Vibration based condition monitoring of rolling element bearing using xg boost algorithm and orange data mining software
1. 8th International Symposium on
Applied Engineering and Sciences (SAES2020)
Virtual Conference / 12th
–19th
December 2020
[1] J. Demsar et al., “Orange: Data Mining Toolbox in Python”, Journal of Machine Learning Research, Vol. 14, pp.
2349-2353, 2013.
[2] https://csegroups.case.edu/bearingdatacenter/pages/welcome-case-western-reserve-university-bearing-data-center-
website (Access on October 2020)
Vibration-based Condition Monitoring of Rolling Element Bearing using
XGBoost algorithm and Orange Data Mining Software
Tuan A. Z. Rahman* and Nordin Ramli
1
Electrical Motor Propulsion Laboratory, Wireless Innovation, MIMOS Berhad,
Technology Park Malaysia 57000 Kuala Lumpur, Malaysia
*Corresponding author’s e-mail: zahidi.rahman@mimos.my
Keywords: condition monitoring, bearing, machine learning, vibration
Introduction: In rotating machinery such as industrial electrical motor and wind turbine, rolling
element bearing is an important component. The faults detection of this key component at early
stage is crucial process in order to avoid catastrophic event and to reduce economic losses. This
paper presents a condition monitoring approach of rolling element bearing based on vibration
signal using open source tools which are XGBoost algorithm and a data mining software namely
Orange [1]. Previously, these open source tools show great results in solving various types of real-
world classification problems such as in biomedical, computer science and etc. Based on literature
study performed, there is no record on the implementation of these useful tools in vibration-based
condition monitoring of rolling element bearing.
Approach: The simulation study is performed within Python 3.83 environment. The established
dataset of rolling element bearing from Case Western Reserve University (CWRU) is used in this
study [2]. There are two approaches are employed to extract damage sensitive features from initial
time-domain vibration signal which are statistical and image embedding methods. Eight statistical
features are computed such as mean, standard deviation, kurtosis, skewness and etc. for both
time and frequency domains. By using image embedding widget in Orange, the plotted vibration
signals are transformed into high dimension features. The obtained features data are plotted as in
Figure 1 with the help of Principle Component Analysis (PCA) method. The dataset is divided into
two groups for training (60%) and testing (40%) stages. A comparative study is conducted to
explore the best combinations of features selection and machine learning algorithms to ensure
highest classification accuracy can be achieved.
Results and Discussion: Figure 2 highlights the performance of four classifier algorithms based
on two feature extraction approaches. Based on this study, image embedding approach improved
the classification accuracy of machine learning algorithms in comparison to statistical approach.
The results show that the XGBoost algorithm outperformed other learners in Orange in term of
classification accuracy with 91.25% achieved using both features extraction approaches.
Fig. 1: Data visualization using PCA
approach.
Fig. 2: Performance comparison due to features
and machine learning algorithms selection.