Chapter 46
Bearing Fault Diagnosis Using
Grad-CAM and Acoustic Emission
Signals
Jaeyoung Kim and Jong-Myon Kim
Abstract Bearing failure generates burst signals when the rolling elements touch a
crack on the outer race or inner race of the bearing. Over a decade, the acoustic emission (AE) has been used for detecting low-speed bearing failures. Since high sampling
rate of the AE signal makes it difficult to design and extract good features, deep
neural network-based approaches have been proposed in several recent researches.
However, the weak point of deep neural network is uninterpretable. Gradientweighted class activation mapping (Grad-CAM) is used for visual explanation of deep
neural networks in computer vision domain. This paper implements and validates
a deep neural networks-based bearing fault diagnosis method with understandable
visualization using Grad-CAM.
46.1 Introduction
Bearing is a vital component of heavy rotating machines to reduce friction between
rotating shaft and the fixed component such as bearing housing. Since the role of
bearing is important, it is highly required to detect bearing failure in earlier stage
for preventing secondary failure. 45–55% of broken machines are caused by bearing
faults [1]. For a decade, many AE-based bearing fault diagnosis techniques have been
developed. Acoustic emission (AE) means the generation of transient elastic waves
from sudden cyclic fatigue, fraction, impacting, etc. [2]. The advantages of AE-based
analysis compared with vibration-based analysis is that AE sensor can measure very
low energy signals caused by bearing failures in early stage or slow-speed operation [2]. However, since the sampling rate of AE signal is higher than 1 MHz, it is
difficult to analyze the AE signal because of the complexity of large data and computational time for analysis. To overcome this issue, DNN-based bearing fault diagnosis
J. Kim · J.-M. Kim (B)
Department of Electrical, Electronics and Computer Engineering, University of Ulsan, Ulsan
44610, South Korea
e-mail: jongmyon.kim@gmail.com
J. Kim
e-mail: kjy7097@gmail.com
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021
G. Shen et al. (eds.), Advances in Acoustic Emission Technology, Springer Proceedings
in Physics 259, https://doi.org/10.1007/978-981-15-9837-1_46
527
Bearing Fault Diagnosis Using
Grad-CAM and Acoustic Emission
Signals
Jaeyoung Kim and Jong-Myon Kim
Abstract Bearing failure generates burst signals when the rolling elements touch a
crack on the outer race or inner race of the bearing. Over a decade, the acoustic emission (AE) has been used for detecting low-speed bearing failures. Since high sampling
rate of the AE signal makes it difficult to design and extract good features, deep
neural network-based approaches have been proposed in several recent researches.
However, the weak point of deep neural network is uninterpretable. Gradientweighted class activation mapping (Grad-CAM) is used for visual explanation of deep
neural networks in computer vision domain. This paper implements and validates
a deep neural networks-based bearing fault diagnosis method with understandable
visualization using Grad-CAM.
46.1 Introduction
Bearing is a vital component of heavy rotating machines to reduce friction between
rotating shaft and the fixed component such as bearing housing. Since the role of
bearing is important, it is highly required to detect bearing failure in earlier stage
for preventing secondary failure. 45–55% of broken machines are caused by bearing
faults [1]. For a decade, many AE-based bearing fault diagnosis techniques have been
developed. Acoustic emission (AE) means the generation of transient elastic waves
from sudden cyclic fatigue, fraction, impacting, etc. [2]. The advantages of AE-based
analysis compared with vibration-based analysis is that AE sensor can measure very
low energy signals caused by bearing failures in early stage or slow-speed operation [2]. However, since the sampling rate of AE signal is higher than 1 MHz, it is
difficult to analyze the AE signal because of the complexity of large data and computational time for analysis. To overcome this issue, DNN-based bearing fault diagnosis
J. Kim · J.-M. Kim (B)
Department of Electrical, Electronics and Computer Engineering, University of Ulsan, Ulsan
44610, South Korea
e-mail: jongmyon.kim@gmail.com
J. Kim
e-mail: kjy7097@gmail.com
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021
G. Shen et al. (eds.), Advances in Acoustic Emission Technology, Springer Proceedings
in Physics 259, https://doi.org/10.1007/978-981-15-9837-1_46
527