整星隔振用磁流变阻尼器神经网络模型

NEURAL NETWORK MODEL OF MAGNETORHEOLOGICAL DAMPER FOR VIBRATION ISOLATION PLATFORM OF WHOLE-SPACECRAFT

为改善星箭界面振动环境,设计六杆隔振平台,采用磁流变阻尼器作为半主动控制元件,替代原有锥壳过渡支架.对整星隔振平台用磁流变阻尼器进行性能测试,得到反映磁流变阻尼器阻尼特性的实验数据.建立具有两个隐含层的反向传播神经网络对阻尼器进行建模,用于预测磁流变阻尼器阻尼特性以及控制系统设计.提出一种串行算法优化网络结构、权值和阈值,保证网络具有较好的泛化能力和稳定性.仿真结果表明,与参数化模型相比,提出的神经网络模型具有较小的训练误差和较强的泛化能力,能够很好地预测阻尼器的阻尼特性.

To improve the interface vibration environment of the satellite and the rocket,a six-pole vibrationisolation platform is built,with the magnetorheological damper as the semi-active control device.The magne-torheological damper for the whole vibration isolation platform is tested to obtain the experimental data forthe damping characteristics of the magnetorheological damper.The BP(back propagation)neural networkwith two hidden layers is established to model the damper for predicting the damping characteristics of themagnetorheological damper and for the design of the control system.A sequential algorithm is proposed tooptimize the network structure,the weight and the threshold to ensure that the network has a better general-ization ability and the stability of the network training.The simulation results show that compared with theparametric model,the proposed neural network model has less training error and higher generalization ability,and can well predict the damping characteristics of the damper.