Method for improving metal mine LIBS element quantitative analysis performance

Abstract

The invention discloses a method for improving metal ore LIBS element quantitative analysis performance. The method comprises the steps that an original metal ore sample is collected, crushed and ground, and rock powder with the particle size not larger than the set mesh number is obtained; selecting a binder with forming auxiliary capability, calculating the addition amount of the binder according to the amount of the rock powder and a set proportion, mixing, and tabletting to prepare a sample; sample LIBS data acquisition is carried out on a to-be-detected sample under the same LIBS experiment condition; the method comprises the following steps: preprocessing collected LIBS data, and establishing a convolutional neural network quantitative analysis model of LIBS elements; predicting through a convolutional neural network quantitative analysis model to obtain prediction results of different LIBS data of each element in the sample; and averaging prediction results of different LIBS data of the same sample and the same element as a final prediction value. According to the method, the reliability and practicability of the LIBS technology in quantitative analysis application in the field of metal mineral exploitation are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of atomic emission spectrometry analysis, and particularly relates to a method for improving the performance of LIBS elemental quantitative analysis of metal ores. Background Art

[0002] Laser-induced breakdown spectrometry (LIBS) is an atomic emission spectrometry technique. With its significant advantages such as non-contact, rapid detection, and simultaneous multi-element analysis, it is widely used in material composition analysis. In the field of mineral extraction, accurately determining the content of elements in metal ores can determine the type, scale, and distribution range of ore deposits, provide important scientific basis for mineral resource exploration, and guide the rational formulation of mining plans.

[0003] Currently, LIBS technology still faces challenges in quantitative analysis, and the accuracy and precision of elemental quantitative data need to be improved. When analyzing bulk samples, affected by factors such as sample inhomogeneity and ubiquitous matrix effects, the repeatability and stability of LIBS calibration data are poor, increasing the difficulty of elemental quantitative analysis. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for improving the performance of LIBS elemental quantitative analysis of metal ores, improve the reliability and practicality of LIBS technology in quantitative analysis applications in the field of metal mineral extraction, and effectively solve the above problems existing in the background art.

[0005] The technical solution adopted by the present invention is as follows:

[0006] A method for improving the performance of LIBS elemental quantitative analysis of metal ores, which includes the following steps:

[0007] Step 1, collecting the original metal ore samples and crushing and grinding them to obtain rock powders with a particle size not greater than a set mesh number;

[0008] Step 2, selecting a binder with forming assistance ability, calculating the addition amount of the binder according to the amount of the rock powder according to a set ratio, and mixing and pressing to make samples;

[0009] Step 3, collecting sample LIBS data for the sample to be measured under the same LIBS experimental conditions;

[0010] Further, when collecting sample LIBS data in Step 3, for each sample piece in different series, the same number of acquisitions are performed at different positions to obtain the corresponding number of LIBS data

[0011] Specifically, as a feasible implementation method, 10 measurements were taken at different positions for each sample piece of different series, and a total of 130 sets of effective spectral data were collected.

[0012] As another feasible implementation method, 10 sites were randomly selected on the surface of each sample, and 60 pulsed lasers were applied to each site to collect one LIBS data. 10 such LIBS data were collected for each sample. The instrument conditions for LIBS data collection were: pulsed laser wavelength 1064 nm, pulsed laser energy 88.5 mJ, spectrometer spectral acquisition delay 0.6 μs, gate width 10 μs, and pulsed laser frequency 10 Hz.

[0013] Step 4, after preprocessing the collected LIBS data, establish a convolutional neural network quantitative analysis model for LIBS elements;

[0014] Step 5, obtain the prediction results of different LIBS data of each element in the sample through the convolutional neural network quantitative analysis model; and take the average of the prediction results of different LIBS data of the same element in the same sample as the final prediction value.

[0015] Further, in step 1, the particle size of the rock powder is less than or equal to 200 mesh.

[0016] Specifically, in step 1, a crushing instrument is used to crush and grind it, and the particle size is strictly controlled to ensure that the particle size of the finally obtained rock powder is less than or equal to 200 mesh. The purpose is to ensure the uniformity of the sample and make the laser generate a more stable plasma when acting on the sample during subsequent LIBS detection.

[0017] Further, in step 2, the original sample that has not been crushed and ground is additionally used as a control group for the control experiment.

[0018] Further, in step 2, different series of sample tablets are made according to the sample incorporation ratios of 80%, 60%, 40%, 20%, 10%, and 5% respectively according to requirements.

[0019] Further, step 2 specifically includes the following steps:

[0020] Step 2-1, put the binder and the crushed ore powder into a stirring device in a set ratio and stir well;

[0021] Step 2-2, use a precision tablet pressing mold to press the mixed powder to obtain a tablet sample with a smooth surface and uniform thickness.

[0022] Specifically, select a suitable binder, accurately calculate the addition amount, and press the tablets. The selected binder needs to have good forming assistance ability;

[0023] Further, step 4 has the following steps:

[0024] Step 4-1, process the collected LIBS data with the multiplicative scatter correction (MSC) algorithm to reduce the scattering effect caused by the surface characteristics of the sample or the physical properties of the particle distribution;

[0025] Step 4-2, reasonably divide the data set into a training set and a test set;

[0026] Further, in step 4-2,

[0027] Step 4-3, establish a random forest (RF) model for different elements respectively;

[0028] Step 4-4, sort according to the importance score of each feature in the spectral data according to the random forest (RF) model; the sum of the importance scores of all features is 1;

[0029] Step 4-5, select features from all features and ensure that the cumulative importance score of the selected features is not less than the set threshold;

[0030] Step 4-6, establish a convolutional neural network (CNN) quantitative analysis model with the training set data and the selected feature variables.

[0031] Further, the threshold set in step 4-5 is 0.97.

[0032] Further, step 5 evaluates the convolutional neural network quantitative analysis model by calculating the fitting coefficient (R 2 ) and the root mean square error (RMSE) in order to optimize the parameters of the convolutional neural network quantitative analysis model.

[0033] Further, the specific steps of step 5 are as follows:

[0034] Step 5-1, take the average of the prediction results of different LIBS data of the same sample and the same element as the final prediction value;

[0035] Step 5-2, calculate the relative standard deviation (RSD) of the spectral data of the same sample,

[0036] Step 5-3, use the formula and to evaluate the model, where n represents the number of samples, y i represents the reference value, represents the predicted value, represents the average value of the reference values.

[0037] Further, the specific steps of step 5-2 are:

[0038] Step 5-2-1: First, calculate the total spectral line intensity where x ij represents the peak intensity value at the i-th wavelength point of the j-th measurement; m represents the number of spectral wavelength points;

[0039] Step 5-2-1: Secondly, calculate the average total intensity where n represents the number of measurements;

[0040] Step 5-2-1: Calculate the standard deviation of the total intensity

[0041] Step 5-2-1: Calculate the relative standard deviation

[0042] The present invention adopts the above technical solutions, performs a series of treatments such as crushing and grinding on the original ore sample. First, the sample matrix difference is reduced through sample preparation, and further combined with the multivariate statistical regression analysis technology to improve the reliability and practicability of the LIBS technology in the quantitative analysis application in the field of metal mineral mining. Brief Description of the Drawings

[0043] The following further describes the present invention in detail with reference to the drawings and specific embodiments;

[0044] Figure 1 is a schematic flow chart of a method for improving the LIBS element quantitative analysis performance of a metal ore in the present invention;

[0045] Figure 2 is a schematic architecture diagram of the convolutional neural network (CNN) quantitative analysis model in the present invention;

[0046] Figure 3 is a schematic diagram of the result comparison of the relative standard deviation RSD of the LIBS data of different series of samples in the present invention;

[0047] Figure 4 is a schematic diagram of the performance comparison of the quantitative analysis model in the present invention on the test sets of different series of samples;

[0048] Figure 5 is a schematic diagram of the deviation degree (40%) of the predicted value of the quantitative analysis model in the present invention from the reference value;

[0049] Figure 6 is a schematic diagram of the deviation degree (Cu) of the predicted value of the original sample of the quantitative analysis model in the present invention from the reference value.

[0050] Figure 7 is a schematic diagram of the deviation degree (Cu) of the predicted value of the tablet sample model of the quantitative analysis model in the present invention from the reference value; Detailed Description of the Embodiments

[0051] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this application.

[0052] Based on the limitations of existing metal ore LIBS element analysis technology and starting from the factors affecting the performance of LIBS quantitative analysis, such as Figures 1 to 7 shown, the present invention proposes a method for improving the performance of LIBS element quantitative analysis of metal ore, that is, a strategy and method of powder mixing sample preparation. The binder is evenly mixed with the metal ore powder and pressed into tablets to prepare sheet samples for LIBS analysis.

[0053] Example 1: For ease of description and to improve the calibration accuracy of the model, 13 groups of Mn ore samples were selected, and the sample reference contents were provided by the National Standard Material Center. The reference contents are shown in Table 1.

[0054] Table 1 Reference content of Mn element in the samples

[0055] Number Mn (mass percentage) 01 33.23991 02 23.69966 03 41.30139 04 12.74444 05 3.85097 06 36.24848 07 5.52765 08 20.57384 09 33.1914 10 32.33121 11 40.87706 12 27.98206 13 11.06708

[0056] A method for improving the performance of LIBS element quantitative analysis of metal ore according to the present invention has a flowchart as Figure 1 , and includes the following steps:

[0057] Step 1: Collect the original ore samples;

[0058] Step 2: Use a crushing device to crush and grind them, strictly control the particle size to ensure that the particle size of the final obtained ore powder is less than or equal to 200 mesh. The purpose is to ensure the uniformity of the sample and make the laser generate a more stable plasma when acting on the sample during subsequent LIBS detection;

[0059] Step 3: Use paraffin as the binder, determine six different series of sample incorporation ratios, which are 80%, 60%, 40%, 20%, 10%, and 5% respectively. Put them into a powder mixing blender, stir evenly, weigh 1 g of the mixed powder, and use a tablet press to press the mixed powder at a pressure of 4 MPa;

[0060] Step 4: Optimize the LIBS experimental conditions to ensure the consistency of the experimental conditions for the samples to be measured;

[0061] Step 5: Collect LIBS data of the samples. Each sample tablet of different series is measured 10 times at different positions to collect a total of 130 groups of effective spectral data;

[0062] Step 6: Divide the dataset in a reasonable proportion, and separate the test set spectral data, test set label data, training set spectral data, and training set label data from the spectral data. Among them, 10 sample data are used as the training set, and 3 sample data are used as the test set. Process the collected LIBS data using the MSC algorithm, establish an RF model using the training set data, score its feature importance, sort the data according to the score value, set the threshold to 0.97, and according to this standard, the number of features selected for the 80%, 60%, 40%, 20%, 10%, and 5% series are 351, 442, 487, 291, 305, and 511 respectively;

[0063] Step 7: According to the training set divided in Step 6 and the selected feature variables, establish and train a CNN analysis model. The CNN model architecture is as Figure 2 shown;

[0064] Step 8: Calculate the relative standard deviation (RSD) of the spectral data of the same sample. First, calculate the total spectral line intensity Secondly, calculate the average total intensity Then, calculate the standard deviation of the total intensity Finally, calculate the relative standard deviation where x ij represents the peak intensity value of the i-th wavelength point in the j-th measurement, m is the number of spectral wavelengths, and n is the number of measurements. The RSD of the LIBS data of different series of samples is as Figure 3 shown. According to the formulas and calculate the fitting coefficient (R 2 ) and the root mean square error (RMSE) to evaluate the model. Among them, n represents the number of samples, y i represents the reference value, represents the predicted value, represents the average value of the reference values. Taking the 40% series as an example, the deviation degree between the model predicted value and the true value is as Figure 5 shown, where R2_train and RMSEC represent the R 2 and RMSE of the model with the training data as the input; R2_test and RMSEP represent the R 2 and RMSE of the model with the test data as the input.

[0065] Example 2: Quantitative analysis of Cu element in ore:

[0066] Select 55 groups of ore samples, and the National Standard Material Center provides the reference content of these samples. The reference content is shown in Table 2.

[0067] Table 2 Reference content of Cu element in samples

[0068] Number Cu (mass percentage) 01 4.56 02 0.41 03 1.01 04 0 05 0.67 06 0 07 0.16 …… …… 51 8.46 52 8.53 53 10.71 54 12.59 55 12.79

[0069] A method for improving the performance of quantitative analysis of elements in metal ores by LIBS, comprising the following steps:

[0070] Step 1, collect the original ore samples;

[0071] Step 2, use a crushing device to crush and grind the samples, strictly control the particle size to ensure that the particle size of the finally obtained ore powder is less than or equal to 200 mesh, aiming to ensure the uniformity of the samples and enable the laser to generate a more stable plasma when acting on the samples during subsequent LIBS detection;

[0072] Step 3, use paraffin as the binder, with a sample incorporation ratio of 40%, put it into a powder mixing and stirring machine, stir well, weigh 1 g of the mixed powder, and use a tablet press to press the mixed powder at a pressure of 4 MPa; in addition, use the uncrushed and unground original samples as the control group;

[0073] Step 4, optimize the LIBS experimental conditions to ensure the consistency of the experimental conditions for the samples to be measured;

[0074] Step 5, collect LIBS data of the samples. Randomly select 10 sites on the surface of each sample, and collect one LIBS data for each site under the action of 60 pulsed lasers. A total of 10 such LIBS data are collected for each sample. The instrument conditions for LIBS data collection are: pulsed laser wavelength 1064 nm, pulsed laser energy 88.5 mJ, spectrometer spectral acquisition delay 0.6 μs, gate width 10 μs, pulsed laser frequency 10 Hz.

[0075] Step 6, divide the data set in a reasonable proportion, separate the test set spectral data, test set label data, training set spectral data, and training set label data from the spectral data. Among them, 48 sample data are used as the training set, and 7 sample data are used as the test set. Process the collected LIBS data using the MSC algorithm, establish an RF model using the training set data, score the importance of its features, sort the data according to the size of the score value, set the threshold to 0.97, and according to this standard, the number of features selected for the tableted samples and the original samples are 135 and 203 respectively;

[0076] Step 7, respectively train a CNN analysis model according to the training set and the selected feature variables divided in Step 6;

[0077] Step 8, according to the formula and calculate the fitting coefficient (R 2 ) and the root mean square error (RMSE) to evaluate the model, where n represents the number of samples, yi represents the reference value, represents the predicted value, represents the average value of the reference values. The deviation degree between the model predicted value and the true value of the original sample is as Figure 6 shown, and the deviation degree between the model predicted value and the true value of the tablet sample is as Figure 7 shown, where R2_train and RMSEC represent the R 2 and RMSE of the model with the training data as the input; R2_test and RMSEP represent the R 2 and RMSE of the model with the test data as the input.

[0078] The present invention can make full use of the improvement effect of powder mixing treatment on the complex mechanism characteristics of the sample, making the matrix characteristics of the tablet sample closer and the surface morphology more consistent, effectively improving the interaction between the pulsed laser and the sample surface, and fundamentally enhancing the stability of pulsed laser ablation sampling. As can be seen from Figure 3 , the RSD of different series of samples is all below 20%, which indicates that after powder mixing sample preparation, the collected LIBS data has good stability, effectively solving the problems of poor repeatability and stability of LIBS data caused by surface inhomogeneity and element distribution differences in bulk samples.

[0079] The present invention also provides a method for quantitative analysis of metal ore LIBS elements. By exploring different sample and binder incorporation ratios, a three-layer convolutional neural network model is established, and finally through data evaluation, the purpose of improving the performance of quantitative analysis of metal ore LIBS elements is achieved. As can be seen from Figure 4 , using paraffin as the binder and incorporating 40% of the sample, the model has the largest R 2 and the smallest RMSE on the test set, indicating that the powder mixing and tablet pressing sample preparation strategy for the sample incorporating 40% fundamentally enhances the stability of pulsed laser ablation sampling, improves the quality of the plasma light source, and to a certain extent reduces the influence of matrix effects, improves the quality of LIBS data, and provides support for constructing a more accurate quantitative analysis model.

[0080] The present invention adopts the above technical solutions to perform a series of treatments such as crushing and grinding on the original ore sample. First, the sample matrix difference is reduced through sample preparation, and further combined with multivariate statistical regression analysis technology to improve the reliability and practicality of LIBS technology in quantitative analysis applications in the field of metal ore mining.

[0081] Obviously, the described embodiments are some, but not all, of the embodiments of the present application. Without conflict, the embodiments in the present application and the features in the embodiments can be combined with each other. Generally, the components of the embodiments of the present application described and illustrated in the accompanying drawings here can be arranged and designed in various different configurations. Therefore, the detailed description of the embodiments of the present application is not intended to limit the scope of the present application claimed, but merely represents selected embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts fall within the scope of protection of the present application.

Claims

1. A method for improving the performance of LIBS element quantitative analysis of metal ores, characterized by: It includes the following steps: Step 1, collecting the original metal ore sample and crushing and grinding it to obtain rock powder with a particle size not greater than a set mesh size; Step 2: Select a binder with molding auxiliary ability, calculate the amount of binder to be added according to the amount of rock powder according to the set ratio, and mix and press the mixture into tablets to make samples; Step 3, collecting sample LIBS data for the sample to be tested under the same LIBS experimental conditions; Step 4, after preprocessing the collected LIBS data, a convolutional neural network quantitative analysis model of LIBS elements is established; Step 5: The prediction results of different LIBS data of each element in the sample are predicted by the convolutional neural network quantitative analysis model; and the prediction results of different LIBS data of the same sample and the same element are averaged as the final prediction value.

2. A method for improving the performance of LIBS element quantitative analysis of metal ores according to claim 1, characterized in that: In step 1, the particle size of the rock powder is less than or equal to 200 mesh.

3. A method for improving the performance of LIBS element quantitative analysis of metal ores according to claim 1, characterized in that: In step 2, an original sample that has not been crushed or ground is used as a control group for a control experiment.

4. The method for improving the performance of LIBS element quantitative analysis of metal ores according to claim 1, characterized in that: In step 2, different series of sample tablets are prepared according to the sample incorporation ratios of 80%, 60%, 40%, 20%, 10%, and 5% respectively.

5. The method for improving the performance of LIBS element quantitative analysis of metal ores according to claim 1, characterized in that: Step 2 specifically includes the following steps: Step 2-1, putting the binder and the crushed ore powder into a mixing device according to a set ratio and mixing them thoroughly; Step 2-2, use a precision tableting mold to tablet the mixed powder, and the obtained tablet sample should have a smooth surface and uniform thickness.

6. The method for improving the performance of LIBS element quantitative analysis of metal ores according to claim 1, characterized in that: When collecting sample LIBS data in step 3, each sample piece of different series is collected the same number of times at different positions to obtain a corresponding amount of LIBS data.

7. The method for improving the performance of LIBS element quantitative analysis of metal ores according to claim 1, characterized in that: Step 4 has the following steps: Step 4-1, processing the collected LIBS data with a multivariate scatter correction algorithm to reduce the scattering effect; Step 4-2, reasonably divide the data set into training set and test set; Step 4-3, establish random forest models for different elements; Step 4-4, sort the importance score of each feature in the spectral data according to the random forest model; the sum of the importance scores of all features is 1; Step 4-5, select features from all features and ensure that the cumulative importance score of the selected features is not less than the set threshold; Steps 4-6: Use the training set data and the selected feature variables to establish a convolutional neural network quantitative analysis model.

8. The method for improving the performance of LIBS element quantitative analysis of metal ores according to claim 1, characterized in that: Step 5: Calculate the fitting coefficient R 2 The convolutional neural network quantitative analysis model is evaluated by the root mean square error (RMSE) in order to optimize the parameters of the convolutional neural network quantitative analysis model.

9. A method for improving the performance of LIBS element quantitative analysis of metal ores according to claim 8, characterized in that: The specific steps of step 5 are as follows: Step 5-1, take the average of the prediction results of different LIBS data of the same sample and the same element as the final prediction value; Step 5-2, calculate the relative standard deviation RSD of the spectrum data of the same sample, Step 5-3, use the formula and Evaluate the model, where n represents the number of samples, yi represents the reference value, represents the predicted value, Indicates the average value of the reference value.

10. A method for improving the performance of LIBS element quantitative analysis of metal ores according to claim 9, characterized in that: The specific steps of step 5-2 are: Step 5-2-1, first calculate the total intensity of the spectrum lines where x ij represents the peak intensity value of the i-th wavelength point at the j-th measurement; m represents the number of spectral wavelength points; Step 5-2-1, then calculate the average total intensity Where n represents the number of measurements; Step 5-2-1, calculate the standard deviation of the total intensity Step 5-2-1, calculate the relative standard deviation

Images