Cross-domain mechanical fault diagnosis method based on multi-channel feature fusion of cbam and use thereof

Patent metadata

Field	Value
Publication number	US-2024142342-A1
Application number	US-202318494396-A
Country	US
Kind code	A1
Filing date	Oct 25, 2023
Priority date	Oct 26, 2022
Publication date	May 2, 2024
Grant date	—

Abstract

The present invention belongs to the technical field of mechanical fault data recognition, and discloses a cross-domain mechanical fault diagnosis method based on multi-channel feature fusion of an improved CBAM and use thereof, wherein the method comprises: conducting preliminary feature extraction in a grey-scale graph formed by original signals with convolutional neural network, obtaining high-level features, and compressing the high-level features with a full-connection layer module; conducting deep-level multi-sensor feature extraction with an improved convolutional block attention module (CBAM); conducting fusion for multi-sensor features extracted with an improved convolutional block attention module and obtaining multi-sensor fusion features; and inputting the multi-sensor fusion features into a tag assignor for fault diagnosis results. In the present invention, the latest multi-channel domain adaptation fault diagnosis method is used to realize efficiently intelligent fault diagnosis tasks of bearings in different working states.

First claim

1 . A cross-domain mechanical fault diagnosis method based on multi-channel feature fusion of an improved CBAM, wherein the method comprises:
S1: conducting preliminary feature extraction in a grey-scale graph formed by original signals with convolutional neural network, obtaining high-level features, and compressing the high-level features with a full-connection layer module; S2: conducting deep-level multi-sensor feature extraction with an improved convolutional block attention module (CBAM); S3: conducting fusion for multi-sensor features extracted with an improved convolutional block attention module and obtaining multi-sensor fusion features; and S4: inputting the multi-sensor fusion features into a tag assignor for fault diagnosis results.

2 . The cross-domain mechanical fault diagnosis method based on multi-channel feature fusion of an improved CBAM according to claim 1 , wherein in the step S1, conducting preliminary feature extraction in the grey-scale graph formed by the original signals with a convolutional neural network, obtaining the high-level features and compressing the high-level features with the full-connection layer module comprises specifically the following steps:
(1) obtaining bearing vibration signals under different working conditions and health states by experiments on a test platform; (2) conducting aligned equal distance sampling for the obtained multi-sensor bearing vibration signals, and building a tagged source domain dataset and an untagged target domain dataset; (3) using signals in a temporal domain as an input of a cross-domain mechanical fault model based on multi-channel feature fusion of the improved CBAM, wherein data of the signals in the temporal domain are aligned to be a grid as per a principle of row-major.

3 . The cross-domain mechanical fault diagnosis method based on multi-channel feature fusion of an improved CBAM according to claim 2 , wherein in the step (1), obtaining the bearing vibration signals under different working conditions and health states, wherein a sampling frequency is 12800 Hz;
in the step (2), obtaining the untagged target domain dataset by not generating a tag set for target domain data.

4 . The cross-domain mechanical fault diagnosis method based on multi-channel feature fusion of an improved CBAM according to claim 1 , wherein in the step S2, extracting deep-level multi-sensor features with the improved convolutional block attention module, comprising, building a multi-sensor feature extractor with a multi-channel convolutional neural network, a fully connected layer and the convolutional block attention module;
conducting deep-level multi-sensor feature extraction with the improved convolutional block attention module comprises conducting feature extraction in signal space by features of a local receptive field of a convolutional neural network and acquiring features of the signal space; converting the features extracted by the convolutional neural network into one-dimensional data and inputting the one-dimensional data into the full-connection layer can help to acquire corresponding features of time series signals and finally conducting deep extraction and fusion for signal features with the improved convolutional block attention module.

5 . The cross-domain mechanical fault diagnosis method based on multi-channel feature fusion of an improved CBAM according to claim 1 , wherein in the step S3, obtaining the multi-sensor fusion features comprises the following steps:
1) conducting classification of health states to the multi-sensor features by a tag assignor, judging fault diagnosis performance in the source domain and the target domain via cross-entropy loss; wherein a function of the cross-entropy loss is:

L
C

(

x
s

,

y
s

)

=

1
n

⁢

∑

i
=
1

n
s

⁢

∑

k
=
1

K

⁢

l
⁡
(

y
i
s

=
k

)

·
log

⁢

C
⁡
(

F
⁡
(

x
i
s

)

)

k

(
1
)

wherein, l(y i s =k) is an indicator function, C(F(x i s )) k ) is a kth value as predicted, and k is a number of health classification;
2) conducting parallel extraction for features of different sensors by a multi-channel feature extractor, obtaining multi-sensor fusion features, and adopting a sequence of the convolutional neural network, the full-connection layer neural network and the final improved CBAM multi-sensor feature fusion module as a sequence of the feature extraction module;
M ( F )=σ( W 1 ( W 0 ( F avg c )))+ W 1 ( W 0 ( F max c ))+σ(ƒ n×n [F avg s :F max s ])   (2)

wherein, σ is a sigmoid activation function, W is a weight of the full-connection neural network, and f{circumflex over ( )}(n×n) is a convolutional operation with a cnonvolutional kernel of n×n.

6 . The cross-domain mechanical fault diagnosis method based on multi-channel feature fusion of an improved CBAM according to claim 1 , wherein in the step S4, inputting the multi-sensor fusion features into a tag assignor and obtaining t

Assignees

• Univ Shandong Science & Tech

Inventors

• Bao Huaiqian
• Wang Jinrui
• Zhang Zongzhen
• Han Baokun
• Xing Shuo

Classifications

• G01M13/028

  Acoustic or vibration analysis · CPC title

• G01M13/045 Primary

  Acoustic or vibration analysis · CPC title

• G06F18/253 Primary

  of extracted features · CPC title

• G06N3/08 Primary

  Learning methods · CPC title