A CNN–LSTM–Attention Model for Fatigue-State Assessment of Helicopter Rotor Using FBG Sensors

Accurate fatigue state assessment of helicopter rotors is essential for structural health monitoring, enabling early identification of degradation and improved operational safety. However, existing fatigue assessment methods often suffer from insufficient nonlinear modeling capability, heavy reliance on simulated data, and limited treatment of temporal dynamics and long-range dependencies across the fatigue assessment. To address these limitations, we propose an optimized hybrid 1D CNN-LSTM-Attention model for fatigue state assessment from fiber Bragg grating strain measurements. A full life-cycle strain-fatigue dataset is established through ground calibration and dynamic flapping fatigue experiments on a full-scale GFRP rotor beam instrumented with multiple FBG channels. The proposed architecture extracts local temporal patterns using 1D CNN, captures long-range dependencies using LSTM, and emphasizes fatigue-critical segments via attention-based reweighting. Experimental results under six-fold leave-one-sensor-out evaluation achieve 95.25% accuracy, 95.52% precision, 95.51% recall, and a 95.50% F1 score, outperforming conventional machine-learning baselines. Ablation studies further verify the contributions of each module. Additional evaluations under sensor-wise and section-wise holdout provide quantitative evidence of robustness to spatial variability within the same rotor-beam design under the tested conditions.