Feature Based Bearing Fault Detection With Phase Current Sensor Signals Under Different Operating Conditions



Published Jun 29, 2021
Tobias Wagner Sara Sommer


To avoid sudden motor failures, which might lead to unexpected downtimes of entire production plants, predictive machinery maintenance is of great importance. Since bearings
are the only connection between rotor and stator and for this
reason, among others, more than 40 percent of all motor failures are reasoned by bearing defects, bearing defect detection
is of particular importance for predictive maintenance of motor failures. In the past, mainly external vibration sensors
were used for fault detection. However, in recent years, there
has been increasing research into using phase currents for
bearing damage detection, which has the advantage of saving
additional sensors and thus costs. The practical use of phase
currents is justified by the fact that bearing damage is accompanied by eccentricities between the rotor and stator, which
affect the magnetic flux and thus also the phase currents. For
these reasons, we focus on bearing damage detection using
phase currents.
We show in the paper that variations of the rotational shaft
frequency, torque loads and radial forces applied to the bearings have a strong influence of the applicability of a defect
detection model trained with phase current data. This means,
a defect detection model, which was trained with labeled data
collected under a fixed operating condition is not able to make
reliable classifications if the operating conditions differ from
the training conditions. This is due to discrepancies in feature distributions between the different operating conditions.
However, it is of utmost importance for practical applicability that a fault detection model is able to make reliable classifications under a variety of previously unknown operating conditions.
To solve this problem, we present a deep learning approach
for diminishing the influence of different operating conditions
by utilizing unsupervised domain adaptation. As input data
for the neural network, we propose to extract hand-crafted
features, both from time- and frequency domain, from the
phase current signals. This greatly reduces training time and
resource utilization compared to training a neural network as
feature extractor on the raw time series sensor signals. The
basic idea is then to train a neural network with labeled data
from the source domain, i.e. data of one defined operating
condition, and at the same time consider unlabeled data of
the new, previously unknown, target operating condition in
order to achieve higher classification accuracies for this operating condition as well. Thus, labeled data is only necessary
from the source domain to train the Softmax classification
layer of the deep neural network. As the features computed
from the differing working conditions are the same, domain
discrepancy can be reduced in an unsupervised manner by
weight sharing and formulating a specific loss term (we use
Multi-Layer Maximum Mean Discrepancy) that captures the
discrepancy between the different operating conditions. By
applying this additional loss term to the backpropagation process, discrepancies between the feature means of source- and
target domain are reduced iteratively.
The effectiveness of the proposed method is evaluated with a
benchmark dataset from the University of Paderborn, which is
well known in the bearing damage community. The proposed
feature based deep unsupervised domain adaptation method
significantly increases classification accuracies for all transfer scenarios. With the best of our knowledge, it is the first
time one deals with phase current based domain adaptation
on statistical fault-features.

How to Cite

Wagner, T., & Sommer, S. (2021). Feature Based Bearing Fault Detection With Phase Current Sensor Signals Under Different Operating Conditions. PHM Society European Conference, 6(1), 9. https://doi.org/10.36001/phme.2021.v6i1.2852
Abstract 260 | PDF Downloads 647



bearing fault detection, domain discrepancy reduction, maximum-mean-discrepancy, deep-learning

Technical Papers