摘要
获取执行环节的模型是进行系统分析、预测和控制器设计的基础。该文针对工业中常用的磁粉制动器进行了深入研究,通过分析其工作机理,指出转矩控制非线性因素和不确定因素的来源;进而采用神经网络辨识的方法获取磁粉制动器激磁电流与制动扭矩之间的映射关系。为了避免参数陷入局部极小并提高学习效率,引入误差饱和抑制函数来防止网络学习的早熟。实验结果表明,本文所设计的神经网络辨识器辨识精度较高,弥补了此前文献中近似拟合的不足,为随后设计随动负载模拟器中的摩擦阻力矩控制器奠定了良好基础。
In general,the first and the most important step in system analysis,prediction and controller design is to acquire the proper model of the drive element. In order to design the force controller in the load simulator,the magnetic particle brake （MPB） is studied in this paper,which has widespread used in industry field. Through analyzing the working principle of MPB,the origin of nonlinear and uncertain factors are indicated,and then obtain the map function between exciting current and brake torque of MPB by using the developed iden-tifier of neural networks. In order to avoid the learning process traps into local minimum and improve the learning efficiency,an error saturation prevention （ESP） function is proposed to prevent the premature saturation（PS） phenomenon. Simulation and experimental results show that the developed identifier has high precision,not only cover the shortage of approximate fitting in former thesis,but also lays a solid foundation for the following design of frictional resistance moment controller,which in the load simulator。
出处
《电气自动化》
2010年第5期55-58,共4页
Electrical Automation
关键词
磁粉制动器
建模
神经网络
误差饱和抑制
magnetic particle brake modeling neural networks error saturation prevention
