基于有限元和神经网络的杂交建模

针对复杂非线性结构动力学系统提出了一种基于有限元与神经网络相结合的杂交建模方法。依据该方法,首先将系统中的线性结构部分采用有限元建模,非线性或难以机理建模的结构部件采用神经网络描述。其次,再通过力和位移边界联接条件将有限元模型部分和神经网络模型部分结合从而得到整个系统的杂交模型,且杂交模型的物理结构明确,精度较高,网络规模较小。在一非线性隔振系统的杂交建模算例仿真中,用所建杂交模型对正弦及宽带随机激励进行了预测检验分析,结果良好,该杂交建模方法为主体结构为线弹性结构而又包含有强非线性器件的非线性动力学系统提供了一种有效的建模途径。

Hybrid modeling based on finite element method and neural network

For complicated nonlinear structural dynamic systems,a hybrid modeling strategy combining the finite element method and neural network is proposed.The finite element method is used to model linear and elastic structure components,while the dynamic characteristic of nonlinear or hardly modeling components is described by neural network.Two models are integrated into hybrid dynamic model by satisfying the force and displacement consistency conditions at the connections.And the hybrid dynamic model is of distinct physical structure,high modeling precision,small network dimension.In the numerical simulation of hybrid modeling for a nonlinear vibration isolation system,the predictive analysis under sine and wide-band random excitations are conducted by using hybrid model,and the satisfactory simulation results show that the method provided an effective modeling way for complex elastic structures equipped with strongly nonlinear components.