Research On Ash Content Prediction Of Coal Flotation Tailings Based On Machine Learning

The tailings ash content is an important indicator for evaluating the performance of coal flotation,and real-time online detection of ash content is crucial for closed loop control of coal flotation systems.At present,the ash content of flotation tailings in coal preparation plants mainly relies on manual experience estimation,which causes serious material loss in the flotation system.In recent years,machine vision and machine learning technologies have provided the possibility for online detection of the ash content of coal flotation tailings.This thesis extracts the image features of coal flotation tailings,establishes a prediction model for tailings ash content based on machine learning algorithms,and finally develops an online detection software for flotation tailings ash content to achieve online detection of flotation tailings ash content.The main research work is as follows:A machine vision system for coal flotation tailings was built.This thesis designs a machine vision system for coal flotation tailings,which is arranged under the tailings overflow of the flotation machine,mainly including industrial computer,industrial camera,light source,and overflow weir.Research has found that the flow rate of tailings affects the image quality,and the flow control valve of the acquisition device is optimized.When the opening of the regulating valve is 60%,the overflow liquid level is stable and the image quality is the best.On this basis,2261 tailings images were collected,and the range of ash content of all samples in the dataset was between 38.74%and 75.08%.The multi-dimensional features of flotation tailings images were extracted,and the relationship between multi-dimensional features and ash content were studied.This thesis extracts RGB color features(3 features),grayscale features(5 features),grayscale co-occurrence matrix(5 features),HSI color space(18 features),and Lab color space(18 features)from tailings images.The research results indicate that the mean values of RGB components,grayscale median values,grayscale mean values,“I” component mean values,and “L” component mean values of tailings images are significantly correlated with ash content.Reduction of information redundancy between multi-dimensional features of tailings images based on principal component analysis.This thesis uses principal component analysis to reduce the dimensionality of 49 dimensional features and constructs a tailings ash prediction model based on PCA-SVR.Six different feature combinations are used to compare and verify the effectiveness of the proposed features,and the optimal support vector parameters are found by random search and crossvalidation algorithms.The prediction results show that PCA can effectively improve the accuracy of the model,and when the feature dimension is reduced to 16,the optimal fitting coefficient of the model is 0.9504 and the average absolute error is 1.4088.The CNN-SVR ash content prediction model is established based on convolutional neural network.The convolutional neural network is selected to automatically extract the features of tailings image,and the data enhancement operation is performed on the dataset to prevent the model from overfitting.Five optimization methods were compared to select the optimal Res Net18 model,and the classic SVR was used to replace the Soft Max classifier in Res Net18.The model prediction results indicate that the Res Net18-SVR model has the highest prediction accuracy,and the predicted ash content is closer to the actual,which is better than the traditional method.Online detection software for ash content of coal flotation tailings was developed.Designed and developed online detection software for tailings ash content based on Py Qt5.The software can monitor the status of tailings and predict ash content online,which has the advantages of safety and efficiency.This study realizes the online detection of ash content at coal flotation tailings,provides important feedback parameters for the flotation control system,and can provide theoretical support for the intelligent construction of coal flotation system.This thesis contains 48 figures,8 tables and 94 references.