基于神经网络的悬架试验系统研究

针对现有悬架设计及实验不一致问题以及基本型Elman网络忽略了输出层节点的反馈,只能满足一阶线性动态系统信号处理,而不能满足多层网络、多阶系统的需求,提出了一种改进型Elman网络;改进后的网络增强了关联层以及输出层的反馈,把反馈增益作当作连接权值来实施网络训练,训练后的网络不仅比例系数和积分系数具有时变性,还具有自适应性强、学习效率高、逼近精度高等特点。建立了基于改进型Elman神经网络PAC控制器,对六自由度悬架试验平台系统进行控制研究,分析了悬架阻尼、非悬挂质量、悬架刚度、轮胎刚度等参数对时域内轮荷利用率和频域内相位角的影响,通过整车实验证明:悬架参数的匹配可以有效地改善车身振动,降低悬架动挠度、轮胎动载荷,提高乘坐舒适性和车辆行驶安全性,协调整车综合性能。

Suspension test system based on the neural network

According to the difference of suspension design and experiment, the basic Elman network ignores output layer nodes feedback, and only the first order linear dynamic signal processing system is meet, which can''t meet multi network and multi order system. An modified Elman neural network is presented. The improved network enhances the correlation layer and the feedback of output layer, and the feedback gain is taken as the connection weight to train the network, which not only has time varying of proportion coefficient and integral coefficient, but also has strong adaptability, high learn efficiency and approximation accuracy. PAC controller based on modified Elman neural network is established, and the 6 DOF suspension system is used to test the platform control. The suspension damping, un sprung mass, suspension stiffness, tire stiffness parameters are analyzed, the effect on utilization rate of tire wheel load and phase angle is discussed. Suspension parameter optimization design is provided.