Open Access
Review
Table 7
Comparison of accuracy, robustness, and computational resource requirements of mainstream deep learning models in nuclear power pump fault diagnosis.
| Model type | Highest accuracy (%) | Accuracy degradation under operating condition disturbances (%) | Sample requirement | Computational cost (FLOPs) |
|---|---|---|---|---|
| Basic CNN | 87.2 | Δ-12.5 | 1,500 | 3.2 × 106 [28] |
| Spatiotemporal CNN | 93.5 | Δ-8.3 | 2,200 | 7.1 × 106 [18] |
| Multi-scale Self-Attention | 99.5 | Δ-2.1 | 3,800 | 12.8 × 106 [11] |
| CNN-LSTM Optimization | 98.2 | Δ-3.8 | 2,500 | 9.3 × 106 [16] |
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.