基于神经网络的六维力传感器静态标定方法研究

利用BP神经网络研究了六维力传感器的静态标定方法.良好的泛化能力是基于神经网络标定方法的关键,而现有基于神经网络的传感器标定方法对网络的泛化能力鲜有考虑.针对上述问题,以提高神经网络泛化能力为目的,对神经网络的训练样本、训练期望误差值和网络隐层单元数等的选取进行了研究.实验及计算机仿真表明,当选取样本数据具有遍历性时,利用神经网络标定后进行补偿计算可提高六维力传感器的测量精度.

By use of neural network,the method of static demarcation of six-dimensional force sensor has been studied.The good capability of generalization is the key to the best use of the demarcation method based on neural network.Yet it is seldom taken into consideration.Thus,in order to develop the capability of generalization,the selection of network sample,expected training error and number of units in hidden layers of network is studied.Experiments and computer simulation show that,when the selected data of sample has ergodicity,.the measuring precision of six-dimensional force sensor can be improved if compensating calculation is made after calibration by means of neural network

国家自然科学基金(50275143)资助课题