Hysteresis Compensation in Force/Torque Sensors Using Time Series Information

Ryuichiro Koike,Sho Sakaino,Toshiaki Tsuji

doi:10.3390/s19194259

Ryuichiro Koike, Sho Sakaino + Show 1 more

Open Access

https://doi.org/10.3390/s19194259

Copy DOI

Journal: Sensors	Publication Date: Sep 30, 2019
Citations: 7	License type: CC BY 4.0

Affiliation: University of Tsukuba, Saitama University

Abstract

The purpose of this study is to compensate for the hysteresis in a six-axis force sensor using signal processing, thereby achieving high-precision force sensing. Although mathematical models of hysteresis exist, many of these are one-axis models and the modeling is difficult if they are expanded to multiple axes. Therefore, this study attempts to resolve this problem through machine learning. Since hysteresis is dependent on the previous history, this study investigates the effect of using time series information in machine learning. Experimental results indicate that the performance is improved by including time series information in the linear regression process generally utilized to calibrate six-axis force sensors.

Highlights

Force control has long been a fundamental technology for robotics and is still being actively studied
Most of these are designed such that the deformation volume of the strain body and the force have a linear relationship, and this relationship is often estimated through linear regression
As a method to compensate for the creep phenomenon was proposed in Reference [22], and as hysteresis is a more critical issue, this study focuses on hysteresis compensation

Summary

Introduction

Force control has long been a fundamental technology for robotics and is still being actively studied. Force sensors enable the acquisition of force information in a simple manner. They typically face problems of hysteresis, where measurements are affected by previously applied loads. Force sensors acquire the deformation volume of a strain body using sensing elements (e.g., strain gauges [3,4,5], capacitance sensors [6,7], resistive sensors [8] or optical sensors [9,10,11]) and estimate the force applied to the deformation volume. Most of these are designed such that the deformation volume of the strain body and the force have a linear relationship, and this relationship is often estimated through linear regression

Objectives

Methods

Results

Discussion

Conclusion