基于堆叠学习的六维力传感器解耦方法研究  

Research on Decoupling Method of Six-axis Force Sensor Based on Stacking Learning

作　　者：戴裕强 王鹏 张昌明[1,2,3] 李威 石宇航 吴佳敏 付麒瑞 孙茹雪 DAI Yuqiang;WANG Peng;ZHANG Changming;LI Wei;SHI Yuhang;WU Jiamin;FU Qirui;SUN Ruxue(School of Mechanical Engineering,Shaanxi University of Technology;Shaanxi Key Laboratory of Industrial Automation;Engineering Research Center of Manufacturing and Testing for Landing Gear and Aircraft Structural Parts,Universities of Shaanxi Province)

机构地区：[1]陕西理工大学机械工程学院 [2]陕西省工业自动化重点实验室 [3]起落架及飞机结构件加工检测陕西省高校工程研究中心

出　　处：《仪表技术与传感器》2025年第3期20-24,共5页Instrument Technique and Sensor

基　　金：陕西省重点研发计划(2024GX-YBXM-198);陕西省重点产业链项目(2023-ZDLGY-28);陕西省秦创原“科学家+工程师”队伍项目(2022KXJ-139);陕西省教育厅青年创新团队项目(23JP025)。

摘　　要：针对线性解耦算法精度差、改进算法或部分融合算法无法满足指尖六轴力/扭矩传感器静态校准数据高精度解耦需求的问题,提出了一种堆叠集成学习模型以同时提高6个维度的解耦精度并减少串扰。该模型将极端随机树、多层感知机神经网络、装袋集成学习、投票法和自适应提升算法作为元学习器,最小二乘线性回归模型作为集成模型。实验结果表明:堆叠学习模型有效地结合了5种算法的解耦特点,6个维度的解耦精度提高了20.7%~95.1%。To address the issues of poor accuracy in linear decoupling algorithms and the inability of improved or partially integrated algorithms to meet the high-precision decoupling requirements for the static calibration data of fingertip six-axis force/torque sensors,a stacking ensemble learning model was proposed to simultaneously enhance decoupling accuracy across all six dimensions and reduce crosstalk.The proposed model employed extremely randomized trees,multilayer perceptron neural networks,bagging ensemble learning,voting methods,and adaptive boosting as base learners,with a least squares linear regression model serving as the integration model.Experimental results demonstrate that the stacking learning model effectively integrates the decoupling characteristics of five algorithms,leading to an improvement in decoupling accuracy across all six dimensions,ranging from 20.7%to 95.1%.

关 键 词：六维力传感器 集成学习 堆叠学习 解耦方法 

分 类 号：TH823[机械工程—仪器科学与技术]