Blade-tip vibration informed latent physics residual prognostics for fatigue crack growth and remaining useful life prediction

Fatigue crack growth prognosis of blades remains challenging because crack length is difficult to measure continuously, while vibration responses are influenced by both damage evolution and loading conditions. A blade-tip vibration-informed latent physics residual prognostics (LPRP) framework is proposed for crack-growth modeling and remaining useful life (RUL) prediction. Blade-tip displacement and resonance-frequency degradation are used to characterize loading response and structural degradation, respectively, enabling crack-growth-rate prediction without direct crack-length input. Resonance fatigue tests on four compressor blades show that LPRP reduces the mean LOBO RMSE of logarithmic crack-growth-rate prediction from 0.482 to 0.274 and the macro-averaged RUL RMSE from 1.93 × 1 0 5 to 7.25 × 1 0 4 cycles. The results demonstrate the potential of LPRP for non-contact blade crack-growth prognosis. • A blade-tip vibration informed LPRP framework is developed for crack-growth and RUL prediction. • Latent driving-coordinate calibration and regime-local residual correction enhance cross-blade generalization. • LPRP reduces the mean LOBO crack-growth-rate RMSE by 43.2% and the recursive RUL RMSE by 62.5%.