Physically-Constrained Mamba-SDE for Remaining Useful Life Prediction under Irregular Observations
Researchers have developed a new framework called PC-MambaSDE to improve the prediction of remaining useful life (RUL) for industrial machinery, especially when sensor data is irregular or missing. This model integrates physical constraints into its continuous-time dynamics, ensuring that predicted degradation trajectories are physically plausible and adhere to the irreversible nature of damage accumulation. The framework uses a mask-aware encoder to handle observation gaps and a physics-guided latent SDE to enforce monotonic degradation, outperforming existing methods in experiments, particularly under severe data scarcity. AI
IMPACT Enhances predictive maintenance capabilities by providing more reliable RUL predictions from irregular sensor data.