Citation: Dong, H.; Lu, J.; Han, Y.
Multi-Stream Convolutional Neural
Networks for Rotating Machinery
Fault Diagnosis under Noise and
Trend Items. Sensors 2022, 22, 2720.
https://doi.org/10.3390/s22072720
Academic Editor: Jongmyon Kim
Received: 28 February 2022
Accepted: 29 March 2022
Published: 1 April 2022
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2022 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/).
Article
Multi-Stream Convolutional Neural Networks for Rotating
Machinery Fault Diagnosis under Noise and Trend Items
Han Dong
1
, Jiping Lu
2
and Yafeng Han
1,
*
1
School of Mechanical Engineering, Beijing Institute of Technology, Beijing 100081, China;
3120190313@bit.edu.cn
2
Changjiang Delta Institute, Beijing Institute of Technology, Jiaxing 314001, China; jipingLu@bit.edu.cn
* Correspondence: hanyafeng@bit.edu.cn
Abstract:
In recent years, rotating machinery fault diagnosis methods based on convolutional neural
network have achieved much success. However, in real industrial environments, interfering signals
are unavoidable, which may reduce the accuracy of fault diagnosis seriously. Most of the current
fault diagnosis methods are of single input type, which may lead to the information contained in
the vibration signal not being fully utilized. In this study, theoretical analysis and comprehensive
comparative experiments are completed to investigate the time domain input, frequency domain
input, and two types of time–frequency domain input. Based on this, a new fault diagnosis model,
named multi-stream convolutional neural network, is developed. The model takes the time domain,
frequency domain, and time–frequency domain images as input, and it automatically fuses the
information contained in different inputs. The proposed model is tested based on three public
datasets. The experimental results suggested that the model achieved pretty high accuracy under
noise and trend items without the help of signal separation algorithms. In addition, the positive
implications of multiple inputs and information fusion are analyzed through the visualization of
learned features.
Keywords: fault diagnosis; convolutional neural network; interfering signal; information fusion
1. Introduction
Rotating machinery, as key mechanical devices, is ubiquitous in modern industry.
In engineering practice, rotating machinery frequently serves in harsh and complex envi-
ronment with high speed, heavy load, variable working conditions, and elevated tempera-
ture. Generated faults will lead to unexpected downtime, enormous economic loss, and
sometimes security incidents. Machine fault diagnosis, which is designed to detect faults
before failure happens, is one of the most essential systems in a wide range of rotating
machinery. However, in practical industrial situations, the acquired data are significantly
affected by the operating conditions, environment, and data acquisition devices, which may
lead to unreliable diagnostic results [
1
,
2
]. Therefore, how to perform diagnosis efficiently
and precisely is a challenging and worthwhile problem.
Traditional intelligent diagnosis methods mainly consist of three main stages: data
collection, artificial feature extraction, and health state recognition [
2
,
3
]. However, artificial
feature extraction greatly relies on the engineers’ specialized prior knowledge, and it is
difficult to manually design a set of features that are applicable for all conditions. Further-
more, it is difficult for the generalization performance of traditional diagnosis models to
bridge the relationship between massive data and health states [2–4].
Deep learning (DL) methods provide effective solutions to overcome the above limita-
tions. Deep learning methods are able to automatically select discriminative features that
are useful for making accurate predictions and learning nonlinear representation of the raw
signal to a higher level of abstraction according to the training data [
5
]. Different kinds of
Sensors 2022, 22, 2720. https://doi.org/10.3390/s22072720 https://www.mdpi.com/journal/sensors
