【】 Full article: (Authored by Shen Liu and Fei Meng, from University of Shanghai for Science and Technology, China.)

Rubber supports are essential in automotive, heavy machinery, and aerospace engineering. They offer excellent hyper elasticity, viscoelastic dissipation, and noise reduction. However, their fatigue evolution under coupled thermo-mechanical loading is exceptionally complex. This study develops an LSTM-Physics-Informed Neural Network (PINN) framework that integrates prior physical knowledge transfer with Partial Differential Equation (PDE) constraints, to address the challenge of predicting the fatigue life of rubber_isolators under thermo-mechanical-damage coupling.

Abstract

Rubber supports are ubiquitous in modern vibration isolation systems. Their fatigue evolution under coupled thermo-mechanical loading is exceptionally complex. Traditional life prediction methods rely heavily on empirical formulas. These methods often lack accuracy and extrapolation capabilities under varying temperatures. To address this, we propose a novel LSTM-PINN architecture. This framework integrates physical constitutive relations and temperature effects into a neural network. We used transfer learning to extract baseline physical data across wide temperature ranges. Long Short-Term Memory (LSTM) layers capture sequential loading features. We embedded partial differential equations (PDEs) into the loss function. These PDEs are based on strain energy density (SED) and Arrhenius thermodynamics. This approach ensures strict adherence to physical laws. Results demonstrate that LSTM-PINN achieves high precision even with small datasets. It also exhibits superior out-of-distribution (OOD) generalization. This framework provides a new paradigm for evaluating the reliability of rubber components.

Rubber Isolator, Fatigue Life, PINN, LSTM, Thermo–Mechanical Coupling