Instrumentation & Measurement Magazine 26-2 - 46

Fig. 6. Confusion matrix corresponding the prediction results under the noise interference.
Table 2 - The prediction results
under the noise interference
SNR
BF007
BF014
IR007
IR014
Normal
10 dB
90
90
90
90
90
5 dB
90
87
89
90
90
2 dB
90
87
90
88
90
0 dB
88
85
90
86
90
dB. The model accuracy declines noticeably with the growing
effect of the noise. Comparatively, the RGB-TDResNet model
provides a 100% prediction accuracy for the test samples with
noise of 10 dB, and 99.11% prediction accuracy of 5 dB. For the
case of noise of 2 dB, only five samples out of the 450 test samples
were predicted incorrectly with a diagnosis accuracy of
98.89%. The existence of the fault states can be successfully
detected, and all the misjudgments occur between different
fault types. Similar experimental results were derived for test
samples with a 0 dB noise interference where 11 samples were
predicted incorrectly.
Conclusions
To solve the problem of insufficient samples and the accuracy
decline in the cross-domain diagnosis tasks, an improved
DResNet model, i.e., RGB-DResNet, is proposed in this paper.
The features extracted from the RGB mapped images consider
the temporal dependence and spatial properties of the original
time domain signal and are more robust in cross-domain classification
tasks. Meanwhile, the introduced dense connection can
solve the overfitting of the model when the sample is insufficient
and strengthen the robustness of the model by means of an
implicit 'deep supervision.' The deep supervision was realized
by a single classifier at the top of the network to provide direct
supervision of all layers through at most three to four layers. It
enables the middle network layer to learn discriminative features
by a classifier attached to each hidden layer. Moreover, by
the transfer learning mechanism, the derived RGB-TDResNet
model demands a small amount of target domain information
to adapt to the feature distribution of the target domain.
46
It should be noted that the proposed method still requires
the target domain samples to fine-tune the model. When faced
with unknown fault samples in the target domain, it is insufficient
to deal with the effects of feature drift in cross-domain
tasks. Domain adaptation by extracting domain-invariant
features to minimize the impact of feature drift may be an alternative
approach for such a challenge. Therefore, subsequent
research will focus on domain adaptation and domain generalization
and further optimize the model to deal with unknown
faults as well as new types of faults in the target domain.
Acknowledgment
This work has been supported by the Fundamental Research
Funds for the Universities of Henan Province (NSFRF210305)
and Scientific and Technological Research Projects in Henan
Province (222102210274, 212102210244).
References
[1] Z. Wang et al., " Application of an improved ensemble local mean
decomposition method for gearbox composite fault diagnosis, "
Complexity, vol. 2019, no. 4, pp. 1-17, 2019.
[2] Y. Chang, J. Chen, C. Qu, and T. Pan, " Intelligent fault diagnosis
of wind turbines via a deep learning network using parallel
convolution layers with multi-scale kernels, " Renewable Energy,
vol. 153, no. 8, pp. 205-213, 2020.
[3] L. Cui, X. Wang, H. Wang, and N. Wu, " Improved fault size
estimation method for rolling element bearings based on
concatenation dictionary, " IEEE Access, vol. 99, 2019.
[4] Z. Wang et al., " Data-driven fault diagnosis method based on
the conversion of erosion operation signals into images and
convolutional neural network, " Process Safety and Environmental
Protection, vol. 149, no. 12, pp. 591-601, 2021.
[5] S. Aydın, " Deep learning classification of neuro-emotional
phase domain complexity levels induced by affective video film
clips, " IEEE J. Biomedical and Health Informatics, vol. 24, no. 6, pp.
1695-1702, 2019.
[6] Y. Xu et al., " A hybrid deep-learning model for fault diagnosis of
rolling bearings, " Measurement, vol. 169, no. 6, article108502, 2021.
[7] H. Liang, and X. Zhao, " Rolling bearing fault diagnosis based
on one-dimensional dilated convolution network with residual
connection, " IEEE Access, vol. 99, 2021.
IEEE Instrumentation & Measurement Magazine
April 2023

Instrumentation & Measurement Magazine 26-2

Table of Contents for the Digital Edition of Instrumentation & Measurement Magazine 26-2