|
| 基于深度学习的摩擦电法向切向载荷解耦测量方法 |
| Triboelectric decoupling measurement of normal and shear loads based on deep learning |
| 投稿时间：2024-02-19 修订日期：2024-06-29 |
| DOI： |
| 中文关键词: 摩擦电 法向力 切向力 测量 深度学习 |
| 英文关键词:triboelectric normal force shear force measurement deep learning |
| 基金项目:国家自然科学基金项目（面上项目，重点项目，重大项目） |
|
| 摘要点击次数: 29 |
| 全文下载次数: 0 |
| 中文摘要: |
| 基于摩擦电的多向力柔性薄膜传感器件不依赖于外部供电,在机器人触觉感知等领域应用中具有微型化、便携性、持续性等优势。然而,现有的标定方法主要依赖力-摩擦电特性曲线的线性区间来实现法向切向载荷的解耦表征,面临线性区间较短的问题,限制了测量范围。本文提出了一种基于深度学习的摩擦电法向切向载荷解耦测量方法。首先,研制了基于差分阵列的摩擦电多向力柔性薄膜传感器件,采用自研的多方向力电测试台开展了力-摩擦电标定实验。随后,采用卷积神经网络方法学习原始的力-摩擦电标定数据,对法向切向载荷进行解耦预测,特别是针对极端数据引入了注意力机制。结果表明,相较于传统标定方法,卷积神经网络方法将法向切向载荷测量范围从0 N ～ 8 N拓展到0 N ～ 28 N,外力倾角测量范围从0° ～ 30°拓展到0° ～ 50°,法向切向载荷预测均方根误差分别为1.0851和0.7448。 |
| 英文摘要: |
| Flexible thin film sensors of multidirectional force measurements based on triboelectricity is free of external power supply, and show the advantages of miniaturization, portability, and sustainability in applications such as robotic tactile perception. However, existing calibration methods mainly rely on the linear interval of the force-triboelectric characteristic curve to realize the decoupled characterization of normal and shear loads, facing the challenge of a short linear interval to limit the measurement range. In this study, a triboelectric decoupling measurement of normal and shear loads based on deep learning is proposed. A triboelectric flexible thin-film sensor with a differential-mode array is designed and fabricated, and its force-triboelectric calibration experiment is performed on a self-built multidirectional force-electric test rig. A convolutional neural network is used to learn the raw force-triboelectric data so as to achieve a decoupled prediction of normal and shear loads, especially with the introduction of an attention mechanism for some extreme data. In comparison to the traditional calibration method, the convolutional neural network method extends the normal and shear load measurement range from 0 N ～ 8 N to 0 N ～ 28 N, and the force inclination measurement range from 0° ～ 30° to 0° ～ 50°, with the root mean square errors of normal- and shear-load predictions at 1.0851 and 0.7448, respectively. |
| 下载PDF阅读器 |
| 关闭 |
|
|
|
|
|
