Alumina content detection device and method

Abstract

The application discloses an alumina substance content detection device and method, and belongs to the technical field of alumina substance content detection. The method of the application is as follows: 1, the alumina substance content detection device is used to configure the spectrum collection on the surface of a sample; 2, a spectrum matrix M of an nth sample is constructed n ; 3, the spectrum data obtained by the alumina sample is preprocessed; 4, a prediction model for detecting the alumina substance content of the in-model sample composed of 75% training spectrum data and 25% training spectrum data is constructed by a PCA-PLS linear regression sub-model and a preliminary model, the prediction model for detecting the alumina substance content is compared and verified by using out-of-model samples, and a trained prediction model for detecting the alumina substance content is obtained; 5, test data is input into the trained prediction model for detecting the alumina substance content, and a prediction result of the alumina substance content is obtained; compared with the prior art, the application improves the efficiency of the existing alumina detection.

Description

Technical Field

[0001] This invention relates to an alumina content detection device and method, belonging to the field of alumina content detection technology, and is applied to the detection of alumina components in industrial applications. Background Technology

[0002] Alumina, as a crucial basic raw material, plays an indispensable role in metallurgy, ceramics, electronics, chemicals, and new materials. Its purity and the content of specific impurity elements directly determine the performance, quality, and grade of the final product. For example, in the electrolytic aluminum industry, the impurity content (such as SiO2, Fe2O3, Na2O, etc.) of metallurgical-grade alumina used as raw material directly affects current efficiency, energy consumption, and the purity of the primary aluminum. In the production of high-end ceramics, artificial gemstones, and semiconductor substrates (such as sapphire crystals), the purity requirements for high-purity alumina are extremely stringent; even trace amounts of specific impurities can become defect centers in the product, leading to a sharp decline in performance. Therefore, rapid and accurate detection of the substance content in alumina is not only a key link in quality control and cost accounting during industrial production, but also a core technological support for ensuring the performance of downstream high-end products and promoting the development of new materials technologies, possessing extremely important industrial and economic significance.

[0003] Currently, the detection of substances in alumina mainly relies on classical wet chemical analysis and modern large-scale instrumental analysis. Wet chemical analysis is the most traditional and fundamental method, such as EDTA titration for alumina content, gravimetric determination of SiO2 content, and colorimetric determination of Fe2O3 content. Its biggest advantage is its relatively low equipment cost, and it is widely accepted as the arbitration method. However, its drawbacks are quite obvious: the operation process is cumbersome and lengthy, involving multiple steps of sample digestion, precipitation, filtration, washing, ignition, and weighing, requiring high operational skills and experience from analysts; the analysis cycle is too long, typically requiring several days to complete a full elemental analysis, which is seriously lagging behind the fast-paced demands of modern industrial production; reagent consumption is high, easily generating a large amount of waste liquid, which is environmentally unfriendly; and there are many human interference factors, making automation difficult, and the repeatability and stability of the results are sometimes difficult to guarantee. Modern large-scale instrumental analysis mainly includes inductively coupled plasma optical emission spectrometry (ICP-OES) and X-ray fluorescence spectrometry (XRF). For the ICP-OES method, the solid alumina sample needs to be completely dissolved and converted into a liquid. This process is time-consuming and carries the risk of incomplete dissolution or contamination. The accuracy of the XRF method heavily relies on a set of standard samples that are highly matched with the sample to be tested in terms of matrix, particle size, and structure to establish a calibration curve.

[0004] In existing laser-induced breakdown spectroscopy, the single opposite-side collection method leads to large fluctuations in the collected spectral signal and requires high accuracy of the optical path. This is mainly because the excitation point of the laser and the collection direction of the spectral signal are located on both sides of the sample. During the cooling process, the main expansion direction of the plasma differs from the collection direction, and the cooling rate at each location is inconsistent, which in turn affects the stability of the spectral signal.

[0005] Therefore, in the context of raw material testing in factories, improving the efficiency and accuracy of existing alumina testing methods has become an urgent problem to be solved. Summary of the Invention

[0006] The purpose of this invention is to solve the technical problem of efficiency in existing alumina detection in the context of raw material testing in factories, and to propose an alumina content detection device and method.

[0007] The working principle of this invention is as follows: Plasma is generated by laser loading of an alumina sample, and the radiation spectrum information of the plasma under multidimensional spatial coupling is collected. Highly stable plasma spectral data is obtained through multidimensional spatial coupling. A multi-directional light-receiving system is used to couple the spatial signal to a spectrometer to obtain the spectral data of each sample under multi-dimensional spatial coupling. The spectral data is preprocessed, and instrument calibration and prediction model establishment are performed using laboratory measurement data. A linear model is established between the preprocessed spectrum and known sample information using principal component analysis (PCA) combined with partial least squares regression (PLS) algorithm to obtain the final result. The established model has high accuracy in detecting the substance content of samples outside the model, with a prediction coefficient of determination R0. 2 With high accuracy, low root mean square error (RMSE), and low mean absolute error (MAE), it can achieve high-precision and rapid detection of the substance content in alumina.

[0008] The objective of this invention is achieved through the following technical solution:

[0009] On one hand, the present invention discloses a method for detecting the content of alumina, comprising the following steps:

[0010] Step 1: Configure the spectral acquisition of the sample surface using the alumina content detection device, and use the laser height stabilization instrument in the device to adjust the height between the sample surfaces to be equal.

[0011] Step 2: Construct the spectral matrix M of the nth sample as shown in equation (1). n ;

[0012]

[0013] The spectrometer has j pixels corresponding to j wavelengths, I k,j Represents the spectrum of the k-th row I kThe spectral intensity corresponding to the j-th wavelength, M n Represents all spectral information of the nth training set sample;

[0014] Step 3: Preprocess the spectral data obtained from the alumina sample, including removing saturated spectra from the spectral matrix sequentially, and using the σ criterion to remove spectra outside the [μ-σ,μ+σ] interval that deviate from the mean by more than one standard deviation. After removing the abnormal spectrum to form the abnormal spectrum and removing the background spectrum by differential baseline correction, the sample preliminary feature vector is formed by normalizing the background intensity of each channel.

[0015] Step 3.1: Remove the spectral matrix M n The spectrum with intensity higher than the saturation intensity of the spectrometer is used to form a desaturated spectrum M' as shown in equation (2). n ;

[0016]

[0017] Where g≤k; I k,j Spectrum I greater than the saturation threshold k Unable to represent sample information, the entire line of I k From M n Remove from the middle to obtain the removed saturation spectrum M' n ;I g This is the spectrum of the g-th row;

[0018] Step 3.2: Identify and screen valid data for abnormal spectra. Use the σ criterion to exclude data that deviates from the mean by more than one standard deviation and lies outside the [μ-σ, μ+σ] interval. Remove abnormal spectra;

[0019] Step 3.2.1: Average spectral intensity of the nth training set sample after multiple pulses. The total average value μ of the average spectral intensity data is formed as shown in equation (3);

[0020]

[0021] in, Let g be the average spectral intensity of the g-th pulse in the n-th training set sample;

[0022] Step 3.2.2: Using the σ criterion shown in equation (4), items outside the interval [μ-σ, μ+σ] that deviate from the mean by more than one standard deviation are excluded. The abnormal spectrum is removed to form the removed abnormal spectrum M” as shown in equation (5). n ;

[0023]

[0024] Where h≤g;

[0025] Step 3.3: Remove anomalous spectral M” using cubic spline interpolation. n The original spectrum is subjected to minimum value segmentation spectral fitting, and the baseline is corrected by difference between the original spectrum and the original spectrum to form a background-removed spectrum M”'. n ;

[0026] Step 3.3.1: Remove anomalous spectrum M” n Extract the original spectrum, segment the original spectrum according to the window width w1 and the window movement distance w2, and extract the minimum value segmented spectrum P. f ;

[0027] Step 3.3.2: Use cubic spline interpolation to segment the spectrum P by the minimum value. f Fitting is performed to form the background spectrum I hb ;

[0028] Step 3.3.3: Utilize the original spectrum I h Compared with background spectrum I hb Baseline correction is performed on the difference, and the original spectrum I h Compared with background spectrum I hb The difference is the baseline-corrected spectrum after removing the continuous background, and the background-removed spectrum M”' of the nth sample. n As shown in equation (6);

[0029]

[0030] Step 3.4: Use the channel-specific background intensity normalization method to process the background-de-background spectrum M”' n Normalization is performed to standardize the spectral intensity, forming a normalized spectrum M. n The initial feature vector of all samples is obtained by taking the mean value.

[0031] Step 3.4.1: Remove background spectrum M”' n Normalize the background intensity of each channel by dividing its spectral intensity at each wavelength by the integral value of the background spectral intensity of that channel, to obtain the normalized spectrum M”” as shown in equation (7). n ;

[0032]

[0033] Step 3.4.2: Normalize the spectrum M”” n The average characteristic spectrum is obtained by taking the mean value. The initial feature vector of the training set sample as shown in equation (8) is obtained as X;

[0034]

[0035] in,

[0036] Step 4: Construct a predictive model for detecting alumina content in in-model samples consisting of a PCA-PLS linear regression sub-model and a preliminary model. The predictive model for detecting alumina content is then compared and validated using out-of-model samples to obtain the trained predictive model for detecting alumina content.

[0037] Step 4.1: Divide the initial feature vectors into in-model samples and out-of-model samples by random partitioning;

[0038] Step 4.2: Construct a predictive model for detecting alumina content in in-model samples, consisting of a PCA-PLS linear regression sub-model and a preliminary model; validate the predictive model for detecting alumina content using out-of-model samples.

[0039] Step 4.3: Obtain the hyperparameter npc by cross-validating the in-model samples, and use the hyperparameter npc to adjust the hyperparameters of the PCA-PLS linear regression sub-model;

[0040] Step 4.3.1: Divide the samples in the model as shown in Equation (9) into 75% training spectral data and 25% test spectral data according to the total amount of data;

[0041]

[0042] Where L ′ N,o Refers to the spectral intensity at the wavelength of the o-th pixel of the N-th sample;

[0043] Step 4.3.2: Use 75% of the training spectral data to model the PCA-PLS linear regression sub-model to generate regression spectral data;

[0044] Step 4.3.3: Input 25% of the test spectral data and regression spectral data into the preliminary model for cross-validation to obtain the hyperparameter npc;

[0045] Step 4.3.4: Use hyperparameter npc to tune the hyperparameters of the PCA-PLS linear regression sub-model;

[0046] Step 4.4: Input the off-model samples into the prediction model for alumina content detection, compare and verify with the estimated alumina content true value, and obtain the trained prediction model for alumina content detection.

[0047] Step 5: Input the test data into the trained prediction model for alumina content detection to obtain the prediction results for alumina content.

[0048] This invention discloses an alumina content detection device for implementing the above-mentioned method. The alumina content detection device disclosed in this invention comprises a pulsed laser, a reflector, a plano-convex lens group, a sample stage, an optical fiber group, a spectrometer, and a timing controller.

[0049] The pulsed laser is used to emit a parallel laser beam in a pulsed manner with a wavelength of 1064 nm.

[0050] The reflector is used to reflect the pulsed laser emitted by the pulsed laser to the first plano-convex lens;

[0051] The plano-convex lens group is used to focus the plasma formed by the laser onto the sample surface and connect it to the spectrometer via optical fiber after optical fiber coupling. It consists of the first to the seventh plano-convex lenses.

[0052] Furthermore, the first plano-convex lens is located above the sample stage and is used to focus the laser onto the sample surface;

[0053] Furthermore, the second to fifth plano-convex lenses are used to receive the plasma generated by laser irradiation on the sample surface and couple the spectrum into the first optical fiber;

[0054] Furthermore, the sixth and seventh plano-convex lenses are used to focus and couple the multidimensional spatial light signal emitted from the first optical fiber to the second optical fiber;

[0055] The optical fiber assembly consists of a first optical fiber and a second optical fiber. The first optical fiber is used to couple the multidimensional spatial optical signal obtained from the opposite side of the light reception, and the second optical fiber is used to transmit the optical signal to the spectrometer connected to the optical fiber.

[0056] The sample stage is a three-dimensional adjustable electric displacement stage used to support the sample;

[0057] The spectrometer is used to convert optical signals into electrical signals;

[0058] The timing controller is used to perform delayed synchronization between the laser and the spectrometer;

[0059] Furthermore, the aforementioned three-dimensional adjustable electric displacement stage is computer-controlled, transmitting real-time height data to the computer. The computer then controls the electric displacement stage to automatically adjust the height, ensuring that the sample is at the same height for each test. The fixed height of the displacement stage is selected as 3mm above the laser focusing position on the sample surface.

[0060] The optical path of this device is as follows:

[0061] After the timing controller performs a delay synchronization on the laser and the spectrometer, the pulsed laser emits a laser beam that is reflected by a mirror to a first plano-convex lens. The first plano-convex lens then focuses the beam onto the sample surface on the sample stage, thereby exciting the sample to generate plasma. The plasma signal is then focused into the first optical fiber by a second, third, fourth, and fifth plano-convex lens, respectively. The spectral signal collected in the multidimensional space after coupling through the first optical fiber is focused into the second optical fiber by a sixth and seventh plano-convex lens. The second optical fiber transmits the spectral signal to the spectrometer, where the spectral image is displayed by a computer.

[0062] Compared with existing technologies, it has the following beneficial effects:

[0063] 1. The present invention discloses a rapid alumina substance content detection device and method, which has rapid data acquisition and can realize the simultaneous and rapid detection of multiple substances in alumina, solving the problems of separate testing of different substances and complex experimental operation in traditional methods.

[0064] 2. The present invention discloses a rapid alumina content detection device and method, which can control the movement and height of the sample stage in real time by computer, ensuring a fixed sample height, improving the spectral stability of each detection, and improving the representativeness of the spectrum, which is of great significance to promoting industrial development.

[0065] 3. This invention effectively reduces spectral fluctuations and improves spectral stability by coupling plasma spectral information from multiple directions using a multi-dimensional spatial heterogeneous light-receiving device. Simultaneously, through algorithm improvements, the mean relative error (MRE) for predicting the material content of out-of-model samples can be below 4%, with the MRE for Al2O3 reaching as low as 0.02%, demonstrating high detection accuracy. Attached Figure Description

[0066] Figure 1 This is a flowchart illustrating the present invention;

[0067] Figure 2 This is a schematic diagram of the device of the present invention;

[0068] Figure 3 This is a graph showing the relationship between the actual and predicted values ​​of Al2O3 and NaO2 content in alumina samples.

[0069] Among them, 1-pulse laser, 2-reflector, 3-first plano-convex lens, 4-sample stage, 5-second plano-convex lens, 6-third plano-convex lens, 7-fourth plano-convex lens, 8-fifth plano-convex lens, 9-first optical fiber, 10-sixth plano-convex lens, 11-seventh plano-convex lens, 12-second optical fiber, 13-spectrometer, and 14-timing controller. Detailed Implementation

[0070] To better illustrate the purpose and advantages of this invention, the invention will be further described below with reference to the accompanying drawings and examples. It should be noted that the implementation of this invention is not limited to the following embodiments, and any modifications or alterations made to this invention will fall within the scope of protection of this invention.

[0071] Example

[0072] like Figure 1 As shown in the figure, the specific implementation steps of the alumina content detection method in this embodiment are as follows:

[0073] Step 1: Configure the spectral acquisition of the sample surface using the alumina content detection device, and use the laser height stabilization instrument in the device to adjust the height between the sample surfaces to be equal.

[0074] Step 2: Construct the spectral matrix M of the nth sample as shown in equation (1). n ;

[0075]

[0076] The spectrometer has j pixels corresponding to j wavelengths, I k,j Represents the spectrum of the k-th row I k The spectral intensity corresponding to the j-th wavelength, M n Represents all spectral information of the nth training set sample;

[0077] Step 3: Preprocess the spectral data obtained from the alumina sample, including removing saturated spectra from the spectral matrix sequentially, and using the σ criterion to remove spectra outside the [μ-σ,μ+σ] interval that deviate from the mean by more than one standard deviation. After removing the abnormal spectrum to form the abnormal spectrum and removing the background spectrum by differential baseline correction, the sample preliminary feature vector is formed by normalizing the background intensity of each channel.

[0078] Step 3.1: Remove the spectral matrix M n The spectrum with intensity higher than the saturation intensity of the spectrometer is used to form a desaturated spectrum M' as shown in equation (2). n ;

[0079]

[0080] Where g≤k; I k,j Spectrum I greater than the saturation threshold k Unable to represent sample information, the entire line of I k From M n Remove from the middle to obtain the removed saturation spectrum M' n ;I g This is the spectrum of the g-th row;

[0081] Step 3.2: Identify and screen valid data for abnormal spectra. Use the σ criterion to exclude data that deviates from the mean by more than one standard deviation and lies outside the [μ-σ, μ+σ] interval. Remove abnormal spectra;

[0082] Step 3.2.1: Average spectral intensity of the nth training set sample after multiple pulses. The total average value μ of the average spectral intensity data is formed as shown in equation (3);

[0083]

[0084] in, Let g be the average spectral intensity of the g-th pulse in the n-th training set sample;

[0085] Step 3.2.2: Using the σ criterion shown in equation (4), items outside the interval [μ-σ, μ+σ] that deviate from the mean by more than one standard deviation are excluded. The abnormal spectrum is removed to form the removed abnormal spectrum M” as shown in equation (5). n ;

[0086]

[0087] Where h≤g;

[0088] Step 3.3: Remove anomalous spectral M” using cubic spline interpolation. n The original spectrum is subjected to minimum value segmentation spectral fitting, and the baseline is corrected by difference between the original spectrum and the original spectrum to form a background-removed spectrum M”'. n ;

[0089] Step 3.3.1: Remove anomalous spectrum M” n Extract the original spectrum, segment the original spectrum according to the window width w1 and the window movement distance w2, and extract the minimum value segmented spectrum P. f ;

[0090] Step 3.3.2: Use cubic spline interpolation to segment the spectrum P by the minimum value. f Fitting is performed to form the background spectrum I hb ;

[0091] Step 3.3.3: Utilize the original spectrum I h Compared with background spectrum I hb Baseline correction is performed on the difference, and the original spectrum I h Compared with background spectrum I hb The difference is the baseline-corrected spectrum after removing the continuous background, and the background-removed spectrum M”' of the nth sample. n As shown in equation (6);

[0092]

[0093] Step 3.4: Use the channel-specific background intensity normalization method to process the background-de-background spectrum M”' n Normalization is performed to standardize the spectral intensity, forming a normalized spectrum M. n The initial feature vector of all samples is obtained by taking the mean value.

[0094] Step 3.4.1: Remove background spectrum M”' n Normalize the background intensity of each channel by dividing its spectral intensity at each wavelength by the integral value of the background spectral intensity of that channel, to obtain the normalized spectrum M”” as shown in equation (7). n ;

[0095]

[0096] Step 3.4.2: Normalize the spectrum M”” n The average characteristic spectrum is obtained by taking the mean value. The initial feature vector of the training set sample as shown in equation (8) is obtained as X;

[0097]

[0098] in,

[0099] Step 4: Construct a predictive model for detecting alumina content in in-model samples consisting of a PCA-PLS linear regression sub-model and a preliminary model. The predictive model for detecting alumina content is then compared and validated using out-of-model samples to obtain the trained predictive model for detecting alumina content.

[0100] Step 4.1: Divide the initial feature vectors into in-model samples and out-of-model samples by random partitioning;

[0101] Step 4.2: Construct a predictive model for detecting alumina content in in-model samples, consisting of a PCA-PLS linear regression sub-model and a preliminary model; validate the predictive model for detecting alumina content using out-of-model samples.

[0102] Step 4.3: Obtain the hyperparameter npc by cross-validating the in-model samples, and use the hyperparameter npc to adjust the hyperparameters of the PCA-PLS linear regression sub-model;

[0103] Step 4.3.1: Divide the samples in the model as shown in Equation (9) into 75% training spectral data and 25% test spectral data according to the total amount of data;

[0104]

[0105] Where L ′ N,o Refers to the spectral intensity at the wavelength of the o-th pixel of the N-th sample;

[0106] Step 4.3.2: Use 75% of the training spectral data to model the PCA-PLS linear regression sub-model to generate regression spectral data;

[0107] Step 4.3.3: Input 25% of the test spectral data and regression spectral data into the preliminary model for cross-validation to obtain the hyperparameter npc;

[0108] Step 4.3.4: Use hyperparameter npc to tune the hyperparameters of the PCA-PLS linear regression sub-model;

[0109] Step 4.4: Input the off-model samples into the prediction model for alumina content detection, compare and verify with the true values ​​of alumina content, and obtain the trained prediction model for alumina content detection.

[0110] Step 5: Input the test data into the trained prediction model for alumina content detection to obtain the prediction results for alumina content.

[0111] In this embodiment, alumina samples are randomly divided, and a small number of samples are selected as out-of-model samples to verify the applicability and robustness of the model. These out-of-model samples are not used in building the prediction model. The remaining samples are then divided into training and test sets, with 75% as training samples and 25% as test samples. The spectral data of the training and prediction sets are preprocessed according to steps two and three. A PCA-PLS linear regression sub-model is built based on the training set, and the hyperparameters are adjusted through multiple cross-validations during model building. After training, the feature data of the test set samples are input into the initially built prediction model for prediction. The hyperparameters of the prediction model are fine-tuned based on the prediction results of the test set samples until the optimal prediction model is obtained. The plasma spectra of the out-of-model samples are collected using the constructed alumina content detection device, and the spectral data are preprocessed according to steps two and three. The feature data of the out-of-model samples are then input into the established prediction model to obtain the prediction results of the alumina content, thus realizing the detection of the alumina sample content.

[0112] like Figure 2As shown, this embodiment provides an alumina content detection device for implementing the above method. The alumina content detection device disclosed in this invention comprises a pulsed laser, a reflector, a plano-convex lens group, a sample stage, an optical fiber group, a spectrometer, and a timing controller.

[0113] The pulsed laser is used to emit a parallel laser beam in a pulsed manner with a wavelength of 1064 nm.

[0114] The reflector is used to reflect the pulsed laser emitted by the pulsed laser to the first plano-convex lens;

[0115] The plano-convex lens group is used to focus the plasma formed by the laser onto the sample surface and connect it to the spectrometer via optical fiber after optical fiber coupling. It consists of the first to the seventh plano-convex lenses.

[0116] Furthermore, the first plano-convex lens is located above the sample stage and is used to focus the laser onto the sample surface;

[0117] Furthermore, the second to fifth plano-convex lenses are used to receive the plasma generated by laser irradiation on the sample surface and couple the spectrum into the first optical fiber;

[0118] Furthermore, the sixth and seventh plano-convex lenses are used to focus and couple the multidimensional spatial light signal emitted from the first optical fiber to the second optical fiber;

[0119] The optical fiber assembly consists of a first optical fiber and a second optical fiber. The first optical fiber is used to couple the multidimensional spatial optical signal obtained from the opposite side of the light reception, and the second optical fiber is used to transmit the optical signal to the spectrometer connected to the optical fiber.

[0120] The sample stage is a three-dimensional adjustable electric displacement stage used to support the sample;

[0121] The spectrometer is used to convert optical signals into electrical signals;

[0122] The timing controller is used to perform delayed synchronization between the laser and the spectrometer;

[0123] Furthermore, the aforementioned three-dimensional adjustable electric displacement stage is computer-controlled, transmitting real-time height data to the computer. The computer then controls the electric displacement stage to automatically adjust the height, ensuring that the sample is at the same height for each test. The fixed height of the displacement stage is selected as 3mm above the laser focusing position on the sample surface.

[0124] The optical path of this device is as follows:

[0125] After the timing controller performs a delay synchronization on the laser and the spectrometer, the pulsed laser emits a laser beam that is reflected by a mirror to a first plano-convex lens. The first plano-convex lens focuses the laser beam onto the sample surface on the sample stage, thereby exciting the sample to generate plasma. The plasma signal is then focused into the first optical fiber by a second, third, fourth, and fifth plano-convex lens, respectively. The spectral signal collected in the multidimensional space after coupling through the first optical fiber is focused into the second optical fiber by a sixth and seventh plano-convex lens. The second optical fiber transmits the spectral signal to the spectrometer, where the spectral image is displayed by a computer.

[0126] To further illustrate the advantages of this invention, a laser-induced plasma experimental method is used for explanation; such as... Figure 3 As shown, for Al2O3, the root mean square error (RMSE) between the true and predicted values ​​in the training set is 0.017, and the coefficient of determination R0 is... 2 The root mean square error (RMSE) between the actual and predicted values ​​on the test set was as high as 0.978, and the coefficient of determination (R²) was 0.029. 2 The root mean square error (RMSE) between the true and predicted values ​​in the training set for NaO2 is as high as 0.851. 2 The root mean square error (RMSE) between the actual and predicted values ​​on the test set was as high as 0.983, and the coefficient of determination (R²) was 0.02. 2 The mean relative error (MRE) for out-of-model samples was as high as 0.913. For Al2O3, the mean relative error (MRE) between the actual and predicted values ​​was as low as 0.0191%, and for NaO2, the mean relative error (MRE) between the actual and predicted values ​​was as low as 1.4415%.

Claims

1. A method for detecting the content of alumina, characterized in that: Includes the following steps, Step 1: Configure the spectral acquisition of the sample surface using the alumina content detection device, and use the laser height stabilization instrument in the device to adjust the height between the sample surfaces to be equal. Step 2: Construct the spectral matrix M of the nth sample as shown in equation (1). n ; The spectrometer has j pixels corresponding to j wavelengths, I k,j Represents the spectrum of the k-th row I k The spectral intensity corresponding to the j-th wavelength, M n Represents all spectral information of the nth training set sample; Step 3: Preprocess the spectral data obtained from the alumina sample, including removing saturated spectra from the spectral matrix sequentially, and using the σ criterion to remove spectra outside the [μ-σ,μ+σ] interval that deviate from the mean by more than one standard deviation. After removing the abnormal spectrum to form the abnormal spectrum and removing the background spectrum by differential baseline correction, the sample preliminary feature vector is formed by normalizing the background intensity of each channel. Step 4: Construct a predictive model for detecting alumina content in in-model samples consisting of a PCA-PLS linear regression sub-model and a preliminary model. The predictive model for detecting alumina content is then compared and validated using out-of-model samples to obtain the trained predictive model for detecting alumina content. Step 4.1: Divide the initial feature vectors into in-model samples and out-of-model samples by random partitioning; Step 4.2: Construct a predictive model for detecting alumina content in in-model samples, consisting of a PCA-PLS linear regression sub-model and a preliminary model; validate the predictive model for detecting alumina content using out-of-model samples. Step 4.3: Obtain the hyperparameter npc by cross-validating the in-model samples, and use the hyperparameter npc to adjust the hyperparameters of the PCA-PLS linear regression sub-model; Step 4.3.1: Divide the samples in the model as shown in Equation (9) into 75% training spectral data and 25% test spectral data according to the total amount of data; Where L ′ N,o Refers to the spectral intensity at the wavelength of the o-th pixel of the N-th sample; Step 4.3.2: Use 75% of the training spectral data to model the PCA-PLS linear regression sub-model to generate regression spectral data; Step 4.3.3: Input 25% of the test spectral data and regression spectral data into the preliminary model for cross-validation to obtain the hyperparameter npc; Step 4.3.4: Use hyperparameter npc to tune the hyperparameters of the PCA-PLS linear regression sub-model; Step 4.4: Input the off-model samples into the prediction model for alumina content detection, compare and verify with the estimated alumina content true value, and obtain the trained prediction model for alumina content detection. Step 5: Input the test data into the trained prediction model for alumina content detection to obtain the prediction results for alumina content.

2. The method for detecting alumina content as described in claim 1, characterized in that: Step 3 is implemented as follows: Step 3.1: Remove the spectral matrix M n The spectrum with intensity higher than the saturation intensity of the spectrometer is used to form a desaturated spectrum M' as shown in equation (2). n ; Where g≤k; I k,j Spectrum I greater than the saturation threshold k Unable to represent sample information, the entire line of I k From M n Remove from the middle to obtain the removed saturation spectrum M' n ;I g This is the spectrum of the g-th row; Step 3.2: Identify and screen valid data for abnormal spectra. Use the σ criterion to exclude data that deviates from the mean by more than one standard deviation and lies outside the [μ-σ, μ+σ] interval. Remove abnormal spectra; Step 3.3: Remove anomalous spectral M” using cubic spline interpolation. n The original spectrum is subjected to minimum value segmentation spectral fitting, and the baseline is corrected by difference between the original spectrum and the original spectrum to form a background-removed spectrum M”'. n ; Step 3.4: Use the channel-specific background intensity normalization method to process the background-de-background spectrum M”' n Normalization is performed to standardize the spectral intensity, forming a normalized spectrum M. n The initial feature vector of all samples is obtained by taking the mean value.

3. The method for detecting alumina content as described in claim 2, characterized in that: Step 3.2 is implemented as follows: Step 3.2.1: Average spectral intensity of the nth training set sample after multiple pulses. The total average value μ of the average spectral intensity data is formed as shown in equation (3); in, Let g be the average spectral intensity of the g-th pulse in the n-th training set sample; Step 3.2.2: Using the σ criterion shown in equation (4), items outside the interval [μ-σ, μ+σ] that deviate from the mean by more than one standard deviation are excluded. The abnormal spectrum is removed to form the removed abnormal spectrum M” as shown in equation (5). n ; Where h≤g.

4. The method for detecting alumina content as described in claim 2, characterized in that: Step 3.3 is implemented as follows: Step 3.3.1: Remove anomalous spectrum M” n Extract the original spectrum, segment the original spectrum according to the window width w1 and the window movement distance w2, and extract the minimum value segmented spectrum P. f ; Step 3.3.2: Use cubic spline interpolation to segment the spectrum P by the minimum value. f Fitting is performed to form the background spectrum I hb ; Step 3.3.3: Utilize the original spectrum I h Compared with background spectrum I hb Baseline correction is performed on the difference, and the original spectrum I h Compared with background spectrum I hb The difference is the baseline-corrected spectrum after removing the continuous background, and the background-removed spectrum M”' of the nth sample. n As shown in equation (6).

5. The method for detecting alumina content as described in claim 2, characterized in that: Step 3.4 is implemented as follows: Step 3.4.1: Remove background spectrum M”' n Normalize the background intensity of each channel by dividing its spectral intensity at each wavelength by the integral value of the background spectral intensity of that channel, to obtain the normalized spectrum M”” as shown in equation (7). n ; Step 3.4.2: Normalize the spectrum M”” n The average characteristic spectrum is obtained by taking the mean value. The initial feature vector of the training set sample as shown in equation (8) is obtained as X; in, 6. An alumina content detection device for implementing the method described in claim 1, characterized in that: It consists of a pulsed laser, a mirror, a plano-convex lens group, a sample stage, an optical fiber group, a spectrometer, and a timing controller; The pulsed laser is used to emit a parallel laser beam in a pulsed manner with a wavelength of 1064 nm. The reflector is used to reflect the pulsed laser emitted by the pulsed laser to the first plano-convex lens; The plano-convex lens group is used to focus the plasma formed by laser onto the sample surface and connect it to the spectrometer via optical fiber after optical fiber coupling. It consists of the first to the seventh plano-convex lenses. Furthermore, the first plano-convex lens is located above the sample stage and is used to focus the laser onto the sample surface; Furthermore, the second to fifth plano-convex lenses are used to receive the plasma generated by laser irradiation on the sample surface and couple the spectrum into the first optical fiber; Furthermore, the sixth and seventh plano-convex lenses are used to focus and couple the multidimensional spatial light signal emitted from the first optical fiber to the second optical fiber; The optical fiber assembly consists of a first optical fiber and a second optical fiber. The first optical fiber is used to couple the multidimensional spatial optical signal obtained from the opposite side of the light reception, and the second optical fiber is used to transmit the optical signal to the spectrometer connected to the optical fiber. The sample stage is a three-dimensional adjustable electric displacement stage used to support the sample; The spectrometer is used to convert optical signals into electrical signals; The timing controller is used to perform delayed synchronization of the laser and the spectrometer.

7. The alumina content detection device as described in claim 6, characterized in that: The aforementioned three-dimensional adjustable electric displacement stage is computer-controlled, and can transmit the real-time height to the computer. The computer controls the electric displacement stage to automatically adjust the height, ensuring that the sample is at the same height for each test. The fixed height of the displacement stage is selected to be 3mm above the laser focusing position on the sample surface.

Images