Evaluation of the Use of the Angular Domain and Order Domain in
a Bearing Fault Detection Framework using Deep Learning
Racquel Knust Domingues
1
, J
´
ulio A. Cordioli
2
, Danilo Silva
3
, Danilo de Souza Braga
4
and Guilherme Cartagena Miron
5
1,2,3
Federal University of Santa Catarina, Florian
´
opolis, Santa Catarina, 88040-900, Brazil
racquel.knust@lva.ufsc.br
julio.cordioli@ufsc.br
danilo.silva@ufsc.br
4,5
Dynamox SA, Florian
´
opolis, Santa Catarina, 88034-110, Brazil
danilo@dynamox.net
guilherme.miron@dynamox.net
ABSTRACT
Bearing failures are very common in the industrial envi-
ronment, requiring effective fault detection methods, which
can be categorized into physics-based, knowledge-based and
data-driven types. Data-driven methods are efficient in differ-
entiating healthy conditions from faulty conditions by char-
acterizing machine signals, involving stages of data acquisi-
tion, feature extraction, and condition determination. Tradi-
tionally, feature extraction and condition determination were
manual, but advances in artificial intelligence and machine
learning, especially deep learning, have automated this pro-
cess. Although deep learning can automatically learn fea-
tures from input data, the signal domain can affects model
performance. Time and frequency domain representations are
widely used in fault detection methodologies using vibration
signals. In contrast, angular and order domains are more com-
mon in variable operating conditions, but their direct use with
deep learning remains rare in the literature. Considering this,
this study evaluates a bearing fault detection methodology us-
ing vibration signals in different domains (time, frequency,
angular, and order) under various rotational conditions. Three
distinct approaches were tested to assess the effectiveness of
these representations. Results showed the frequency domain
had the best performance, and the study concluded that angu-
lar and order domains offer no significant advantage over it.
Nonetheless, it is recommended to conduct a more in-depth
analysis with more diverse datasets, especially those contain-
ing early-stage bearing fault signals.
Racquel Domingues et al. This is an open-access article distributed under the
terms of the Creative Commons Attribution 3.0 United States License, which
permits unrestricted use, distribution, and reproduction in any medium, pro-
vided the original author and source are credited.
1. INTRODUCTION
Industrial plants are commonly composed of rotating machin-
ery, resulting in the extensive use of bearings, as they are
fundamental elements for reducing friction in rotary motion
(Lei, 2016). In addition to this extensive use, bearings are of-
ten exposed to harsh operating conditions, such as high loads
or high temperatures. Consequently, bearing failures account
for a significant portion of mechanical failures in the industry,
creating the need for methodologies to detect these failures,
which can contribute to efficient and optimized maintenance.
Failure detection methodologies can be classified into three
categories: model-based methodologies, knowledge-based
methodologies, and data-based methodologies (X. Zhang,
Zhao, & Lin, 2021). Unlike the first two, which require a
deep understanding of the behavior of the element, the data-
based methodology offers a more efficient approach by dis-
tinguishing between a healthy and a faulty condition through
the characterization of a signal measured from the machine.
Data-based methodologies are basically composed of three
main stages: data acquisition, extraction of useful features,
and determination of the current condition of the element
(Mushtaq, Islam, & Sohaib, 2021). In the first stage, an in-
ternal parameter of the element must be measured and stored.
Machine vibration is a widely used parameter in this appli-
cation because any change in the element will cause an im-
mediate modification in its dynamic response (Domingues,
2023). In the feature extraction process, the goal is to ex-
tract representations and metrics that allow for the distinction
between data from different conditions (Guyon & Elisseeff,
2006). From these extracted features, the current condition of
the element can then be determined, which will serve as the
basis for the proper planning of maintenance actions.
1
