Author Affiliations
College of Electrical and Information Engineering, Lanzhou Jiaotong University, Lanzhou, Gansu 730070, Chinashow less
Fig. 1. Structure of the CNN
Fig. 2. Flow chart of the model
Fig. 3. CWRU rolling bearing data acquisition system
[22] Fig. 4. Ten kinds of simulation signal diagrams. (a) Normal signal; (b) OR 0.1778 nm; (c) OR 0.3556 mm; (d) OR 0.5334 mm; (e) B 0.1778 nm; (f) B 0.3556 mm; (g) B 0.5334 mm; (h) IR 0.1778 nm; (i) IR 0.3556 mm; (j) IR 0.5334 mm
Fig. 5. Time frequency transformation results of 10 kinds of time domain signals. (a) Normal signal; (b) OR 0.1778 nm; (c) OR 0.3556 mm; (d) OR 0.5334 mm; (e) B 0.1778 nm; (f) B 0.3556 mm; (g) B 0.5334 mm; (h) IR 0.1778 nm; (i) IR 0.3556 mm; (j) IR 0.5334 mm
Fig. 6. Graying results of 10 kinds frequency images. (a) Normal signal; (b) OR 0.1778 nm; (c) OR 0.3556 mm; (d) OR 0.5334 mm; (e) B 0.1778 nm; (f) B 0.3556 mm; (g) B 0.5334 mm; (h) IR 0.1778 nm; (i) IR 0.3556 mm; (j) IR 0.5334 mm
Fig. 7. Experimental results of two activation functions
Fig. 8. Experimental results of different networks
Fig. 9. Performance comparison of 4 types of networks
Type | Bearing condition | Fault diameter /mm | Data length | Number of samples | Load | Speed /(r·min-1) |
---|
1 | norm | 0 | 1024 | 2000 | 2 | 1797 | 2 | slight damage to inner ring | 0.1778 | 1024 | 2000 | 2 | 1797 | 3 | moderate damage of inner ring | 0.3556 | 1024 | 2000 | 2 | 1797 | 4 | severe damage of inner ring | 0.5334 | 1024 | 2000 | 2 | 1797 | 5 | slight damage of rolling element | 0.1778 | 1024 | 2000 | 2 | 1797 | 6 | moderate damage of rolling element | 0.3556 | 1024 | 2000 | 2 | 1797 | 7 | rolling weight injury | 0.5334 | 1024 | 2000 | 2 | 1797 | 8 | slight damage of outer ring | 0.1778 | 1024 | 2000 | 2 | 1797 | 9 | slight damage of outer ring | 0.3556 | 1024 | 2000 | 2 | 1797 | 10 | slight damage of outer ring | 0.5334 | 1024 | 2000 | 2 | 1797 |
|
Table 1. Experimental data
Fault location | Fault diameter /mm | Code name | Label |
---|
None | 0 | a | 100000000 | Rolling element+inner ring | 0.1778 | b | 010000000 | Rolling element+outer ring | 0.3556 | c | 001000000 | Rolling element+outer ring | 0.5334 | d | 001000000 | Inner ring | 0.1778 | e | 000100000 | Inner ring | 0.3556 | f | 000010000 | Inner ring | 0.5334 | g | 000001000 | Outer ring | 0.1778 | i | 000000100 | Outer ring | 0.3556 | j | 000000010 | Outer ring | 0.5334 | k | 000000001 |
|
Table 2. Fault code
Learning rate | Accuracy /% | Training time /s |
---|
1 | 69.64 | 134 | 0.1 | 34.42 | 128 | 0.05 | 73.03 | 132 | 0.005 | 99.87 | 127 | 0.001 | 91.01 | 144 |
|
Table 3. Effects of different learning rates on network performance
Batchsize | Accuracy /% | Training time /s |
---|
8 | 99.81 | 226 | 16 | 99.62 | 141 | 32 | 91.38 | 237 | 64 | 88.76 | 124 | 128 | 80.52 | 138 |
|
Table 4. Effects of different Batchsize on network performance
Convolution kernel size | Accuracy /% | Training time /s |
---|
3×3 | 98.7 | 84 | 5×5 | 93.2 | 98 | 8×8 | 92.4 | 79 | 12×12 | 91.8 | 77 |
|
Table 5. Effects of different convolution kernel on network performance
Network layer type | Specific parameter | Network layer output |
---|
Input | input of time-frequency diagram | 32×32 | C1 | 3×3 convolution kernels(32), in steps of 1 | 30×30×32 | S1 | maximum pool 2×2 cores, in steps of 1 | 15×15×32 | C2 | 3×3 convolution kernels(32), in steps of 2 | 13×13×32 | S2 | maximum pool 2×2 cores, in steps of 2 | 7×7×32 | FC | 1568 nodes | 1×1568 | Softmax | 10-classification | 1×10 |
|
Table 6. Structural parameters of the CNN