PREDICTING THE PARAMETERS OF THE MULTI VORTEX CLASSIFIER USING MACHINE LEARNING IN THE ORANGE DATA MINING ENVIRONMENT
Authors:
1.
Kazan National Research Technological University
(Department of Food Production Equipment, assistant professor)
from 01.01.2020 until now
Russian Federation
from 01.01.2020 until now
Russian Federation
2.
Kazanskiy gosudarstvennyy energeticheskiy universitet
(ATPP, assistent)
from 01.01.2018 until now
Russian Federation
from 01.01.2018 until now
Russian Federation
3.
Kazanskiy gosudarstvennyy energeticheskiy universitet
(ATPP, zaveduyuschiy kafedroy)
from 01.01.2015 until now
Russian Federation
from 01.01.2015 until now
Russian Federation
Type:
Article
Pages:
from 141
to 146
Status:
In work
Language:
Russian
Keywords:
PARTICLE CLASSIFICATION, CLASSIFICATION EFFICIENCY, MACHINE LEARNING, DATA ANALYSIS, REGRESSION MODEL ACCURACY
Abstract (English):
The classification of dry particles by size has many applications in many industries. During the preparation process, powders must have a narrow particle size distribution, so the use of classifiers allows you to ensure the specified parameters and prevent excessive grinding. Here, the operational characteristics are evaluated because of experimental studies and numerical modeling. A multivortex classifier was modeled in Ansys Fluent to study the influence of various factors on classification efficiency. As the amount of data is constantly increasing, it becomes difficult for a person to analyze it manually to make strategic decisions. The aim of the research is to develop a model for data analysis using Orange Data Mining software with a visual interface. The process of evaluating the comparative results of the tested forecasting model is presented. The "Testing and Evaluation" widget displays the results of nine classification algorithms, where the results of comparing the prediction of the effectiveness of particle classification are shown with the values of MSE, RMSE, MAE, MAPE and R2. The evaluation results showed that performance forecasting models have different levels of accuracy. The Support Vector Machine (SVM), neural network method, and gradient boosting have lower regression coefficients for linear MAE and RMSE compared to kNN and other classification algorithms. The prediction result using the Neural Network method corresponds to the prediction of the effectiveness of the multi-vortex classifier with a small error. The best methodology used in this study can be re-applied in future studies to get optimal multivortex separator designs that ensure high classification accuracy.
The classification of dry particles by size has many applications in many industries. During the preparation process, powders must have a narrow particle size distribution, so the use of classifiers allows you to ensure the specified parameters and prevent excessive grinding. Here, the operational characteristics are evaluated because of experimental studies and numerical modeling. A multivortex classifier was modeled in Ansys Fluent to study the influence of various factors on classification efficiency. As the amount of data is constantly increasing, it becomes difficult for a person to analyze it manually to make strategic decisions. The aim of the research is to develop a model for data analysis using Orange Data Mining software with a visual interface. The process of evaluating the comparative results of the tested forecasting model is presented. The "Testing and Evaluation" widget displays the results of nine classification algorithms, where the results of comparing the prediction of the effectiveness of particle classification are shown with the values of MSE, RMSE, MAE, MAPE and R2. The evaluation results showed that performance forecasting models have different levels of accuracy. The Support Vector Machine (SVM), neural network method, and gradient boosting have lower regression coefficients for linear MAE and RMSE compared to kNN and other classification algorithms. The prediction result using the Neural Network method corresponds to the prediction of the effectiveness of the multi-vortex classifier with a small error. The best methodology used in this study can be re-applied in future studies to get optimal multivortex separator designs that ensure high classification accuracy.
Keywords:
PARTICLE CLASSIFICATION, CLASSIFICATION EFFICIENCY, MACHINE LEARNING, DATA ANALYSIS, REGRESSION MODEL ACCURACY
PARTICLE CLASSIFICATION, CLASSIFICATION EFFICIENCY, MACHINE LEARNING, DATA ANALYSIS, REGRESSION MODEL ACCURACY
