卷积神经网络的光电编码器误差补偿  

Error compensation of photoelectric encoder based on convolutional neural network

作　　者：魏艳平[1] 肖海柳[1] WEI Yanping;XIAO Hailiu(Nanchang Institute of Science and Technology,Nanchang 330108,China)

机构地区：[1]南昌工学院人工智能学院,南昌330108

出　　处：《激光杂志》2022年第8期174-179,共6页Laser Journal

基　　金：2020年江西省教育厅科学技术研究项目(No.GJJ161229);南昌工学院科技项目(No.NGKJ-19-05)。

摘　　要：光电编码器在各个领域应用广泛,测量精度会直接影响对应系统的运行情况,因此,研究卷积神经网络的光电编码器误差补偿方法。构建具有预测能力和训练能力的卷积神经网络模型,为使该卷积神经网络的层数降低,提高模型压缩率并节省计算时间,混合量化卷积神经网络各层权重,获得不同精度的权重结果,得到改进后的卷积神经网络模型;模型输入为采样点位置的光栅角度值,高精度检测仪的检测结果作为输出,通过输入结果与输出结果之间的误差补偿混合量化后的权重,提高检测结果精度实现光电编码器误差补偿。经验证,该方法模型学习和训练过程中能够具备较强泛化能力,补偿光电编码器误差后使得编码器精度比原精度高出3倍,与同类方法相比能够更加有效降低光电编码器误差。The photoelectric encoder is widely used in various fields,and the measurement accuracy will directly affect the operation of the corresponding system,so the error compensation of the photoelectric encoder of the convolution neural network is studied.In order to reduce the number of layers,improve the compression rate of the convolution neural network and save the calculation time,the weight of each layer of the convolution neural network is mixed,and the improved convolution neural network model is obtained.It is proved that the model can have strong generalization ability in the process of learning and training,and the accuracy of encoder is at least 3 times higher than that of original encoder after compensating the error of photoelectric encoder.Compared with the same method,the error of photoelectric encoder can be reduced more effectively.

关 键 词：卷积神经网络 光电 编码器 误差 补偿 权重 

分 类 号：TN248.1[电子电信—物理电子学]