Research On Key Technology Of Coal Quality Detection Based On Machine Vision

Due to the rapid promotion of mechanical mining,the output of fine-grained coal has accounted for more than 20% of the total raw coal.The level of flotation technology in coal preparation plant is directly related to the recovery and utilization rate of coal resources and economics.With the rapid development of computer technology,it has become possible for machine vision technology to replace human vision to automatically process flotation froth images.By extracting the characteristic parameters of froth images timely and accurately,the froth in the flotation process can be detected and identified in real time.However,the continuous change of various operation variables in the flotation process and the inferiority of the on-site environment are great challenges in foam image acquisition,segmentation and surface visual feature extraction.Therefore,it is important to study an effective foam image processing and visual feature extraction and establish a complete and efficient online flotation index detection system.This will improve the efficiency of flotation production,reduce the labor intensity of workers,and have important practical significance for improving the recovery rate of mineral resources.This paper first summarizes the environmental and process characteristics of coal slime flotation,and the visual characteristics of the froth image.Based on the laboratory flotation test,image denoising,enhancement and segmentation algorithms for slime flotation froth are proposed respectively.Based on machine vision,the relevant characteristic parameters that can effectively reflect the coal yield and ash are extracted from the image,and the soft measurement model of coal ash.On this basis,combined with the actual situation of the flotation workshop of the coal preparation plant,a slime flotation foam image control system was designed.The system achieves on-site detection of clean coal ash and verifies the effectiveness of this method.The research content and results are summarized below:(1)Based on total variation theory,a hybrid denoising model is proposed.The model with the principle of dark primary color prior,dual platform histogram equalization and the recursive layered connected domain equalization principle,a comprehensive image enhancement is studied.This technology can effectively remove foam image noise and enhance image contrast while preserving image details.Combining the above preprocessing techniques,a foam image segmentation algorithm based on valley bottom edge detection is designed.While retaining the non-edge pixels,the edges of the obtained image also retain the real foam edge pixels.Through the refinement of the foam boundary pixels,the true mineral foam image edge is obtained.(2)The foam images of flotation experiments were collected under different concentrations of flotation concentration,amount of collector,amount of foaming agent and material particles,and the experimental indicators of the samples were analyzed.The texture,color and shape characteristics of foam images under different flotation experimental conditions were extracted,and the regression methods in machine learning were used to predict the clean coal output,clean coal ash and tail ash during the flotation process.The results show that the proposed multi-feature fusion flotation foam image yield and ash prediction method can achieve better prediction performance.(3)In order to improve the prediction accuracy of clean coal ash,a total of 11 feature parameters were extracted based on the gray image histogram,gray level cooccurrence matrix and size characteristics.The comprehensive analysis between each feature value and the flotation operating state is analyzed.The results show that there is a significant correlation between the variance extracted based on the gray histogram,the energy extracted based on the gray level co-occurrence matrix,and the foam size and the clean coal ash value.However,the smoothness and third-order moment extracted based on the gray histogram are highly correlated with the ash value.The entropy extracted based on the gray histogram and the contrast,entropy extracted by the gray level co-occurrence matrix are moderately related to the ash value.The consistency based on the gray histogram extraction is not sensitive to ash,and the correlation is extremely weak.(4)The multiple linear regression analysis method is used to model the sample data.The multiple linear regression equations when different independent variables are used as the model input values are compared,and the fitting precision and model complexity of the model are comprehensively considered to determine the selection variance.The smoothness,third-order moment,energy,and actual foam area are used as independent variables to establish a soft-sensing model of clean coal ash.Comparing the actual test results of the sample with the prediction results of the model,it is found that the absolute error of the model in predicting the ash content of the sample clean coal can be guaranteed within ±5%.Through empirical formulas and related experiments,the optimal number of hidden layer neurons in the BP neural network was determined,and a flotation concentrate ash prediction model was established.The model is verified through the test set.The results show that the relative error of the predicted value is basically within ±10%,and the absolute error can reach ±1%.Samples with an absolute error of less than ±0.5% accounted for 72% of the total forecast.On the basis of the above research,according to the basic configuration of machine vision,combined with the actual situation of the flotation plant of the coal preparation plant,a flotation foam image control system was built in the paper.The system mainly includes an image acquisition system and an image processing system.The image acquisition system is composed of a color CCD camera,a fixed focus lens and ring light.The image processing system is composed of four modules: image preview,image acquisition,image processing and data storage.It realizes the online detection of ash content of clean coal.Based on industrial test observation,it is found that the error between the data predicted by this system and the actual test data can reach ±1.5%,achieving the real-time accurate prediction of flotation concentrate ash content and timely guidance of on-site production.This paper contains graph 62 pieces,table 20 pieces,and reference 146 pieces.