Adversarial adaptive 1-D convolutional neural networks for

bearing fault diagnosis under varying working condition

China

Republic of China

Abstract

Traditional intelligent fault diagnosis of rolling bearings work well only under a common as-

sumption that the labeled training data (source domain) and unlabeled testing data (target do-

main) are drawn from the same distribution. However, in many real-world applications, this

assumption does not hold, especially when the working condition varies. In this paper, a new

adversarial adaptive 1-D CNN called A2CNN is proposed to address this problem. A2CNN

consists of four parts, namely, a source feature extractor, a target feature extractor, a label clas-

sifier and a domain discriminator. The layers between the source and target feature extractor are

partially untied during the training stage to take both training e

fficiency and domain adaptation

into consideration. Experiments show that A2CNN has strong fault-discriminative and domain-

invariant capacity, and therefore can achieve high accuracy under di

fferent working conditions.

We also visualize the learned features and the networks to explore the reasons behind the high

performance of our proposed model.

Keywords:

intelligent fault diagnosis; convolutional neural networks; domain adaptation;

adversarial network

1. Introduction

Machine health monitoring is of great importance in modern industry. Failure of these ma-

chines could cause great economical loss, and sometimes poses threats to the people who work

with the machines. Therefore, in order to keep the industrial machines working properly and

reliably, demand for better and more intelligent machine health monitoring technique has never

ceased [1, 2]. Rolling element bearings are the most commonly used components in rotating

machinery, and bearing faults may result in significant breakdowns, and even casualties [3, 4].

Therefore, e

ffective fault diagnosis plays a highly significant role in increasing the safety and

reliability of machinery and preventing possible damage [5].

In recent years, deep learning techniques have achieved huge success in computer vision

[6, 7] and speech recognition [8, 9]. Some deep learning techniques have already found their

way into machine health monitoring systems. For example, Jia et al. took the frequency spectra

generated by fast Fourier transform (FFT) as the input of a stacked autoencoder (SAE) with three

hidden layers for fault diagnosis of rotary machinery components [10]. Zhu et al. proposed a

May 10, 2018