Table 3.
Average testing accuracy of comparative methods with single sensory data.
| Sensory Data | Model | Feature Learning from Raw Data | Manual Feature Extraction | ||
|---|---|---|---|---|---|
| Time-Domain Features | Frequency-Domain Features | Handcraft Features | |||
| Vibration signal | DCNN | 81.45% | 55.84% | 70.74% | 73.64% |
| BPNN | 42.56% | 55.62% | 69.03% | 72.36% | |
| SVM | 45.11% | 56.35% | 72.23% | 73.86% | |
| Acoustic signal | DCNN | 66.23% | 31.42% | 76.45% | 76.02% |
| BPNN | 19.80% | 35.89% | 76.04% | 75.79% | |
| SVM | 26.54% | 33.62% | 77.36% | 76.32% | |
| Current signal | DCNN | 85.68% | 60.73% | 61.45% | 76.85% |
| BPNN | 52.36% | 60.47% | 61.21% | 76.43% | |
| SVM | 51.64% | 63.74% | 63.53% | 78.76% | |
| Instantaneous angular speed (IAS) signal | DCNN | 90.23% | 75.34% | 84.42% | 88.34% |
| BPNN | 51.37% | 75.36% | 85.22% | 89.82% | |
| SVM | 48.22% | 75.68% | 85.65% | 89.85% | |
DCNN = deep convolutional neural network; BPNN = back-propagation neural networks; SVM = support vector machine.