International Society for Industrial Process Tomography

2D Spatial-1D Temporal Carbon Black Volume Fraction Detection in Slurry Composition Mixing by Electrical Impedance Tomography coupled with Convolutional Neural Network (EIT-CNN)

Y. Ashida¹, Y. A. K. Prayitno^1,2.*, D.Kawashima^1*, M. Takei²

¹Graduate School of Engineering, Chiba University, Chiba, Japan

²Department of Mechanical Engineering, Vocational College, Universitas Gadjah Mada, Yogyakarta, Indonesia

^*Email: yosephus.ardean@ugm.ac.id

ABSTRACT

2D spatial-1D temporal carbon black (CB) volume fraction α in slurry composition mixing has been detected by electrical impedance tomography coupled with convolutional neural network (EIT-CNN). The EIT-CNN has two main sections which are I) training section and II) prediction section. In the I) training section, four components are set as I-⓪ frequency response analysis, I-① mixing simulation, I-② electromagnetic simulation, and I- ③ CNN training. In the II) prediction section, two components are explained in the experiments which are II-① reactance measurement, II-② prediction by trained CNN. In the 1^st section, the characteristic frequency of EIT f_c^L and f_c^H are determined by evaluating the deviation of the pre-experimental reactance ^preX under known concentrations of LiCoO₂-CB powders. Then, the reactanc ^simX is calculated under the determined f_c^L and f_c^H by the electromagnetic simulation from the input of 2D spatial-1D temporal CB α mixing simulation by liquid-liquid model solved in OpenFOAM. In the end of the 1^st section, the convolutional layer of CNN is optimized by evaluating the output of 2D spatial-1D temporal CB α with the input of ^simX. Next, in the 2^nd section, a lab-scale mixing experiments with the determined f_c^L = 76.28 Hz and f_c^H = 36.73 kHz were conducted under a constant mixing speed ω = 720rpm and mixing time t = 0 - 60s for obtaining reactance ^expX as the input into the trained CNN. As a result, the EIT-CNN has the best performance for detecting the 2D spatial-1D temporal CB α by the average mean square error MSE = 0.30% for CB powders. The EIT-CNN offers a modern approach to detect the CB α in complex conditions such as during slurry composition mixing.

Keywords: Carbon Black (CB) detection, 2Dspatial-1D temporal, slurry composition mixing, EIT, CNN

Industrial Application: Battery Slurry Processing, Powder Processing.