【摘要】 凿岩机器人的结构具有重载荷、大跨度和关节冗余等特点,关节运动时会产生较大的非几何误差。通过简化结构建立凿岩机器人DH正运动学模型,在正运动学的基础上建立机械臂末端运动轨迹关于时间、空间的特征工程。基于LSTM神经网络方法进行数据挖掘,通过对数据归一化处理,设置交叉验证,进行dropout和L2正则化操作,提高LSTM神经网络算法的性能,减少过拟合问题,实现凿岩机器人末端定位的误差补偿。试验数据表明,基于LSTM误差补偿模型可将凿岩机器人机械臂末端两个观测位置的绝对误差分别降低64.13%和63.05%。更多还原

【Abstract】 The structure of rock drilling robot has the characteristics of heavy load, large span and redundant joint, and the joint movement will produce large non-geometric error. By simplifying the structure, the DH forward kinematics model of the rock drilling robot is established, and the characteristic engineering of the end motion trajectory of the manipulator with respect to time and space is established on the basis of the forward kinematics. Based on LSTM neural network method for data mining, through data normalization processing, cross validation is set, dropout and L2 regularization operations are performed, the performance of LSTM neural network algorithm is improved, over fitting problemsare reduced, and error compensation of end positioning of rock drilling robot is realized. The test data shows that the absolute errors of two observation positions at the end of the robot arm can be reduced by 64.13% and 63.05% respectively based on the LSTM error compensation model.更多还原