改进型BP神经网络的3-UPS-RCR并联机构位置正解  

Forward Kinematics Solution Analysis of 3-UPS-RCR Parallel Mechanism Based on Improved BP Neural Network

作　　者：林杰克 施光林[1] 汤澍 LIN Jieke;SHI Guanglin;TANG Shu(School of Mechanical Engineering,ShanghaiJiaotong University,Shanghai 200240,China)

机构地区：[1]上海交通大学机械与动力工程学院,上海200240

出　　处：《机械设计与研究》2020年第3期30-34,49,共6页Machine Design And Research

摘　　要：针对立式旋压机的加工要求并结合并联机构的结构特点,提出利用3-UPS-RCR并联结构实现立式旋压加工操作。用Adams建立结构模型并联合Matlab实现并联机构动态仿真。基于加工轨迹利用动态仿真获得大量BP神经网络并联机构正解学习样本,并在Python环境下进行神经网络的搭建和训练。通过动量参数更新法和位移补偿法改进BP神经网络,并将结果和传统的梯度下降法进行比较,发现动量参数更新法能加快神经网络的训练速度,位移补偿法能进一步提高神经网络的拟合精度。3-UPS-RCR parallel structure is proposed to realize the vertical spinning operation according to the spinning processing requirements and the structural characteristics of the parallel mechanism.Combine Adams with Matlab to achieve the dynamic simulation of parallel mechanism,which can help to get large number of learning samples under the target processing track.Use Python to build and train the neural network.Improve the neural network by using momentum parameters update algorithm and displacement compensation algorithm.Comparing the improved algorithm with the traditional gradient descent method,it is found that the momentum update algorithm can accelerate the training speed of the neural network and the displacement compensation algorithm can further improve the accuracy of the fitting accuracy.

关 键 词：并联机构 运动学正解 BP神经网络 动量参数更新法 位移补偿法 Adams/Matlab Python 

分 类 号：TH112[机械工程—机械设计及理论] TP242[自动化与计算机技术—检测技术与自动化装置]