Toward accurate RUL and SoH estimation using reinforced graph-based physics-informed neural networks enhanced with dynamic weights
Researchers have developed a new framework called Reinforced Graph-based Physics-informed Networks with Dynamic Weighting (RGPD) to improve the accuracy of Remaining Useful Life (RUL) and State of Health (SoH) estimations. This model combines data-driven learning with physics-based regularization, adapting its approach using dynamic loss weights to better handle different degradation patterns across various assets. RGPD demonstrated significant improvements in accuracy, reducing RMSE by up to 12 percent and MAPE by 20 percent on benchmark datasets for engine, bearing, and battery degradation. AI
IMPACT This new framework could lead to more reliable industrial operations through improved predictive maintenance capabilities.