Citation: Salimy, A.; Mitiche, I.;
Nesbit, A.; Boreham, P.; Morison, G.
Dynamic Noise Reduction with Deep
Residual Shrinkage Networks for
Online Fault Classification. Sensors
2022, 22, 515. https://doi.org/
10.3390/s22020515
Academic Editors: Nunzio Cennamo,
Yangquan Chen, Subhas
Mukhopadhyay, M. Jamal Deen,
Junseop Lee and Simone Morais
Received: 22 November 2021
Accepted: 6 January 2022
Published: 10 January 2022
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2022 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
Article
Dynamic Noise Reduction with Deep Residual Shrinkage
Networks for Online Fault Classification
Alireza Salimy
1
, Imene Mitiche
1
, Philip Boreham
2
, Alan Nesbitt
1
and Gordon Morison
1,∗
1
School of Computing, Engineering and Built Environment, Glasgow Caledonian University,
70 Cowcaddens Road, Glasgow G4 0BA, UK; alireza.salimy@gcu.ac.uk (A.S.); imene.mitiche@gcu.ac.uk (I.M.);
a.nesbitt@gcu.ac.uk (A.N.)
2
Innovation Centre for Online Systems, 7 Townsend Business Park, Bere Regis BH20 7LA, UK;
pboreham@doble.com
* Correspondence: gordon.morison@gcu.ac.uk; Tel.: +44-141-3313352
Abstract:
Fault signals in high-voltage (HV) power plant assets are captured using the electromag-
netic interference (EMI) technique. The extracted EMI signals are taken under different conditions,
introducing varying noise levels to the signals. The aim of this work is to address the varying noise
levels found in captured EMI fault signals, using a deep-residual-shrinkage-network (DRSN) that
implements shrinkage methods with learned thresholds to carry out de-noising for classification,
along with a time-frequency signal decomposition method for feature engineering of raw time-series
signals. The approach will be to train and validate several alternative DRSN architectures with
previously expertly labeled EMI fault signals, with architectures then being tested on previously
unseen data, the signals used will firstly be de-noised and a controlled amount of noise will be added
to the signals at various levels. DRSN architectures are assessed based on their testing accuracy in
the varying controlled noise levels. Results show DRSN architectures using the newly proposed
residual-shrinkage-building-unit-2 (RSBU-2) to outperform the residual-shrinkage-building-unit-1
(RSBU-1) architectures in low signal-to-noise ratios. The findings show that implementing threshold-
ing methods in noise environments provides attractive results and their methods prove to work well
with real-world EMI fault signals, proving them to be sufficient for real-world EMI fault classification
and condition monitoring.
Keywords:
shrinkage function; thresholding; EMI method; classification; machine-learning; condition
monitoring; de-noising
1. Introduction
Power generation equipment and assets used in high-voltage (HV) power produc-
tion plants are prone to developing faults; if these faults are undetected they can lead to
breakdowns, in turn causing health and safety hazards, legal issues and incurring major
losses such as fines and large-scale power outages [
1
]. Condition monitoring is carried
out on HV assets for early fault detection and breakdown prevention, preventing afore-
mentioned losses, and other unwanted outcomes. Condition monitoring is carried out
manually by experts [
2
] observing electromagnetic interference (EMI) data in differing
forms based upon individual preferences then following the observations present faults
are classified, this method is often used to detect partial-discharge (PD) in assets [
3
]. The
current expert-led manual approach to condition monitoring is problematic operationally
due to high cost, sole reliance on experts to detect and classify faults, and lack of contin-
uous monitoring. This leads to faults going unnoticed when experts are not available to
carry out condition monitoring practices. An automated approach to condition monitoring
will not only reduce dependence on experts but will also allow condition monitoring to
be practiced continuously; this will prevent faults from going unnoticed and becoming
malfunctions. Reference [4]
outlines that the spectrum analysis of specific conditions can
lead to more automated methods of data analysis. A continuous automated approach to
Sensors 2022, 22, 515. https://doi.org/10.3390/s22020515 https://www.mdpi.com/journal/sensors
