论文部分内容阅读
提出多层前向神经网络求解6自由度并联机构位置正解的方法,将位置反解结果作为训练样本,采用Levenberg-Marquardt训练方法,实现了机构位置从关节变量空间到工作变量空间的非线性映射,从而求得并联机构运动学正解值。结果表明:与数值分析法相比,该方法计算精度高、耗时少、计算过程简洁,可应用于该机构的任务空间实时控制或求解机构的工作空间。
A multi-layer forward neural network is proposed to solve the 6-DOF parallel mechanism position positive solution. The position inverse solution is used as the training sample. The Levenberg-Marquardt training method is used to realize the nonlinear mapping of the mechanism position from the joint variable space to the working variable space , So as to obtain parallel mechanism kinematics positive solution value. The results show that compared with the numerical analysis method, this method has the advantages of high precision, less time consuming and simple calculation process. It can be applied to the task space real-time control or solution organization’s work space.