| Copper concentrate is an indispensable mineral resource for modern industry and is also one of the most important commodities in the international trade of non-ferrous metals.Since China is the largest importer of copper concentrate in the world,copper concentrate inspection management aims at preventing the potential risk involving safety,health,environmental protection,and fraud.The management including the conformance verification about the origin of imported copper concentrate,and test the content of each element can be filtered the phenomenon such as doping,adulterated,shoddy,helps to risk classification,early warning,to ensure the safety of entry copper concentrate is of great significance.X-ray fluorescence spectroscopy and X-ray diffraction spectroscopy analysis both have the advantages of simple sample preparation,nondestructive analysis,good stability and fast analysis and so on,which got widely attention in the detection of copper concentrate.In this paper,the characteristics of different copper concentrate producing areas are analyzed by X-ray fluorescence and diffraction spectrum,and the identification model of copper concentrate producing areas is established by combining chemometrics.The main research contents are as follows:(1)The element and phase characteristics of copper concentrates from different sources were analyzed.X-ray fluorescence spectroscopy(XRF),X-ray diffraction(XRD)and polarizing microscopy were used to observe the representative samples of copper concentrate imported from12 batches of different mining areas in 8 countries to carry out comprehensive analysis.The characteristics and differences of element content,physical phase composition were compared.The mineralogical characteristics of copper concentrate samples of different geological origin types were further discussed.The XRF analysis showed that the main elements of the copper concentrates were Cu,Fe,S,O,and generally contain Zn,Si,Al,Mg,Ca,Pb.The XRD phase analysis showed that the main phase of the copper concentrate sample is chalcopyrite,and often contains pyrite and sphalerite.Polarization microscope shows that the content of chalcopyrite in copper concentrates was between 88%and 98%,and most of the samples are chalcopyrite(main part)and pyrite(general less than 5%)composition,content over 90%.Combining the analysis of different metallogenic types of copper concentrates,the samples of porphyry,skarn and volccanogenic massive sulfide deposits had common mineral compositions of chalcopyrite,pyrite,sphalerite and special mineral compositons of biotite,weddellite and lead anglesite separately.The main minerals of the iron oxide-copper–gold deposit samples were chalcopyrite,pyrrhotite and talc.(2)The classification model of copper concentrate was established based on the results of non-standard analysis of X-ray fluorescence spectrum.The elemental composition and content of280 batches of copper concentrate from 8 countries were determined by XRF without standard sample analysis,and the difference of the elemental content of copper concentrate from different producing areas was analyzed.The 280 copper concentrate samples were divided into 226modeling samples and 54 prediction samples to establish the classification and identification model of copper concentrate.The BP neural network classification and identification model of importing copper concentrate was established by using 17 elements of O,Mg,Al,Si,P,S,K,Ca,Ti,Fe,Cu,Zn,Mn,As,Mo,Ag and Pb As variables.Moreover,13 elements including O,Mg,Al,Si,P,S,K,Ca,Cu,Zn,Mo,Ag,Pb were screened out as valid variables by F-score,and the Fisher discriminant analysis prediction model and BP neural network prediction model were established for importing copper concentrate countries respectively.Three prediction models were used to establish the classification model of copper concentrate.The results showed as follows:(1)The accurate recognition rates of the Fisher discriminant analysis model using F-score for the modeled sample was 94.2%,the cross-validation accuracy was 92.8%,and the accurate recognition rate of the predicted sample reached 96.8%.(2)The training sets,calibration sets,validation sets,modeling sets,and predicted samples of the BP neural network with 17 variables in the input layer were accurately identified as 100%,97.1%,94.1%,98.2%,and 100%,respectively.(3)The third model was established by BP neural network using variables selected by f-score with Fisher discriminant analysis.The accurate recognition rates of the training set,calibration set,verification set,modeling set and predicted samples of the BP neural network model with 13 variables(selected by f-score)were 100%,97.1%,100%,99.6%,and 100%,respectively.Comparing the results of three times of modeling,it can be seen that the model established by f-score combined BP neural network had the highest accurate recognition rate.The f-score combined BP neural network method can not only reduce the input variables of modeling,but also improve the recognition accuracy.(3)The classification model of copper concentrate was established by X-ray diffraction spectrum(XRD).X-ray diffraction technology was used to analyze the physical phase characteristics of 138 copper concentrate samples from 3 countries with the largest number of copper concentrates in China.The characteristic XRD data of copper concentrate were extracted by combining principal component analysis and random forest characteristic importance methods to establish a random forest classification model.Principal component analysis(PCA)was used to reduce the dimension of the data of X-ray diffraction spectra of copper concentrate.The random forest classification model was further established based on the number of principal component analysis.The results showed that the first16 principal components(PC1-16)were used in the random forest classification model,and the highest accurate recognition rate was 91.42%.Since PCA model cannot reflect the characteristics of the X-ray diffraction spectra of copper concentrate,122 characteristic spectral data from the first 16 principal components(PC1-16)were extracted by principal component load threshold method to establish a random forest(RF)classification model.As compared with the PCA classification model,the accurate recognition rate was improved by RF model and reached94.28%.The method of importance ranking of random forest characteristics is further adopted for classification model.The results show that the accurate recognition rate of classification model by RF importance ranking was 94.28%by selecting the first 34 data of characteristics importance.Compared with the principal component load threshold,the RF importance ranking method not only effectively reduced the number of feature input variables,but also achieved good classification and recognition effect. |
PDF Full Text Request