基于视觉标定与BiLSTM-RF的FBG形状传感末端误差补偿方法
DOI:
CSTR:
作者:
作者单位:

华北电力大学

作者简介:

通讯作者:

中图分类号:

xx

基金项目:


End Error Compensation Method for FBG Shape Sensing Based on Visual Calibration and BiLSTM-RF
Author:
Affiliation:

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    针对光纤布拉格光栅(Fiber Bragg Gating,FBG)形状传感中因传统积分递推导致的末端误差累积问题,本文提出了一种基于视觉标定与双向长短期记忆网络(Bidirectional Long Short-Term Memory,BiLSTM)及随机森林(Random Forest,RF)的末端误差补偿方法。该方法以视觉图像处理与中心骨架提取技术获取的形变坐标为基准,利用BiLSTM网络对沿光纤轴向应变序列进行双向建模,提取光栅节点间的空间相关性及轴向误差传递特征。然后,结合随机森林的非线性回归与抗噪能力,建立原始应变数据到末端位置偏差之间的精确映射模型,对传统Frenet-Serret几何重构的末端位置进行了非线性补偿与修正。实验结果表明,该方法有效克服了递推过程中的误差累积效应,将传统Frenet标架法的均方根误差(RMSE)和平均绝对误差(MAE)分别由1.585 mm和1.549 mm降低至0.172 mm和0.097 mm,显著提升了传感器的末端定位精度。

    Abstract:

    To address the end-point error accumulation caused by recursive integration in Fiber Bragg Grating (FBG) shape sensing, this paper proposes a compensation method combining visual calibration, a BiLSTM network, and Random Forest (RF). Taking deformation coordinates obtained via visual image processing and centerline skeleton extraction as a reference, the BiLSTM network bidirectionally models the axial strain sequence along the fiber to extract spatial correlations among grating nodes and axial error propagation features. Subsequently, leveraging the nonlinear regression and anti-noise capabilities of Random Forest (RF), an accurate mapping model between the original strain data and end-point position deviation is established, enabling nonlinear compensation and correction of traditional Frenet–Serret geometric reconstruction results. Experimental results demonstrate this method effectively overcomes recursive error accumulation. The RMSE and MAE drop from 1.585 mm and 1.549 mm to 0.170 mm and 0.097 mm, respectively, significantly improving the sensor"s end-point positioning accuracy.

    参考文献
    相似文献
    引证文献
引用本文
分享
相关视频

文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2026-07-07
  • 最后修改日期:2026-07-07
  • 录用日期:2026-07-08
  • 在线发布日期:
  • 出版日期:
文章二维码