Non-destructive testing method and system for the content of harmful elements in mineral products

Abstract

The application discloses a nondestructive testing method and system for harmful element content in mineral products, and relates to the technical field of material nondestructive testing. The method comprises the following steps: collecting a characteristic fluorescence net intensity sequence and a scattering background intensity sequence of a target harmful element; calculating the relative standard deviation of the relative fluorescence yield as the surface state dispersion; fitting the spectral background baseline slope of each point, and calculating the relative standard deviation thereof among multiple points as the background fluctuation dispersion; performing double-weighted fusion on the characteristic fluorescence net intensity of each point to obtain effective characteristic intensity; inputting the effective characteristic intensity into a mineral species correlation model, outputting the content value of the harmful element, and comparing the content value with a dynamic limit threshold to output an over-limit determination result. The application solves the technical problem that the existing XRF detection method is easily disturbed by the surface state and matrix effect of the mineral product, and thus the detection result has a large deviation.

Description

Technical Field

[0001] This invention relates to the field of nondestructive testing technology for materials, specifically to a nondestructive testing method and system for the content of harmful elements in mineral products. Background Technology

[0002] During customs clearance inspection of mineral products, rapid testing for controlled heavy metals such as lead, mercury, and cadmium is required in imported mineral products such as iron ore, chromium ore, and copper-gold ore. Currently, on-site inspections often employ handheld X-ray fluorescence spectrometers, which directly irradiate the mineral products and determine the content of harmful elements by analyzing the characteristic X-ray fluorescence produced.

[0003] However, this direct irradiation detection method is susceptible to interference from factors such as surface moisture, particle size, and differences and inhomogeneities in the mineral matrix. The roughness of the mineral stack surface, intergranular gaps, and local component segregation can lead to significant differences in X-ray excitation efficiency and fluorescence yield between measurement points. Different mineral matrices exhibit varying scattering and absorption characteristics of incident X-rays, causing fluctuations in the spectral background baseline between measurement points. The combined effect of these factors often results in large deviations in detection results, potentially leading to the wrongful release of goods exceeding hazardous element limits or the wrongful detention of qualified goods, thereby causing trade disputes and safety hazards. Summary of the Invention

[0004] This invention provides a non-destructive testing method and system for the content of harmful elements in mineral products, which solves the technical problem that existing XRF testing methods are easily affected by the surface condition and matrix effect of mineral products, resulting in large deviations in the test results.

[0005] In a first aspect, the present invention provides a non-destructive testing method for the content of harmful elements in mineral products, the method comprising: Multiple measurement points are set up on the target surface of the stack of mineral products to be tested. The characteristic fluorescence net intensity sequence of the target harmful element is excited at each point and collected. The background scattering intensity sequence of each point is obtained simultaneously. Based on the characteristic net fluorescence intensity and scattering background intensity of each site, the relative fluorescence yield of each site is calculated, and the relative standard deviation of the relative fluorescence yield among multiple sites is calculated as the surface state dispersion. Using the count values ​​of multiple energy channels in the scattering background intensity sequence at each point as the dependent variable and the corresponding energy values ​​as the independent variable, the slope of the spectral background baseline at each point is fitted, and the relative standard deviation of the slope of the spectral background baseline at multiple points is calculated as the background fluctuation dispersion. Based on the surface state dispersion and the background fluctuation dispersion, a penalty coefficient is calculated, and based on the penalty coefficient and the spectral background baseline slope of each point, the net characteristic fluorescence intensity of each point is double-weighted and fused to obtain the effective characteristic intensity. The effective feature intensity is input into the mineral type association model, the content value of harmful elements in the mineral product to be tested is output, and the content value is compared with the dynamic limit threshold to output the result of exceeding the standard.

[0006] Preferably, multiple measurement points are set up on the target surface of the stack of mineral products to be tested, the characteristic net fluorescence intensity sequence of the target harmful element is collected at each point, and the background scattering intensity sequence corresponding to each point is obtained simultaneously, including: The top surface of the stack of mineral products to be tested exposed to the outside world is used as the target surface. Multiple measurement points are arranged in a grid pattern within the target surface, with equal spacing between adjacent measurement points. The excitation source of the X-ray fluorescence spectrometer is vertically aligned with each measurement point, and the same excitation time is continuously applied at each measurement point to collect the energy dispersive X-ray fluorescence spectrum corresponding to each measurement point. From the energy dispersive X-ray fluorescence spectra of each measurement point, locate the characteristic X-ray fluorescence peak of the target harmful element, perform area integration on the characteristic X-ray fluorescence peak, and subtract the baseline background of linear interpolation on both sides of the peak to obtain the net intensity of characteristic fluorescence corresponding to each measurement point. In a continuous scattering spectrum region that does not overlap with the characteristic X-ray fluorescence peak, an equal-width energy channel is extracted, and the count value of each energy channel is extracted to form a scattering background intensity sequence corresponding to each measurement point. The center energy value of each energy channel in the scattering background intensity sequence is recorded synchronously. The characteristic fluorescence net intensities corresponding to each measurement point are arranged in the order of the measurement points to form a characteristic fluorescence net intensity sequence. Arrange the scattering background intensity sequence and the center energy value of the corresponding energy channel at each measurement point in the same order to form the scattering background intensity sequence and the corresponding energy independent variable sequence.

[0007] Preferably, the target harmful element includes at least one of lead, mercury, cadmium, chromium, and arsenic. The continuous scattering spectrum region that does not overlap with the characteristic X-ray fluorescence peak is the continuous scattering spectrum region located on the high-energy side of the characteristic X-ray fluorescence peak. The number of energy channels intercepted in the continuous scattering spectrum region is not less than 3, and when the energy distance between the characteristic X-ray fluorescence peak and the adjacent high-energy side characteristic X-ray fluorescence peak is greater than 1 keV, 5 energy channels are intercepted.

[0008] Preferably, based on the characteristic net fluorescence intensity and scattering background intensity of each site, the relative fluorescence yield of each site is calculated, and the relative standard deviation of the relative fluorescence yield among multiple sites is calculated as the surface state dispersion, including: The net intensity of each characteristic fluorescence in the net intensity sequence of the characteristic fluorescence is extracted point by point, and the count value of each energy channel in the scattering background intensity sequence corresponding to the same measurement point is extracted. For the same measurement point, the count values ​​of each energy channel in the scattering background intensity sequence are weighted by the center energy value of the same position in the energy independent variable sequence to obtain the energy-weighted total scattering intensity. The characteristic net fluorescence intensity at the same measurement point is normalized using the energy-weighted total scattering intensity to obtain the relative fluorescence yield at each measurement point. By traversing all measurement points, a relative fluorescence yield sequence corresponding to each measurement point is obtained. The relative standard deviation of the relative fluorescence yield sequence is calculated as the surface state dispersion.

[0009] Preferably, the slope of the spectral background baseline at each point is fitted using the count values ​​of multiple energy channels in the scattering background intensity sequence as the dependent variable and the corresponding energy values ​​as the independent variable. The relative standard deviation of the spectral background baseline slope among multiple points is calculated as the background fluctuation dispersion, including: For each measurement point, count values ​​are extracted from the background scattering intensity sequence channel by channel, and center energy values ​​of the same order are extracted from the energy independent variable sequence. The center energy values ​​and count values ​​of the same channel are combined into a data pair. After traversing all energy channels, a set of data pairs for the same measurement point is formed. For each data pair in the data pair set, an energy weighting factor is determined based on the central energy value, and a counting statistical weighting factor is determined based on the count value; The product of the energy weighting factor and the count statistics weighting factor is used as the composite weight of the data pair in the linear regression. Using the center energy value as the independent variable and the count value as the dependent variable, apply the corresponding composite weights to all data pairs in the data pair set, perform a weighted least squares linear regression, and solve for the slope of the regression line, which is used as the slope of the spectral background baseline corresponding to the same measurement point. All the spectral background baseline slopes obtained by traversing point by point are arranged in the order of the points to form a spectral background baseline slope sequence. Calculate the arithmetic mean of the spectral background baseline slope sequence; Calculate the normalized deviation of the slope of each spectral background baseline relative to the arithmetic mean, and obtain the relative deviation of the slope corresponding to each spectral background baseline slope; Using the minimum value among the relative deviations of the slope as the benchmark deviation, calculate the normalized ratio of each relative deviation of the slope to the benchmark deviation; The normalization ratios are sorted in ascending order of value. The spectral background baseline slopes corresponding to the first two-thirds of the normalization ratios in the sort are retained, while the spectral background baseline slopes corresponding to the last one-third of the normalization ratios in the sort are removed. Calculate the relative standard deviation of the slope of each retained spectral background baseline as the background fluctuation dispersion.

[0010] Preferably, a penalty coefficient is calculated based on the surface state dispersion and the background fluctuation dispersion, and the net characteristic fluorescence intensity of each point is double-weighted and fused based on the penalty coefficient and the slope of the spectral background baseline at each point to obtain the effective characteristic intensity, including: Extract the surface state dispersion and the background fluctuation dispersion, calculate the sum of the two, and calculate the exponential function value with the natural constant as the base and the negative of the sum of the two as the exponent; The value of the exponential function is normalized and mapped to the interval between 0 and 1, and the mapping result is used as the penalty coefficient. Extract the net intensity of the characteristic fluorescence and the slope of the spectral background baseline corresponding to each measurement point; Using the reciprocal of the absolute value of the slope of each spectral background baseline as the initial weight, the net intensity of the characteristic fluorescence is weighted and averaged to obtain a reference weighted mean. Calculate the relative deviation between the net intensity of each characteristic fluorescence and the reference weighted mean, and use the penalty coefficient as the base and the relative deviation as the exponent to calculate the weight adjustment factor corresponding to each measurement point; Each initial weight is multiplied by the weight adjustment factor at the same measurement point to obtain the corrected weight. The net intensity of the characteristic fluorescence is then weighted and averaged using the corrected weight to obtain the effective characteristic intensity.

[0011] Preferably, the steps for constructing the mineral type association model include: Collect multiple historical mineral samples of the same type as the mineral product to be tested, and obtain the historical characteristic fluorescence net intensity sequence and historical scattering background intensity sequence corresponding to each historical mineral sample; Calculate the historical surface state dispersion corresponding to each historical mineral product sample, and fit the historical spectral background baseline slope of each historical mineral product sample to obtain the historical background fluctuation dispersion. Based on the historical surface state dispersion and historical background fluctuation dispersion of each historical mineral sample, the historical penalty coefficient is calculated, and the corresponding historical effective feature intensity is determined. Laboratory chemical analysis was performed on each historical mineral product sample to determine the reference content value of harmful elements in each historical mineral product sample. The historical effective characteristic intensity and reference content value of the same historical mineral product sample were combined to form a training sample pair. Training sample pairs of the same matrix type are aggregated into a matrix training subset. For each matrix type of matrix training subset, a regression mapping relationship between the historical effective feature intensity and the reference content value is fitted and established. The regression mapping relationship corresponding to each matrix type is used as the mineral type association model of the corresponding matrix.

[0012] Preferably, the calculation steps for the dynamic limit threshold include: Extract the surface state dispersion, the background fluctuation dispersion, and the penalty coefficient; Extract the reference weighted mean and extract the slope parameter of the regression mapping relationship between the mineral type association model and the matrix type of the mineral product to be tested; The square root of the sum of the squares of the surface state dispersion and the background fluctuation dispersion is calculated as the composite coefficient of variation. Using the penalty coefficient as an iterative shrinkage factor, the synthetic coefficient of variation is iteratively weighted until convergence, including: The current threshold factor is weighted and shrunken using the penalty coefficient, and a correction term is formed by weighting the synthetic coefficient of variation with the complement of the penalty coefficient to obtain the updated threshold factor. Determine the deviation between the updated threshold factor and the current threshold factor; When the deviation is less than the product of the penalty coefficient and the surface state dispersion, the iteration is terminated, and the update threshold factor is used as the stable threshold factor. Otherwise, replace the current threshold factor with the updated threshold factor and continue iterating; The basic dynamic threshold is obtained by multiplying the reference weighted mean, the slope parameter, and the stability threshold factor. The dynamic limit threshold is determined by integrating the basic dynamic threshold, the reference weighted mean, the slope parameter, the background fluctuation dispersion, and the penalty coefficient. The dynamic limit threshold is composed of the basic dynamic threshold and a broadening term that is jointly weighted by the background fluctuation dispersion and the penalty coefficient.

[0013] Preferably, after outputting the result of exceeding the standard, it also includes: Calculate the difference between the content value and the dynamic limit threshold, and divide the absolute value of the difference by the dynamic limit threshold to obtain the normalized deviation. Based on the penalty coefficient and the background fluctuation dispersion, the confidence base factor is determined, and the confidence score is calculated by combining the normalized deviation. A decision threshold is determined based on the penalty coefficient and the surface state dispersion, wherein the decision threshold is proportional to both the penalty coefficient and the surface state dispersion. When the confidence score is greater than or equal to the judgment threshold, the current exceeding judgment result is output as the final judgment result. When the confidence score is less than the judgment threshold, additional measurement points are set up on the target surface of the stack of mineral products to be tested, and the net intensity of characteristic fluorescence and the background intensity of scattering are collected at the additional points. The net characteristic fluorescence intensity corresponding to the supplementary points is incorporated into the existing net characteristic fluorescence intensity sequence, and the background scattering intensity corresponding to the supplementary points is simultaneously incorporated into the existing background scattering intensity sequence. The surface state dispersion and background fluctuation dispersion are recalculated, and double weighted fusion is performed again to obtain the supplemented effective characteristic intensity. The supplemented effective feature intensity is input into the mineral type association model to obtain the supplemented content value, which is then compared with the dynamic limit threshold. The supplemented exceedance judgment result is output as the final judgment result.

[0014] Secondly, the present invention also provides a non-destructive testing system for the content of harmful elements in mineral products, the system comprising: The data acquisition module is used to set up multiple measurement points on the target surface of the stack of mineral products to be tested, excite each point and collect the characteristic fluorescence net intensity sequence of the target harmful elements, and simultaneously acquire the background scattering intensity sequence corresponding to each point. The dispersion calculation module is used to calculate the relative fluorescence yield of each point based on the characteristic net fluorescence intensity and scattering background intensity of each point, and to calculate the relative standard deviation of the relative fluorescence yield among multiple points as the surface state dispersion. The background fluctuation fitting module is used to fit the spectral background baseline slope of each point with the count values ​​of multiple energy channels in the scattering background intensity sequence of each point as the dependent variable and the corresponding energy value as the independent variable, and to calculate the relative standard deviation of the spectral background baseline slope among multiple points as the background fluctuation dispersion. The dual-weighted fusion module is used to calculate the penalty coefficient based on the surface state dispersion and the background fluctuation dispersion, and to perform dual-weighted fusion of the net characteristic fluorescence intensity of each point based on the penalty coefficient and the spectral background baseline slope of each point to obtain the effective characteristic intensity. The content output module is used to input the effective feature intensity into the mineral type association model, output the content value of harmful elements in the mineral product to be tested, compare the content value with the dynamic limit threshold, and output the result of exceeding the standard.

[0015] One or more technical solutions provided in this invention have at least the following technical effects or advantages: First, this invention uses the relative standard deviation of the relative fluorescence yield across multiple locations as the surface state dispersion, and normalizes the net characteristic fluorescence intensity using the scattering background intensity to eliminate the influence of surface physical state differences on fluorescence excitation efficiency, thus solving the problem of fluctuations in the net characteristic fluorescence intensity caused by differences in surface roughness and particle size of mineral products. It also uses the relative standard deviation of the spectral background baseline slope across multiple locations as the background fluctuation dispersion, and utilizes the baseline slope fluctuation characteristics of the scattering background spectral region to quantify the influence of mineral matrix differences on the spectral background at different locations. This allows for independent and quantitative assessment of the surface inhomogeneity of mineral product stacks from two dimensions: surface physical state and matrix spectral response, avoiding the problem of incomplete assessment of complex interference sources using a single index.

[0016] Second, the present invention generates a penalty coefficient from the surface state dispersion and the background fluctuation dispersion. Using the inverse of the baseline slope as the initial weight and an exponential function with the penalty coefficient as the base as the adjustment factor, the net fluorescence intensity of each point is weighted and fused in a dual manner. When the dispersion is large, the contribution of the deviated points is automatically reduced, and when the dispersion is small, the balanced contribution of each point is retained, thus obtaining a more robust and effective feature intensity under non-uniform surface conditions. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating a non-destructive testing method for the content of harmful elements in mineral products provided in an embodiment of the present invention. Figure 2 This is a logic diagram of a non-destructive testing method for the content of harmful elements in mineral products provided in an embodiment of the present invention; Figure 3 This is a table showing an example of the layout parameters and the number of acquisition parameters for a non-destructive testing method for the content of harmful elements in mineral products provided in this embodiment of the invention. Figure 4 This is a structural diagram of a non-destructive testing system for the content of harmful elements in mineral products provided in an embodiment of the present invention; The diagram is labeled as follows: Data acquisition module 11, Discreteness calculation module 12, Background fluctuation fitting module 13, Dual weighted fusion module 14, Content output module 15. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0020] Example 1, as Figure 1 , Figure 2 As shown, the present invention provides a non-destructive testing method for the content of harmful elements in mineral products, the method comprising: S10: Multiple measurement points are set up on the target surface of the stack of mineral products to be tested, each point is excited and the characteristic fluorescence net intensity sequence of the target harmful element is collected, and the scattering background intensity sequence corresponding to each point is obtained simultaneously. At the customs inspection site for mineral products, handheld X-ray fluorescence spectrometers can directly perform non-destructive testing on mineral products without sample preparation. However, the surface of stacked mineral products exhibits non-uniform characteristics such as uneven particle size, local component segregation, and differences in surface roughness. If only a single point measurement is performed at a single location, the test results are easily affected by the local surface condition at that point, making it difficult to represent the true content of harmful elements in the entire batch of mineral products. By deploying multiple measurement points on the stacked surface and collecting characteristic fluorescence signals and scattered background signals at multiple points, the fluctuation characteristics of the signals at each point can be statistically analyzed in subsequent steps to identify the degree of surface non-uniformity and reduce its interference with the final test results.

[0021] Step S10 provided in this embodiment of the invention includes: The top surface of the stack of mineral products to be tested exposed to the outside world is used as the target surface. Multiple measurement points are arranged in a grid pattern within the target surface, with equal spacing between adjacent measurement points. The excitation source of the X-ray fluorescence spectrometer is vertically aligned with each measurement point, and the same excitation time is continuously applied at each measurement point to collect the energy dispersive X-ray fluorescence spectrum corresponding to each measurement point. From the energy dispersive X-ray fluorescence spectra of each measurement point, locate the characteristic X-ray fluorescence peak of the target harmful element, perform area integration on the characteristic X-ray fluorescence peak, and subtract the baseline background of linear interpolation on both sides of the peak to obtain the net intensity of characteristic fluorescence corresponding to each measurement point. In a continuous scattering spectrum region that does not overlap with the characteristic X-ray fluorescence peak, an equal-width energy channel is extracted, and the count value of each energy channel is extracted to form a scattering background intensity sequence corresponding to each measurement point. The center energy value of each energy channel in the scattering background intensity sequence is recorded synchronously. The characteristic fluorescence net intensities corresponding to each measurement point are arranged in the order of the measurement points to form a characteristic fluorescence net intensity sequence. Arrange the scattering background intensity sequence and the center energy value of the corresponding energy channel at each measurement point in the same order to form the scattering background intensity sequence and the corresponding energy independent variable sequence.

[0022] The target harmful element includes at least one of lead, mercury, cadmium, chromium and arsenic. The continuous scattering spectrum region that does not overlap with the characteristic X-ray fluorescence peak is the continuous scattering spectrum region located on the high-energy side of the characteristic X-ray fluorescence peak. The number of energy channels intercepted in the continuous scattering spectrum region is not less than 3. When the energy distance between the characteristic X-ray fluorescence peak and the adjacent high-energy side characteristic X-ray fluorescence peak is greater than 1 keV, 5 energy channels are intercepted.

[0023] The specific implementation method is as follows: The top surface of the mineral product stack exposed to the outside environment is used as the target surface. Specifically, mineral products are usually stored in stacks during customs inspection, with the top surface of the stack being the uppermost exposed surface. This allows for vertical access and measurement by a handheld X-ray fluorescence spectrometer from above, without needing to disturb the stack. Multiple measurement points are arranged in a grid pattern within the target surface, with equal spacing between adjacent measurement points. The number of measurement points is determined based on the area of ​​the stack's top surface and the typical non-uniformity scale of the mineral product surface. The grid spacing is determined by taking the minimum typical non-uniformity scale of the mineral product surface and setting the grid spacing to half of this minimum value, ensuring that each non-uniform region contains at least two measurement points, thus guaranteeing statistical representativeness of the point coverage. The minimum typical non-uniformity scale of the mineral product surface is determined by observing the scale of particle aggregation areas and component segregation patches on the stack surface.

[0024] For example, the top surface area of ​​a stack of iron ore products to be tested is about 2 meters × 2 meters. After observing the scale of the surface particle aggregation area and component segregation patches, the minimum typical non-uniform scale is about 0.4 meters. The grid spacing is set to 0.2 meters. The grids are arranged at equal intervals of 0.2 meters on the 2 meter × 2 meter top surface, resulting in a total of 100 measurement points in 10 rows × 10 columns.

[0025] The excitation source of the X-ray fluorescence spectrometer is vertically aligned with each measurement point. Specifically, the excitation source of the X-ray fluorescence spectrometer is generated by an X-ray tube, and the voltage and current of the excitation source are set according to the excitation efficiency requirements of different mineral matrices. Excitation is continued at each measurement point for the same duration, which is the preset single-point measurement duration of the X-ray fluorescence spectrometer. This parameter is set during instrument operation and is usually reasonably determined based on the relationship between the characteristic X-ray fluorescence intensity of the detected element and the detection time, ensuring that sufficient fluorescence signals are counted within a reasonable time. Maintaining a consistent excitation duration at each point ensures comparability of the number of spectrometers at each point on the same time scale. Energy-dispersive X-ray fluorescence spectra are collected for each measurement point. The horizontal axis of the energy-dispersive X-ray fluorescence spectrum represents energy (keV), and the vertical axis represents the count value.

[0026] For example, the excitation source voltage was set to 40 kV and the current to 20 μA, depending on the different mineral matrix types. The excitation time at each point was set to the instrument's preset 60 seconds. A total of 100 energy-dispersive X-ray fluorescence spectra were collected from 100 points. For example, Figure 3 This is an example table and diagram of the layout parameters and number of acquisition parameters for a non-destructive testing method for the content of harmful elements in mineral products provided by an embodiment of the present invention. It shows the grid layout parameters and spectral acquisition parameters of the measurement points on the surface of the stack of mineral products to be tested.

[0027] The characteristic X-ray fluorescence peaks of the target harmful elements are located from the energy-dispersive X-ray fluorescence spectra at each measurement point. Specifically, the target harmful elements include at least one of lead, mercury, cadmium, chromium, and arsenic, each of which has characteristic X-ray fluorescence peaks with different energies in the spectrum. The location of the characteristic X-ray fluorescence peaks is automatically completed by spectral analysis software or the analysis program built into the instrument, identifying the corresponding peak positions in the spectrum based on the characteristic energy values ​​of the target elements. The area of ​​the characteristic X-ray fluorescence peaks is integrated, and the baseline background of linear interpolation on both sides of the peak is subtracted to obtain the net intensity of the characteristic fluorescence at each measurement point, with units consistent with the spectral count values.

[0028] For example, taking the Lα characteristic X-ray fluorescence peak of lead as an example, the energy position of the peak is located at approximately 10.55 keV in the spectrum corresponding to the first measurement point. The count value is 15200 by integrating the peak area. Two points in the non-overlapping baseline region are selected on both sides of the peak, and the baseline value under the peak is determined by linear interpolation to be 430. After subtracting the baseline, the net count value is 14770.

[0029] In the continuous scattering spectrum region that does not overlap with the characteristic X-ray fluorescence peak, equal-width energy channels are extracted, and the count values ​​of each energy channel are extracted to construct the scattering background intensity sequence corresponding to each measurement point. Specifically, the continuous scattering spectrum region is selected as a pure scattering region located on the high-energy side of the characteristic X-ray fluorescence peak without any overlapping characteristic peaks. The count values ​​in this region mainly originate from the scattering of incident X-rays on the surface and matrix of the mineral product, reflecting the surface physical state and matrix scattering characteristics of the measurement point. The center energy value of each energy channel in this scattering background intensity sequence is recorded simultaneously. The width of the energy channel is usually determined by the energy resolution of the X-ray fluorescence spectrometer, which is a fixed technical specification of the instrument and can be directly adopted.

[0030] At least three energy channels should be selected within the continuous scattering spectrum region to ensure statistical stability. When the energy gap between the characteristic X-ray fluorescence peak and the adjacent high-energy side characteristic X-ray fluorescence peak is greater than 1 keV, it indicates a wide high-energy side scattering region; in this case, five energy channels should be selected to increase the number of data points for fitting the spectral background baseline slope. When the energy gap is no greater than 1 keV, it indicates a narrow high-energy side scattering region; in this case, three energy channels should be selected to avoid overlap and interference with characteristic peaks of adjacent elements.

[0031] For example, the Lα characteristic peak energy of lead is about 10.55 keV. The energy spacing between adjacent characteristic peaks on its high-energy side is greater than 1 keV and the scattering region is relatively wide. Five equally wide energy channels located on the high-energy side of this peak (for example, the channel center energies are 10.8, 11.0, 11.2, 11.4 and 11.6 keV respectively) are selected, and the count values ​​of each channel are extracted to form the scattering background intensity sequence of this measurement point. At the same time, the center energy value of each channel is recorded.

[0032] The characteristic fluorescence net intensities corresponding to each measurement point are arranged in order of measurement point location to form a characteristic fluorescence net intensity sequence. Specifically, the characteristic fluorescence net intensity sequence is a one-dimensional array with a length equal to the number of measurement points, and the i-th element corresponds to the characteristic fluorescence net intensity of the i-th measurement point in the sequence. The scattering background intensity sequence corresponding to each measurement point and the center energy value of the corresponding energy channel are arranged in the same order to form a scattering background intensity sequence and a corresponding energy independent variable sequence. The scattering background intensity sequence is a two-dimensional array with the number of rows equal to the number of measurement points and the number of columns equal to the number of energy channels, and each row corresponds to the scattering background intensity of the measurement point with the corresponding row number.

[0033] The following technical effects were achieved through this step: Spectral data were collected from multiple points on the top surface of the mineral product stack at equal intervals in a grid pattern. The grid spacing was dynamically determined based on the surface non-uniformity scale to ensure statistical representativeness of the detection coverage. Characteristic net fluorescence intensity sequences and scattering background intensity sequences were simultaneously extracted from the spectra of each point. The signals from each point were organized into sequence data in a unified order, providing scanning-style multi-dimensional raw data for subsequent calculations of surface state dispersion and background fluctuation dispersion.

[0034] S20: Based on the characteristic net fluorescence intensity and scattering background intensity of each point, calculate the relative fluorescence yield of each point, and calculate the relative standard deviation of the relative fluorescence yield among multiple points as the surface state dispersion. The surface of stacked mineral products exhibits intergranular gaps and localized roughness variations. These inhomogeneities in surface physical states lead to differences in the coupling efficiency between incident X-rays and the mineral product surface at different locations. For mineral products with uniform elemental content, the characteristic fluorescence yield is lower in rougher areas due to enhanced scattering, while it is higher in denser areas. The background scattering intensity originates from Compton scattering of incident X-rays on the mineral product surface. Its response direction to changes in surface physical state is consistent with the characteristic fluorescence, i.e., scattering is enhanced when the surface is rough and weakened when the surface is dense. Therefore, the background scattering intensity can be used as an internal standard parameter to assess the differences in surface physical states at various locations.

[0035] Step S20 provided in this embodiment of the invention includes: The net intensity of each characteristic fluorescence in the net intensity sequence of the characteristic fluorescence is extracted point by point, and the count value of each energy channel in the scattering background intensity sequence corresponding to the same measurement point is extracted. For the same measurement point, the count values ​​of each energy channel in the scattering background intensity sequence are weighted by the center energy value of the same position in the energy independent variable sequence to obtain the energy-weighted total scattering intensity. The characteristic net fluorescence intensity at the same measurement point is normalized using the energy-weighted total scattering intensity to obtain the relative fluorescence yield at each measurement point. By traversing all measurement points, a relative fluorescence yield sequence corresponding to each measurement point is obtained. The relative standard deviation of the relative fluorescence yield sequence is calculated as the surface state dispersion.

[0036] The specific implementation method is as follows: The net fluorescence intensity of each measurement point is extracted from the net fluorescence intensity sequence. Specifically, the net fluorescence intensity sequence is a one-dimensional array with a length equal to the number of measurement points, where the i-th element corresponds to the net fluorescence intensity of the i-th measurement point in the sequence. The net fluorescence intensity of each measurement point is extracted sequentially from the net fluorescence intensity sequence. Simultaneously, the count values ​​of each energy channel in the background scattering intensity sequence corresponding to the same measurement point are extracted. The background scattering intensity sequence is a two-dimensional array, with the number of rows equal to the number of measurement points and the number of columns equal to the number of energy channels, where the i-th row corresponds to the background scattering intensity sequence of the i-th measurement point in the sequence.

[0037] For the same measurement point, the count values ​​of each energy channel in the scattering background intensity sequence are weighted using the center energy values ​​of the same position in the energy independent variable sequence to obtain the energy-weighted total scattering intensity. Specifically, the energy independent variable sequence is a one-dimensional array composed of the center energy values ​​of each energy channel in the scattering background intensity sequence recorded synchronously in step S10, with a length equal to the number of energy channels N. The energy-weighted total scattering intensity = count value 1 ÷ center energy value 1 + count value 2 ÷ center energy value 2 + ... + count value N ÷ center energy value N. The purpose of using the center energy value as the weight denominator is that, in the continuous scattering spectrum region on the high-energy side, the Compton scattering cross section decreases with increasing energy, and the scattering background intensity naturally exhibits a decreasing trend with energy. Weighting by energy value can unify the count values ​​of each channel to the equivalent scattering cross section scale, making the scattering contribution of different energy channels have a comparable physical position in the energy-weighted total scattering intensity.

[0038] For example, the background scattering intensity sequence at a certain measurement point contains 5 energy channels, with center energy values ​​of 10.8 keV, 11.0 keV, 11.2 keV, 11.4 keV, and 11.6 keV, respectively. The count values ​​of the 5 channels are 1850, 1720, 1610, 1520, and 1450, respectively. The energy-weighted total scattering intensity = 1850 ÷ 10.8 + 1720 ÷ 11.0 + 1610 ÷ 11.2 + 1520 ÷ 11.4 + 1450 ÷ 11.6 ≈ 171.3 + 156.4 + 143.8 + 133.3 + 125.0 = 729.8.

[0039] The net characteristic fluorescence intensity at the same measurement point is normalized using the energy-weighted total scattering intensity to obtain the relative fluorescence yield for each measurement point. Specifically, the relative fluorescence yield = net characteristic fluorescence intensity at the same measurement point ÷ energy-weighted total scattering intensity at the same measurement point. The relative fluorescence yield is a dimensionless ratio representing the efficiency of the net characteristic fluorescence intensity production at a unit scattering level at that measurement point. A higher relative fluorescence yield indicates a stronger fluorescence response of harmful elements at that point after deducting the influence of surface physical state; fluctuations in the relative fluorescence yield between different points reflect differences caused by factors other than surface state (such as local component segregation). For example, if the net characteristic fluorescence intensity at this measurement point is 14770, the relative fluorescence yield = 14770 ÷ 729.8 ≈ 20.24.

[0040] Repeat the normalization operation above for all measurement sites to obtain a relative fluorescence yield sequence that corresponds one-to-one with each measurement site. The length of the relative fluorescence yield sequence is equal to the number of measurement sites.

[0041] The relative standard deviation of the relative fluorescence yield sequence is calculated as the surface state dispersion. Specifically, surface state dispersion = standard deviation of the relative fluorescence yield sequence ÷ arithmetic mean of the relative fluorescence yield sequence × 100%. Surface state dispersion represents the relative fluctuation range of the normalized fluorescence yield among each measurement point. The smaller the surface state dispersion, the better the consistency of the surface state of each region of the mineral product, and the more stable and reliable the measurement results of the characteristic fluorescence net intensity. The larger the surface state dispersion, the more significant the differences in surface state. The measurement results of individual points may deviate significantly due to local roughness anomalies or component segregation, requiring subsequent correction based on information such as background fluctuation dispersion. For example, the mean of the relative fluorescence yield sequence of 100 measurement points is 20.58, and the standard deviation is 3.62. Surface state dispersion = 3.62 ÷ 20.58 × 100% ≈ 17.6%.

[0042] The following technical effects were achieved through this step: The net intensity of characteristic fluorescence at each site was normalized using the background scattering intensity as an internal standard parameter. Utilizing the synchronous response of the background scattering to changes in surface physical state, the influence of surface particle size and roughness differences on fluorescence excitation efficiency was eliminated from the characteristic fluorescence signal. This ensures that the normalized relative fluorescence yield reflects only the relative level of harmful element content. By calculating the relative standard deviation of the relative fluorescence yield across multiple sites, the surface state dispersion was obtained as an indicator of the stability of the detection results under a unified dimension, providing an assessment basis for subsequent penalty coefficient calculation and dual-weighted fusion to evaluate surface consistency.

[0043] S30: Using the count values ​​of multiple energy channels in the scattering background intensity sequence at each point as the dependent variable and the corresponding energy values ​​as the independent variable, fit the spectral background baseline slope at each point, and calculate the relative standard deviation of the spectral background baseline slope among multiple points as the background fluctuation dispersion. The matrix composition varies across different regions of mineral products. For example, the types and proportions of gangue minerals in iron ore differ across regions, leading to local variations in the X-ray scattering cross-section and absorption coefficient. This manifests as fluctuations in the baseline slope of the background scattering spectrum at each location. These matrix composition differences also affect the detection results of the net intensity of characteristic fluorescence: regions with higher heavy element content in the matrix absorb incident X-rays and characteristic fluorescence more strongly, resulting in a lower net intensity of characteristic fluorescence; conversely, regions with higher light element content absorb less strongly, leading to a higher net intensity of characteristic fluorescence.

[0044] Step S30 provided in this embodiment of the invention includes: For each measurement point, count values ​​are extracted from the background scattering intensity sequence channel by channel, and center energy values ​​of the same order are extracted from the energy independent variable sequence. The center energy values ​​and count values ​​of the same channel are combined into a data pair. After traversing all energy channels, a set of data pairs for the same measurement point is formed. For each data pair in the data pair set, an energy weighting factor is determined based on the central energy value, and a counting statistical weighting factor is determined based on the count value; The product of the energy weighting factor and the count statistics weighting factor is used as the composite weight of the data pair in the linear regression. Using the center energy value as the independent variable and the count value as the dependent variable, apply the corresponding composite weights to all data pairs in the data pair set, perform a weighted least squares linear regression, and solve for the slope of the regression line, which is used as the slope of the spectral background baseline corresponding to the same measurement point. All the spectral background baseline slopes obtained by traversing point by point are arranged in the order of the points to form a spectral background baseline slope sequence. Calculate the arithmetic mean of the spectral background baseline slope sequence; Calculate the normalized deviation of the slope of each spectral background baseline relative to the arithmetic mean, and obtain the relative deviation of the slope corresponding to each spectral background baseline slope; Using the minimum value among the relative deviations of the slope as the benchmark deviation, calculate the normalized ratio of each relative deviation of the slope to the benchmark deviation; The normalization ratios are sorted in ascending order of value. The spectral background baseline slopes corresponding to the first two-thirds of the normalization ratios in the sort are retained, while the spectral background baseline slopes corresponding to the last one-third of the normalization ratios in the sort are removed. Calculate the relative standard deviation of the slope of each retained spectral background baseline as the background fluctuation dispersion.

[0045] The specific implementation method is as follows: For each measurement point, count values ​​are extracted channel by channel from the background scattering intensity sequence, and center energy values ​​of the same order are extracted from the energy independent variable sequence. The background scattering intensity sequence is a two-dimensional array, with the i-th row corresponding to the i-th measurement point, and each column of the row representing the count values ​​of each energy channel at that point. The energy independent variable sequence is a one-dimensional array, with each element representing the center energy value of each energy channel. The center energy value and count value of the same channel are paired to form a data pair. After traversing all energy channels, a set of data pairs for the same measurement point is formed. Let N be the number of energy channels extracted in step S10, then the data pair set for each measurement point contains N data pairs.

[0046] For each data pair in the dataset, an energy weighting factor is determined based on the central energy value, and a count statistical weighting factor is determined based on the count value. Specifically, the energy weighting factor = 1 ÷ central energy value. This energy weighting factor compensates for the systematically low count value of high-energy channels caused by the decrease in the high-energy sidescattering cross-section with increasing energy. This ensures a more balanced statistical contribution of data pairs from different energy channels in the weighted regression, preventing the low count values ​​of high-energy channels from being overwhelmed by the high count values ​​of low-energy channels, thereby improving the accuracy of baseline slope fitting. The count statistical weighting factor = 1 ÷ square root of the count value. This factor reflects the relative uncertainty of each channel's count value under Poisson counting statistics; the lower the count value, the greater the relative uncertainty, and the smaller the assigned weight. The product of the energy weighting factor and the count statistical weighting factor is used as the composite weight of the data pair in the linear regression.

[0047] Using the center energy as the independent variable and the count value as the dependent variable, a weighted least squares linear regression is performed on all data pairs in the data pair set, applying the corresponding composite weights. The slope of the regression line is then calculated and used as the slope of the spectral background baseline corresponding to the same measurement point. Specifically, the weighted least squares linear regression solves for the slope and intercept of the regression line by minimizing the weighted sum of squared residuals: Weighted sum of squared residuals = Σ[composite weights × (count value - slope × center energy value - intercept)]. 2 The slope obtained from the regression represents the direction and magnitude of the trend of the scattering background intensity at that point with energy. The larger the absolute value of the slope, the stronger the energy dependence of the scattering background spectrum, and the more significant the matrix scattering characteristics at that point. When the slope approaches zero, it indicates that the scattering background intensity has no obvious trend of change among different energy channels, and the matrix scattering characteristics at that point tend to be uniform. The slope can be positive or negative, and the positive or negative value reflects the upward or downward trend of the scattering spectrum within the intercepted energy range.

[0048] For example, in step S10, 5 energy channels are intercepted. The data pairs corresponding to the 5 energy channels at a certain measurement point are (10.8 keV, 1850 count value), (11.0 keV, 1720 count value), (11.2 keV, 1610 count value), (11.4 keV, 1520 count value), and (11.6 keV, 1450 count value). The count values ​​of each data pair show a trend of gradually decreasing from low energy to high energy. After weighted least squares regression fitting, the slope of this point is -498.5 per keV, and the intercept is 7246.2 count values. A negative slope indicates that the background scattering intensity decreases with increasing energy within this energy range, which is consistent with the physical law of the Compton scattering cross section. This measurement point is the first of 100 points.

[0049] All spectral background baseline slopes obtained by iterating through each measurement point are arranged in order to form a spectral background baseline slope sequence. Specifically, the spectral background baseline slope sequence is a one-dimensional array with a length equal to the number of measurement points, and the i-th element corresponds to the spectral background baseline slope of the i-th measurement point in the sequence. For example, in the spectral background baseline slope sequence of 100 measurement points, the first slope is -498.5, the second is -485.3, the third is -512.8, and so on, up to the 100th slope which is -476.2.

[0050] Calculate the arithmetic mean of the spectral background baseline slope sequence. For example, the arithmetic mean of 100 spectral background baseline slopes is approximately -492.0.

[0051] The normalized deviation of the slope of each spectral background baseline from its arithmetic mean is calculated one by one to obtain the relative deviation of the slope for each spectral background baseline. For example, the slope of the first spectral background baseline is -498.5, the arithmetic mean is -492.0, and the relative deviation of the slope is |(-498.5)-(-492.0)|÷|-492.0|=6.5÷492.0≈0.0132. The slope of the 100th spectral background baseline is -476.2, and the relative deviation of the slope is |(-476.2)-(-492.0)|÷492.0=15.8÷492.0≈0.0321.

[0052] Using the minimum value among the relative deviations of the slope as the benchmark deviation, calculate the normalized ratio of each relative deviation of the slope to the benchmark deviation. For example, if the minimum value among the 100 relative deviations of the slope is 0.002, and the benchmark deviation is 0.002, then the normalized ratio of the first deviation is 0.0132 ÷ 0.002 = 6.6, and the normalized ratio of the 100th deviation is 0.0321 ÷ 0.002 ≈ 16.1.

[0053] When the baseline deviation approaches zero, it indicates that the slope of the spectral background baseline at each point is almost completely consistent, and the measurement results at all points are highly similar. At this time, the normalization ratios are also close to consistent, and the elimination operation has no practical distinguishing significance. Therefore, the elimination step is skipped, and all elements of the spectral background baseline slope sequence are directly used in subsequent calculations.

[0054] When the baseline deviation is non-zero, the normalized ratios are sorted from smallest to largest. The spectral background baseline slopes corresponding to the first two-thirds of the normalized ratios are retained, while the spectral background baseline slopes corresponding to the last one-third are removed. The purpose of retaining the first two-thirds is that the spectral background baseline slopes corresponding to the points with smaller normalized ratios are relatively close to the mean level, representing the concentration trend of the overall matrix scattering background on the surface of the mineral product stack. Removing the last one-third of points with larger normalized ratios removes the relatively discrete spectral background baseline slopes from the dataset, avoiding their excessive contribution to the standard deviation calculation.

[0055] The relative standard deviation of the slopes of each retained spectral background baseline is calculated as the background fluctuation dispersion. Specifically, the background fluctuation dispersion = standard deviation of the retained slopes ÷ absolute value of the arithmetic mean of the retained slopes × 100%. Background fluctuation dispersion reflects the degree to which the scattered background baselines at each measurement point tend to be consistent overall. The smaller the background fluctuation dispersion, the better the homogeneity of the matrix composition of the mineral product, and the less interference from matrix effects on the net intensity of characteristic fluorescence. For example, 100 normalized ratios are sorted from smallest to largest. The spectral background baseline slopes corresponding to the first approximately 67 normalized ratios are retained, and approximately 33 are discarded. The standard deviation of the slopes of the 67 retained spectral background baselines is 8.5, and the absolute value of the mean is approximately 491.2. The background fluctuation dispersion = 8.5 ÷ 491.2 × 100% ≈ 1.73%.

[0056] The following technical effects were achieved through this step: By fitting the spectral background baseline slope to the high-energy side-scattering background region that does not overlap with the characteristic fluorescence net intensity peak, the computational interference of the characteristic fluorescence peak on the baseline fitting is avoided. A weighted least squares regression with a composite weight of the product of an energy weighting factor and a count statistics weighting factor is introduced when fitting the slope. The energy weighting factor compensates for the decrease in count values ​​caused by the attenuation of the high-energy side-scattering cross section, while the count statistics weighting factor reflects the relative uncertainty of the count values ​​of each channel under Poisson counting statistics. Compared with unweighted regression, this improves the balance of contributions from different energy channels in the baseline fitting.

[0057] The relative standard deviation is calculated as the background fluctuation dispersion by statistically filtering the relatively discrete spectral background baseline slope of the calculation results through normalized deviation and normalized ratio. This reduces the interference of extreme values ​​of the baseline slope caused by abrupt changes in matrix composition at abnormal points on the overall evaluation, and makes the background fluctuation dispersion more robust in reflecting the overall uniformity of the matrix composition of mineral products.

[0058] S40: Calculate the penalty coefficient based on the surface state dispersion and the background fluctuation dispersion, and perform double weighted fusion of the net characteristic fluorescence intensity of each point based on the penalty coefficient and the spectral background baseline slope of each point to obtain the effective characteristic intensity. Surface condition dispersion and background fluctuation dispersion independently assessed the non-uniformity of the mineral product stack surface from two dimensions: surface physical state and matrix spectral response. For uniform, consistent stacks, both dispersions are small, indicating low fluctuations in measurement results at each point and a high reliability of each point's contribution to the overall result. In this case, all points should participate in the calculation of the final effective characteristic intensity in a more balanced manner, without excessive bias towards the measurement results of a particular point. For non-uniform stacks, at least one of the two dispersions will increase significantly, indicating that the measurement results at some points deviate severely from the overall dispersion trend due to local surface roughness anomalies or compositional segregation. In this case, it is necessary to reduce the contribution of these discrete points.

[0059] Step S40 provided in this embodiment of the invention includes: Extract the surface state dispersion and the background fluctuation dispersion, calculate the sum of the two, and calculate the exponential function value with the natural constant as the base and the negative of the sum of the two as the exponent; The value of the exponential function is normalized and mapped to the interval between 0 and 1, and the mapping result is used as the penalty coefficient. Extract the net intensity of the characteristic fluorescence and the slope of the spectral background baseline corresponding to each measurement point; Using the reciprocal of the absolute value of the slope of each spectral background baseline as the initial weight, the net intensity of the characteristic fluorescence is weighted and averaged to obtain a reference weighted mean. Calculate the relative deviation between the net intensity of each characteristic fluorescence and the reference weighted mean, and use the penalty coefficient as the base and the relative deviation as the exponent to calculate the weight adjustment factor corresponding to each measurement point; Each initial weight is multiplied by the weight adjustment factor at the same measurement point to obtain the corrected weight. The net intensity of the characteristic fluorescence is then weighted and averaged using the corrected weight to obtain the effective characteristic intensity.

[0060] The specific implementation method is as follows: The exponential function value is calculated using the natural constant as the base and the negative of the sum of the two discrete values ​​as the exponent. Specifically, the natural constant is e. When the sum of the two discrete values ​​is large, the negative of the negative value is large, and the exponential function value approaches zero; when the sum of the two discrete values ​​is small, the negative of the negative value is small, and the exponential function value approaches 1. The exponential function value is normalized and mapped to the interval between 0 and 1, and the mapping result is used as the penalty coefficient. Penalty coefficient = (exponential function value + 1) ÷ 2. When the sum of the two discrete values ​​approaches 0, the exponential function value approaches 1, and the penalty coefficient approaches 1; when the sum of the two discrete values ​​increases, the exponential function value approaches 0, and the penalty coefficient approaches 0.5. The penalty coefficient decreases (approaches 0.5) as the sum of the surface state discrete value and the background fluctuation discrete value increases, and increases (approaches 1) as the sum of the two decreases.

[0061] For example, the surface state dispersion is 17.6%, and the background fluctuation dispersion is 1.73%, the sum of which is 17.6% + 1.73% = 19.33% = 0.1933. -0.1933 It is approximately equal to 0.824. The penalty coefficient = (0.824 + 1) ÷ 2 = 0.912. The penalty coefficient of 0.912 is relatively close to 1, indicating that the overall level of surface condition dispersion and background fluctuation dispersion is moderately low, and the consistency of measurement results at each point is acceptable.

[0062] Extract the characteristic fluorescence net intensity and spectral background baseline slope corresponding to each measurement point. Specifically, the characteristic fluorescence net intensity of each measurement point is collected in step S10 and constitutes a characteristic fluorescence net intensity sequence, and the spectral background baseline slope of each measurement point is calculated in step S30 and constitutes a spectral background baseline slope sequence, with the point order of both being the same. The characteristic fluorescence net intensity of the i-th measurement point is Fi, and the spectral background baseline slope of the i-th measurement point is Ki.

[0063] Using the reciprocal of the absolute value of the slope of each spectral background baseline as the initial weight, a weighted average of the net intensity of characteristic fluorescence is calculated to obtain a reference weighted average. Specifically, the larger the absolute value of the slope of the spectral background baseline, the stronger the energy dependence of the scattering background spectrum at that point, and the more significant the influence of matrix scattering characteristics on the measurement at that point, thus assigning a smaller initial weight; conversely, the smaller the absolute value of the slope, the smaller the influence of matrix scattering characteristics on the measurement at that point, thus assigning a larger initial weight. The initial weight Wiinitial for the i-th measurement point is 1 ÷ |Ki|. The reference weighted average is calculated as Σ(Wiinitial × Fi) ÷ ΣWiinitial. The reference weighted average represents the weighted average level of the net intensity of characteristic fluorescence at each point, weighted by matrix scattering characteristics, and serves as a reference benchmark for subsequent calculations of the relative deviation values ​​at each point.

[0064] Calculate the relative deviation of the net fluorescence intensity of each feature from the reference weighted mean. Specifically, the relative deviation is calculated as |Fi - reference weighted mean| ÷ reference weighted mean. The relative deviation represents the proportion by which the net fluorescence intensity of the feature at that location deviates from the reference standard. The larger the deviation, the more significant the influence of local anomalies on the surface or matrix at that location.

[0065] The weight adjustment factor for each measurement point is calculated using the penalty coefficient as the base and the relative deviation value as the exponent. Specifically, the penalty coefficient ranges from 0.5 to 1. When the penalty coefficient is close to 1, the change in the relative deviation value has little impact on the weight adjustment factor, and the weight adjustment factor for each point approaches 1, with each point participating in the fusion in a manner close to its initial weight. When the penalty coefficient is close to 0.5, the change in the relative deviation value has an exponential amplification effect on the weight adjustment factor. The weight adjustment factor for points with larger deviations decays rapidly, while the weight adjustment factor for points with smaller deviations remains at a high level.

[0066] For example, the first measurement point is selected, with an initial weight of W1, a net characteristic fluorescence intensity of F1, and a reference weighted mean of Fref. The relative deviation is calculated as |F1 - Fref| ÷ Fref. The weight adjustment factor is 0.912 raised to the power of the relative deviation. If the relative deviation is 0.15, the weight adjustment factor is 0.912 to the power of 0.15, approximately equal to 0.986; if the relative deviation is 0.50, the weight adjustment factor is 0.912 to the power of 0.50, approximately equal to 0.955.

[0067] The corrected weight is obtained by multiplying each initial weight by the weight adjustment factor for the same measurement point. Corrected weight Wicorrected = Wiinitial × weight adjustment factor.

[0068] The effective feature intensity is obtained by weighting the net intensity of the characteristic fluorescence with corrected weights. Effective feature intensity = Σ(Wicorrected × Fi) ÷ ΣWicorrected. The effective feature intensity is the final output value after double weighted fusion, which integrates the differences in matrix scattering characteristics at each measurement point (first-layer initial weights) and the degree of signal deviation (second-layer adjustment factor), enabling robust estimation of elemental signal characteristic intensity under inhomogeneous surface conditions.

[0069] The following technical effects were achieved through this step: The penalty coefficient, generated from surface state dispersion and background fluctuation dispersion, serves as a high-level control factor for regulating the contribution intensity of the deviation signal at each point, achieving a mathematical synthesis of the quality assessment results across two dimensions: surface physical state and matrix scattering characteristics. Using the inverse of the slope of the spectral background baseline at each point as the initial weight, points that significantly affect matrix scattering characteristics are assigned lower weights during fusion. A weight adjustment factor, with the penalty coefficient as the base and the relative deviation value as the exponent, ensures balanced contributions from each point when surface uniformity is good (penalty coefficient close to 1), and automatically reduces the contribution of points with larger deviations when the surface is inhomogeneous (penalty coefficient close to 0.5). The reduction magnitude is adaptively correlated with the degree of deviation and the overall dispersion level, enabling a more robust and effective feature intensity estimate under inhomogeneous surface conditions.

[0070] S50: Input the effective feature intensity into the mineral type association model, output the content value of harmful elements in the mineral product to be tested, compare the content value with the dynamic limit threshold, and output the result of exceeding the standard.

[0071] The effective characteristic intensity integrates the characteristic fluorescence signals from multiple measurement points and eliminates the influence of surface condition differences and matrix spectral background fluctuations, thus representing the fluorescence response level of the entire batch of mineral products in terms of the target harmful element. However, the effective characteristic intensity is a fluorescence signal intensity value related to element content, and the correspondence between it and element content needs to be calibrated using known samples, which involves the construction of a mineral type association model. Furthermore, mineral products require content exceeding limits during customs inspection, and the threshold for this determination is dynamically changing and cannot use a fixed limit value. This is because when the detection quality of the surface and matrix is ​​poor, the reliability of the detection results will also decrease. In this case, it is necessary to appropriately adjust the determination threshold to avoid misjudgments that could lead to trade disputes.

[0072] Step S50 provided in this embodiment of the invention includes: The steps for constructing the mineral type association model include: Collect multiple historical mineral samples of the same type as the mineral product to be tested, and obtain the historical characteristic fluorescence net intensity sequence and historical scattering background intensity sequence corresponding to each historical mineral sample; Calculate the historical surface state dispersion corresponding to each historical mineral product sample, and fit the historical spectral background baseline slope of each historical mineral product sample to obtain the historical background fluctuation dispersion. Based on the historical surface state dispersion and historical background fluctuation dispersion of each historical mineral sample, the historical penalty coefficient is calculated, and the corresponding historical effective feature intensity is determined. Laboratory chemical analysis was performed on each historical mineral product sample to determine the reference content value of harmful elements in each historical mineral product sample. The historical effective characteristic intensity and reference content value of the same historical mineral product sample were combined to form a training sample pair. Training sample pairs of the same matrix type are aggregated into a matrix training subset. For each matrix type of matrix training subset, a regression mapping relationship between the historical effective feature intensity and the reference content value is fitted and established. The regression mapping relationship corresponding to each matrix type is used as the mineral type association model of the corresponding matrix. The calculation steps for the dynamic limit threshold include: Extract the surface state dispersion, the background fluctuation dispersion, and the penalty coefficient; Extract the reference weighted mean and extract the slope parameter of the regression mapping relationship between the mineral type association model and the matrix type of the mineral product to be tested; The square root of the sum of the squares of the surface state dispersion and the background fluctuation dispersion is calculated as the composite coefficient of variation. Using the penalty coefficient as an iterative shrinkage factor, the synthetic coefficient of variation is iteratively weighted until convergence, including: The current threshold factor is weighted and shrunken using the penalty coefficient, and a correction term is formed by weighting the synthetic coefficient of variation with the complement of the penalty coefficient to obtain the updated threshold factor. Determine the deviation between the updated threshold factor and the current threshold factor; When the deviation is less than the product of the penalty coefficient and the surface state dispersion, the iteration is terminated, and the update threshold factor is used as the stable threshold factor. Otherwise, replace the current threshold factor with the updated threshold factor and continue iterating; The basic dynamic threshold is obtained by multiplying the reference weighted mean, the slope parameter, and the stability threshold factor. The dynamic limit threshold is determined by integrating the basic dynamic threshold, the reference weighted mean, the slope parameter, the background fluctuation dispersion, and the penalty coefficient. The dynamic limit threshold is composed of the basic dynamic threshold and a broadening term that is jointly weighted by the background fluctuation dispersion and the penalty coefficient.

[0073] The specific implementation method is as follows: The mineral type association model is pre-built before the actual samples are tested, and its construction process is as follows: Collect multiple historical mineral samples of the same type as the mineral being tested. For example, mineral types include matrix types such as iron ore, chromium ore, and copper-gold ore. For each mineral type, collect a number of historical mineral samples of that type; for example, collect 50 iron ore samples, 40 chromium ore samples, and 30 copper-gold ore samples.

[0074] Obtain the historical characteristic fluorescence net intensity sequence and historical scattering background intensity sequence corresponding to each historical mineral product sample, and calculate the historical surface state dispersion, historical background fluctuation dispersion and historical effective feature intensity corresponding to each historical mineral product sample in the same way as steps S10 to S40.

[0075] Laboratory chemical analysis was performed on each historical mineral sample to determine the reference content values ​​of harmful elements in each sample. Specifically, the historical effective characteristic intensity and reference content value of the same historical mineral sample were combined into a training sample pair. Training sample pairs of the same matrix type were aggregated into a matrix training subset. For each matrix type's matrix training subset, a linear regression mapping relationship between the historical effective characteristic intensity and the reference content value was fitted. The form of the linear regression is: Reference content value = Slope parameter a × Historical effective characteristic intensity + Intercept parameter b. The solution method uses the least squares method, i.e., fitting a straight line that minimizes the sum of the squares of the vertical distances from all training points to this line. The regression mapping relationship corresponding to each matrix type is used as the mineral type association model for the corresponding matrix.

[0076] For example, laboratory chemical analysis was performed on 50 iron ore samples to determine the reference content of lead in each sample using inductively coupled plasma mass spectrometry (ICP-MS), with units on the order of milligrams per kilogram. The historical effective characteristic intensity of each sample was plotted on the x-axis, and the reference content value on the y-axis. The regression line was fitted with a slope parameter a of 0.082 and an intercept parameter b of 15.6.

[0077] The calculation steps for the dynamic limit threshold include: The surface state dispersion is extracted in step S20, the background fluctuation dispersion is extracted in step S30, the penalty coefficient and the reference weighted mean are extracted in step S40, and the slope parameter 'a' of the regression mapping relationship corresponding to the matrix type of the mineral product to be tested is extracted from the mineral type association model. For example, the mineral product to be tested is iron ore, and the slope parameter 'a' of the corresponding model is 0.082.

[0078] The square root of the sum of the squares of surface state dispersion and background fluctuation dispersion is used as the composite coefficient of variation. Specifically, the composite coefficient of variation integrates the detection quality fluctuations from both surface physics and matrix spectral dimensions; a higher value indicates lower overall reliability of the current detection. For example, with a surface state dispersion of 17.6% and a background fluctuation dispersion of 1.73%, the composite coefficient of variation = It is approximately equal to 0.177.

[0079] Using the penalty coefficient as an iterative shrinkage factor, the synthetic coefficient of variation is iteratively weighted until convergence, resulting in a stable threshold factor for determining the threshold width. Specifically, the iterative process is as follows: Assuming the initial threshold factor is 0, in each iteration, the updated threshold factor = penalty coefficient × current threshold factor + (1 - penalty coefficient) × synthetic coefficient of variation. The deviation before and after the update is calculated as |updated threshold factor - current threshold factor|. When the deviation is less than the product of the penalty coefficient and the surface state dispersion, the iteration terminates, and the updated threshold factor at this point is taken as the stable threshold factor. For example, the penalty coefficient is 0.912, the synthetic coefficient of variation is 0.177, and the surface state dispersion is 0.176. After multiple iterations, the deviation meets the preset condition, and the output stable threshold factor is approximately 0.156.

[0080] The basic dynamic threshold is obtained by multiplying the reference weighted mean, the slope parameter 'a', and the stability threshold factor. Specifically, this basic dynamic threshold reflects a fundamental judgment limit for anomaly detection signals at the current detected surface and matrix fluctuation levels. For example, with a reference weighted mean of 14850, a slope parameter 'a' of 0.082, and a stability threshold factor of 0.156, the basic dynamic threshold = 14850 × 0.082 × 0.156 ≈ 190.0.

[0081] Based on the basic dynamic threshold, necessary expansion is made to obtain the final dynamic limit threshold. Dynamic limit threshold = Basic dynamic threshold + Expansion term. Expansion term = Reference weighted mean × Slope parameter a × Background fluctuation dispersion × (1 - Penalty coefficient). The significance of introducing the expansion term is: when background fluctuation is large, matrix consistency is poor (large background fluctuation dispersion), or surface condition is poor, or measurement is unreliable (penalty coefficient close to 0.5), the threshold needs to be appropriately relaxed to avoid misjudging qualified goods as exceeding the limit. For example, the reference weighted mean is 14850, the slope parameter a is 0.082, the background fluctuation dispersion is 0.0173, the penalty coefficient is 0.912, and the expansion term = 14850 × 0.082 × 0.0173 × (1 - 0.912) ≈ 1.85. The dynamic limit threshold = 190.0 + 1.85 ≈ 191.85 mg / kg.

[0082] The effective feature intensity obtained in step S40 is input into the mineral type association model, substituted into the regression equation, and the content value of harmful elements in the tested mineral product is output. Content value = slope parameter a × effective feature intensity + intercept parameter b. For example, if the effective feature intensity is 14820, a is 0.082, b is 15.6, the content value = 0.082 × 14820 + 15.6 = 1215.24 + 15.6 = 1230.84 mg / kg. The calculated content value is compared with the dynamic limit threshold, and the result of exceeding the limit is output. For example, if the content value is 1230.84 mg / kg, which is far higher than the dynamic limit threshold of 191.85 mg / kg, it is determined to be exceeding the limit.

[0083] Step S50 provided in this embodiment of the invention further includes: After outputting the result of the out-of-limit judgment, it also includes: Calculate the difference between the content value and the dynamic limit threshold, and divide the absolute value of the difference by the dynamic limit threshold to obtain the normalized deviation. Based on the penalty coefficient and the background fluctuation dispersion, the confidence base factor is determined, and the confidence score is calculated by combining the normalized deviation. A decision threshold is determined based on the penalty coefficient and the surface state dispersion, wherein the decision threshold is proportional to both the penalty coefficient and the surface state dispersion. When the confidence score is greater than or equal to the judgment threshold, the current exceeding judgment result is output as the final judgment result. When the confidence score is less than the judgment threshold, additional measurement points are set up on the target surface of the stack of mineral products to be tested, and the net intensity of characteristic fluorescence and the background intensity of scattering are collected at the additional points. The net characteristic fluorescence intensity corresponding to the supplementary points is incorporated into the existing net characteristic fluorescence intensity sequence, and the background scattering intensity corresponding to the supplementary points is simultaneously incorporated into the existing background scattering intensity sequence. The surface state dispersion and background fluctuation dispersion are recalculated, and double weighted fusion is performed again to obtain the supplemented effective characteristic intensity. The supplemented effective feature intensity is input into the mineral type association model to obtain the supplemented content value, which is then compared with the dynamic limit threshold. The supplemented exceedance judgment result is output as the final judgment result.

[0084] The specific implementation method is as follows: Calculate the difference between the content value and the dynamic limit threshold, divide the absolute value of the difference by the dynamic limit threshold to obtain the normalized deviation, which is used to measure the relative degree of exceeding the threshold. For example, normalized deviation = |1230.84-192.7|÷191.85≈5.41.

[0085] Based on the penalty coefficient and background fluctuation dispersion, a basic confidence factor is determined, and a confidence score is calculated by combining it with the normalized deviation. The confidence score = normalized deviation × penalty coefficient × (1 - background fluctuation dispersion), used to comprehensively assess the reliability of the detection exceeding the standard. A higher value indicates greater confidence in the determination of exceeding the standard. For example, the confidence score = 5.41 × 0.912 × (1 - 0.0173) ≈ 4.85.

[0086] A decision threshold is determined based on the penalty coefficient and surface state dispersion. Decision threshold = penalty coefficient × surface state dispersion. The smaller the surface state dispersion, the better the consistency of fluorescence yield at the measurement sites; the lower the decision threshold, the less likely it is to trigger supplementary measurements. For example, decision threshold = 0.912 × 0.176 ≈ 0.161.

[0087] When the confidence score is greater than or equal to the decision threshold, it indicates that the current exceedance decision has sufficient confidence, and the current exceedance decision is output as the final decision result. For example, a confidence score of 4.85 is much greater than the decision threshold of 0.161, indicating a high confidence level in the decision result. No further measurement is required, and the final decision result is output directly.

[0088] When the confidence score is less than the judgment threshold, it indicates that the confidence of the current exceedance judgment result is insufficient. Additional measurement points are then set up on the target surface of the mineral product stack, and the net intensity of characteristic fluorescence and the background scattering intensity are collected at these additional points. The net intensity of characteristic fluorescence corresponding to the additional points is merged into the existing net intensity sequence of characteristic fluorescence, and simultaneously, the background scattering intensity corresponding to the additional points is merged into the existing background scattering intensity sequence. The surface state dispersion and background fluctuation dispersion are recalculated, and a double-weighted fusion is performed again to obtain the supplemented effective characteristic intensity. The supplemented effective characteristic intensity is input into the mineral type association model to obtain the supplemented content value, which is then compared with the dynamic limit threshold. The supplemented exceedance judgment result is output as the final judgment result.

[0089] The following technical effects were achieved through this step: A regression mapping relationship between effective feature intensities and laboratory chemical analysis reference values ​​was established for different historical samples of the same mineral type, converting fluorescence signal intensity into physically meaningful elemental content values. Simultaneously, the judgment threshold was dynamically adjusted based on the currently measured surface state and matrix fluctuation information, making the judgment limit adaptive to the reliability of the current measurement conditions. When the detection quality is poor, the judgment criteria were appropriately relaxed to reduce the risk of misjudgment. When the judgment confidence is insufficient, automatic supplementary measurements were used to improve the statistical representativeness and reliability of the detection results, realizing an adaptive closed-loop detection system driven by confidence, moving from single static judgment to confidence-based judgment.

[0090] Example 2, as Figure 4 As shown, based on the same inventive concept provided in Embodiment 1, this embodiment of the invention also provides a non-destructive testing system for the content of harmful elements in mineral products, the system comprising: Data acquisition module 11 is used to set up multiple measurement points on the target surface of the stack of mineral products to be tested, excite each point and collect the characteristic fluorescence net intensity sequence of the target harmful elements, and simultaneously acquire the scattering background intensity sequence corresponding to each point. The dispersion calculation module 12 is used to calculate the relative fluorescence yield of each point based on the characteristic net fluorescence intensity and scattering background intensity of each point, and to calculate the relative standard deviation of the relative fluorescence yield among multiple points as the surface state dispersion. Background fluctuation fitting module 13 is used to fit the spectral background baseline slope of each point with the count value of multiple energy channels in the scattering background intensity sequence of each point as the dependent variable and the corresponding energy value as the independent variable, and calculate the relative standard deviation of the spectral background baseline slope among multiple points as the background fluctuation dispersion. The dual-weighted fusion module 14 is used to calculate the penalty coefficient based on the surface state dispersion and the background fluctuation dispersion, and to perform dual-weighted fusion of the net characteristic fluorescence intensity of each point based on the penalty coefficient and the spectral background baseline slope of each point to obtain the effective characteristic intensity. The content output module 15 is used to input the effective feature intensity into the mineral type association model, output the content value of harmful elements in the mineral product to be tested, compare the content value with the dynamic limit threshold, and output the result of exceeding the standard.

[0091] In one embodiment, the data acquisition module 11 is further configured to use the top surface of the stack of mineral products to be tested exposed to the outside as the target surface, and to arrange multiple measurement points in a grid-like path within the target surface, with the spacing between adjacent measurement points being equal. The excitation source of the X-ray fluorescence spectrometer is vertically aligned with each measurement point, and the same excitation time is continuously applied at each measurement point to collect the energy dispersive X-ray fluorescence spectrum corresponding to each measurement point. From the energy dispersive X-ray fluorescence spectra of each measurement point, locate the characteristic X-ray fluorescence peak of the target harmful element, perform area integration on the characteristic X-ray fluorescence peak, and subtract the baseline background of linear interpolation on both sides of the peak to obtain the net intensity of characteristic fluorescence corresponding to each measurement point. In a continuous scattering spectrum region that does not overlap with the characteristic X-ray fluorescence peak, an equal-width energy channel is extracted, and the count value of each energy channel is extracted to form a scattering background intensity sequence corresponding to each measurement point. The center energy value of each energy channel in the scattering background intensity sequence is recorded synchronously. The characteristic fluorescence net intensities corresponding to each measurement point are arranged in the order of the measurement points to form a characteristic fluorescence net intensity sequence. Arrange the scattering background intensity sequence and the center energy value of the corresponding energy channel at each measurement point in the same order to form the scattering background intensity sequence and the corresponding energy independent variable sequence.

[0092] The target harmful element includes at least one of lead, mercury, cadmium, chromium and arsenic. The continuous scattering spectrum region that does not overlap with the characteristic X-ray fluorescence peak is the continuous scattering spectrum region located on the high-energy side of the characteristic X-ray fluorescence peak. The number of energy channels intercepted in the continuous scattering spectrum region is not less than 3. When the energy distance between the characteristic X-ray fluorescence peak and the adjacent high-energy side characteristic X-ray fluorescence peak is greater than 1 keV, 5 energy channels are intercepted.

[0093] In one embodiment, the discreteness calculation module 12 is further used to extract the net intensity of each characteristic fluorescence in the net intensity sequence of characteristic fluorescence point by point, and to extract the count value of each energy channel in the scattering background intensity sequence corresponding to the same measurement point. For the same measurement point, the count values ​​of each energy channel in the scattering background intensity sequence are weighted by the center energy value of the same position in the energy independent variable sequence to obtain the energy-weighted total scattering intensity. The characteristic net fluorescence intensity at the same measurement point is normalized using the energy-weighted total scattering intensity to obtain the relative fluorescence yield at each measurement point. By traversing all measurement points, a relative fluorescence yield sequence corresponding to each measurement point is obtained. The relative standard deviation of the relative fluorescence yield sequence is calculated as the surface state dispersion.

[0094] In one embodiment, the background fluctuation fitting module 13 is also used to extract count values ​​from the scattering background intensity sequence channel by channel for each measurement point, and extract the center energy value of the same order from the energy independent variable sequence, and form a data pair by combining the center energy value and the count value of the same channel, and form a set of data pairs for the same measurement point after traversing all energy channels. For each data pair in the data pair set, an energy weighting factor is determined based on the central energy value, and a counting statistical weighting factor is determined based on the count value; The product of the energy weighting factor and the count statistics weighting factor is used as the composite weight of the data pair in the linear regression. Using the center energy value as the independent variable and the count value as the dependent variable, apply the corresponding composite weights to all data pairs in the data pair set, perform a weighted least squares linear regression, and solve for the slope of the regression line, which is used as the slope of the spectral background baseline corresponding to the same measurement point. All the spectral background baseline slopes obtained by traversing point by point are arranged in the order of the points to form a spectral background baseline slope sequence. Calculate the arithmetic mean of the spectral background baseline slope sequence; Calculate the normalized deviation of the slope of each spectral background baseline relative to the arithmetic mean, and obtain the relative deviation of the slope corresponding to each spectral background baseline slope; Using the minimum value among the relative deviations of the slope as the benchmark deviation, calculate the normalized ratio of each relative deviation of the slope to the benchmark deviation; The normalization ratios are sorted in ascending order of value. The spectral background baseline slopes corresponding to the first two-thirds of the normalization ratios in the sort are retained, while the spectral background baseline slopes corresponding to the last one-third of the normalization ratios in the sort are removed. Calculate the relative standard deviation of the slope of each retained spectral background baseline as the background fluctuation dispersion.

[0095] In one embodiment, the dual-weighted fusion module 14 is further used to extract the surface state dispersion and the background fluctuation dispersion, calculate the sum of the two, and calculate the exponential function value with the natural constant as the base and the negative of the sum of the two as the exponent. The value of the exponential function is normalized and mapped to the interval between 0 and 1, and the mapping result is used as the penalty coefficient. Extract the net intensity of the characteristic fluorescence and the slope of the spectral background baseline corresponding to each measurement point; Using the reciprocal of the absolute value of the slope of each spectral background baseline as the initial weight, the net intensity of the characteristic fluorescence is weighted and averaged to obtain a reference weighted mean. Calculate the relative deviation between the net intensity of each characteristic fluorescence and the reference weighted mean, and use the penalty coefficient as the base and the relative deviation as the exponent to calculate the weight adjustment factor corresponding to each measurement point; Each initial weight is multiplied by the weight adjustment factor at the same measurement point to obtain the corrected weight. The net intensity of the characteristic fluorescence is then weighted and averaged using the corrected weight to obtain the effective characteristic intensity.

[0096] In one embodiment, the content output module 15 is also used in the mineral type association model construction step, which includes: Collect multiple historical mineral samples of the same type as the mineral product to be tested, and obtain the historical characteristic fluorescence net intensity sequence and historical scattering background intensity sequence corresponding to each historical mineral sample; Calculate the historical surface state dispersion corresponding to each historical mineral product sample, and fit the historical spectral background baseline slope of each historical mineral product sample to obtain the historical background fluctuation dispersion. Based on the historical surface state dispersion and historical background fluctuation dispersion of each historical mineral sample, the historical penalty coefficient is calculated, and the corresponding historical effective feature intensity is determined. Laboratory chemical analysis was performed on each historical mineral product sample to determine the reference content value of harmful elements in each historical mineral product sample. The historical effective characteristic intensity and reference content value of the same historical mineral product sample were combined to form a training sample pair. Training sample pairs of the same matrix type are aggregated into a matrix training subset. For each matrix type of matrix training subset, a regression mapping relationship between the historical effective feature intensity and the reference content value is fitted and established. The regression mapping relationship corresponding to each matrix type is used as the mineral type association model of the corresponding matrix.

[0097] The calculation steps for the dynamic limit threshold include: Extract the surface state dispersion, the background fluctuation dispersion, and the penalty coefficient; Extract the reference weighted mean and extract the slope parameter of the regression mapping relationship between the mineral type association model and the matrix type of the mineral product to be tested; The square root of the sum of the squares of the surface state dispersion and the background fluctuation dispersion is calculated as the composite coefficient of variation. Using the penalty coefficient as an iterative shrinkage factor, the synthetic coefficient of variation is iteratively weighted until convergence, including: The current threshold factor is weighted and shrunken using the penalty coefficient, and a correction term is formed by weighting the synthetic coefficient of variation with the complement of the penalty coefficient to obtain the updated threshold factor. Determine the deviation between the updated threshold factor and the current threshold factor; When the deviation is less than the product of the penalty coefficient and the surface state dispersion, the iteration is terminated, and the update threshold factor is used as the stable threshold factor. Otherwise, replace the current threshold factor with the updated threshold factor and continue iterating; The basic dynamic threshold is obtained by multiplying the reference weighted mean, the slope parameter, and the stability threshold factor. The dynamic limit threshold is determined by integrating the basic dynamic threshold, the reference weighted mean, the slope parameter, the background fluctuation dispersion, and the penalty coefficient. The dynamic limit threshold is composed of the basic dynamic threshold and a broadening term that is jointly weighted by the background fluctuation dispersion and the penalty coefficient.

[0098] Calculate the difference between the content value and the dynamic limit threshold, and divide the absolute value of the difference by the dynamic limit threshold to obtain the normalized deviation. Based on the penalty coefficient and the background fluctuation dispersion, the confidence base factor is determined, and the confidence score is calculated by combining the normalized deviation. A decision threshold is determined based on the penalty coefficient and the surface state dispersion, wherein the decision threshold is proportional to both the penalty coefficient and the surface state dispersion. When the confidence score is greater than or equal to the judgment threshold, the current exceeding judgment result is output as the final judgment result. When the confidence score is less than the judgment threshold, additional measurement points are set up on the target surface of the stack of mineral products to be tested, and the net intensity of characteristic fluorescence and the background intensity of scattering are collected at the additional points. The net characteristic fluorescence intensity corresponding to the supplementary points is incorporated into the existing net characteristic fluorescence intensity sequence, and the background scattering intensity corresponding to the supplementary points is simultaneously incorporated into the existing background scattering intensity sequence. The surface state dispersion and background fluctuation dispersion are recalculated, and double weighted fusion is performed again to obtain the supplemented effective characteristic intensity. The supplemented effective feature intensity is input into the mineral type association model to obtain the supplemented content value, which is then compared with the dynamic limit threshold. The supplemented exceedance judgment result is output as the final judgment result.

Claims

1. A non-destructive testing method for the content of harmful elements in mineral products, characterized in that, The method includes: Multiple measurement points are set up on the target surface of the stack of mineral products to be tested. The characteristic fluorescence net intensity sequence of the target harmful element is excited at each point and collected. The background scattering intensity sequence of each point is obtained simultaneously. Based on the characteristic net fluorescence intensity and scattering background intensity of each site, the relative fluorescence yield of each site is calculated, and the relative standard deviation of the relative fluorescence yield among multiple sites is calculated as the surface state dispersion. Using the count values ​​of multiple energy channels in the scattering background intensity sequence at each point as the dependent variable and the corresponding energy values ​​as the independent variable, the slope of the spectral background baseline at each point is fitted, and the relative standard deviation of the slope of the spectral background baseline at multiple points is calculated as the background fluctuation dispersion. Based on the surface state dispersion and the background fluctuation dispersion, a penalty coefficient is calculated. Then, based on the penalty coefficient and the slope of the spectral background baseline at each location, the net characteristic fluorescence intensity at each location is double-weighted and fused to obtain the effective characteristic intensity, including: Extract the surface state dispersion and the background fluctuation dispersion, calculate the sum of the two, and calculate the exponential function value with the natural constant as the base and the negative of the sum of the two as the exponent; The value of the exponential function is normalized and mapped to the interval between 0 and 1, and the mapping result is used as the penalty coefficient. Extract the net intensity of the characteristic fluorescence and the slope of the spectral background baseline corresponding to each measurement point; Using the reciprocal of the absolute value of the slope of each spectral background baseline as the initial weight, the net intensity of the characteristic fluorescence is weighted and averaged to obtain a reference weighted mean. Calculate the relative deviation between the net fluorescence intensity of each feature and the reference weighted mean, and use the penalty coefficient as the base and the relative deviation as the exponent to calculate the weight adjustment factor corresponding to each measurement point; Each initial weight is multiplied by the weight adjustment factor at the same measurement point to obtain the corrected weight. The net intensity of the characteristic fluorescence is then weighted and averaged using the corrected weight to obtain the effective characteristic intensity. The effective feature intensity is input into the mineral type association model, the content value of harmful elements in the mineral product to be tested is output, and the content value is compared with the dynamic limit threshold to output the result of exceeding the standard.

2. The non-destructive testing method for the content of harmful elements in mineral products according to claim 1, characterized in that, Multiple measurement points were set up on the target surface of the stack of mineral products to be tested. The characteristic net fluorescence intensity sequence of the target harmful elements was collected at each point, and the corresponding background scattering intensity sequence was simultaneously acquired, including: The top surface of the stack of mineral products to be tested exposed to the outside world is used as the target surface. Multiple measurement points are arranged in a grid pattern within the target surface, with equal spacing between adjacent measurement points. The excitation source of the X-ray fluorescence spectrometer is vertically aligned with each measurement point, and the same excitation time is continuously applied at each measurement point to collect the energy dispersive X-ray fluorescence spectrum corresponding to each measurement point. From the energy dispersive X-ray fluorescence spectra of each measurement point, the characteristic X-ray fluorescence peaks of the target harmful elements are located. The area of ​​the characteristic X-ray fluorescence peaks is integrated, and the baseline background of linear interpolation on both sides of the peak is subtracted to obtain the net intensity of the characteristic fluorescence corresponding to each measurement point. In a continuous scattering spectrum region that does not overlap with the characteristic X-ray fluorescence peak, an equal-width energy channel is extracted, and the count value of each energy channel is extracted to form a scattering background intensity sequence corresponding to each measurement point. The center energy value of each energy channel in the scattering background intensity sequence is recorded synchronously. The characteristic fluorescence net intensities corresponding to each measurement point are arranged in the order of the measurement points to form a characteristic fluorescence net intensity sequence. Arrange the scattering background intensity sequence and the center energy value of the corresponding energy channel at each measurement point in the same order to form the scattering background intensity sequence and the corresponding energy independent variable sequence.

3. The non-destructive testing method for the content of harmful elements in mineral products according to claim 2, characterized in that, The target harmful element includes at least one of lead, mercury, cadmium, chromium and arsenic. The continuous scattering spectrum region that does not overlap with the characteristic X-ray fluorescence peak is the continuous scattering spectrum region located on the high-energy side of the characteristic X-ray fluorescence peak. The number of energy channels intercepted in the continuous scattering spectrum region is not less than 3. When the energy distance between the characteristic X-ray fluorescence peak and the adjacent high-energy side characteristic X-ray fluorescence peak is greater than 1 keV, 5 energy channels are intercepted.

4. The non-destructive testing method for the content of harmful elements in mineral products according to claim 1, characterized in that, Based on the characteristic net fluorescence intensity and scattering background intensity at each site, the relative fluorescence yield at each site is calculated. The relative standard deviation of the relative fluorescence yield across multiple sites is then calculated as the surface state dispersion, including: The net intensity of each characteristic fluorescence in the net intensity sequence of the characteristic fluorescence is extracted point by point, and the count value of each energy channel in the scattering background intensity sequence corresponding to the same measurement point is extracted; For the same measurement point, the count values ​​of each energy channel in the scattering background intensity sequence are weighted by the center energy value of the same position in the energy independent variable sequence to obtain the energy-weighted total scattering intensity. The characteristic net fluorescence intensity at the same measurement point is normalized using the energy-weighted total scattering intensity to obtain the relative fluorescence yield at each measurement point. By traversing all measurement points, a relative fluorescence yield sequence corresponding to each measurement point is obtained. The relative standard deviation of the relative fluorescence yield sequence is calculated as the surface state dispersion.

5. The non-destructive testing method for the content of harmful elements in mineral products according to claim 1, characterized in that, Using the count values ​​of multiple energy channels in the scattering background intensity sequence at each location as the dependent variable and the corresponding energy values ​​as the independent variables, the slope of the spectral background baseline at each location is fitted. The relative standard deviation of the spectral background baseline slope among multiple locations is calculated as the background fluctuation dispersion, including: For each measurement point, count values ​​are extracted from the background scattering intensity sequence channel by channel, and center energy values ​​of the same order are extracted from the energy independent variable sequence. The center energy values ​​and count values ​​of the same channel are combined into a data pair. After traversing all energy channels, a set of data pairs for the same measurement point is formed. For each data pair in the data pair set, an energy weighting factor is determined based on the central energy value, and a counting statistical weighting factor is determined based on the count value; The product of the energy weighting factor and the count statistics weighting factor is used as the composite weight of the data pair in the linear regression. Using the center energy value as the independent variable and the count value as the dependent variable, apply the corresponding composite weights to all data pairs in the data pair set, perform a weighted least squares linear regression, and solve for the slope of the regression line, which is used as the slope of the spectral background baseline corresponding to the same measurement point. All the spectral background baseline slopes obtained by traversing point by point are arranged in the order of the points to form a spectral background baseline slope sequence. Calculate the arithmetic mean of the spectral background baseline slope sequence; Calculate the normalized deviation of the slope of each spectral background baseline relative to the arithmetic mean, and obtain the relative deviation of the slope corresponding to each spectral background baseline slope; Using the minimum value among the relative deviations of the slope as the benchmark deviation, calculate the normalized ratio of each relative deviation of the slope to the benchmark deviation; The normalization ratios are sorted in ascending order of value. The spectral background baseline slopes corresponding to the first two-thirds of the normalization ratios in the sort are retained, while the spectral background baseline slopes corresponding to the last one-third of the normalization ratios in the sort are removed. Calculate the relative standard deviation of the slope of each retained spectral background baseline as the background fluctuation dispersion.

6. The non-destructive testing method for the content of harmful elements in mineral products according to claim 1, characterized in that, The steps for constructing the mineral type association model include: Collect multiple historical mineral samples of the same type as the mineral product to be tested, and obtain the historical characteristic fluorescence net intensity sequence and historical scattering background intensity sequence corresponding to each historical mineral sample; Calculate the historical surface state dispersion corresponding to each historical mineral product sample, and fit the historical spectral background baseline slope of each historical mineral product sample to obtain the historical background fluctuation dispersion. Based on the historical surface state dispersion and historical background fluctuation dispersion of each historical mineral sample, the historical penalty coefficient is calculated, and the corresponding historical effective feature intensity is determined. Laboratory chemical analysis was performed on each historical mineral product sample to determine the reference content value of harmful elements in each historical mineral product sample. The historical effective characteristic intensity and reference content value of the same historical mineral product sample were combined to form a training sample pair. Training sample pairs of the same matrix type are aggregated into a matrix training subset. For each matrix type of matrix training subset, a regression mapping relationship between the historical effective feature intensity and the reference content value is fitted and established. The regression mapping relationship corresponding to each matrix type is used as the mineral type association model of the corresponding matrix.

7. The non-destructive testing method for the content of harmful elements in mineral products according to claim 1, characterized in that, The calculation steps for the dynamic limit threshold include: Extract the surface state dispersion, the background fluctuation dispersion, and the penalty coefficient; Extract the reference weighted mean and extract the slope parameter of the regression mapping relationship between the mineral type association model and the matrix type of the mineral product to be tested; The square root of the sum of the squares of the surface state dispersion and the background fluctuation dispersion is calculated as the composite coefficient of variation. Using the penalty coefficient as an iterative shrinkage factor, the synthetic coefficient of variation is iteratively weighted until convergence, including: The current threshold factor is weighted and shrunken using the penalty coefficient, and a correction term is formed by weighting the synthetic coefficient of variation with the complement of the penalty coefficient to obtain the updated threshold factor. Determine the deviation between the updated threshold factor and the current threshold factor; When the deviation is less than the product of the penalty coefficient and the surface state dispersion, the iteration is terminated, and the update threshold factor is used as the stable threshold factor. Otherwise, replace the current threshold factor with the updated threshold factor and continue iterating; The basic dynamic threshold is obtained by multiplying the reference weighted mean, the slope parameter, and the stability threshold factor. The dynamic limit threshold is determined by integrating the basic dynamic threshold, the reference weighted mean, the slope parameter, the background fluctuation dispersion, and the penalty coefficient. The dynamic limit threshold is composed of the basic dynamic threshold and a broadening term that is jointly weighted by the background fluctuation dispersion and the penalty coefficient.

8. The non-destructive testing method for the content of harmful elements in mineral products according to claim 1, characterized in that, After outputting the result of the out-of-limit judgment, it also includes: Calculate the difference between the content value and the dynamic limit threshold, and divide the absolute value of the difference by the dynamic limit threshold to obtain the normalized deviation. Based on the penalty coefficient and the background fluctuation dispersion, the confidence base factor is determined, and the confidence score is calculated by combining the normalized deviation. A decision threshold is determined based on the penalty coefficient and the surface state dispersion, wherein the decision threshold is proportional to both the penalty coefficient and the surface state dispersion. When the confidence score is greater than or equal to the judgment threshold, the current exceeding judgment result is output as the final judgment result. When the confidence score is less than the judgment threshold, additional measurement points are set up on the target surface of the stack of mineral products to be tested, and the net intensity of characteristic fluorescence and the background intensity of scattering are collected at the additional points. The net characteristic fluorescence intensity corresponding to the supplementary points is incorporated into the existing net characteristic fluorescence intensity sequence, and the background scattering intensity corresponding to the supplementary points is simultaneously incorporated into the existing background scattering intensity sequence. The surface state dispersion and background fluctuation dispersion are recalculated, and double weighted fusion is performed again to obtain the supplemented effective characteristic intensity. The supplemented effective feature intensity is input into the mineral type association model to obtain the supplemented content value, which is then compared with the dynamic limit threshold. The supplemented exceedance judgment result is output as the final judgment result.

9. A non-destructive testing system for the content of harmful elements in mineral products, characterized in that, The system is used to implement the non-destructive testing method for the content of harmful elements in mineral products according to any one of claims 1 to 8, the system comprising: The data acquisition module is used to set up multiple measurement points on the target surface of the stack of mineral products to be tested, excite each point and collect the characteristic fluorescence net intensity sequence of the target harmful elements, and simultaneously acquire the background scattering intensity sequence corresponding to each point. The dispersion calculation module is used to calculate the relative fluorescence yield of each point based on the characteristic net fluorescence intensity and scattering background intensity of each point, and to calculate the relative standard deviation of the relative fluorescence yield among multiple points as the surface state dispersion. The background fluctuation fitting module is used to fit the spectral background baseline slope of each point with the count values ​​of multiple energy channels in the scattering background intensity sequence of each point as the dependent variable and the corresponding energy value as the independent variable, and to calculate the relative standard deviation of the spectral background baseline slope among multiple points as the background fluctuation dispersion. A dual-weighted fusion module is used to calculate a penalty coefficient based on the surface state dispersion and the background fluctuation dispersion, and to perform dual-weighted fusion of the net characteristic fluorescence intensity of each point based on the penalty coefficient and the slope of the spectral background baseline at each point to obtain the effective characteristic intensity, including: Extract the surface state dispersion and the background fluctuation dispersion, calculate the sum of the two, and calculate the exponential function value with the natural constant as the base and the negative of the sum of the two as the exponent; The value of the exponential function is normalized and mapped to the interval between 0 and 1, and the mapping result is used as the penalty coefficient. Extract the net intensity of the characteristic fluorescence and the slope of the spectral background baseline corresponding to each measurement point; Using the reciprocal of the absolute value of the slope of each spectral background baseline as the initial weight, the net intensity of the characteristic fluorescence is weighted and averaged to obtain a reference weighted mean. Calculate the relative deviation between the net fluorescence intensity of each feature and the reference weighted mean, and use the penalty coefficient as the base and the relative deviation as the exponent to calculate the weight adjustment factor corresponding to each measurement point; Each initial weight is multiplied by the weight adjustment factor at the same measurement point to obtain the corrected weight. The net intensity of the characteristic fluorescence is then weighted and averaged using the corrected weight to obtain the effective characteristic intensity. The content output module is used to input the effective feature intensity into the mineral type association model, output the content value of harmful elements in the mineral product to be tested, compare the content value with the dynamic limit threshold, and output the result of exceeding the standard.

Images