Abstract: Aiming at the forward kinematics of 3-PPR parallel mechanism,the results of inverse kinematics were utilized combining with Levenberg-Marquardt training method.By adopting BP neural network and improved BP neural network in succession,the nonlinear mapping from joint-variable-space to operation-variable-space of the mechanism was accomplished,and the forward kinematics was achieved.To improve the accuracy of positive solution,a displacement compensation algorithm was put forward to optimize the BP neural network.By applying this method in the mechanism with MATLAB tool,the results illustrated that the precision of positive solution was enhanced from level 10-3 to level  10-6 by iterative computations with 2.17 milliseconds,and the effectiveness as well as the correctness of  proposed algorithm were proved.Thus the high accuracy and real-time controls of the 3-PPR parallel mechanism were realized.

Key words: parallel mechanism, forward kinematics, displacement compensation algorithm, BP neural network

摘要： 针对3-PPR并联机构的正运动学问题,利用运动学逆解结果,结合Levenberg-Marquardt训练方法,先后采用BP神经网络和改进型BP神经网络完成了该机构位姿从关节变量空间到工作变量空间的非线性映射,从而求得其运动学正解。为进一步提高正解精度,提出一种位移补偿算法对BP神经网络进行优化。将所提方法应用于该机构,并利用MATLAB进行求解,结果显示经过2.17ms的迭代计算,正解结果的精度由10－3级提高到10－6级,从而验证了该算法的有效性和正确性,实现了3-PPR并联机构位姿的高精度和实时性控制。

关键词: 并联机构, 运动学正解, 位移补偿算法, BP神经网络