Gearbox Fault Diagnosis Based on MSCNN-LSTM-CBAM-SE

Abstract

Ensuring the safety of mechanical equipment, gearbox fault diagnosis is crucial for the stable operation of the whole system. However, existing diagnostic methods still have limitations, such as the analysis of single-scale features and insufficient recognition of global temporal dependencies. To address these issues, this article proposes a new method for gearbox fault diagnosis based on MSCNN-LSTM-CBAM-SE. The output of the CBAM-SE module is deeply integrated with the multi-scale features from MSCNN and the temporal features from LSTM, constructing a comprehensive feature representation that provides richer and more precise information for fault diagnosis. The effectiveness of this method has been validated with two sets of gearbox datasets and through ablation studies on this model. Experimental results show that the proposed model achieves excellent performance in terms of accuracy and F1 score, among other metrics. Finally, a comparison with other relevant fault diagnosis methods further verifies the advantages of the proposed model. This research offers a new solution for accurate fault diagnosis of gearboxes.

Keywords: convolutional block attention module; fault diagnosis; gearbox; long short-term memory networks; multi-scale feature extraction; squeeze-and-excitation.

Figure 1

Internal structure of an LSTM with a forget gate.

Figure 2

Structure of CBAM.

Figure 3

Architecture of the CAM.

Figure 4

Architecture of the SAM.

Figure 5

The principal flowchart of CBAM–SE.

Figure 6

MSCNN–LSTM–CBAM–SE network structure.

Figure 7

Intelligent fault diagnosis process.

Figure 8

Illustration of overlapping sampling.

Figure 9

Main structures of the experimental models: (a) CNN1–CBAM–SE; (b) CNN2–CBAM–SE; (c) CNN1–LSTM–CBAM–SE; (d) CNN2–LSTM–CBAM–SE; (e) MSCNN–LSTM–CBAM–SE (w/o BN); (f) MSCNN–LSTM–CBAM–SE.

Figure 10

Wind power generator drivetrain testing setup.

Figure 11

(a) Configuration of planetary gearbox internals; (b) healthy condition; (c) broken tooth; (d) worn tooth; (e) cracked tooth; (f) missing tooth.

Figure 12

Vibration signal time–domain waveforms under different states: (a,b) broken tooth, (c,d) normal, (e,f) missing tooth, (g,h) cracked tooth, (i,j) worn tooth. Odd labels indicate disassembly, and even labels indicate installation conditions.

Figure 12

Vibration signal time–domain waveforms under different states: (a,b) broken tooth, (c,d) normal, (e,f) missing tooth, (g,h) cracked tooth, (i,j) worn tooth. Odd labels indicate disassembly, and even labels indicate installation conditions.

Figure 13

Accuracy and loss curves of Experiment 1: (a) training accuracy; (b) test accuracy; (c) training loss; (d) test loss.

Figure 14

Confusion matrix for MSCNN-LSTM-CBAM-SE fault classification in Experiment 1.

Figure 15

Diagnostic results of six methods in Experiment 1.

Figure 16

Feature distribution visualization via T–SNE in Experiment 1: (a) original test set; (b) CNN1–CBAM–SE; (c) CNN2–CBAM–SE; (d) CNN1–LSTM–CBAM–SE; (e) CNN2–LSTM–CBAM–SE; (f) MSCNN–LSTM–CBAM–SE (w/o BN); (g) MSCNN–LSTM–CBAM–SE.

Figure 16

Feature distribution visualization via T–SNE in Experiment 1: (a) original test set; (b) CNN1–CBAM–SE; (c) CNN2–CBAM–SE; (d) CNN1–LSTM–CBAM–SE; (e) CNN2–LSTM–CBAM–SE; (f) MSCNN–LSTM–CBAM–SE (w/o BN); (g) MSCNN–LSTM–CBAM–SE.

Figure 17

(a) Experimental setup for the HUST gearbox dataset: ① speed control; ② motor; ③ acceleration sensor; ④ gearbox; ⑤ data acquisition board. (b) Images of the gearbox.

Figure 18

Photographs of the faulty gears.

Figure 19

Vibration signal time–domain waveforms for (a) broken tooth, (b) normal, and (c) missing tooth states.

Figure 20

Accuracy and loss curves of Experiment 2: (a) training accuracy; (b) test accuracy; (c) training loss; (d) test loss.

Figure 21

Confusion matrix for MSCNN-LSTM-CBAM-SE fault classification in Experiment 2.

Figure 22

Diagnostic results of six methods in Experiment 2.

Figure 23

Feature distribution visualization via T–SNE in Experiment 2: (a) original test set; (b) CNN1–CBAM–SE; (c) CNN2–CBAM–SE; (d) CNN1–LSTM–CBAM–SE; (e) CNN2–LSTM–CBAM–SE; (f) MSCNN–LSTM–CBAM–SE (w/o BN); (g) MSCNN–LSTM–CBAM–SE.

Figure 24

Ablation study comparison chart.