Physics-guided hybrid learning framework (IMAGE)
Caption
Physics-guided hybrid learning framework. This schematic illustrates a mechanism-data hybrid-driven strategy for predicting the dynamics of multibody systems. Large datasets are used for neural-network fitting, while smaller mechanism-informed datasets introduce governing mechanical equations and kinematic constraints into the training process. By combining data-driven learning with differential-algebraic equations, the framework helps the physics-informed neural network improve prediction accuracy, preserve system constraints and enable faster dynamic simulation for complex mechanical systems.
Credit
Acta Mechanica Sinica
Usage Restrictions
Acta Mechanica Sinica
License
Original content