An Imbalanced Fault Diagnosis Method Based on TFFO and CNN for Rotating Machinery

Abstract

Deep learning-based fault diagnosis usually requires a rich supply of data, but fault samples are scarce in practice, posing a considerable challenge for existing diagnosis approaches to achieve highly accurate fault detection in real applications. This paper proposes an imbalanced fault diagnosis of rotatory machinery that combines time-frequency feature oversampling (TFFO) with a convolutional neural network (CNN). First, the sliding segmentation sampling method is employed to primarily increase the number of fault samples in the form of one-dimensional signals. Immediately after, the signals are converted into two-dimensional time-frequency feature maps by continuous wavelet transform (CWT). Subsequently, the minority samples are expanded again using the synthetic minority oversampling technique (SMOTE) to realize TFFO. After such two-fold data expansion, a balanced data set is obtained and imported to an improved 2dCNN based on the LeNet-5 to implement fault diagnosis. In order to verify the proposed method, two experiments involving single and compound faults are conducted on locomotive wheel-set bearings and a gearbox, resulting in several datasets with different imbalanced degrees and various signal-to-noise ratios. The results demonstrate the advantages of the proposed method in terms of classification accuracy and stability as well as noise robustness in imbalanced fault diagnosis, and the fault classification accuracy is over 97%.

Keywords: continuous wavelet transform; convolution neural network; data expansion; imbalanced data; synthetic minority oversampling technique.

Figure 1

Illustration of the sliding segmentation. It mainly contains four key factors, including window size, sliding step and starting point, and sliding direction.

Figure 2

Illustration of SMOTE algorithm. The blue balls, red asterisks, and black triangles, respectively represent the majority classes, the minority classes, and the generation points.

Figure 3

The architecture of LeNet-5-based CNN. It mainly contains two multiple convolutional, two pooling layers, and two fully connected layers. The time-frequency images are input to the first convolutional layer, and the classification of the output layer is achieved by the softmax function.

Figure 4

The Sigmoid and ReLU activation function.

Figure 5

Imbalanced fault diagnosis flow chart of rotating machinery based on TFFO and CNN. First, the bearing and gearbox raw vibration signals are collected. Second, sliding segmentation is used for repeated sampling, and CWT is applied to generate time−frequency images. Third, the SMOTE is utilized to generate minority samples again. Finally, an improved CNN based on LeNet−5 is established to achieve intelligent fault diagnosis while the features are visualized by t−SNE, and results are displayed by a confusion matrix.

Figure 6

The locomotive bearing test rig. It is from a locomotive depot of the China Railway Administration. It mainly contains a hydraulic system, a spindle box, hydraulic loading, and three accelerometers at different locations.

Figure 7

Different types of defective bearings: (a) F1; (b) F2; (c) F3; (d) F4; (e) F5; (f) F6; (g) F7; (h) F8. The red circle in the figure indicates the location of the defect.

Figure 8

Time domain signal of F1−F8. It mainly contains eight types of fault signals in Table 1.

Figure 9

Time-frequency images of the original samples and generated samples. It mainly contains a healthy-bearing sample and seven fault-bearing samples, and seven generated samples.

Figure 10

Experimental results for balanced bearing Dataset 1, Dataset 2, and Dataset 3: (a) train accuracy; (b) train loss; (c) validation accuracy; (d) validation loss.

Figure 10

Experimental results for balanced bearing Dataset 1, Dataset 2, and Dataset 3: (a) train accuracy; (b) train loss; (c) validation accuracy; (d) validation loss.

Figure 11

The confusion matrix under different datasets: (a) Dataset 1; (b) Dataset 2; (c) Dataset 3.

Figure 12

The visualization by t−SNE of the learned features in the Conv2D layer and Fully connected layer of Dataset 3: (a) layer C1 in the balanced Dataset 3; (b) layer C2 in the balanced Dataset 3; (c) layer F1 in the balanced Dataset 3; (d) layer F1 in the imbalanced Dataset 3.

Figure 13

Accuracy curves of four models with different SNRs: (a) SNR = −4 dB; (b) SNR = −2 dB; (c) SNR = 0 dB; (d) SNR = 2 dB; (e) SNR = 4 dB.

Figure 14

Comparison of the performance of different models with different SNRs. It mainly contains four models, including the proposed method, CWT−CNN, CWT−GAN−CNN, and LSTM−CNN.

Figure 15

The gear test rig, which is from Zhejiang University and primarily contains a motor, three gears, and three accelerometers, and a data acquisition board.