基于神经网络的力矩补偿动力学参数辨识

Dynamics Parameter Identification of Torque Compensation Based on Neural Networks

针对机械臂刚柔耦合动力学模型拟合精度不高的问题,在分析动力学参数对各关节力矩影响的基础上,提出了一种基于神经网络力矩补偿的动力学参数辨识方法.首先,对动力学模型进行线性化分析,得到最小惯性参数集,设计机械臂激励轨迹,并采集各关节数据集.其次,搭建神经网络架构,对不同隐藏层数的神经网络模型的训练效果比较,证明了本文模型的准确性.将关节位置、速度、加速度数据集作为网络架构输入,经过神经网络学习后提取出各关节力矩.最后,对算法辨识出来的模型进行验证,以模型预测力矩的均方根误差为评判标准,关节拟合力矩结果表明,本文所用方法相较刚体逆动力学有更好的拟合精度,减少了关节摩擦等非线性因素对辨识实验的影响,得到更精准的动力学模型,对机械臂系统有更好的控制效果.

To address the issue of low fitting accuracy of the rigid flexible coupling dynamic model of a mechanical arm,a dynamics parameter identification method based on neural network torque compensation was proposed by analyzing the effect of dynamic parameters on joint torque.First,the dynamical model was linearized,the minimum inertial parameter was established,the mechanical arm excitation trajectory was optimized,and data sets for each joint were collected.Second,a multilayer neural network architecture was built,and the training effects of the neural network models with different numbers of hidden layers were compared,confirming the accuracy of the proposed model.The joint position,velocity and acceleration data sets were used as the input of the network architecture,the torque was calculated after neural network.Finally,the identification model was verified.The root-mean-square error of the predicted moment was taken as the evaluation standard.The findings of joint fitting torque reveal that the proposed method has better fitting accuracy than inverse dynamics and control effect on the mechanical arm system.Fur-ther,the model reduces the impact of nonlinear factors,such as joint friction,on the identification experiments,obtains a more accurate dynamical model,and enhances precision in the manipulation of the robotic arm.