基于RBF神经网络的关节转角误差补偿

Error Compensation of Joint Angles Based on RBF Neural Networks

关节转角误差对关节臂式坐标测量机的精度有非常重要的影响,影响关节转角误差的因素众多,难以用准确的数学模型来描述,为此提出一种采用三坐标测量机标定关节转角误差、基于径向基函数(Radial basis function,RBF)神经网络进行关节转角误差补偿的方法。应用该方法对关节臂式坐标测量机6个关节的转角误差进行离散标定,标定数据训练各关节的RBF神经网络,使用经过训练的RBF神经网络分别对6个关节进行转角误差补偿。试验结果表明经过补偿后关节转角精度提高了约两个数量级,基于RBF神经网络的补偿效果优于正弦函数补偿模型,且其适用范围更广,具有很强的工程应用价值。

Joint angle errors have important influence on the accuracy of articulated arm coordinate measuring machines （AACMMs）.There are many factors causing joint angle errors,so it is difficult to describe them accurately with mathematic models.A compensation method based on radial basis function （RBF） neural networks is presented to solve the problem.A practical method to calibrate errors of joint angles with a coordinate measuring machine is designed,thereafter the angle errors of six joints of an AACMM are calibrated discretely.The calibration data is used to train the RBF neural networks,after training the RBF neural networks are used to compensate the angle errors of six joints respectively.Experimental tests are carried out to evaluate the efficiency of the RBF neural network compensation method,showing that after compensation the rotation accuracy of the joints angle is improved greatly,and the RBF neural network compensation method is more efficient than a compensation model of sine function,therefore,it is valuable for engineering applications.