Laser-induced breakdown spectroscopy analysis temperature self-calibration method and system based on standard light source

Abstract

The invention discloses a laser-induced breakdown spectroscopy analysis temperature self-calibration method based on a standard light source. The method comprises the following steps: establishing a laser-induced breakdown spectroscopy analysis system; measuring the standard light source at different temperatures, and establishing a temperature-response deviation mapping model; analyzing the sample to obtain a sample analysis result; a sample analysis result is corrected through a temperature-response deviation mapping model; dividing sample correction results into a training set, a verification set and a test set, inputting the training set, the verification set and the test set into a neural network model Transform prediction model for training, and performing feature extraction; inputting a feature extraction result into a random forest model RF for element concentration prediction; the steps are repeated until the number of times of repetition reaches a preset value and the concentration prediction accuracy reaches a preset value; and obtaining a component detection output model. According to the invention, the spectral response characteristics of the LIBS system can be monitored and calibrated in real time at different environment temperatures, so that the detection precision, stability and reliability of the LIBS system in a complex and dynamic environment are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of laser-induced breakdown spectroscopy, and in particular to a method and system for temperature self-calibration of laser-induced breakdown spectroscopy based on a standard light source. Background Technology

[0002] Laser-induced breakdown spectroscopy (LIBS) is a technique that uses high-energy pulsed lasers to ablate the surface of a sample, generating plasma, and then analyzes the plasma emission spectrum to achieve qualitative and quantitative elemental analysis. When the ambient temperature fluctuates, LIBS may encounter the following problems:

[0003] 1. Deformation of optical components: The expansion and contraction of lenses, gratings, mirrors, etc. due to thermal changes cause the optical path to shift, resulting in wavelength drift;

[0004] 2. Detector response changes: The dark current and quantum efficiency of CCD or ICCD detectors change with temperature, resulting in signal strength fluctuations and a decrease in signal-to-noise ratio;

[0005] 3. Electronic system drift: The parameters of electronic components such as amplifier circuits and ADC converters change with temperature, affecting the stability of signal acquisition;

[0006] 4. Laser output fluctuations: Laser energy and pulse width are affected by temperature, which leads to a decrease in plasma excitation stability.

[0007] In existing technologies, compressed air is typically introduced into the LIBS system cavity to regulate the internal ambient temperature, combined with temperature sensors and software compensation models for temperature control or correction. Compressed air can, to some extent, suppress local temperature rise and stabilize the operating temperature of optical components, while the software compensation model empirically corrects the spectral signal based on the measured temperature data. However, existing methods still have the following shortcomings:

[0008] (1) The supply of compressed air requires an external air source and pipeline system, which increases the complexity of the equipment and the operating cost; and the temperature of the compressed air used varies with the local environment and is significantly affected by region, season and climate. The environmental temperature varies greatly in different regions or in different seasons in the same region, making it difficult to achieve a stable and controllable temperature control effect.

[0009] (2) Software compensation models based on temperature sensors often rely on empirical formulas or simplified physical assumptions, which are difficult to fully reflect the complex nonlinear response characteristics in optical systems, detectors and electronic links, resulting in limited correction accuracy.

[0010] (3) The lack of a real-time, direct and traceable spectral calibration method makes it impossible to synchronously and accurately quantitatively correct wavelength drift and intensity response changes, resulting in a decrease in the accuracy of element identification and quantitative analysis.

[0011] Therefore, there is an urgent need for a technical solution that can monitor and calibrate the spectral response characteristics of the LIBS system in real time under different ambient temperatures, so as to effectively improve its detection accuracy, stability and reliability in complex and dynamic environments. Summary of the Invention

[0012] This invention aims to at least partially address one of the technical problems in related technologies. To this end, one objective of this invention is to propose a temperature self-calibration method and system for laser-induced breakdown spectral analysis based on a standard light source. This method can monitor and calibrate the spectral response characteristics of the LIBS system in real time under different ambient temperatures, thereby effectively improving its detection accuracy, stability, and reliability in complex and dynamic environments.

[0013] In a first aspect, the present invention proposes a self-calibration method for laser-induced breakdown spectral analysis temperature based on a standard light source, the method comprising the following steps:

[0014] S1: Establish a laser-induced breakdown spectral analysis system;

[0015] S2: Using the laser-induced breakdown spectroscopy analysis system built in step S1, measure the standard light source at different temperatures, obtain the wavelength shift and spectral intensity response factor of the measured values ​​at each temperature compared with the standard light source, and establish a temperature-response deviation mapping model.

[0016] S3: Use the laser-induced breakdown spectroscopy analysis system built in step S1 to analyze the sample and obtain the sample analysis results;

[0017] S4: Apply the temperature-response deviation mapping model obtained in step S2 to correct the sample analysis results obtained in step S3, and obtain the sample correction results;

[0018] S5: Apply the sample correction results obtained in step S4, divide the sample correction results into training set, validation set and test set, and input them into the neural network model Transformer prediction model for training and feature extraction.

[0019] S6: Obtain the feature extraction results from step S5 and input them into the Random Forest (RF) model for element concentration prediction;

[0020] S7: Repeat steps S3-S6 until the number of repetitions reaches the preset value and the concentration prediction accuracy reaches the preset value, then proceed to step S8.

[0021] S8: Obtain the Transformer-RF model at this time as the component detection output model of the laser-induced breakdown spectroscopy analysis system.

[0022] Preferably, in step S2, a third-order polynomial fitting method is used to establish the temperature-response deviation mapping model:

[0023] S21: Within the set temperature range, the standard light source is measured at preset temperature intervals using a laser-induced breakdown spectral analysis system. The measured wavelength position and spectral intensity of the standard light source are recorded.

[0024] S22: Calculate the wavelength shift at each temperature. and spectral intensity response factor

[0025]

[0026]

[0027] in, The measured wavelength at temperature T. The standard wavelength of a standard light source

[0028] The measured spectral intensity is at temperature T. The spectral intensity of a standard light source at a reference temperature;

[0029] S23: Use the least squares method to fit the data with a third-order polynomial to establish the mapping relationship between temperature and deviation;

[0030]

[0031]

[0032] in, These are all fitting coefficients, solved using the least squares method. Different wavelengths have different thermal expansion effects, and each characteristic spectral line has an independent fitting coefficient. Each characteristic spectral line is fitted independently to obtain a unique set of coefficients.

[0033] Preferably, in step S21, the temperature range is set to 0-55℃, the temperature interval is set to 1℃, and the measured wavelength position of the mercury-argon lamp and the spectral intensity of the deuterium-tungsten lamp are recorded.

[0034] Preferably, the reference temperature in step S22 is 25°C.

[0035] Preferably, the corrected sample wavelength is calculated in step S4. and the spectral intensity of the corrected sample

[0036]

[0037]

[0038] This ensures that the corrected spectral data is consistent with the spectral data at the reference temperature, eliminating the effects of temperature drift.

[0039] Preferably, in step S5, the sample calibration results are extracted to obtain the corresponding wavelength, intensity, concentration and temperature. The extracted results are divided into training set, validation set and test set and input into the neural network model Transformer prediction model for training and feature extraction.

[0040] Preferably, in step S7, the root mean square error (RMSE) and the coefficient of determination (R²) are used. 2 Mean relative error (MRE) and concentration prediction accuracy As an evaluation indicator;

[0041]

[0042]

[0043]

[0044]

[0045] in, To predict concentration, This represents the actual concentration. This represents the number of samples.

[0046] Secondly, the present invention proposes a laser-induced breakdown spectral analysis temperature self-calibration system based on a standard light source, applying any of the above-mentioned laser-induced breakdown spectral analysis temperature self-calibration methods based on a standard light source, the system comprising:

[0047] Spectrometer;

[0048] Laser emission module: used to emit high-energy pulsed lasers and focus them onto the surface of the sample to excite plasma;

[0049] The light collection module has its light-incident end facing the sample to collect the plasma emitted light, and its light-out end is connected to the incident light path of the spectrometer.

[0050] Detection and processing module: Connected to the output of the spectrometer, used to acquire spectral signals and analyze elemental composition;

[0051] Standard light source module: Provides standard spectral output for calibration;

[0052] Optical path switching module: switches the optical path output to the spectrometer to enable input of standard spectrum or detection spectrum of the sample to be tested;

[0053] Temperature sensing module: used to monitor the internal ambient temperature of the system in real time;

[0054] Component detection output module: Outputs the component detection results of the sample based on the component detection output model;

[0055] Self-calibration control module: Triggered when preset trigger conditions are met, the above-mentioned laser-induced breakdown spectral analysis temperature self-calibration method based on standard light source is applied to correct the component detection output model.

[0056] Preferably, the self-calibration control module is triggered under one or more of the following conditions:

[0057] The current temperature T deviates from the set reference temperature;

[0058] The preset calibration cycle has been reached;

[0059] A significant decrease in signal-to-noise ratio or spectral abnormalities occurred during the detection process;

[0060] Sample quantification analysis failed.

[0061] The beneficial effects of this invention are:

[0062] (1) Improve wavelength measurement accuracy

[0063] By utilizing highly stable standard light sources such as mercury argon lamps, spectral lines are collected periodically or in real-time and compared with traceable standard wavelength tables to dynamically correct wavelength drift in the spectrometer. This mechanism significantly improves spectral line positioning accuracy, reduces spectral line shift errors caused by high temperatures or long-term use, ensures accurate element identification, and is suitable for qualitative analysis of various materials, including metals, non-metals, and multi-component alloys.

[0064] (2) Enhance spectral intensity stability

[0065] The system utilizes highly stable standard light sources such as deuterium-tungsten lamps to periodically or in real-time acquire spectral intensity. It integrates temperature response modeling and real-time gain correction mechanisms to automatically compensate for temperature-induced shifts in detector sensitivity and electronic system gain, reducing intensity fluctuations and signal-to-noise ratio degradation caused by environmental changes. Especially in high-temperature or high-humidity environments, it effectively maintains spectral intensity consistency, improving the repeatability and accuracy of quantitative analysis.

[0066] (3) Achieve adaptive correction under temperature changes

[0067] By establishing a mapping model (which can be a polynomial or machine learning model) between temperature and system response deviation, adaptive correction of the spectrum under continuous temperature changes can be achieved without relying on external temperature control measures such as constant temperature chambers and compressed air. Compared with traditional solutions, it is more suitable for deployment in complex application scenarios with drastic temperature fluctuations, such as in the field, industrial sites, high-temperature kiln openings, and mining vehicle platforms.

[0068] (3) Possesses predictive and self-healing abilities

[0069] By integrating Transformer and Random Forest (RF) algorithms, this system, based on traditional static calibration, possesses the ability to predict and replace spectra when failure occurs at high temperatures. It constructs a closed-loop self-healing mechanism of "measurement-evaluation-prediction-correction," which significantly improves the long-term operational reliability of the equipment. Attached Figure Description

[0070] In the attached diagram:

[0071] Figure 1 This is a flowchart of the method proposed in this invention;

[0072] Figure 2 This is the low-temperature spectrum of the deuterium-tungsten lamp proposed in this invention;

[0073] Figure 3 This is the high-temperature spectrum of the deuterium-tungsten lamp proposed in this invention;

[0074] Figure 4 This is a graph showing the wavelength deviation of the mercury-argon lamp as a function of temperature, as proposed in this invention.

[0075] Figure 5 This is a diagram of the Transformer-RF model proposed in this invention;

[0076] Figure 6 The flowchart for the elemental concentration and spectral prediction using the Transformer-RF model proposed in this invention is shown below;

[0077] Figure 7 This is a system diagram proposed in this invention;

[0078] Figure 8 This is the LIBS quantitative analysis spectrum of the standard sample block proposed in this invention;

[0079] Figures 9-11 The curves showing the relationship between the out-of-bag error and the number of decision trees in the random forest model for Ni, Mn, and Cu elements proposed in this invention are shown.

[0080] Figures 12-14 This is a distribution diagram showing the importance of OOB substitution features for Ni, Mn, and Cu elements proposed in this invention.

[0081] Figures 15-17This is a comparison of the temperature-concentration distribution of Ni, Mn, and Cu elements before and after calibration, as proposed in this invention.

[0082] Figures 18-20 This is a comparison chart of the quantitative analysis accuracy of Ni, Mn, and Cu elements before and after calibration, as proposed in this invention. Detailed Implementation

[0083] Example 1:

[0084] Reference Figure 1 A self-calibration method for laser-induced breakdown spectral analysis temperature based on a standard light source, the method comprising the following steps:

[0085] S1: Establish a laser-induced breakdown spectral analysis system;

[0086] It should be noted that the laser-induced breakdown spectroscopy analysis system is a mature technology that has been widely applied in the industry.

[0087] S2: Using the laser-induced breakdown spectroscopy analysis system built in step S1, measure the standard light source at different temperatures, obtain the wavelength shift and spectral intensity response factor of the measured values ​​at each temperature compared with the standard light source, and establish a temperature-response deviation mapping model.

[0088] Specifically: A temperature-response deviation mapping model is established using a third-order polynomial fitting method.

[0089] S21: Within the set temperature range, the standard light source is measured at preset temperature intervals using a laser-induced breakdown spectral analysis system. The measured wavelength position and spectral intensity of the standard light source are recorded.

[0090] In this embodiment, the measured wavelength position of the mercury-argon lamp and the spectral intensity of the deuterium-tungsten lamp are recorded.

[0091] Reference Figure 2 and Figure 3 When the standard light source is a deuterium-tungsten lamp, the full-band spectral intensity is tested. The change of spectral intensity with temperature shows that when the temperature is above 25℃, the spectral intensity of the deuterium-tungsten lamp decreases as the temperature increases, and when the temperature is below 25℃, the spectral intensity of the standard light source decreases as the temperature decreases. Therefore, the LIBS test can be used to compensate for the attenuation of spectral intensity with temperature by observing the change of spectral intensity of the deuterium-tungsten lamp with temperature.

[0092] Reference Figure 4The standard light source is a mercury argon lamp for resolution testing. The signal-to-noise ratio is optimal at around 15℃. The wavelength deviation increases with increasing temperature. By analyzing the relationship between the wavelength deviation of the standard light source and temperature, the wavelength axis of the LIBS spectrum is calibrated during the calibration stage to eliminate the misalignment of characteristic peaks caused by thermal drift, ensuring accurate positioning of elemental analysis lines, and thus more accurately identifying the substances corresponding to each wavelength.

[0093] S22: Calculate the wavelength shift at each temperature. and spectral intensity response factor ;

[0094]

[0095]

[0096] in, The measured wavelength at temperature T. The standard wavelength of a standard light source

[0097] The measured spectral intensity is at temperature T. The spectral intensity of a standard light source at a reference temperature;

[0098] In this embodiment, the reference temperature is 25°C.

[0099] S23: Use the least squares method to fit the data with a third-order polynomial to establish the mapping relationship between temperature and deviation;

[0100]

[0101]

[0102] in, These are all fitting coefficients, solved using the least squares method. Because the thermal expansion effect differs at different wavelengths, each characteristic spectral line (e.g., 253.7 nm, 435.8 nm, etc.) has an independent fitting coefficient. Therefore, each characteristic spectral line is fitted independently to obtain a dedicated set of coefficients.

[0103] S3: Use the laser-induced breakdown spectroscopy analysis system built in step S1 to analyze the sample and obtain the sample analysis results;

[0104] S4: Apply the temperature-response deviation mapping model obtained in step S2 to correct the sample analysis results obtained in step S3, and obtain the sample correction results;

[0105] Specifically, calculate the corrected sample wavelength. and the spectral intensity of the corrected sample :

[0106]

[0107]

[0108] This ensures that the corrected spectral data is consistent with the spectral data at the reference temperature, eliminating the effects of temperature drift.

[0109] S5: Apply the sample correction results obtained in step S4, divide the sample correction results into training set, validation set and test set, and input them into the neural network model Transformer prediction model for training and feature extraction.

[0110] The Transformer is a neural network architecture based on an attention mechanism. Attention is a deep learning technique that allows the model to focus on the most relevant parts of the input data when processing it. By calculating the correlation between elements in the input sequence, the model assigns weights to each element, reflecting its importance. Thus, when generating the output, self-attention is a form of attention mechanism that allows the model to consider information from all positions in the same sequence when processing an element at a specific position.

[0111] Attention mechanisms have become a core component of advanced models such as Transformer, significantly improving model performance and efficiency. Figure 1 As shown. In the standard self-attention mechanism, each element of the input sequence X is mapped to three different vector spaces, called Key (K), Query (Q), and Value (V) vectors.

[0112]

[0113] Then, the attention mechanism score is calculated as follows:

[0114]

[0115] in It consists of three trainable parameter matrices.

[0116] The Attention mechanism does not use matrices directly, but rather uses three matrices generated by multiplying them together.

[0117] In this embodiment, as Figure 5As shown, firstly, features (wavelength, intensity, concentration, temperature) are extracted from the corrected spectral data. Next, the extracted features are Z-score normalized; this process helps improve the model's training and convergence speed and ensures fairness. Then, the features, along with positional encoding information, are input into the first layer: a multi-head attention layer, followed by a dropout layer to prevent overfitting and enhance the model's generalization ability. Sequence order information requires additional positional encoding, adding positional information to each input feature. In the multi-head attention mechanism, each head independently calculates attention (self-attention when i=1):

[0118]

[0119] Each head independently computes attention, and after obtaining multiple outputs, the outputs of all heads are concatenated. Then, a parameter matrix is ​​applied to reduce the dimensionality, thus obtaining the final output: .

[0120]

[0121] Multi-head attention (MGA) learns in multiple subspaces, enabling it to simultaneously focus on the relationships between different positions and features in the input sequence. This enhances the model's ability to model complex patterns and improves its fitting performance. In the model, the input sequentially passes through two MGA layers and a dropout layer, and finally the features are connected to the regressor flow (RF) to complete the regression task for LIBS quantitative analysis.

[0122] The Attention mechanism does not use matrices directly, but rather uses three matrices generated by multiplying them together.

[0123] S6: Obtain the feature extraction results from step S5 and input them into the Random Forest (RF) model for element concentration prediction;

[0124] Specifically, different types of models are combined by combining the neural network model Transformer with the efficient random forest model.

[0125] In this embodiment, as Figure 6 As shown, the entire dataset is first divided into training, validation, and test sets. A Transformer model is then trained on the training set. Hyperparameters are initialized and optimized by setting upper and lower bounds and optimizing the number of iterations.

[0126] Random forests (RF) build predictive models by integrating multiple decision trees to improve overall accuracy and robustness. RF performs well in handling complex and nonlinear data, making it widely used for regression problems. Model performance depends on proper tuning of the decision trees and the number of nodes.

[0127] S7: Repeat steps S3-S6 until the number of repetitions reaches the preset value and the concentration prediction accuracy reaches the preset value, then proceed to step S8.

[0128] Specifically, the root mean square error (RMSE) and the coefficient of determination (R²) are used. 2 Mean relative error (MRE) and concentration prediction accuracy As an evaluation indicator;

[0129]

[0130] in, To predict concentration, This represents the actual concentration. This represents the number of samples.

[0131] S8: Obtain the Transformer-RF model at this time as the component detection output model of the laser-induced breakdown spectroscopy analysis system.

[0132] Example 2:

[0133] like Figure 7 As shown, a self-calibration system for laser-induced breakdown spectral analysis temperature based on a standard light source is provided. This system applies any of the aforementioned self-calibration methods for laser-induced breakdown spectral analysis temperature based on a standard light source. The system includes:

[0134] Spectrometer;

[0135] Laser emission module: used to emit high-energy pulsed lasers and focus them onto the surface of the sample to excite plasma;

[0136] Specifically, the laser emission module uses an Nd:YAG 1064 nm laser.

[0137] The light collection module includes a reflector, a focusing lens group, a dichroic mirror, a collecting lens, and an optical fiber coupler. Its light-incident end faces the sample to be tested to collect the plasma emitted light, and its light-out end is connected to the incident light path of the spectrometer.

[0138] Detection and processing module: includes a CCD / ICCD detector and a data processing unit, which is connected to the output of the spectrometer and is used to acquire spectral signals and analyze elemental composition;

[0139] Standard light source module: Provides standard spectral output for calibration;

[0140] Specifically, standard light sources can be mercury argon lamps and deuterium tungsten lamps.

[0141] Optical path switching module: switches the optical path output to the spectrometer to enable input of standard spectrum or detection spectrum of the sample to be tested;

[0142] Temperature sensing module: used to monitor the internal ambient temperature of the system in real time;

[0143] Component detection output module: Outputs the component detection results of the sample based on the component detection output model;

[0144] Self-calibration control module: Triggered when preset trigger conditions are met, the above-mentioned laser-induced breakdown spectral analysis temperature self-calibration method based on standard light source is applied to correct the component detection output model.

[0145] Specifically, the self-calibration control module is triggered when one or more of the following conditions are met:

[0146] The current temperature T deviates from the set reference temperature;

[0147] The preset calibration cycle has been reached;

[0148] A significant decrease in signal-to-noise ratio or spectral abnormalities occurred during the detection process;

[0149] Sample quantification analysis failed.

[0150] Example 3:

[0151] LIBS quantitative analysis was performed on a standard sample (carbon steel), and the spectrum is shown below. Figure 8 As shown, the analysis process and results are illustrated using the detection of three metallic elements, Ni, Mn, and Cu, in the sample as an example.

[0152] Random Forest OOB Error Analysis and Model Stability Evaluation

[0153] To evaluate the stability of the Transformer-RF model in quantitative analysis of LIBS for different elements, Figure 9 , Figure 10 and Figure 11 The out-of-bag (OOB) error of random forest models for Ni, Mn, and Cu elements is presented as a function of the number of decision trees.

[0154] All three elements exhibited highly consistent error evolution patterns: when the number of decision trees was small (approximately 1–20), the OOB error decreased rapidly with increasing tree count, indicating a significant improvement in the model's ensemble capability and rapid correction of prediction bias. When the number of trees further increased to over 50, the OOB error gradually stabilized, and the curve fluctuations significantly decreased, indicating that the model had sufficiently learned the spectral feature distribution, and further increasing the number of trees had limited effect on improving generalization performance. Within the stable range, the OOB errors for Ni, Mn, and Cu all maintained relatively small fluctuations, indicating that the deep features extracted by the Transformer had good discriminative power, and the Random Forest achieved stable and reliable concentration regression in this feature space. Considering both prediction accuracy and computational efficiency, this embodiment uniformly sets the number of decision trees in the Random Forest to 200.

[0155] Importance analysis of key deep features

[0156] To further reveal the discrimination mechanism of the Transformer-RF model, this embodiment performs statistical analysis on the "OOB Permuted Predictor Importance" of the Random Forest output, and selects the top 10 features that contribute the most to concentration prediction for display, such as... Figure 12 , Figure 13 and Figure 14 These correspond to Ni, Mn, and Cu, respectively, as shown in the figure.

[0157] The analysis results show that the feature contributions of different elements exhibit significant sparsity, meaning that only a few deep features dominate concentration prediction, while the contributions of other features are relatively small. This indicates that the Transformer network has effectively compressed and reconstructed high-dimensional spectral information, mapping the original spectrum into a set of deep representations highly correlated with concentration. Furthermore, the Top 10 feature numbers for Ni, Mn, and Cu are not entirely consistent, reflecting the model's ability to adaptively extract differentiated features based on the spectral structure and response characteristics of different elements. This result demonstrates that the Transformer-RF model possesses good adaptability and potential physical interpretability in multi-element LIBS quantitative analysis tasks.

[0158] The effect of temperature on LIBS quantitative analysis and calibration

[0159] Figure 15 , Figure 16 and Figure 17 The original temperature-concentration distributions of Ni, Mn, and Cu under different temperature conditions are presented, along with a comparison of the predicted results after temperature self-calibration.

[0160] Without calibration, the predicted concentrations of all three elements exhibited significant temperature-dependent biases: the predicted results for the same actual concentration sample were discrete under different temperature conditions, with the bias amplified further in the medium-to-high temperature range (approximately 30–50 °C). Among them, Mn showed particularly significant systematic errors due to temperature due to its large concentration range (approximately 1–5 ppm), while Ni and Cu were also quite sensitive to temperature disturbances in the low concentration range.

[0161] After introducing the Transformer–RF+ standard light source-assisted temperature calibration strategy, the prediction results under different temperature conditions were significantly improved. After calibration, the predicted concentration points at each temperature are distributed around the ideal horizontal line, and the correlation between temperature and concentration is significantly weakened; the same true concentration maintains an approximately constant predicted value across the entire temperature range, indicating that the systematic error introduced by temperature has been effectively eliminated.

[0162] Comparative analysis of quantitative prediction performance before and after calibration

[0163] To quantitatively evaluate model performance, root mean square error (RMSE), coefficient of determination (R²), mean relative error (MRE), and concentration prediction accuracy were used. As an evaluation index, the prediction results of the three elements before and after calibration are compared, such as... Figure 18 , Figure 19 and Figure 20 The results are summarized in Table 1.

[0164] Table 1 Summary of parameters before and after calibration

[0165]

[0166] For Cu, the RMSE decreased from 0.0277 ppm to 0.0195 ppm, the R² increased from 0.773 to 0.888, and the MRE decreased from 8.91% to 5.93%, improving prediction accuracy. It increased to 94.07%.

[0167] For Mn, RMSE decreased from 0.424 ppm to 0.333 ppm, R² increased from 0.866 to 0.918, and MRE decreased from 11.79% to 8.22%. It increased to 91.78%.

[0168] For Ni, the calibration effect was particularly significant, with RMSE decreasing from 0.0871ppm to 0.0240ppm, R² increasing to 0.987, MRE decreasing to 5.58%, and prediction accuracy reaching 94.42%.

[0169] In summary, all three elements showed a consistent trend of significantly reduced RMSE, significantly increased R², effective compression of MRE, and overall improved prediction accuracy after the introduction of temperature self-calibration.

Claims

1. A self-calibration method for laser-induced breakdown spectral analysis temperature based on a standard light source, characterized in that, The method steps are as follows: S1: Establish a laser-induced breakdown spectral analysis system; S2: Using the laser-induced breakdown spectroscopy analysis system built in step S1, measure the standard light source at different temperatures, obtain the wavelength shift and spectral intensity response factor of the measured values ​​at each temperature compared with the standard light source, and establish a temperature-response deviation mapping model. S3: Use the laser-induced breakdown spectroscopy analysis system built in step S1 to analyze the sample and obtain the sample analysis results; S4: Apply the temperature-response deviation mapping model obtained in step S2 to correct the sample analysis results obtained in step S3, and obtain the sample correction results; S5: Apply the sample correction results obtained in step S4, divide the sample correction results into training set, validation set and test set, and input them into the neural network model Transformer prediction model for training and feature extraction. S6: Obtain the feature extraction results from step S5 and input them into the Random Forest (RF) model for element concentration prediction; S7: Repeat steps S3-S6 until the number of repetitions reaches the preset value and then proceed to step S8; S8: Obtain the Transformer-RF model at this time as the component detection output model of the laser-induced breakdown spectroscopy analysis system.

2. The method for temperature self-calibration of laser-induced breakdown spectral analysis based on a standard light source according to claim 1, characterized in that, In step S2, a third-order polynomial fitting method is used to establish a temperature-response deviation mapping model: S21: Within the set temperature range, the standard light source is measured at preset temperature intervals using a laser-induced breakdown spectral analysis system. The measured wavelength position and spectral intensity of the standard light source are recorded. S22: Calculate the wavelength shift Δλ(T) and spectral intensity response factor at each temperature. ; ； Where, λ i,meas (T) is the measured wavelength at temperature T, λ i,std The standard wavelength of a standard light source The measured spectral intensity is at temperature T. The spectral intensity of a standard light source at a reference temperature; S23: Use the least squares method to fit the data with a third-order polynomial to establish the mapping relationship between temperature and deviation; Δλ i (T)=(a 0,i )+(a 1,i )T+(a 2,i )T 2 +(a 3,i )T 3 =(b0, i )+(b1, i )T+(b2, i )T 2 +(b3, i )T 3 Among them, a 0,i a 1,i a 2,i a 3,i and b0, i b1, i b2, i b3, i These are all fitting coefficients, solved using the least squares method. Different wavelengths have different thermal expansion effects, and each characteristic spectral line has an independent fitting coefficient. Each characteristic spectral line is fitted independently to obtain a unique set of coefficients.

3. The method for temperature self-calibration of laser-induced breakdown spectral analysis based on a standard light source according to claim 2, characterized in that: In step S21, the temperature range is set to 0-55℃, the temperature interval is set to 1℃, and the measured wavelength position of the mercury-argon lamp and the spectral intensity of the deuterium-tungsten lamp are recorded.

4. The method for temperature self-calibration of laser-induced breakdown spectral analysis based on a standard light source according to claim 2, characterized in that: In step S22, the reference temperature is 25°C.

5. The method for temperature self-calibration of laser-induced breakdown spectral analysis based on a standard light source according to claim 2, characterized in that, In step S4, the corrected sample wavelength λ is calculated. i,corrected and the spectral intensity I of the corrected sample corrected : ； ； This ensures that the corrected spectral data is consistent with the spectral data at the reference temperature, eliminating the effects of temperature drift.

6. The method for temperature self-calibration of laser-induced breakdown spectral analysis based on a standard light source according to claim 1, characterized in that: In step S5, the sample calibration results are extracted to obtain the corresponding wavelength, intensity, concentration and temperature. The extracted results are divided into training set, validation set and test set and input into the neural network model Transformer prediction model for training and feature extraction.

7. The method for temperature self-calibration of laser-induced breakdown spectral analysis based on a standard light source according to claim 1, characterized in that: In step S7, the root mean square error (RMSE) and the coefficient of determination (R²) are used. 2 The mean relative error (MRE) and concentration prediction accuracy (η) were used as evaluation indicators. RMSE = ; R² = 1- ； MRE = ; η = (1 - MRE) * 100%; in, To predict concentration, This represents the actual concentration. This represents the number of samples.

8. A self-calibration system for laser-induced breakdown spectral analysis temperature based on a standard light source, characterized in that: The system employing the laser-induced breakdown spectral analysis temperature self-calibration method based on a standard light source according to any one of claims 1-7, comprises: Spectrometer; Laser emission module: used to emit high-energy pulsed lasers and focus them onto the surface of the sample to excite plasma; The light collection module has its light-incident end facing the sample to collect the plasma emitted light, and its light-out end is connected to the incident light path of the spectrometer. Detection and processing module: Connected to the output of the spectrometer, used to acquire spectral signals and analyze elemental composition; Standard light source module: Provides standard spectral output for calibration; Optical path switching module: switches the optical path output to the spectrometer to enable input of standard spectrum or detection spectrum of the sample to be tested; Temperature sensing module: used to monitor the internal ambient temperature of the system in real time; Component detection output module: Outputs the component detection results of the sample based on the component detection output model; Self-calibration control module: Triggered when preset trigger conditions are met, the component detection output model is corrected by applying the laser-induced breakdown spectral analysis temperature self-calibration method based on a standard light source as described in any one of claims 1-7.

9. The self-calibration system for laser-induced breakdown spectral analysis temperature based on a standard light source according to claim 8, characterized in that, The self-calibration control module is triggered when one or more of the following conditions are met: The current temperature T deviates from the set reference temperature; The preset calibration cycle has been reached; A significant decrease in signal-to-noise ratio or spectral abnormalities occurred during the detection process; Sample quantification analysis failed.

Images