Page 14 - CE_Industral_Journal_Jan2008

A Novel Approach for Ground Fault Detection

Test Results The higher order statistics based algorithm was tested
Figure 6 shows in simplified form an exemplary labora- using the collected data and the results indicate a
tory model that was developed to experimentally stage probability of detection of about 97.14% with a zero
high impedance faults and to collect data for testing false alarm rate for all cases which includes two arc
and evaluation. The exemplary setup included two welder loads. Thresholds were set such that the false
120/4500 V, 1 kVA transformers connected in paral- alarm rates as=05.0 and ah=05 which corresponds to
lel and energized from a 120 V, 15 A, 60 Hz power an overall false alarm rate of about 0.09. These results
source. As shown in Figure 3, a bare conductor was indicate that higher order signatures are distinguish-
connected to one terminal of the transformer secondar- able from welding and other non-linear loads.
ies to simulate a downed transmission line. The other The wavelet based algorithm delivers about 80%
secondary terminal was connected to a copper plate detection with about a 0.5% false alarm rate in the
buried in soil, thereby simulating the ground electrode absence of arc welding loads. With the lowering of
and the earth. thresholds, the detection rate increases to about 95%
The bare conductor was dropped on a variety of soil with about a 0.1% false alarm rate. The detection per-
surfaces to investigate differences in the resulting cur- formance drops to about 65% in the presence of arc
rents. The current signatures were collected using a welding signals and without lowering the thresholds.
data acquisition system based on the National Instru- The false alarm rate remains under about 1%.
ments data acquisition and signal conditioning boards
with Lab-VIEW software operating on Windows NT.
The data was sampled at 20 kHz, quantized to 14 bits
and stored in binary format. Each trial case was con-
ducted for 50 second duration.
Fifteen cases were run for seven different wet surface
conditions (wet and frozen sod, soil, asphalt, gravel,
sand, and concrete) for a total of 105 high impedance
fault cases. This data acquisition scheme was also used
to collect signatures for currents for single-phase non-
linear loads (e.g., TV, fluorescent lamp, PC, bridge rec-
tifier, a phase-controlled motor drive, and arc welder).
A total of 22 load files were created.

Figure 6. Laboratory model developed to experimen-
tally stage high impedance faults.

Industry Journal 13