Instrumentation & Measurement Magazine 26-2 - 40

A Dense ResNet Model with RGB
Input Mapping for Cross-Domain
Mechanical Fault Diagnosis
Xiaozhuo Xu, Chao Li, Xinliang Zhang, and Yunji Zhao
I
n actual engineering applications, the mechanical machine
is exposed to uncertain conditions such as noise
interference and various loads. The commonly used fault
diagnosis models suffer degradation in the prediction accuracy
in such complex industrial environments where the
available label samples are insufficient and the conditions are
varied. To combat this challenge, a cross-domain mechanical
fault diagnosis method based on the deep-learning networks
is proposed. It utilizes small samples, i.e., 10% of the total, and
operates on the time-series signal collected from the mechanical
equipment. It provides a classification accuracy of more
than 97% on the dataset from Case Western Reserve University
(CWRU) under variable conditions and 97.56% with the
noise interference of 0 dB. The one-dimensional vibration signal
is first converted into an image through RGB mapping.
Then, the derived RGB image is capable of the time dependent
and spatial properties of the time sequence signal and can be
directly used as the input of the deep-learning networks. The
deep-learning networks model, i.e., the ResNet, is adopted
for the fault feature extraction and additional dense connections
are added among the residual blocks to supplement
the insufficient labeled samples within the networks. Then,
an RGB-DResNet is constructed, capable of retaining the robust
features for the classification of the mechanical faults in
different working conditions. Finally, through retraining the
model by use of transfer learning, the derived RGB-TDResNet
model gives a fine adaption to the feature distribution with a
small amount of target domain information. The performance
of the proposed fault diagnosis model was validated on the
dataset from CWRU. The results show that it provides a high
identification accuracy and strong robustness in variable operating
conditions as well as the noise environment. It is a rather
promising approach for dealing with the cross-domain tasks of
mechanical fault diagnosis.
Fault Feature Extraction and Deeplearning
Based Diagnosis Methods
The fundamental components in mechanical machines such as
the rolling bearings will suffer peeling off, wearing and other
40
failures due to long-term operation [1]. Under the complex industrial
environment and various working conditions, it is of
great significance to detect and diagnose mechanical faults
quickly and accurately with limited samples. Based on feature
extraction and consequent pattern recognition, machine learning
approaches had shown a potential for motor fault diagnosis
due to their self-learning optimization. Especially, the deep
learning networks, which employ multi-layered deep convolutional
networks to extract the semantic features, have been
proved experimentally as an effective method for the fault diagnosis
[2]. Deep learning models have been popularly used in
medical imaging, image classification, speech recognition and
other fields. In practical applications, the vibration signal of
mechanical machinery contains rich operation state information
and also a clear physical meaning. Its characteristic value
and fault often have a strong corresponding relationship.
Deep learning networks were originally used to process image
analysis tasks. Therefore, they tried to transform the original
one-dimensional vibration data to a two-dimensional image
as their input through interception and rearrangement of the
vibration signal. However, considering that the fundamental
features of the mechanical vibration signals are governed by
the relatively certain functional points, the randomly selected
sample points may lack the distance dependent representation
in the convolution layers and fail to catch sensitive bands for
fault classification in frequency domain.
An applicable approach is to generate a two dimensional
time-frequency domain representation of the original signal
through Hilbert-Huang transform, continuous wavelet
transform, and short-time Fourier transform. The diagnostic
results had reflected the advantages of two-dimensional
time-frequency samples on the mechanical fault classification.
These methods need to consider the size limit of the finite element
of the functional component, the initial time, and the
time when the defect is active. The settings of these parameters
have a significant effect on the natural frequency and
shock response of the system, resulting in great fluctuation in
the final detection performance [3]. Wang et al. [4] employed
a signal transform image (STI) conversion to construct an
IEEE Instrumentation & Measurement Magazine
1094-6969/23/$25.00©2023IEEE
April 2023
Instrumentation & Measurement Magazine 26-2

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