Bearing Fault Detection in Conveyor Belt Drums Using Machine
Learning.
Victor Afonso Bauler
1
, J
´
ulio A. Cordioli
2
, Danilo Silva
3
, and Danilo Braga
4
1,2,3
Federal University of Santa Catarina, Florian
´
opolis, Brazil
victor.bauler@lva.ufsc.br, julio.cordioli@ufsc.br, danilo.silva@ufsc.br
4
Dynamox SA, Florian
´
opolis, Brazil
danilo@dynamox.net
ABSTRACT
In recent years, the application of machine learning tech-
niques in condition monitoring has significantly advanced the
precision and efficiency of fault detection processes. In par-
ticular, detecting bearing faults in conveyor belt drums is crit-
ical in the mining industry for maintaining operational relia-
bility and productivity. This paper presents a case study using
vibration signals and diagnostic reports provided by the com-
pany Dynamox. After meticulous data cleaning, preprocess-
ing, and feature extraction employing advanced signal pro-
cessing techniques and statistical features, several machine
learning models were trained, optimized and evaluated, with
the best models providing very promising results.
1. INTRODUCTION
The evolution of technology and the rise of Industry 4.0 have
provided new opportunities and challenges in the field of
industrial maintenance. The integration of the Internet of
Things (IoT), cloud computing, and artificial intelligence has
redefined traditional approaches to machine monitoring and
maintenance . In this context, machine condition monitor-
ing (CM), and more specifically, vibration monitoring, has
emerged as one of the most effective techniques, allowing
maintenance teams to identify potential problems in advance
and plan interventions strategically, minimizing impacts on
production and operation (Randall, 2021).
In the mining industry, condition monitoring is confronted
with unique challenges due to the extreme environmental
conditions and the broad geographic distribution of assets,
which can extend over several kilometers (Zimroz & Kr
´
ol,
2015). Due to the risks and complexities involved in mon-
itoring all equipment in the field, techniques that leverage
Victor Bauler et al. This is an open-access article distributed under the terms
of the Creative Commons Attribution 3.0 United States License, which per-
mits unrestricted use, distribution, and reproduction in any medium, provided
the original author and source are credited.
IoT sensors for condition monitoring offer significant advan-
tages. These include the ability to continuously collect data,
which then can be analyzed using algorithms for fault detec-
tion. This approach provides vibration analysts with powerful
tools to assess and maintain equipment health.
The conveyor belt plays a crucial role in the transportation
systems of the industry. It is a mechanical device used for
the continuous movement of materials over short or long
distances. Its main components include the motor, rollers,
drums, belt, loading chute, and pulleys (Zimroz & Kr
´
ol,
2015). Conveyor belt drums, particularly their bearings, stand
out due to their susceptibility to faults (Bortnowski, Kawalec,
Kr
´
ol, & Ozdoba, 2022). These faults, if not detected and rec-
tified timely, can lead to substantial operational downtimes,
affecting the overall productivity and economic efficiency of
mining activities.
This study investigates the application of machine learning
techniques for condition monitoring of conveyor belt drums,
with a focus on detecting bearing faults. We evaluate the per-
formance of four distinct models: Logistic Regression, Sup-
port Vector Machines (SVM), RandomForest, and XGBoost.
Deep learning models are not considered in this analysis, as
our processing pipeline already includes a feature extraction
step, making traditional machine learning models more suit-
able for effectively utilizing the pre-processed data.
The methodology section outlines the comprehensive steps
undertaken in this research, including data acquisition, fea-
ture engineering, the division of data into training and testing
sets, a detailed description of the models, and the process of
hyperparameter optimization. Subsequently, the results sec-
tion presents the outcomes derived from each model based on
the methodology employed. In conclusion, this paper high-
lights the key findings of the study, underscoring its contribu-
tions to the field and suggesting directions for future research.
1
