A quantitative analysis method for laser-induced breakdown spectroscopy

Abstract

The present invention relates to a laser induced breakdown spectroscopy quantitative analysis method, which belongs to the technical field of infrared and plasma emission spectroscopy measurement methods. The technical solution of the present invention is: determining a series of samples to be tested, obtaining laser induced breakdown spectra, using three models of Lasso, ridge and quadratic nonlinear regression as stacking ensemble learning model inputs, and utilizing the preferred spectral lines of each element and the best matching normalized spectral line combination pairs. The beneficial effects of the present invention are: the input spectral lines used by each element model can remain stable after verification and normalization, and the sample preparation is simple, fast and convenient in combination with the laser induced breakdown spectroscopy material discrimination technology, eliminating the unstable disadvantages of a single model, combining weak learners into a strong learner, and combining the laser induced breakdown spectroscopy material discrimination technology to make the sample preparation simple, fast and convenient, thereby realizing rapid and high-precision quantitative detection of steel samples.

Description

Technical Field

[0001] The invention relates to a laser induced breakdown spectrum quantitative analysis method, belonging to the technical field of infrared and plasma emission spectrum measurement methods. Background Art

[0002] Laser-induced breakdown spectroscopy (LIBS) is a novel analytical technique that uses high-energy laser pulses focused directly onto a sample surface to ionize the sample and generate a plasma. The plasma then emits characteristic spectral lines during cooling. It is based on the interaction between lasers and matter, examining the composition and concentration of materials through physics and spectroscopy. LIBS instruments are compact and easy to operate, capable of measuring a wide range of substances, including gases, liquids, and solids. The technique offers advantages such as rapid speed, no sample preparation, and non-contact performance. It holds broad application prospects in metallurgical analysis, potentially optimizing steelmaking processes, improving production efficiency, reducing smelting costs, and conserving resources.

[0003] However, laser-induced breakdown spectroscopy has poor stability, the characteristic spectra of each element affect each other, the spectral data contains a lot of noise interference, and is also affected by self-absorption and matrix effects, resulting in poor repeatability and low precision of quantitative analysis results. The limitation of analytical accuracy has seriously hindered its further development and promotion in practical applications. Therefore, it is of great significance to study methods and applications to improve LIBS measurement accuracy.

[0004] When performing LIBS quantitative analysis, algorithms with higher prediction accuracy and stronger generalization capabilities are needed. Some have proposed that multivariate nonlinear PLS models can improve quantitative analysis accuracy, but these models are complex and require high-dimensional data. As spectrometer resolution continues to improve, data dimensionality increases, including a large amount of redundant information. This increases modeling complexity and makes full-spectrum data modeling prone to overfitting. Summary of the Invention

[0005] The purpose of the present invention is to provide a laser-induced breakdown spectroscopy quantitative analysis method. The input spectral lines used in each element model can remain stable after verification and normalization. The sample preparation is simple, fast and convenient in combination with the laser-induced breakdown spectroscopy material discrimination technology, eliminating the instability of a single model. The weak learners are combined into a strong learner. The sample preparation is simple, fast and convenient in combination with the laser-induced breakdown spectroscopy material discrimination technology, realizing rapid and high-precision quantitative detection of steel samples, effectively solving the above-mentioned problems existing in the background technology.

[0006] The technical solution of the present invention is: a laser induced breakdown spectroscopy quantitative analysis method, comprising the following steps:

[0007] Step 1: Optimize the LIBS experimental conditions to ensure the consistency of the experimental conditions of the samples to be tested;

[0008] Step 2: determine the series of samples to be tested;

[0009] Step 3: Select multiple preferred characteristic spectral line combinations of each element and summarize the normalized spectral line set of the Fe element;

[0010] Step 4, matching the best normalized characteristic spectral line of each element with the Fe normalized line combination;

[0011] Step 5: Extract the characteristic variables of the original spectrum according to the matching of the best normalized Fe line pair, and perform normalization preprocessing on the input spectrum to obtain the characteristic variables and normalize the spectral data;

[0012] Step 6: Normalize the spectral data by merging the category column and the element content column to obtain normalized full spectral data;

[0013] Step 7: Establish an integrated model based on each element;

[0014] In step eight, each element is processed according to step seven, and the overall result is predicted and combined for output.

[0015] In the second step, multiple samples of the same type with known characteristics are selected as a series of samples to be tested. For each calibration sample, laser focusing ablation is performed according to a certain detection method to obtain a database of characteristic spectral line intensities of the calibration samples.

[0016] In the fourth step, multiple preferred characteristic spectral lines of each element and the Fe normalized spectral line are freely combined and correlation scores are calculated, and the best combination pairs of each element are matched according to the scores.

[0017] The step seven includes the following steps:

[0018] (1) extracting the column data of the characteristic spectrum line of the element and the column of the element content from the normalized full spectrum data to obtain the element spectrum data;

[0019] (2) Separate the test set spectral data, test set label data, training set spectral data, and training set label data from the element spectral data in a reasonable proportion, and check the rationality of the divided data;

[0020] (3) Using the training set and test set divided in step (2), establish the Lasso regression model, Ridge regression model and quadratic nonlinear regression model in sequence, and obtain the output results of the three models respectively;

[0021] (4) The output results of the three models are used as the input of the ensemble learning model, and the element content is used as the label to train and optimize the model;

[0022] (5) Prediction of the overall results of the integrated model.

[0023] The beneficial effects of the present invention are as follows: the input spectral lines used in each element model can remain stable after verification and normalization, and the sample preparation is simple, fast and convenient when combined with the laser-induced breakdown spectroscopy material discrimination technology, eliminating the instability disadvantage of a single model, combining weak learners into a strong learner, and the sample preparation is simple, fast and convenient when combined with the laser-induced breakdown spectroscopy material discrimination technology, thereby realizing rapid and high-precision quantitative detection of steel samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 is a flow chart of the present invention;

[0025] Figure 2 These are the first two principal component distribution diagrams for reasonably dividing the training set and the test set according to the present invention;

[0026] Figure 3 It is a prediction graph of the degree of deviation between the Lasso regression predicted value and the true value;

[0027] Figure 4 It is a prediction graph of the degree of deviation between the ridge regression predicted value and the true value;

[0028] Figure 5 It is a prediction graph of the degree of deviation between the predicted value and the true value of the quadratic nonlinear regression;

[0029] Figure 6 It is a prediction graph of the degree of deviation between the overall predicted value of the integrated learning model and the true value. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solutions and advantages of the invention implementation cases clearer, the technical solutions in the invention implementation cases will be clearly and completely described below in conjunction with the drawings in the implementation cases. Obviously, the implementation cases described are only a small part of the implementation cases of the present invention, rather than all the implementation cases. Based on the implementation cases in the present invention, all other implementation cases obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0031] A laser induced breakdown spectroscopy quantitative analysis method comprises the following steps:

[0032] Step 1: Optimize the LIBS experimental conditions to ensure the consistency of the experimental conditions of the samples to be tested;

[0033] Step 2: determine the series of samples to be tested;

[0034] Step 3: Select multiple preferred characteristic spectral line combinations of each element and summarize the normalized spectral line set of the Fe element;

[0035] Step 4, matching the best normalized characteristic spectral line of each element with the Fe normalized line combination;

[0036] Step 5: Extract the characteristic variables of the original spectrum according to the matching of the best normalized Fe line pair, and perform normalization preprocessing on the input spectrum to obtain the characteristic variables and normalize the spectral data;

[0037] Step 6: Normalize the spectral data by merging the category column and the element content column to obtain normalized full spectral data;

[0038] Step 7: Establish an integrated model based on each element;

[0039] In step eight, each element is processed according to step seven, and the overall result is predicted and combined for output.

[0040] In the second step, multiple samples of the same type with known characteristics are selected as a series of samples to be tested. For each calibration sample, laser focusing ablation is performed according to a certain detection method to obtain a database of characteristic spectral line intensities of the calibration samples.

[0041] In the fourth step, multiple preferred characteristic spectral lines of each element and the Fe normalized spectral line are freely combined and correlation scores are calculated, and the best combination pairs of each element are matched according to the scores.

[0042] The step seven includes the following steps:

[0043] (1) extracting the column data of the characteristic spectrum line of the element and the column of the element content from the normalized full spectrum data to obtain the element spectrum data;

[0044] (2) Separate the test set spectral data, test set label data, training set spectral data, and training set label data from the element spectral data in a reasonable proportion, and check the rationality of the divided data;

[0045] (3) Using the training set and test set divided in step (2), establish the Lasso regression model, Ridge regression model and quadratic nonlinear regression model in sequence, and obtain the output results of the three models respectively;

[0046] (4) The output results of the three models are used as the input of the ensemble learning model, and the element content is used as the label to train and optimize the model;

[0047] (5) Prediction of the overall results of the integrated model.

[0048] Example:

[0049] To facilitate description and improve the calibration accuracy of the model, 12 groups of low-alloy steel national standard samples (compositions shown in Table 1) were selected for training and testing, referring to the range of low-alloy steel samples in the spark direct reading spectroscopy calibration model.

[0050] Table 1 Element content in each sample

[0051]

[0052] It uses a independently developed portable LIBS composition analyzer with independent intellectual property rights, a built-in independent intellectual property rights embedded LIBS high-precision calibration and analysis software system based on Rockchip RK3399 pro, a Lapa-80 solid-state pulse laser (pulse energy 80mJ adjustable, frequency 0-20Hz adjustable), two AvaSpec-Mini4096CL small fiber optic spectrometers (wavelength range: 170-300nm, 290-400nm), an independent intellectual property rights high-precision optical path probe, a non-coaxial optical path, and an optimal laser focus to sample surface distance LTSD of -2mm, ensuring detection accuracy.

[0053] The flow chart of the present invention is as follows Figure 1 , which includes the following steps:

[0054] Step 1: First, optimize the LIBS experimental conditions to ensure the consistency of the experimental conditions of the samples to be tested;

[0055] Step 2: Determine the series of samples to be tested: Use 12 similar samples with known characteristics as calibration samples. Perform laser focused ablation on the 12 calibration samples according to a specific detection method, and obtain 30 sets of valid spectral data for each sample, thereby obtaining a 360-set database of characteristic spectral line intensities of the calibration samples.

[0056] Step 3: Select multiple preferred characteristic spectral line combinations of each element; summarize the normalized spectral line set of Fe element, as shown in Table 2:

[0057] Table 2 Spectral lines of each element and normalized spectral line set of Fe element

[0058]

[0059] Step 4: Freely combine multiple preferred characteristic spectral lines and Fe normalized spectral lines of each element and calculate the correlation score. Match the best normalized characteristic spectral line of each element with the Fe normalized line combination according to the score, as shown in Table 3:

[0060] Table 3 Optimal spectral line combinations

[0061]

[0062] Step 5: Extract the characteristic variables of the original spectrum according to the matching of the best normalized Fe line pair, and perform normalization preprocessing on the input spectrum. After obtaining the characteristic variables, normalize the spectrum data gy1: 360*y dimension;

[0063] Step 6: Merge the normalized spectral data gy with the category column (column 1) and the element content column (element type a column) to obtain the normalized full spectral data gy: 360*(y+1+a);

[0064] Step 7: Perform the following steps based on each element, such as Al:

[0065] 1. Extract the element characteristic line column data and the element content column from the normalized full spectrum data gy to obtain the element spectrum data gyys: 360*b;

[0066] 2. According to a reasonable proportion, the test set spectrum data X_Val, the test set label data Y_Val, the training set spectrum data X_Train, and the training set label data Y_Train are reasonably separated from the element spectrum data gyys; and the rationality of the data division is checked according to the patent CN202110711023.3 drawing, such as Figure 2 ;

[0067] 3. Using the training set and test set divided in step 2, establish the Lasso regression model, Ridge regression model, and quadratic nonlinear regression model in turn, and obtain the output results of the three models respectively;

[0068] (1) Lasso regression: Lasso regression uses L1 regularization:

[0069]

[0070] from sklearn.linear_model import LassoCV

[0071] model_1=LassoCV()

[0072] model_1.fit(X_train,Y_train)

[0073] The deviation degree between the predicted value and the true value is plotted, such as Figure 3 ;

[0074] (2) Ridge regression: Ridge regression is a form of linear regression using L2 regularization:

[0075]

[0076] from sklearn.linear_model import RidgeCV

[0077] model_2=RidgeCV()

[0078] model_2.fit(X_train,Y_train)

[0079] The deviation degree between the predicted value and the true value is plotted, such as Figure 4 ;

[0080] (3) Nonlinear fitting regression:

[0081] from sklearn.linear_model import LinearRegression

[0082] from sklearn.preprocessing import PolynomialFeatures quadratic_featurizer=PolynomialFeatures(degree=2)

[0083] X_train_quadratic=

[0084] quadratic_featurizer.fit_transform(X_train)

[0085] regressor_quadratic=LinearRegression()

[0086] regressor_quadratic.fit(X_train_quadratic,Y_train)

[0087] The deviation degree between the predicted value and the true value is plotted, such as Figure 5 ;

[0088] 4. Take the output of the three models as the input of the integrated learning model, use the element content as the label, train and optimize the model, and plot the deviation between the predicted value and the true value. Figure 6 .

[0089] 5. Prediction of the overall results of the integrated model.

[0090] Step eight: Each element is carried out according to step seven, and the overall result is predicted and merged for output. The present invention accurately extracts characteristic variables from a large amount of original spectral data, eliminating nonlinear problems caused by data redundancy, self-absorption and matrix effect; targeted selection of characteristic spectral lines of each element and commonly used Fe normalized spectral lines, and the use of correlation calculation scores to find the best matching normalized spectral line combination pair, data input is more accurate; the best matching normalized spectral line combination pair of each element is used for spectral standardization, and the whole is used as the input spectrum. After normalization, the spectral data can remain stable; the three models of Lasso, ridge and quadratic nonlinear regression are input as stacking integrated learning models, eliminating the unstable disadvantages of a single model, combining weak learners into a strong learner, and combining laser induced breakdown spectroscopy material discrimination technology to make sample preparation simple, fast and convenient, and realize the rapid and high-precision quantitative detection of steel samples.

Claims

1. A method for quantitative analysis of laser-induced breakdown spectroscopy, characterized in that The following steps are involved: Step 1: Optimize the LIBS experimental conditions to ensure the consistency of the experimental conditions of the samples to be tested; Step 2: determine the series of samples to be tested; Step 3: Select multiple preferred characteristic spectral line combinations of each element and summarize the normalized spectral line set of the Fe element; Step 4, matching the best normalized characteristic spectral line of each element with the Fe normalized line combination; Step 5: Extract the characteristic variables of the original spectrum according to the matching of the best normalized Fe line pair, and perform normalization preprocessing on the input spectrum to obtain the characteristic variables and normalize the spectral data; Step 6: Normalize the spectral data by merging the category column and the element content column to obtain normalized full spectral data; Step 7: Establish an integrated model based on each element; this includes the following steps: (1) Extracting the column data of the characteristic spectrum line of the element and the column of the element content from the normalized full spectrum data to obtain the spectrum data of the element; (2) Separate the test set spectral data, test set label data, training set spectral data and training set label data from the element spectral data in a reasonable proportion, and check the rationality of the divided data; (3) Based on the training set and test set divided in step (2), the Lasso regression model, Ridge regression model and quadratic nonlinear regression model are established in sequence, and the output results of the three models are obtained respectively; (4) The output results of the three models are used as the input of the ensemble learning model, and the element content is used as the label to train and optimize the model; (5) Prediction of the overall results of the integrated model; In step eight, each element is processed according to step seven, and the overall result is predicted and combined for output.

2. The laser induced breakdown spectroscopy quantitative analysis method according to claim 1, characterized in that: In the second step, multiple samples of the same type with known characteristics are selected as a series of samples to be tested, and laser focusing ablation is performed on each calibration sample to obtain a database of characteristic spectral line intensities of the calibration samples.

3. The laser induced breakdown spectroscopy quantitative analysis method according to claim 1, characterized in that: In the fourth step, multiple preferred characteristic spectral lines of each element and the Fe normalized spectral line are freely combined and correlation scores are calculated, and the best combination pairs of each element are matched according to the scores.

Images