Development of two-axial strain distribution sensing method using deep learning-based anisotropic electrical impedance tomography(aEIT)

Abstract

In the field of robotics, there is a growing demand for multi-axial strain sensation with high durability and performance. It is able to detect the complex motion of the multi degrees of freedom joint of human and robot through multi-axial strain sensor. In several studies, multi-axial strain sensors based on piezoreistivity were suggested, but those have disadvantages such as low manufacturability and durability. In order to overcome the limitations, studies with Electrical Impedance Tomography(EIT)-based strain sensing method was proposed. EIT is a tomographic imaging method that measures conductivity distribution with only boundary electrodes. In the case of electrodes attached at the boundary of piezoresistive material, the change in conductivity due to mechanical deformation can be measured in the form of a distribution. EIT-based tactile sensing has advantages of high durability due to the location of electrodes and high performance that can detect an arbitrary location of deformation. Anisotropic EIT(aEIT) is a state-of-the-art EIT method that can measure not only the degree of conductivity change but also the direction of conductivity change. This allows the distribution of two-axial strain to be measured. In this study, the performance of the existing aEIT method was quantitatively validated. A novel aEIT method was developed to overcome the limitations. While the existing model is constructed based on linearization assumption, a nonlinear model based on deep learning technology was developed for higher performance. The model constructed through deep learning showed higher performance in reconstructing conductivity tensor distribution than the existing linearization-based model. In addition, research was conducted to incorporate the deep learning model with actual sensing application. First, a method to build a noise-robust deep learning model was developed. Also, a platform to measure the conductivity distribution in real time by applying the deep learning model was developed. The demonstration confirmed that the platform operated successfully by using EIT model and a two-axial strain sensor that fabricated with piezoresistive fabric.