Article
Multilocation and Multiscale Learning Framework with Skip
Connection for Fault Diagnosis of Bearing under Complex
Working Conditions
Hongwei Ban, Dazhi Wang *, Sihan Wang and Ziming Liu
Citation: Ban, H.; Wang, D.; Wang,
S.; Liu, Z. Multilocation and
Multiscale Learning Framework with
Skip Connection for Fault Diagnosis
of Bearing under Complex Working
Conditions. Sensors 2021, 21, 3226.
https://doi.org/10.3390/s21093226
Academic Editor: Kim Phuc Tran
Received: 20 April 2021
Accepted: 2 May 2021
Published: 6 May 2021
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2021 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
School of Information Science and Engineering, Northeastern University, Shenyang 110819, China;
1900703@stu.neu.edu.cn (H.B.); 2010257@stu.neu.edu.cn (S.W.); 1900721@stu.neu.edu.cn (Z.L.)
* Correspondence: wangdazhi@ise.neu.edu.cn
Abstract:
Considering various fault states under severe working conditions, the comprehensive
feature extraction from the raw vibration signal is still a challenge for the diagnosis task of rolling
bearing. To deal with strong coupling and high nonlinearity of the vibration signal, this article
proposes a novel multilocation and multikernel scale learning framework based on deep convolution
encoder (DCE) and bidirectional long short-term memory network (BiLSTM). The procedure of the
proposed method using a cascade structure is developed in three stages. In the first stage, each parallel
branch of the multifeature learning combines the skip connection and the DCE, and uses different
size kernels. The multifeature learning network can automatically extract and fuse global and local
features from different network depths and time scales of the raw vibration signal. In the second
stage, the BiLSTM as the feature protection network is designed to employ the internal calculating
data of the forward propagation and backward propagation at the same network propagation node.
The feature protection network is used for further mining sensitive and complementary features.
In the third stage of bearing diagnosis, the classifier identifies the fault types. Consequently, the
proposed network scheme can perform well in generalization capability. The performance of the
proposed method is verified on the two kinds of bearing datasets. The diagnostic results demonstrate
that the proposed method can diagnose multiple fault types more accurately. Also, the method
performs better in load and speed adaptation compared with other intelligent fault classification
methods.
Keywords:
deep learning; multilocation learning; multikernel learning; multifeature protection;
deep convolution encoder (DCE); bidirectional long short-term memory (BiLSTM); bearing fault
diagnosis scheme
1. Introduction
Rolling bearings are widely used as indispensable components in modern mechanical
equipment. However, the rolling bearings usually work under the severe conditions of
varying speed, heavy load, variable load, and high temperature for a long time. They are
vulnerable to occur deformation, abrasive wear, or other faults. These faults may lead to
equipment performance degradation and even lead to severe economic loss [
1
]. Therefore,
it is critically important to develop a system that can accurately diagnose various bearing
faults under complex operating conditions and working environments.
From the perspective of pattern recognition, an intelligent bearing diagnosis process
based on machine learning generally include three steps: data preparation, feature extrac-
tion, and fault classification. The purpose of feature extraction is to mining or summarize
representative features. This operation can present the health condition of hardware de-
vices and is beneficial to improve the accuracy of downstream fault classification tasks.
Traditional bearing fault classification methods are difficult to extract features from the raw
input signals, such as empirical mode decomposition [
2
], local mean decomposition [
3
],
Sensors 2021, 21, 3226. https://doi.org/10.3390/s21093226 https://www.mdpi.com/journal/sensors
