the threshold
for bad data, then the line has a shunt fault to
ground.
utual coupling. The added weight for
sin
responds to the faulted line. The line with the
fault is always detected.
This fault example has considered all data in the
‘snapshot’, equation 11, to be valid. ‘Bad data
errors could originate in electronic equipment
failures in the voltage transducers, current
transducers, A/D converters, data acquisition,
memory, and other sources. For example, the valid
voltage measurement at bus #2 is 265.972. If this
voltage value is changed to 264.972 or 266.972, then
the normalized residual increases to 4.1 X
-3
-3
6 OTHER FAULTS
Bus faults are detected as bad data for zero injection
values at the bus. For example, if all the
transmission line power flows from a bus are
measured the sum is zero, therefore P
i
= 0, Q
i
= 0,
|I
i
|= 0 for the inje
idual of these injections is the largest of residuals,
the bus has faulted.
Transmission line to ground faults are detected
by means of the residuals at the ends of the line. If
the residuals for real power flow (ac or dc systems)
or curren
7 CONCLUDING REMARKS
The dc example is a special case of the 3-phase ac
case where the line flows and the fault is balanced.
To detect a single phase fault on a 3-phase system,
each phase must be instrumented and the State
Estimator algorithm must be extended to single
phase with m
gle phase transducer measurements may be
prohibitive.
State Estimation power flow depends on small
voltage differences, and small phase angle
differences for the ac case, such that differential
voltage measuring methods should be used.
ICINCO 2004 - ROBOTICS AND AUTOMATION
140
Tolerances of 1% accuracy for line-to-line
transducers cannot be used for faul
t detection,
bec
ipment
(Briere. et.al., 1995) where
monitoring and fault detection are even more
necessary.
An ing, C., “Design of the Boeing 777
Bro
hrough the Sparse Inverse
Ma et.al., “Power Management and
Kus Fault Detection
Ku
report to
Bri
m Airbus 320/330/340 to
Future Military Transport”, IEEE Micropro. And
Microsys., v19, Mar, 1995
ause the measurement errors exceed the small
voltage differences from bus-to-bus.
For dc aircraft systems the weight of the
transducers, when applied to only the high power
primary part, is acceptable for improved monitoring.
The International Space Station accepted the weight
penalty in order to employ State Estimation on its
150 Vdc primary system
(Kusic, 1989). An extension
for fault detection requires voltage difference
measurements or a reference distributed to each
measurement point. Military aircraft use
proportionately more electronics than commercial
aircraft. For dc systems, the dc/dc conversion is
advantageous over ac/dc conversion, so
instrumentation is available for the fault detection
methods. Future aircraft may employ more electrical
power equ
REFERENCES
drade, A. and Tenn
Electrical System”, IEEE-Aerosp-Electron-Syst-Mag
v7, n7, July 1992
Tenning, C., “Evolution of the Boeing 777 Electrical
Power System”, 27
th
IECEC, 1992
Kusic, G.L., “ Computer-Aided Power System Analysis”,
textbook, Prentiss-Hall, 1985, ISBN 0-13-164526-9
ussolle, F., “ State Estimation in Power Systems:
Detecting Bad Data T
Matrix Method”, IEEE Trans. Power Appar. and Sys.,
Vol 97, May/June 1978
ldonado, M.,
Distribution System for a More-Electric Aircraft”, 32
nd
IECEC, 1997
ic, G.L., “Aircraft Electrical System
and System Monitoring”, report to Wright Laboratory,
WPAFB, Dayton, OH, Oct 23,2000
sic, G.L., “System Security Monitoring with State
Estimation Applied to the PMAD Test Beds,
Rocketdyne Div. Rockwell Int., Canoga Park, CA
91303, August 1989, (now Boeing Aircraft)
ere,D. et.al.,”A Family of Fault-Tolerant Systems:
Electrical Flight Controls fro
AIRCRAFT ELECTRICAL MONITORING & FAULT DETECTION
141