基于神经网络的电机摩擦力混合模型研究

摘要

In high-precision servo system, friction effect is a nonlinear and nonsmooth phenomenon, thus it is hard to identify by using traditional neural network. This paper proposed that friction effect is interpreted to be consis-ted of a static part and a mutation part. Connecting flexible sigmoid function to Radial basis function ( FtBF) neural network, a hybrid friction model based on neural network was constructed for motors servo system. Results of simula-tion show that, compared with RBF neural network, the proposed model is of high precision responding to changing input signal, thus the effectiveness of ihe proposed model is demonstrated.%在电机伺服系统优化建模的研究中,要求高精度伺服系统.由于系统摩擦力具有强非线性和非光滑特性,传统的神经网络无法进行有效辨识.将非线性摩擦特性理解成为由稳态部分和突变部分串联构成,以电机伺服系统为对象,引入柔性sigmoid函数描述非线性摩擦特性中的突变部分,并与传统的RBF神经网络串联,构造出描述非线性摩擦特性的神经网络混合模型.仿真结果表明,与传统的RBF神经网络辨识方法相比,模型在输入变化响应下均具有较高的模型精度,从而验证了建模方法的有效性.