Multiscale Convolutional Neural Network Based on Channel Space Attention for Gearbox Compound Fault Diagnosis

Abstract

Gearboxes are one of the most widely used speed and power transfer elements in rotating machinery. Highly accurate compound fault diagnosis of gearboxes is of great significance for the safe and reliable operation of rotating machinery systems. However, traditional compound fault diagnosis methods treat compound faults as an independent fault mode in the diagnosis process and cannot decouple them into multiple single faults. To address this problem, this paper proposes a gearbox compound fault diagnosis method. First, a multiscale convolutional neural network (MSCNN) is used as a feature learning model, which can effectively mine the compound fault information from vibration signals. Then, an improved hybrid attention module, named the channel-space attention module (CSAM), is proposed. It is embedded into the MSCNN to assign weights to multiscale features for enhancing the feature differentiation processing ability of the MSCNN. The new neural network is named CSAM-MSCNN. Finally, a multilabel classifier is used to output single or multiple labels for recognizing single or compound faults. The effectiveness of the method was verified with two gearbox datasets. The results show that the method possesses higher accuracy and stability than other models for gearbox compound fault diagnosis.

Keywords: attentional mechanisms; compound faults; gearboxes; multilabel classification; multiscale feature extraction.

Figure 1

SoftMax classifier and multilabel classifier.

Figure 2

Structure of the CBAM.

Figure 3

Channel Attention Module in the CBAM. C stands for channel and L stands for length.

Figure 4

Improved Channel Attention Module.

Figure 5

Spatial Attention Module in the CSAM.

Figure 6

CSAM-MSCNN architecture diagram.

Figure 7

Fault diagnosis process based on the CSAM-MSCNN.

Figure 8

Wind turbine drive system fault diagnosis test bench.

Figure 9

Structure diagram of parallel shaft gearbox.

Figure 10

Faulty parts in the gearbox. (a) Bearing rolling body wear failure; (b) Bearing inner ring wear failure; (c) Bearing outer ring pitting failure; (d) Gear tooth breakage failure; (e) Missing gear teeth failure; (f) Gear tooth wear failure; (g) Gear tooth root crack failure.

Figure 11

Time domain diagram of the vibration signal.

Figure 12

Time domain diagram of the simulated signal.

Figure 13

Simulation experimental training process of CSAM-MSCNN: (a) Loss function curve; (b) accuracy curve.

Figure 14

Visualization of test set diagnostic results. Labels 0, 1, 2, 3 represent normal, cracked tooth, pitting tooth, and gear crack-pitting compound failure, respectively.

Figure 15

Schematic of the gearbox used in the PHM 2009 Challenge Data [32].

Figure 16

The training process of the CSAM-MSCNN: (a) Loss function curve; (b) accuracy curve.

Figure 17

Comparison of accuracy and standard deviation between the CSAM-MSCNN and other models.

Figure 18

t-SNE visualization of network layer features: (a) Features extracted at the first scale; (b) Features extracted at the second scale; (c) Features extracted at the third scale; (d) Features obtained from feature fusion; (e) Features after CSAM correction; (f) The last layer features of the output layer.