计算力学学报2025，Vol.42Issue(5)：729-736,8.DOI:10.7511/jslx20240708001
基于物理信息神经网络的多点摩擦诱发粘-滑振动问题算法
Physics-informed neural networks algorithm for sloving multi-point friction-induced stick-slip vibration problems
摘要
Abstract
Addressing the challenge of accurately solving unstable stick-slip vibration problems in non-smooth dynamics,this paper proposes a solution algorithm based on Physics-informed Neural Networks(PINN).Firstly,the classical stick-slip vibration problem is dynamically modeled using the linear complementarity theory under unilateral constraints.Then,the linear complementarity relationship is designed as a loss function to guide the training of the neural network,constructing a PINN algorithm for solving multi-point friction-induced stick-slip vibration problems.The accurate simulation of complex responses of multiple sliders' stick-slip vibrations in frictional systems is conducted.By comparing the numerical results with the Switching Model method that includes event detection and the traditional Time-Stepping method without event detection,the accuracy of the PINN algorithm is verified.The proposed PINN algorithm transforms the traditional optimization problem calculation into network training of the machine learning algorithm,making it suitable for stick-slip vibration analysis with multiple contact points.This method achieves accurate nonsmooth state transitions and provides a convenient and easy-to-use new approach for the accurate simulation of complex nonlinear vibration responses in multi-degree-of-freedom frictional systems.关键词
物理信息神经网络/摩擦自激振动/非光滑动力学/粘-滑振动/线性互补问题Key words
physical-informed neural networks/friction-induced vibration/nonsmooth dynamics/stick-slip vibration/linear complementarity problem分类
力学引用本文复制引用
张非凡,李姿琳,白金帅,王伟,卫洪涛,卫荣汉..基于物理信息神经网络的多点摩擦诱发粘-滑振动问题算法[J].计算力学学报,2025,42(5):729-736,8.基金项目
国家自然科学基金青年基金(12402242) （12402242）
河南省高等学校重点科研项目(24A130002)资助. （24A130002）
