Fig 1.
Schematic diagram of sample slicing for multidimensional historical health index sequence data.
Fig 2.
Framework of the CNN-transformer model and sequential evaluation strategy for RUL prediction.
Fig 3.
Schematic diagram of the multi-head latent attention mechanism.
Fig 4.
Workflow diagram of the sequential evaluation method.
Fig 5.
C-MAPSS turbofan engine model schematic and module interconnection diagram.
Table 1.
Specific introduction to the dataset.
Table 2.
Impact of different denoising methods on prediction performance (FD001).
Fig 6.
Data visualization for engine No. 9 in the FD001 dataset.
Fig 7.
Data visualization for engine No. 2 in the FD002 dataset.
Table 3.
Effective data size of each subset after data preprocessing.
Table 4.
RMSE results comparison.
Table 5.
R2 results comparison.
Table 6.
MAE results comparison.
Fig 8.
Comparison of RMSE and R2 results.
Table 7.
Comparison of model computational efficiency.
Table 8.
Performance comparison with graph neural network models (FD001).
Table 9.
Evaluation results of the sequential HI evaluation strategy.
Fig 9.
Health index value distribution comparison for FD001 and FD003 subsets.
Table 10.
Impact of different α values on health index quality (FD001).
Fig 10.
Health index extraction results for FD001 and FD003.
Fig 11.
Health curves for engine No. 9 in FD001 and engine No. 63 in FD003.
Table 11.
Ablation study results on model architectures (mean ± standard deviation).
Table 12.
Ablation study results of health index evaluation strategies (FD001).