融合BO-CNN-BiLSTM的压电式六维力/力矩传感器非线性解耦

打开文本图片集

关键词：六维力传感器；静态解耦；贝叶斯优化；卷积双向长短期记忆网络中图分类号：TP183；TP212.9 文献标识码：Adoi：10.37188/OPE.20253323.3702 CSTR：32169.14.OPE.20253323.3702

Abstract： To address the degradation of force measurement performance caused by interdimensional coupling in piezoelectric six-dimensional force/torque sensors，an integrated decoupling algorithm （BO-CNNBiLSTM） combining Bayesian optimization （BO），convolutional neural networks （CNN），and bidirectional long short-term memory networks（BiLSTM） is proposed. In this algorithm，CNN is first employed to enhance the extraction of spatial coupling features from six-dimensional force signals. BiLSTM is then utilized to exploit bidirectional temporal modeling capabilities and dynamically capture cross-dimensional time-domain dependencies of the loads.Subsequently，BO is introduced to achieve adaptive global optimization of hyperparameters. In this way，the limitations of traditional decoupling methods in terms of real-time performance，generalization ability，and physical consistency are efectively overcome. The proposed BO-CNN-BiLSTM algorithm eliminates the empirical dependence on manually tuned parameters in conventional approaches and enables adaptive modeling of the nonlinear characteristics of sensors.Experimental results demonstrate that the maximum nonlinear error and cross-coupling eror of the six-dimensional force/torque sensor outputs are 0.87% and 0.52% ，respectively. The BO-CNN-BiLSTM decoupling algorithm efectively reduces both intra-dimensional and interdimensional coupling in six-dimensional force sensors，significantly improving measurement accuracy and providing important support for anthropomorphic motion control and environmental interaction in humanoid robots.

Key Words ： six-dimensional force sensors； static decoupling; Bayesian optimization; CNN-BiLSTM

1引言

作为高端装备与智能系统实现高精度力感知的核心器件1，压电式六维力/力矩传感器（以下简称六维力传感器）在大型工业机器人[2、航空航天装备[3]等领域已彰显其不可替代性[4]。（剩余13940字）