Coupled Electro-Mechanical Modeling in Electric Impedance Tomography (EIT) for Tactile Sensors

  • Mohamed Elkhodbi

Student thesis: Master's Thesis

Abstract

This research introduces Finite Element models that can be used to solve the forward EIT problem while simultaneously acquiring a detailed mechanical response of the system. Additionally, this research introduces different neural networks structures that can predict the haptic response of the system in terms of location and various mechanical behavior. Those neural networks are trained on a dataset obtained through solving randomized scenarios of the forward problem models. The Finite Element model consists of a cylindrical polyurethane foam domain underneath a layer of conductive spray. The mechanical response of the sensor body is modeled using the Hyperfoam material model through simultaneous fitting of uni-axial and shear stress experimental data. Different objects are used to simulate the process of touching the sensor body. One object is a human index fingertip that is modeled using a Marlow form material model of BIOSKIN. Another object is a cylindrical steel punch. Furthermore, these Finite Element models have the ability to change the electrical conductivity of the domain caused by strain using the USDFLD subroutine provided in Abaqus which proved to influence the way the voltage is distributed across the system. Additionally, the viscoelasticity of the foam has been investigated and found to be negligible in this kind of application. Artificial neural networks were utilized as an alternative approach to solve the EIT inverse problem instead of using the common image reconstruction method. Several neural networks were made to predict both x and y coordinates of the touch location which were found to be highly accurate in predicting the touch location. Also, other neural networks were made to predict the mechanical behavior of the system using the EIT voltage readings as an input. The results demonstrate the efficiency of the neural networks constructed.
Date of AwardJul 2021
Original languageAmerican English

Keywords

  • Electrical impedance tomography; Machine learning; Electrical conductivity; Bio-skin; Finite element analysis.

Cite this

'