阅读说明：本技术 基于激光诱导击穿光谱的金属牌号鉴定方法 (Metal brand identification method based on laser-induced breakdown spectroscopy ) 是由 孙兰香 于海斌 周中寒 张鹏 郭美亭 于 2018-08-03 设计创作，主要内容包括：本发明涉及一种基于激光诱导击穿光谱的金属牌号鉴定方法,具体步骤为：1)获取牌号库中各牌号分析元素浓度区间；2)扩展各牌号分析元素浓度区间；3)根据新的浓度区间范围,按照一定分布生成大量随机样本；4)对随机样本浓度进行标准化；5)使用标准化后的随机样本数据集训练牌号鉴定模型,存储训练好的牌号鉴定模型；6)使用激光诱导击穿光谱系统获取待测样品的LIBS光谱数据；7)对原始LIBS光谱数据进行预处理；8)定量分析,得到待测样品化学组成元素浓度；9)使用训练好的牌号鉴定模型对步骤8中的定量分析结果进行牌号鉴定,输出鉴定结果。以本方法进行金属牌号鉴定,具有更高的准确率、更好的适应性和可扩展性。(The invention relates to a metal mark identification method based on laser-induced breakdown spectroscopy, which comprises the following specific steps: 1) obtaining the concentration interval of each grade analysis element in the grade library; 2) expanding the concentration interval of each grade analysis element; 3) generating a large number of random samples according to a certain distribution according to the new concentration interval range; 4) normalizing the concentration of the random sample; 5) training a grade identification model by using the standardized random sample data set, and storing the trained grade identification model; 6) acquiring LIBS spectral data of a sample to be detected by using a laser-induced breakdown spectroscopy system; 7) preprocessing original LIBS spectral data; 8) carrying out quantitative analysis to obtain the concentration of chemical composition elements of the sample to be detected; 9) and (4) carrying out grade identification on the quantitative analysis result in the step (8) by using the trained grade identification model, and outputting an identification result. The method for identifying the metal mark has higher accuracy, better adaptability and expandability.)

1. The metal mark identification method based on the laser-induced breakdown spectroscopy is characterized by comprising the following steps of:

the method comprises the following steps of (1) establishing an offline grade identification model: analyzing element concentration intervals according to each grade in the grade library, generating random samples and standardizing, and training the grade identification model by using the standardized random sample data set to obtain an offline grade concentration identification model;

and (3) identifying the actual metal grade: preprocessing and quantitatively analyzing LIBS spectral intensity data of a sample to be detected to obtain the concentration of chemical composition elements of the sample to be detected, and comparing the concentration with an offline grade identification model to obtain an identification result.

2. The method for identifying the metal grade based on the laser-induced breakdown spectroscopy as claimed in claim 1, wherein the step of establishing the off-line grade identification model comprises:

obtaining the concentration interval of each grade analysis element in the grade library;

expanding the element concentration interval of each grade analysis according to the proportion;

generating a random sample according to the new concentration interval range;

normalizing the concentration of the random sample;

and training the grade identification model by using the standardized random sample data set to obtain an offline grade concentration identification model.

3. The method for identifying the metal grade based on the laser-induced breakdown spectroscopy as claimed in claim 1, wherein the step of identifying the actual metal grade comprises:

acquiring LIBS spectral intensity data of a sample to be detected by using a laser-induced breakdown spectroscopy method;

preprocessing original LIBS spectral intensity data to obtain normalized spectral intensity data;

carrying out quantitative analysis on the normalized spectral intensity data to obtain the concentration of chemical composition elements of the sample to be detected;

and (4) performing grade concentration comparison and identification on the quantitative analysis result by using an offline grade concentration identification model, and outputting an identification result.

4. The method for identifying the metal grade based on the laser-induced breakdown spectroscopy as claimed in claim 2, wherein the generating of the random samples according to the new concentration interval range is to generate the random samples by uniform distribution.

5. The method for identifying the metal grade based on the laser-induced breakdown spectroscopy as claimed in claim 2, wherein the normalization of the random sample concentration is specifically as follows: and normalizing the generated data set of the random sample to ensure that the value range of the data on each element dimension is between-1 and 1.

6. The method for identifying the metal grade based on the laser-induced breakdown spectroscopy as claimed in claim 1 or 2, wherein the grade identification model is a Support Vector Machine (SVM) model.

7. The method for identifying the metal mark based on the laser-induced breakdown spectroscopy as claimed in claim 3, wherein the raw LIBS spectral intensity data is preprocessed to obtain normalized spectral intensity data:

spectrum screening: calculating the sum of full spectrum intensity of each spectrum in the original data set, and then processing abnormal points of the sum of full spectrum intensity and the spectrum according to a threshold range to remove the spectrum intensity and the abnormal spectrum which is too low or too high;

normalization: and (3) normalizing the spectrum data after spectrum screening by using the full spectrum intensity sum to compensate the fluctuation of the spectrum intensity, wherein the calculation formula is as follows:

wherein, I'_jRepresenting normalized spectral intensity data, I_jRepresents the original spectral intensity, I, of the spectrum corresponding to the wavelength j after screening_sRepresenting the sum of the full spectral intensities of the spectrally screened original spectra.

8. The method for identifying the metal grade based on the laser-induced breakdown spectroscopy as claimed in claim 3, wherein the quantitative analysis of the normalized spectral intensity data comprises: and establishing a quantitative analysis model by using the metal standard sample, and analyzing the normalized spectral intensity data by using the quantitative analysis model so as to obtain the chemical composition element concentration of the sample to be detected.

9. The method as claimed in claim 8, wherein the metal standard sample is selected such that the concentration of each reference element can cover the concentration range of each reference element in the lot library.

10. The method for identifying the metal grade based on the laser-induced breakdown spectroscopy as claimed in claim 8, wherein the quantitative analysis model is a partial least squares model.

Technical Field

The invention belongs to the field of spectral analysis and composition analysis of material, and particularly relates to a metal mark identification method based on laser-induced breakdown spectroscopy.

Background

Laser-induced breakdown spectroscopy (LIBS) analysis technology is an atomic emission spectroscopy analysis method which uses Laser-induced plasma as an evaporation, atomization and excitation source and can realize qualitative and quantitative analysis of chemical elements of substances. The method has the characteristics of no need of sample preparation, direct and rapid property, small sample loss and the like, and becomes a research hotspot in the fields of metallurgical analysis, cultural relic protection, geochemistry, environmental engineering and the like in recent years, and the identification of a metal mark is also one of the main applications of the method.

The metal mark identification process mainly comprises three parts: 1) acquiring spectral data of a metal material to be detected by using a LIBS system; 2) obtaining content information of each element in the metal material to be identified by using a LIBS quantitative analysis method; 3) and (4) performing grade matching query in the selected grade library by utilizing the quantitative information of the metal to be identified.

In the aspect of grade matching, the prior art discloses a grade matching method based on fuzzy membership, which considers that in the aspect of metal grade identification, firstly, the contents of different elements in metal do not have specific relevance and can be considered to have certain degree of freedom, and the inconsistency causes the ambiguity of the measurement of the matching degree required to be calculated for identification.

However, when the matching degree of the metal material to be measured and each grade in the given grade library is calculated by using the grade matching method based on the fuzzy membership degree, the method only considers the membership degree (between 0 and 1) of the content of various elements in the metal material to be measured, the grades of the metal are generally approximate, the difference of the content range of various elements between a plurality of similar grades is very small, and therefore, when the quantitative analysis result of a sample to be measured is deviated from the true value of the sample to be measured, the membership degree of various elements to each grade is also deviated. And the comprehensive membership finally used for determining the grade matching result is the comprehensive membership obtained by weighting, so that the difference of the comprehensive membership between each grade is very small, and the accuracy of the matching result can be influenced as long as the quantitative analysis result of the metal sample to be detected has a little deviation. On the other hand, undetermined parameters exist in the fuzzy membership function, and the neighborhood size of the boundary of each element content range of each grade needs to be determined in advance.

Disclosure of Invention

Aiming at the defects in the prior art, the invention aims to provide a quick grade identification method capable of quickly and accurately matching metal grades.

The technical scheme adopted by the invention for realizing the purpose is as follows: the metal mark identification method based on the laser-induced breakdown spectroscopy comprises the following steps:

the method comprises the following steps of (1) establishing an offline grade identification model: analyzing element concentration intervals according to each grade in the grade library, generating random samples and standardizing, and training the grade identification model by using the standardized random sample data set to obtain an offline grade concentration identification model;

and (3) identifying the actual metal grade: preprocessing and quantitatively analyzing LIBS spectral intensity data of a sample to be detected to obtain the concentration of chemical composition elements of the sample to be detected, and comparing the concentration with an offline grade identification model to obtain an identification result.

The establishment steps of the off-line brand identification model comprise:

obtaining the concentration interval of each grade analysis element in the grade library;

expanding the element concentration interval of each grade analysis according to the proportion;

generating a random sample according to the new concentration interval range;

normalizing the concentration of the random sample;

and training the grade identification model by using the standardized random sample data set to obtain an offline grade concentration identification model.

The steps of identifying the actual metal grade comprise:

acquiring LIBS spectral intensity data of a sample to be detected by using a laser-induced breakdown spectroscopy method;

preprocessing original LIBS spectral intensity data to obtain normalized spectral intensity data;

carrying out quantitative analysis on the normalized spectral intensity data to obtain the concentration of chemical composition elements of the sample to be detected;

and (4) performing grade concentration comparison and identification on the quantitative analysis result by using an offline grade concentration identification model, and outputting an identification result.

And the random sample is generated according to the new concentration interval range by adopting uniform distribution.

The normalizing the concentration of the random sample specifically comprises the following steps: and normalizing the generated data set of the random sample to ensure that the value range of the data on each element dimension is between-1 and 1.

The brand identification model is a Support Vector Machine (SVM) model.

The preprocessing is performed on the original LIBS spectral intensity data, and the acquired normalized spectral intensity data is as follows:

spectrum screening: calculating the sum of full spectrum intensity of each spectrum in the original data set, and then processing abnormal points of the sum of full spectrum intensity and the spectrum according to a threshold range to remove the spectrum intensity and the abnormal spectrum which is too low or too high;

normalization: and (3) normalizing the spectrum data after spectrum screening by using the full spectrum intensity sum to compensate the fluctuation of the spectrum intensity, wherein the calculation formula is as follows:

wherein, I_j' denotes normalized spectral intensity data, I_jRepresents the original spectral intensity, I, of the spectrum corresponding to the wavelength j after screening_sRepresenting the sum of the full spectral intensities of the spectrally screened original spectra.

The quantitative analysis of the normalized spectral intensity data comprises: and establishing a quantitative analysis model by using the metal standard sample, and analyzing the normalized spectral intensity data by using the quantitative analysis model so as to obtain the chemical composition element concentration of the sample to be detected.

When the metal standard sample is selected, the concentration of each reference element of the metal standard sample can cover the concentration range of each brand reference element in the brand library.

The quantitative analysis model is a partial least squares model.

The invention has the following advantages and beneficial effects:

1. the training sample set is constructed in a mode of generating the virtual sample according to the mark regulation, so that a large number of metal samples are not needed in the establishing process of the mark matching model, and a large number of experiments are not needed to acquire sample spectrums, so that the method has better use convenience and expandability.

2. When the virtual samples are generated according to the designation of the brand, the element concentration interval range specified by the brand is properly expanded, so that the final training set comprises some samples which do not belong to the corresponding brand but are very close to each other. Therefore, the invention can allow certain quantitative analysis error when the marks are matched.

Drawings

FIG. 1 is a flow chart of a method implementation of the present invention;

FIG. 2 is an LIBS original spectrum of a metal sample;

FIG. 3 shows the result of identifying the grade of an aluminum alloy sample to be tested by using the method of the present invention.

Detailed Description

The present invention will be described in further detail with reference to the accompanying drawings and examples.

Aiming at the problem that reference element concentration intervals specified by different grades are different, a virtual sample set training grade matching model is generated according to each grade specification, spectrum data of an aluminum alloy sample to be tested is obtained by an LIBS technology, chemical compositions of the aluminum alloy sample to be tested are obtained by an LIBS quantitative analysis method, the grade is identified according to the grade matching model, and finally the grade identification of the aluminum alloy to be tested is realized.

The implementation flow of the method is shown in fig. 1, and the specific implementation steps are as follows:

step 1: and obtaining the concentration interval of each grade analysis element in the aluminum alloy grade library according to national standards.

Step 2: the concentration interval of the analysis elements of each grade is properly expanded, for example, the upper limit is increased by 10 percent, and the lower limit is decreased by 10 percent.

And step 3: according to the new concentration interval range, a large number of random samples are generated according to a certain distribution, for example, the samples are generated according to a uniform distribution.

And 4, step 4: the concentration of the random sample is normalized, so that the value range of the data in each element dimension is between [ -1,1], for example, the normalization mode can be that the upper limit of the concentration interval range is 1, and the lower limit is-1.

And 5: and training the grade identification model by using the standardized random sample data set, and storing the trained grade identification model. The brand authentication model may select a Support Vector Machine (SVM) model.

Step 6: and acquiring LIBS spectral data of the sample to be detected by using a laser-induced breakdown spectroscopy method.

And 7: the raw LIBS spectral data is pre-processed. The pretreatment mainly comprises spectrum screening and normalization.

Spectrum screening: and calculating the sum of full spectrum intensity of each spectrum in the original data set, and then performing abnormal point processing on the spectrum according to the sum of full spectrum intensity to remove spectrum intensity and some abnormal spectra which are too low or too high.

Normalization: the spectral data is normalized using its full spectral intensity sum to compensate for the spectral intensity fluctuations, as calculated by the following equation:

wherein I_j' denotes normalized spectral intensity data, I_jDenotes the original spectral intensity, I, corresponding to the wavelength j_sRepresenting the sum of the full spectral intensities of the original spectrum.

And 8: and carrying out quantitative analysis to obtain the concentration of the chemical composition elements of the sample to be detected. The quantitative analysis needs to establish a quantitative analysis model by using a metal standard sample, and the spectral intensity data after the normalization processing is analyzed by using the quantitative analysis model, so that the chemical composition element concentration of the sample to be detected is obtained. When the metal standard sample is selected, the concentration of each reference element of the metal standard sample can cover the concentration range of each brand reference element in the brand library. In this example, 54 blocks of 10 aluminum alloy standard samples, namely LY11, LF21, 6005, LF6, 7A04, cast aluminum, 7050, ADC12, 5454 and A356, are used for establishing a quantitative analysis model. The quantitative analysis model may employ a partial least squares model.

And step 9: and (4) carrying out grade identification on the quantitative analysis result in the step (8) by using the trained grade identification model, and outputting an identification result.

The grades of the 24 aluminum alloy samples are identified according to the method, and the following table shows the serial numbers of the 24 aluminum alloy samples and the grades of the 24 aluminum alloy samples.

Fig. 2 shows an original spectrum of an aluminum alloy sample acquired by using the LIBS method, and the spectrum intensity data is preprocessed in the modes of screening, normalization and the like, then an aluminum alloy standard sample is selected to establish a LIBS quantitative analysis model, and the spectrum of the sample to be detected is quantitatively analyzed to obtain the concentration of each element of the chemical composition of the sample to be detected. The following table shows the determination coefficients (R) of each element when the quantitative analysis model employs a partial least squares model²) Relative standard error (RSD), Root Mean Square Error (RMSE), etc. analysis indicators:

according to an aluminum alloy national standard grade library, obtaining each grade analysis element concentration interval in the grade library, expanding each interval range to the left and the right by 10%, generating 2000 virtual samples for each grade to form a data set, and then standardizing to enable the value range of the data on each element dimension to be between [ -1,1 ]. Then, the normalized data set is used as a training set to train the SVM multi-classification model. And finally, performing grade identification on the quantitative analysis result of the sample to be detected by using the SVM model, wherein the identification result is shown in figure 3, wherein 'o' represents the actual grade of the sample, 'x' represents the predicted grade of the sample, and the coincidence of 'o' and 'x' represents the correct prediction. The accuracy of a test sample reaches 96.9375%, the sample with wrong classification is mainly the sample with the brand DC.360Y.6, and the data of the sample with the brand has a part of data classified as HB.SF36, because the concentrations of most elements of the two samples with the brands are similar, for example, the Si content in the sample with the brand DC.360Y.6 is 8.0-10.5, Fe is less than or equal to 0.8, Mg: 0.2 to 0.35; in HB.SF36, the Si content is 9.8-10.4, Fe is less than or equal to 0.13, Mg: 0.25-0.35, so that the prediction accuracy of the DC.360Y.6 is low, and the prediction accuracy of other 4 grades of aluminum alloy basically reaches 100 percent, thus the method can realize the accurate identification of the grades of unknown aluminum alloy samples.