IEEE Electrification - March 2021 - 35

An AE can
reconstruct only
those signals similar
to normal conditions
because an AE is
trained only on the
features in normal
condition.

commanded speed, resulting in current peaking in the load signature. In
the faulty condition in Figure 11(d),
the residual has a positive surge
around 40 s because the reconstructed signal is totally different from the
original signal, which would not
happen in normal cases, as depicted
in Figure 10(b). This obvious surge is
a sign of an anomaly and is used
for differentiating this fault from normal conditions.
Table 2 displays the composite
detector performance compared to
the test data set. The RMSE is 0.13,
and the MLAE is 0.06. In this case, both the RMSE and the
averaging window error equally helped detect current
disturbances. This detector scheme failed to flag several
current disturbance events; however, the overall faultidentification accuracy was relatively high at 91.7%.
Increasing the LSTM AE hidden node size to 100 units
raised the fault-identification accuracy to 95% on the
same test data set.

Conclusion
Pulsed-power loads are being increasingly deployed on
naval all-electric ships. A deep learning neural network
can be used as a tool for load-monitoring purposes in

TABLE 2. The identification-confusion matrix.
Load 1 (Coil Gun), RMSE = 0.09, and MLAE = 0.03
-

-

Network classification

-

-

-

Normal

Fault

-

Actual
Normal
classification
Fault

30

0

100%

1

39

97.5%

-

96.8%

100%

-

-

Load 2 (Fixed Impedance), RMSE = 0.01, and MLAE = 0.3

these systems. In this article, a Fourier-fed LSTM AE method was shown to
achieve a rapid and accurate result on
load monitoring and fault identification of three typical loads operating
under four fault cases. Namely, the
method was v e r i f i e d u s i n g a
pulsed-power load with a shunt
fault and an IGBT gate fault, a fixed
impedance load with a series arcing
fault, and a motor drive load with a
partially blocked rotor fault. In all
of the cases, the method proved
highly accurate in identifying faults.

Acknowledgment
We wish to thank Lynn Petersen at the U.S. Office of Naval
Research (ONR) for support with the work discussed in
this article through ONR Grant N00014-18-1-2315. Distribution A. Approved for public release, distribution is
unlimited. DCN 43-7410-20.

For Further Reading
M. Massi. " Autoencoder schema. " Wikimedia. https://com
mons.wikimedia.org/wiki/File:Autoencoder_schema
.png (accessed July 5, 2019).
G. Chevalier. " The LSTM cell. " Wikimedia. https://commons
.wikimedia.org/wiki/File:The_LSTM_cell.png (accessed May 16,
2018).
P. F. Ribeiro, C. A. Duque, P. M. Silveira, and A. S. Cerqueira,
Power Systems Signal Processing for Smart Grids. Hoboken, NJ:
Wiley, 2014, chs. 9-12.
A. Maqsood, D. Oslebo, K. Corzine, L. Parsa, and Y. Ma,
" STFT cluster analysis for DC pulsed load monitoring and
fault detection on naval shipboard power systems, " IEEE
Trans. Transport. Electrific., vol. 6, no. 2, pp. 821-831, June 2020.
doi: 10.1109/TTE.2020.2981880.
Y. Ma, A. Maqsood, K. Corzine and D. Oslebo, " Long shortterm memory autoencoder neural networks based DC pulsed
load monitoring using short-time Fourier transform feature
extraction, " in Proc. IEEE 29th Int. Symp. Ind. Electron. (ISIE), Delft,
The Netherlands, July 2020, pp. 912-917. doi: 10.1109/
ISIE45063.2020.9152477.
L. J. Petersen, D. J. Hoffman, J. P. Borraccini, and S. B. Swindler, " Next-generation power and energy: Maybe not so next
generation, " Naval Eng. J., vol. 122, no. 4, pp. 59-74, Dec. 2010.
doi: 10.1111/j.1559-3584.2010.00280.x.

-

-

Network Classification

-

-

-

Normal

Fault

-

Actual
Normal
classification
Fault

10

0

100%

Biographies

0

20

100%

Yue Ma (yma97@ucsc.edu) is with the University of Cali-

-

100%

100%

-

-

Load 3 (Propulsion Motor), RMSE = 0.13, and MLAE = 0.06
-

-

Network classification

-

-

-

Normal

Fault

-

Actual
Normal
classification
Fault

100

0

100%

5

55

91.7%

-

95.2%

100%

-

-

fornia, Santa Cruz, Santa Cruz, California, 95060, USA.
Damian Oslebo (damian.summers-osleb@navy.mil) is
with the Naval Sea Systems Command, Washington, D.C.,
20376, USA.
Atif Maqsood (amaqsood@ucsc.edu) is with Dynapower, South Burlington, Vermont, 05403, USA.
Keith Corzine (keith@corzine.net) is with the University
of California, Santa Cruz, Santa Cruz, California, 95060,
USA.

	

IEEE Electrific ation Magazine / MARCH 2 0 2 1

35
https://commons.wikimedia.org/wiki/File:Autoencoder_schema.png https://commons.wikimedia.org/wiki/File:Autoencoder_schema.png https://commons.wikimedia.org/wiki/File:Autoencoder_schema.png https://commons.wikimedia.org/wiki/File:The_LSTM_cell.png https://commons.wikimedia.org/wiki/File:The_LSTM_cell.png
IEEE Electrification - March 2021

Table of Contents for the Digital Edition of IEEE Electrification - March 2021
