The classification of signals measured with electrostatic sensor by using machine learning methods

This dissertation presents a theoretical analysis and experimental validation of classification studies to reveal the flow characteristics of the particulates of pneumatic conveying with multimodal sensors, which are used in a wide range of industries, to continuously monitor and measure a variety of industrial processes, such as coal power plants, cement, steel, and food processing. In the literature, no classification of two-phase mass flow rate has been encountered. The multimodal sensor fusion used in the data-driven study collects data from the electrostatic sensor array channels. Electrostatic sensors used in this test rig consist of both ring and four arc-shaped electrodes for regional monitoring of particles in the pipeline. Electrostatic sensors were preferred because of their simplicity in construction, cost-effectiveness, and suitability for a various installation conditions. In this study, two-phase pneumatic transport particles were classified according to five different mass flow rates, with each class associated with four distinct air velocity values. The features derived from the data were used as input values. As a result, the CVA and SVM methods employed in linear classification have achieved the maximum value in the average classification accuracies of 8-folds as 46\% and 43\% respectively. However, the CNN method, which can be modelled for more complex processes, reached a classification accuracy of 93.6\%.