International Society for Industrial Process Tomography

Multimodal System for Data Analysis and Image Reconstruction in Process Tomography

T. Rymarczyk^1,2*, G. Kłosowski³

¹ Research & Development Centre Netrix S.A., Wojciechowska 31, Lublin, Poland

² University of Economics and Innovation, Projektowa 4, Lublin, Poland

³ Lublin University of Technology, Nadbystrzycka 38A, Lublin, Poland

^*E-mail: tomasz@rymarczyk.com

ABSTRACT

The paper presents the idea of a multimodal system for industrial tomography and shows examples of image reconstruction using various tomographic techniques and reconstruction algorithms. Depending on specific technological tomography, both advantages and disadvantages can be observed in terms of accuracy, frequency and resolution of reproduced images. Knowledge of the characteristics of each tomographic technique allows you to choose the appropriate method of image reconstruction. The proposed solution is based on the construction of a distributed system for acquiring, processing and reconstructing measurement data. The key element was the construction of mobile, inexpensive measuring devices to collect and send data to the Cloud Computing module. The system design includes technologies such as Internet of Things, Big Data and Machine Learning. This is particularly beneficial when monitoring large and dispersed measurement objects. The hardware solution is designed for practical applications, with mobile units equipped with wireless data transfer to Cloud Computing. The research presented in this study focuses on the choice of the method for processing tomographic data and on the design of the system for generating output images. Three methods were tested: Support Vector Machines (SVM), Multivariate Adaptive Regression Splines (MARS) and Artificial Neural Networks (ANNs). Based on ANNs, a simulation model was built and images were generated. The obtained results are interesting due to the high level of fidelity of the mapping, which allows us to believe that the presented method combining ECT and EIT elements has high application values.

Keywords Electrical Impedance Tomography, Inverse Problem, Machine Learning, Regression Modes, Neural Networks.

Industrial Application: General