Deep belief network-augmented adaptive direct simulation Monte Carlo for performance prediction of particle dampers

Abstract

Particle damping technology has been extensively utilized in aerospace, mechanical, and civil engineering fields due to its high energy dissipation efficiency, wide frequency adaptability, and robust performance under severe conditions. However, its complex nonlinear dynamic characteristics make traditional discrete element methods challenging in balancing computational efficiency and accuracy. To address this issue, this study proposes a direct simulation Monte Carlo method incorporating an adaptive stochastic collision handling algorithm and a partitioned network simulation framework. These enhancements substantially improve the computational efficiency and accuracy of particle dampers in high-density systems. Building upon this foundation, a simulation data-driven prediction framework is developed by integrating deep belief networks and their optimized variants to efficiently evaluate the performance of particle dampers and analyze the influence of key parameters on the vibration reduction effects. The obtained results reveal that several vital factors, including shell stiffness, damping, particle quantity, particle size, and excitation frequency, significantly affect the performance of particle damping. Furthermore, optimizing particle parameters and implementing high-frequency excitation conditions can substantially enhance the effectiveness of the vibration control. In addition, the whale optimization algorithm-deep belief network model exhibits superior performance in prediction accuracy, generalization ability, and computational efficiency, providing strong support for the rapid optimization design of complex nonlinear systems.