Physics-informed machine learning for predicting spring remaining servicing life

Unexpected fracture of tension springs in high-voltage switchgear poses serious reliability risks. A physics-informed machine learning framework is proposed, which uses Vickers hardness—reflecting stress-induced work hardening—as a direct health indicator. A dataset of alloy composition, service duration, and hardness measurements from high-manganese and stainless-steel springs was cleansed via the interquartile-range method and enhanced with physics-guided features (including Mn×Ni interaction). A two-layer stacking ensemble model accurately predicts hardness evolution, laying the foundation for quantitative remaining useful life estimation.