Impurity content detection method and system for purification treatment of silicon carbide

Abstract

The invention discloses an impurity content detection method and system for silicon carbide purification treatment, and relates to the technical field of silicon carbide detection, the method comprises the following steps: collecting a preset number of historical samples for silicon carbide detection after purification treatment from a silicon carbide database; obtaining organic impurity data and inorganic impurity data based on historical samples; constructing an impurity content prediction model; and inputting the real-time organic impurity data and the real-time inorganic impurity data into the impurity content prediction model to obtain prediction results of the organic impurity content and the inorganic impurity content. According to the method, more comprehensive and accurate organic impurity data and inorganic impurity data are obtained by collecting historical samples of silicon carbide detection and combining mass spectrum and spectrum analysis technologies, the detection range of a traditional optical spectrum analysis technology is covered, impurity components are deeply analyzed, and the detection accuracy is improved. Therefore, richer, more accurate and more comprehensive impurity information is provided, and the detection accuracy and reliability are greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of silicon carbide detection technology, and in particular to a method and system for detecting impurity content in silicon carbide purification processing. Background Art

[0002] Silicon carbide, as a crystalline material formed by the stable combination of carbon and silicon elements, exhibits many unique structural characteristics and excellent physical and chemical properties. These characteristics make it extremely important in many fields, especially in key areas such as energy, electronics, semiconductors, ceramics and coatings. Silicon carbide is increasingly widely used. For example, in the manufacture of high-temperature furnaces, high-performance semiconductor devices, optoelectronic devices, high-frequency devices, and as ceramic coatings and abrasives, silicon carbide has shown its excellent application value.

[0003] With the rapid development of science and technology, the application areas of silicon carbide are still expanding, bringing unprecedented development opportunities and new possibilities to related industries. In order to meet the needs of different application fields for high-performance, high-purity silicon carbide materials, while improving production efficiency, reducing costs and promoting environmental protection and sustainable development, it is very important to purify silicon carbide. During the purification process, although various measures are taken to reduce the generation of impurities, the content of impurities still needs to be strictly controlled to ensure the purity of the final product. Therefore, accurate impurity content testing of purified silicon carbide is an important step to ensure material quality and meet application requirements.

[0004] In the purification process of silicon carbide materials, optical spectroscopy is an important means of detecting impurity content; however, optical spectroscopy has certain limitations in practical applications, which may include sensitivity limitations for specific impurity detection, complexity of sample impurities, difficulty in data analysis, etc. Due to these factors, optical spectroscopy sometimes cannot provide accurate measurement of impurity content after silicon carbide purification, which affects the accurate evaluation of silicon carbide product performance and causes silicon carbide products to fail to meet high performance requirements.

[0005] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0006] In view of the problems in the related art, the present invention proposes a method and system for detecting impurity content in silicon carbide purification processing to overcome the above-mentioned technical problems existing in the existing related technology.

[0007] To this end, the specific technical solution adopted by the present invention is as follows: According to one aspect of the present invention, a method for detecting impurity content in silicon carbide purification treatment is provided, the method comprising the following steps: S1. Collect a preset number of historical samples of purified silicon carbide testing from the silicon carbide database; S2. Based on the historical samples detected by silicon carbide, mass spectrum data and spectral data of the historical samples are obtained, and the mass spectrum data and spectral data are processed respectively to obtain organic impurity data and inorganic impurity data based on the historical samples; S3. Process the organic impurity data and inorganic impurity data based on historical samples and build an impurity content prediction model in combination with machine learning algorithms; S4, obtaining real-time organic impurity data and real-time inorganic impurity data after purification, inputting the real-time organic impurity data and real-time inorganic impurity data into an impurity content prediction model, and obtaining prediction results of organic impurity content and inorganic impurity content; Among them, based on the historical samples detected by silicon carbide, obtaining mass spectrum data and spectral data of the historical samples, and processing the mass spectrum data and spectral data respectively to obtain organic impurity data and inorganic impurity data based on the historical samples includes the following steps: S21. Based on the historical samples of silicon carbide detection, separate the organic impurities and silicon carbide components in the historical samples of silicon carbide detection to obtain mass spectrum data; and obtain spectrum data by exciting the historical samples of silicon carbide detection through X-rays; S22, using a silicon carbide processing method, respectively processing the mass spectrometry data and the spectral data of the historical samples to identify organic characteristic peaks and inorganic characteristic peaks; S23, combining the organic characteristic peaks with the mass spectrum database to obtain organic impurity content data; S24. Extract the peak height of the inorganic characteristic peak and combine it with the spectrum database to obtain the inorganic impurity content data.

[0008] Furthermore, the mass spectrometry data and spectral data of the historical samples are processed by the silicon carbide processing method, and the identification of organic characteristic peaks and inorganic characteristic peaks includes the following steps: S221, screening the mass spectrum data and the spectrum data using a data screening algorithm; S222, using different preprocessing combinations to process the screened mass spectrum data and spectrum data to obtain a number of pairs of mass spectrum characteristic peak points and spectrum characteristic peak points; S223, based on a plurality of pairs of mass spectrum characteristic peak points and spectrum characteristic peak points, selecting the optimal mass spectrum characteristic peak point and spectrum characteristic peak point; S224, analyzing the best mass spectrum characteristic peak point, determining the mass-to-charge ratio of each mass spectrum characteristic peak point, combining the mass-to-charge ratio of known organic impurities, and identifying the organic characteristic peak; S225, analyzing the optimal spectral characteristic peak point, using a peak detection algorithm to identify the characteristic peak point associated with the known inorganic impurities, and obtaining the inorganic characteristic peak.

[0009] Furthermore, using a data screening algorithm to screen the mass spectrometry data and the spectral data includes the following steps: S2211, normalizing the mass spectrum data and the spectral data, and reducing the dimensions of the mass spectrum data and the spectral data using a linear discriminant analysis method; S2212, using Manhattan distance, calculating the mass spectrum distance between the mass spectrum data after dimensionality reduction processing and the spectral distance between the spectral data; S2213. Based on the mass spectrum distance and the spectral distance, filter out the mass spectrum data and spectral data distance that are higher than a preset threshold.

[0010] Furthermore, the preprocessing combination includes baseline correction, scatter correction, smoothing and scaling.

[0011] Furthermore, extracting the peak height of the inorganic characteristic peak and combining it with the spectrum database to obtain the inorganic impurity content data includes the following steps: S241, based on the inorganic characteristic peak, extracting the peak height of the inorganic characteristic peak by Fourier transform; S242, using a peak height matching algorithm, matching the peak height of the inorganic characteristic peak with the peak height of the known inorganic impurity spectrum data in the spectrum database to obtain inorganic impurity content data.

[0012] Furthermore, using a peak height matching algorithm, the peak height of the inorganic characteristic peak is matched with the peak height of the known inorganic impurity spectrum data in the spectrum database, and obtaining the inorganic impurity content data includes the following steps: S2421, based on the peak height of the extracted inorganic characteristic peak, selecting a number of spectral data below a preset distance to form a subsequence; S2422, dividing the subsequence into a number of intervals, and fitting each interval using the least squares method to obtain the slope of the interval; S2423, integrating the slopes of all intervals into a descriptor, and matching the descriptor with the descriptor in the spectral database using a dynamic time warping algorithm to obtain a matching path of the descriptor; S2424, based on the matching path of the descriptor, combined with the cost function of constructing the sum of squares, obtaining the optimal path of the cost function; S2425, obtaining the corresponding relationship between the peak height of the inorganic characteristic peak and the peak height of the known inorganic impurity in the spectrum database according to the index matrix corresponding to the optimal path of the cost function; S2426. Obtain inorganic impurity content data based on the correspondence between the peak height of the inorganic characteristic peak and the peak height of known inorganic impurities in the spectrum database.

[0013] Furthermore, the formula of the cost function is: In the formula, N represents the cost function; X Represents the index relationship of the descriptor; k Indicates the number of steps required to complete the description subsequence matching; r Indicates the first r step; d Indicates the peak height of the inorganic characteristic peak; t Indicates time.

[0014] Furthermore, the organic impurity data and inorganic impurity data based on historical samples are processed and combined with machine learning algorithms to construct an impurity content prediction model, including the following steps: S31. Perform data cleaning, data coding and feature engineering on organic impurity data and inorganic impurity data based on historical samples; S32, dividing the processed organic impurity data and inorganic impurity data into a training set and a test set; S33. Construct an impurity content prediction model based on machine learning algorithm; S34, using the training set to train the impurity content prediction model, and using the test set to evaluate the trained impurity content prediction model.

[0015] Furthermore, obtaining real-time organic impurity data and real-time inorganic impurity data after purification, inputting the real-time organic impurity data and real-time inorganic impurity data into an impurity content prediction model, and obtaining prediction results of organic impurity content and inorganic impurity content include the following steps: S41, collecting a preset number of real-time samples of purified silicon carbide for detection, and obtaining real-time mass spectrometry data by liquid chromatography-mass spectrometry; obtaining real-time spectrum data by X-ray fluorescence spectroscopy; S42, respectively processing the acquired real-time spectrum data and real-time mass spectrum data to obtain real-time organic impurity data and real-time inorganic impurity data based on the real-time sample; S43, processing the real-time organic impurity data and the real-time inorganic impurity data, and inputting the processed real-time organic impurity data and the real-time inorganic impurity data into the impurity content prediction model to obtain the prediction results of the organic impurity content and the inorganic impurity content.

[0016] According to another aspect of the present invention, there is also provided an impurity content detection system for silicon carbide purification treatment, the impurity content detection system for silicon carbide purification treatment comprising a sample collection module, an impurity data acquisition module, a model building module and an impurity content prediction module; The sample collection module, the impurity data acquisition module, the model building module and the impurity content prediction module are connected in sequence; A sample collection module, used to collect a preset number of purified historical samples of silicon carbide testing from a silicon carbide database; An impurity data acquisition module is used to acquire mass spectrum data and spectral data of historical samples based on historical samples detected by silicon carbide, and process the mass spectrum data and spectral data respectively to obtain organic impurity data and inorganic impurity data based on historical samples; Model building module, which is used to process organic impurity data and inorganic impurity data based on historical samples and build an impurity content prediction model in combination with machine learning algorithms; The impurity content prediction module is used to obtain real-time organic impurity data and real-time inorganic impurity data after purification, input the real-time organic impurity data and real-time inorganic impurity data into the impurity content prediction model, and obtain the prediction results of organic impurity content and inorganic impurity content.

[0017] The beneficial effects of the present invention are: 1. The present invention collects historical samples of silicon carbide detection and combines mass spectrometry and spectral analysis technology to obtain more comprehensive and accurate organic impurity data and inorganic impurity data. It not only covers the detection range of traditional optical spectral analysis technology, but also deeply analyzes the impurity composition, thereby providing richer, more accurate and comprehensive impurity information, greatly improving the accuracy and reliability of detection.

[0018] 2. The present invention combines the impurity content prediction model constructed by the machine learning algorithm to obtain the impurity content prediction results through real-time organic impurity data and real-time inorganic impurity data. It is not only highly real-time, but also can accurately reflect the impurity content in the silicon carbide sample, thereby greatly improving the production efficiency and quality control level.

[0019] 3. The present invention can reduce the scrap rate caused by substandard product performance by accurately detecting the impurity content in the silicon carbide purification process, thereby reducing production costs; at the same time, accurate impurity content detection can also help manufacturers to effectively control the impurity content of silicon carbide, ensure the overall quality and performance of the product, and further enhance the market competitiveness of the product. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0021] Figure 1 is a flow chart of a method for detecting impurity content in silicon carbide purification processing according to an embodiment of the present invention; Figure 2 The present invention is a block diagram of an impurity content detection system for silicon carbide purification according to an embodiment of the present invention.

[0022] In the figure: 1. Sample collection module; 2. Impurity data acquisition module; 3. Model building module; 4. Impurity content prediction module. DETAILED DESCRIPTION

[0023] To further illustrate each embodiment, the present invention provides drawings, which are part of the disclosure of the present invention and are mainly used to illustrate the embodiments. They can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these contents, ordinary technicians in this field should be able to understand other possible implementation methods and advantages of the present invention.

[0024] According to an embodiment of the present invention, a method and system for detecting impurity content in silicon carbide purification processing are provided.

[0025] The present invention is further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, according to the impurity content detection method for silicon carbide purification processing according to an embodiment of the present invention, the method comprises the following steps: S1. Collect a preset number of historical samples of purified silicon carbide testing from a silicon carbide database.

[0026] It should be noted that the number of historical samples to be collected should be determined based on the characteristics, quality requirements and accuracy requirements of the silicon carbide purification process. This number should be large enough to cover various possible impurity contents and distributions, while also taking into account the efficiency and feasibility of data acquisition.

[0027] A specific database query language (such as SQL) or a database management tool is used to access the silicon carbide database, and silicon carbide test samples that have been purified are screened out from the silicon carbide database, wherein the silicon carbide test samples contain complete test data and related processing information, such as purification method, processing time, processing conditions, etc.

[0028] Check the screened silicon carbide test samples to ensure their integrity and accuracy; from the checked silicon carbide test samples, randomly or according to specific rules, extract samples according to the preset number. If the number of samples in the silicon carbide database is much larger than the preset number, a random sampling method can be used. If sampling is required based on certain specific conditions, such as time sequence, processing conditions, etc., the corresponding sampling strategy needs to be used.

[0029] Export the extracted sample data to a local or specified storage location for subsequent data processing and model building.

[0030] S2. Based on historical samples detected by silicon carbide, mass spectrometry data and spectral data of the historical samples are obtained, and the mass spectrometry data and spectral data are processed respectively to obtain organic impurity data and inorganic impurity data based on the historical samples.

[0031] It should be noted that organic impurities mainly include some hydrocarbons, oils and other organic substances. These organic impurities may affect the performance of silicon carbide products, especially under high temperature or high pressure conditions. They may decompose to produce gas or liquid, causing defects or cracks inside the product.

[0032] Inorganic impurities include various metal and non-metal elements, such as iron, aluminum, magnesium, calcium, silicon, oxygen, etc. The presence of these impurities in silicon carbide products will affect their physical and chemical properties, such as hardness, strength, thermal stability, electrical conductivity, etc.; among them, metal impurities such as iron, aluminum, magnesium, etc. are common inorganic impurities in silicon carbide products, which may come from impurities in the raw materials, pollution in the smelting process or pollution on the surface of the equipment, etc. Non-metallic impurities such as silicon and oxygen mainly come from the components of silicon carbide itself.

[0033] Specifically, based on historical samples of silicon carbide detection, obtaining mass spectrum data and spectral data of the historical samples, and processing the mass spectrum data and spectral data respectively, to obtain organic impurity data and inorganic impurity data based on the historical samples includes the following steps: S21. Based on the historical samples of silicon carbide detection, separate the organic impurities and silicon carbide components in the historical samples of silicon carbide detection to obtain mass spectrometry data; and obtain spectral data by exciting the historical samples of silicon carbide detection through X-rays.

[0034] It should be noted that appropriate liquid chromatography conditions, such as mobile phase, flow rate, column temperature, etc., are set, and the historical samples of silicon carbide detection are processed by liquid chromatography-mass spectrometry to separate the organic impurities and silicon carbide components in the historical samples of silicon carbide detection, and obtain mass spectrometry data, including peak area, retention time, mass-to-charge ratio, etc.

[0035] Set appropriate X-ray fluorescence spectrometer parameters, such as excitation source energy, detector sensitivity, etc., process historical samples of silicon carbide detection by X-ray fluorescence spectroscopy, and obtain spectral data, including peak height, intensity, wavelength, etc., of historical samples detected by silicon carbide through X-ray excitation.

[0036] S22. Using the silicon carbide processing method, the mass spectrometry data and spectral data of the historical samples are processed respectively to identify the organic characteristic peaks and the inorganic characteristic peaks.

[0037] Specifically, the mass spectrometry data and spectral data of the historical samples are processed by the silicon carbide processing method, and the identification of organic characteristic peaks and inorganic characteristic peaks includes the following steps: S221. Filter the mass spectrometry data and the spectral data using a data filtering algorithm.

[0038] Specifically, using the data screening algorithm to screen the mass spectrometry data and the spectral data includes the following steps: S2211, normalizing the mass spectrum data and the spectral data, and reducing the dimensions of the mass spectrum data and the spectral data using a linear discriminant analysis method; S2212, using Manhattan distance, calculating the mass spectrum distance between the mass spectrum data after dimensionality reduction processing and the spectral distance between the spectral data; S2213. Based on the mass spectrum distance and the spectral distance, filter out the mass spectrum data and spectral data distance that are higher than a preset threshold.

[0039] S222. Process the filtered mass spectrum data and spectrum data using different preprocessing combinations to obtain a plurality of pairs of mass spectrum characteristic peak points and spectrum characteristic peak points.

[0040] Specifically, the preprocessing combination includes baseline correction, scatter correction, smoothing and scaling.

[0041] It should be noted that the baseline correction, scattering correction, smoothing and scaling in the preprocessing combination are fully arranged and combined, for example, baseline correction-scattering correction-smoothing-scaling, scattering correction-baseline correction-smoothing-scaling, etc., to form different preprocessing sequences; for each preprocessing sequence, the mass spectrometry data is preprocessed in this sequence, and after the preprocessing is completed, characteristic peak points are extracted from the mass spectrometry data, and corresponding characteristic peak points are extracted from the spectral data; the characteristic peak points in the mass spectrum and the spectrum are compared to obtain several pairs of mass spectrum characteristic peak points and spectral characteristic peak points.

[0042] For example, the preprocessing combination of baseline correction-scattering correction-smoothing processing-scale scaling is used to preprocess mass spectrometry data: first, baseline correction is performed to eliminate baseline offset or drift; then scattering correction is performed to reduce the scattering effect caused by sample particle size and uneven distribution; then smoothing processing is performed to reduce noise and improve the signal-to-noise ratio; finally, scale scaling is performed to eliminate the impact of scale differences.

[0043] For example, the preprocessing combination of baseline correction-scattering correction-smoothing processing-scale scaling is used to preprocess spectral data: first, baseline correction is performed to eliminate baseline offset or drift; then scattering correction is performed to reduce the scattering effect caused by sample particle size and uneven distribution; then smoothing processing is performed to reduce noise and improve the signal-to-noise ratio; finally, scale scaling is performed to eliminate the impact of scale differences.

[0044] S223. Based on several pairs of mass spectrum characteristic peak points and spectral characteristic peak points, select the optimal mass spectrum characteristic peak points and spectral characteristic peak points.

[0045] S224, analyzing the best mass spectrum characteristic peak point, determining the mass-to-charge ratio of each mass spectrum characteristic peak point, combining the mass-to-charge ratio of known organic impurities, and identifying the organic characteristic peak.

[0046] S225, analyzing the optimal spectral characteristic peak point, using a peak detection algorithm to identify the characteristic peak point associated with the known inorganic impurities, and obtaining the inorganic characteristic peak.

[0047] S23. Combining the organic characteristic peaks with the mass spectrum database, obtaining organic impurity content data.

[0048] It should be noted that the identified organic characteristic peaks are compared with a known mass spectrometry database, which usually contains a large number of mass spectra of known compounds and their corresponding structural information. By comparing, the organic impurity molecules corresponding to the characteristic peaks in the mass spectrometry data are determined.

[0049] For the identified organic impurity molecules, the intensity of the characteristic peak in the mass spectrum is used for quantification; in mass spectrometry analysis, the intensity of the characteristic peak is usually proportional to the concentration of the corresponding molecule in the historical sample; therefore, by measuring the intensity of the characteristic peak and combining it with appropriate quantitative methods (such as peak area method, external standard method, etc.), the content of organic impurities in the historical sample can be calculated.

[0050] In addition, the peak area can be obtained by the integration algorithm, that is, the area under the peak is numerically integrated. The calculation formula of the peak area is: A=h*W; where A represents the peak area, h represents the peak height, and W represents the peak width.

[0051] S24. Extract the peak height of the inorganic characteristic peak and combine it with the spectrum database to obtain the inorganic impurity content data.

[0052] Specifically, extracting the peak height of the inorganic characteristic peak and combining it with the spectrum database to obtain the inorganic impurity content data includes the following steps: S241. Based on the inorganic characteristic peak, the peak height of the inorganic characteristic peak is extracted using Fourier transform.

[0053] S242, using a peak height matching algorithm, matching the peak height of the inorganic characteristic peak with the peak height of the known inorganic impurity spectrum data in the spectrum database to obtain inorganic impurity content data.

[0054] Specifically, using a peak height matching algorithm, the peak height of the inorganic characteristic peak is matched with the peak height of the known inorganic impurity spectrum data in the spectrum database, and obtaining the inorganic impurity content data includes the following steps: S2421. Based on the peak height of the extracted inorganic characteristic peak, select a number of spectral data below a preset distance to form a subsequence.

[0055] S2422. Divide the subsequence into several intervals, and use the least squares method to fit each interval to obtain the slope of the interval.

[0056] S2423. Integrate the slopes of all intervals into a descriptor, and use a dynamic time warping algorithm to match the descriptor with the descriptor in the spectral database to obtain a matching path of the descriptor.

[0057] S2424. Based on the matching path of the descriptor, combined with the cost function of constructing the sum of squares, obtain the optimal path of the cost function.

[0058] Specifically, the formula of the cost function is: In the formula, N represents the cost function; X Represents the index relationship of the descriptor; k Indicates the number of steps required to complete the description subsequence matching; r Indicates the first r step; d Indicates the peak height of the inorganic characteristic peak; t Indicates time.

[0059] S2425. Obtain the correspondence between the peak height of the inorganic characteristic peak and the peak height of the known inorganic impurity in the spectrum database according to the index matrix corresponding to the optimal path of the cost function.

[0060] S2426. Obtain inorganic impurity content data based on the correspondence between the peak height of the inorganic characteristic peak and the peak height of known inorganic impurities in the spectrum database.

[0061] S3. Process the organic impurity data and inorganic impurity data based on historical samples, and build an impurity content prediction model by combining machine learning algorithms.

[0062] Specifically, the organic impurity data and inorganic impurity data based on historical samples are processed and combined with machine learning algorithms to construct an impurity content prediction model, including the following steps: S31. Perform data cleaning, data encoding and feature engineering on organic impurity data and inorganic impurity data based on historical samples.

[0063] It should be noted that data cleaning includes missing value processing, outlier processing, duplicate value processing and data format unification; check whether there are missing values ​​or null values ​​in the organic impurity data and inorganic impurity data. If so, it is necessary to decide whether to delete the records containing missing values ​​and use statistical values ​​such as the mean, median, and mode to fill them according to the characteristics of the organic impurity data and inorganic impurity data; identify outliers in the organic impurity data and inorganic impurity data by drawing statistical charts such as histograms and box plots. For outliers, it can be decided whether to retain, delete, or replace them with the mean or other statistical values; check whether there are duplicate records in the organic impurity data and inorganic impurity data. If so, it is necessary to decide whether to retain unique records or delete duplicate records according to the needs of the organic impurity data and inorganic impurity data; ensure that the data formats of the organic impurity data and inorganic impurity data are unified, such as date format, number format, etc.

[0064] Data encoding includes categorical variable encoding and text data encoding. If the organic impurity data and inorganic impurity data contain categorical variables (such as the type of silicon carbide impurities, treatment methods, etc.), they need to be converted into numerical data for processing by machine learning algorithms. Common encoding methods include label encoding and one-hot encoding. If the organic impurity data and inorganic impurity data contain text data (such as the name of silicon carbide impurities, description of treatment methods, etc.), text mining technology (such as TF-IDF, Word2Vec, etc.) is needed to convert them into numerical features.

[0065] Feature engineering includes feature selection, feature construction, feature scaling, and feature dimensionality reduction. According to the requirements and data characteristics of organic impurity data and inorganic impurity data, select features that are useful for the impurity content prediction model. The importance of features can be evaluated through statistical tests (such as chi-square test, information gain, etc.) or machine learning algorithms (such as random forest, gradient boosting machine, etc.). New features can be constructed according to the requirements and data characteristics of organic impurity data and inorganic impurity data. For example, derived features (such as mean, standard deviation, ratio, etc.) can be calculated based on existing features, or new features can be created through feature combination, polynomial features, etc. In order to eliminate the dimensional differences between different features, features can be scaled. Common scaling methods include minimum-maximum scaling, Z-score scaling, etc. If the number of features is too large, it may cause the impurity content prediction model to overfit or the computational complexity to be too high. Feature dimensionality reduction methods can be used to reduce the number of features while retaining most of the information in the data. Common dimensionality reduction methods include principal component analysis and t-distributed random neighbor embedding.

[0066] S32, dividing the processed organic impurity data and inorganic impurity data into a training set and a test set.

[0067] S33. Build an impurity content prediction model based on machine learning algorithm.

[0068] It should be supplemented that, according to the characteristics of organic impurity data and inorganic impurity data, a decision tree regression model was selected to construct an impurity content prediction model; According to the characteristics of organic impurity data and inorganic impurity data, the organic impurity data and inorganic impurity data are divided into several subsets, each subset represents a specific feature value range or category; the above steps are recursively applied to each subset, that is, in each subset, the organic impurity data and inorganic impurity data are again divided according to the characteristics of the organic impurity data and inorganic impurity data, and subtrees are continuously constructed until the stopping condition is met, such as reaching the maximum number of characteristics of the organic impurity data and inorganic impurity data, the number of samples in the feature judgment of the organic impurity data and inorganic impurity data is less than a threshold, etc.; a complete decision tree model, that is, an impurity content prediction model, is constructed.

[0069] S34, using the training set to train the impurity content prediction model, and using the test set to evaluate the trained impurity content prediction model.

[0070] It should be noted that the training set is used as input and the corresponding impurity content is used as the target variable; the training set is passed to the impurity content prediction model for training, and the impurity content prediction model adjusts its internal parameters according to the relationship between the input training set and the target variable to minimize the prediction error.

[0071] The test set is passed as input to the impurity content prediction model, and the impurity content prediction value output by the impurity content prediction model is obtained; the impurity content predicted by the impurity content prediction model is compared with the actual impurity content in the test set, and evaluation indicators such as mean square error, mean absolute error, root mean square error or R² score are calculated, and the performance of the model is judged according to the value of the evaluation indicator; if the performance is not good, the hyperparameters in the training process of the impurity content prediction model can be returned and adjusted.

[0072] S4. Obtaining real-time organic impurity data and real-time inorganic impurity data after purification, inputting the real-time organic impurity data and real-time inorganic impurity data into an impurity content prediction model to obtain prediction results of the organic impurity content and the inorganic impurity content.

[0073] Specifically, obtaining real-time organic impurity data and real-time inorganic impurity data after purification, inputting the real-time organic impurity data and real-time inorganic impurity data into the impurity content prediction model, and obtaining the prediction results of the organic impurity content and the inorganic impurity content include the following steps: S41. Collect a preset number of real-time samples of purified silicon carbide for detection, and obtain real-time mass spectrometry data using liquid chromatography-mass spectrometry; and obtain real-time spectral data using X-ray fluorescence spectroscopy.

[0074] S42, respectively processing the acquired real-time spectrum data and real-time mass spectrum data to obtain real-time organic impurity data and real-time inorganic impurity data based on the real-time sample.

[0075] It should be noted that the real-time spectral data and real-time mass spectrometry data are screened using a data screening algorithm; the screened real-time spectral data and real-time mass spectrometry data are processed using different preprocessing combinations to obtain several pairs of real-time mass spectrometry characteristic peak points and real-time spectral characteristic peak points; based on several pairs of real-time mass spectrometry characteristic peak points and real-time spectral characteristic peak points, the optimal real-time mass spectrometry characteristic peak points and real-time spectral characteristic peak points are selected; the optimal real-time mass spectrometry characteristic peak points are analyzed to determine the mass-to-charge ratio of each real-time mass spectrometry characteristic peak point, and the real-time organic characteristic peak is identified in combination with the mass-to-charge ratio of known organic impurities; the optimal real-time spectral characteristic peak point is analyzed, and the characteristic peak points associated with known inorganic impurities are identified using a peak detection algorithm to obtain real-time inorganic characteristic peaks; the real-time organic characteristic peaks are combined with the mass spectrum database to obtain real-time organic impurity content data; the peak height of the real-time inorganic characteristic peak is extracted, and the real-time inorganic impurity content data is obtained in combination with the spectrum database.

[0076] S43, processing the real-time organic impurity data and the real-time inorganic impurity data, and inputting the processed real-time organic impurity data and the real-time inorganic impurity data into the impurity content prediction model to obtain the prediction results of the organic impurity content and the inorganic impurity content.

[0077] It should be noted that the real-time organic impurity data and the real-time inorganic impurity data are cleaned, encoded and feature engineered; the processed real-time organic impurity data and the real-time inorganic impurity data are used as feature vectors and input into the input layer of the impurity content prediction model; according to the feature values ​​in the real-time organic impurity data and the real-time inorganic impurity data, the decision path is traversed downward, and in each feature of the real-time organic impurity data and the real-time inorganic impurity data, according to the corresponding threshold represented by the feature, it is determined which sub-feature the real-time organic impurity data and the real-time inorganic impurity data should enter; when the final predicted value, i.e., the impurity content, is traversed, the impurity content will contain one or more predicted values; according to the impurity content, the predicted value corresponding to the impurity content is extracted as the prediction result of the organic impurity content and the inorganic impurity content.

[0078] like Figure 2 As shown, according to another embodiment of the present invention, there is also provided an impurity content detection system for silicon carbide purification treatment, the impurity content detection system for silicon carbide purification treatment comprises a sample collection module 1, an impurity data acquisition module 2, a model building module 3 and an impurity content prediction module 4; The sample collection module 1, the impurity data acquisition module 2, the model building module 3 and the impurity content prediction module 4 are connected in sequence; The sample collection module 1 is used to collect a preset number of purified historical samples of silicon carbide detection from the silicon carbide database; The impurity data acquisition module 2 is used to acquire mass spectrum data and spectral data of historical samples based on historical samples detected by silicon carbide, and process the mass spectrum data and spectral data respectively to obtain organic impurity data and inorganic impurity data based on historical samples; Model building module 3, used to process organic impurity data and inorganic impurity data based on historical samples, and build an impurity content prediction model in combination with a machine learning algorithm; The impurity content prediction module 4 is used to obtain real-time organic impurity data and real-time inorganic impurity data after purification, input the real-time organic impurity data and real-time inorganic impurity data into the impurity content prediction model, and obtain the prediction results of the organic impurity content and the inorganic impurity content.

[0079] In summary, by means of the above technical solution of the present invention, by collecting historical samples of silicon carbide detection, and combining mass spectrometry and spectral analysis technology, more comprehensive and accurate organic impurity data and inorganic impurity data are obtained, which not only covers the detection range of traditional optical spectral analysis technology, but also deeply analyzes the impurity components, thereby providing more abundant, accurate and comprehensive impurity information, greatly improving the accuracy and reliability of detection. By combining the impurity content prediction model constructed by machine learning algorithm, it is possible to obtain the impurity content prediction result through real-time organic impurity data and real-time inorganic impurity data, which not only has a high degree of real-time performance, but also can accurately reflect the impurity content in the silicon carbide sample, thereby greatly improving the production efficiency and quality control level. By accurately detecting the impurity content in the silicon carbide purification process, the scrap rate caused by substandard product performance can be reduced, thereby reducing production costs; at the same time, accurate impurity content detection also helps manufacturers to achieve effective control of the impurity content of silicon carbide, ensure the overall quality and performance of the product, and further enhance the market competitiveness of the product.

[0080] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for detecting impurity content in silicon carbide purification, characterized in that: The method comprises the following steps: S1. Collect a preset number of historical samples of purified silicon carbide testing from the silicon carbide database; S2. Based on the historical samples detected by silicon carbide, mass spectrum data and spectral data of the historical samples are obtained, and the mass spectrum data and spectral data are processed respectively to obtain organic impurity data and inorganic impurity data based on the historical samples; S3. Process the organic impurity data and inorganic impurity data based on historical samples and build an impurity content prediction model in combination with machine learning algorithms; S4, obtaining real-time organic impurity data and real-time inorganic impurity data after purification, inputting the real-time organic impurity data and real-time inorganic impurity data into an impurity content prediction model, and obtaining prediction results of organic impurity content and inorganic impurity content; Wherein, the historical samples detected based on silicon carbide, obtaining mass spectrum data and spectral data of the historical samples, and processing the mass spectrum data and spectral data respectively to obtain organic impurity data and inorganic impurity data based on the historical samples include the following steps: S21. Based on the historical samples of silicon carbide detection, separate the organic impurities and silicon carbide components in the historical samples of silicon carbide detection to obtain mass spectrum data; and obtain spectrum data by exciting the historical samples of silicon carbide detection through X-rays; S22, using a silicon carbide processing method, respectively processing the mass spectrometry data and the spectral data of the historical samples to identify organic characteristic peaks and inorganic characteristic peaks; S23, combining the organic characteristic peaks with the mass spectrum database to obtain organic impurity content data; S24. Extract the peak height of the inorganic characteristic peak and combine it with the spectrum database to obtain the inorganic impurity content data.

2. The impurity content detection method for silicon carbide purification according to claim 1, characterized in that: The silicon carbide processing method is used to process the mass spectrum data and the spectrum data of the historical samples respectively, and the identification of the organic characteristic peak and the inorganic characteristic peak includes the following steps: S221, screening the mass spectrum data and the spectrum data using a data screening algorithm; S222, using different preprocessing combinations to process the screened mass spectrum data and spectrum data to obtain a number of pairs of mass spectrum characteristic peak points and spectrum characteristic peak points; S223, based on a plurality of pairs of mass spectrum characteristic peak points and spectrum characteristic peak points, selecting the optimal mass spectrum characteristic peak point and spectrum characteristic peak point; S224, analyzing the best mass spectrum characteristic peak point, determining the mass-to-charge ratio of each mass spectrum characteristic peak point, combining the mass-to-charge ratio of known organic impurities, and identifying the organic characteristic peak; S225, analyzing the optimal spectral characteristic peak point, using a peak detection algorithm to identify the characteristic peak point associated with the known inorganic impurities, and obtaining the inorganic characteristic peak.

3. The impurity content detection method for silicon carbide purification according to claim 2, characterized in that: The method of using the data screening algorithm to screen the mass spectrum data and the spectrum data comprises the following steps: S2211, normalizing the mass spectrum data and the spectral data, and reducing the dimensions of the mass spectrum data and the spectral data using a linear discriminant analysis method; S2212, using Manhattan distance, calculating the mass spectrum distance between the mass spectrum data after dimensionality reduction processing and the spectral distance between the spectral data; S2213. Based on the mass spectrum distance and the spectral distance, filter out the mass spectrum data and spectral data distance that are higher than a preset threshold.

4. The impurity content detection method for silicon carbide purification according to claim 2, characterized in that: The preprocessing combination includes baseline correction, scatter correction, smoothing and scaling.

5. The impurity content detection method for silicon carbide purification according to claim 1, characterized in that: The step of extracting the peak height of the inorganic characteristic peak and combining it with the spectrum database to obtain the inorganic impurity content data comprises the following steps: S241, based on the inorganic characteristic peak, extracting the peak height of the inorganic characteristic peak by Fourier transform; S242, using a peak height matching algorithm, matching the peak height of the inorganic characteristic peak with the peak height of the known inorganic impurity spectrum data in the spectrum database to obtain inorganic impurity content data.

6. The method for detecting impurity content in silicon carbide purification according to claim 5, characterized in that: The peak height matching algorithm is used to match the peak height of the inorganic characteristic peak with the peak height of the known inorganic impurity spectrum data in the spectrum database to obtain the inorganic impurity content data, which includes the following steps: S2421, based on the peak height of the extracted inorganic characteristic peak, selecting a number of spectral data below a preset distance to form a subsequence; S2422, dividing the subsequence into a number of intervals, and fitting each interval using the least squares method to obtain the slope of the interval; S2423, integrating the slopes of all intervals into a descriptor, and matching the descriptor with the descriptor in the spectral database using a dynamic time warping algorithm to obtain a matching path of the descriptor; S2424, based on the matching path of the descriptor, combined with the cost function of constructing the sum of squares, obtaining the optimal path of the cost function; S2425, obtaining the corresponding relationship between the peak height of the inorganic characteristic peak and the peak height of the known inorganic impurity in the spectrum database according to the index matrix corresponding to the optimal path of the cost function; S2426. Obtain inorganic impurity content data based on the correspondence between the peak height of the inorganic characteristic peak and the peak height of known inorganic impurities in the spectrum database.

7. The method for detecting impurity content in silicon carbide purification according to claim 6, characterized in that: The formula of the cost function is: In the formula, N represents the cost function; X Represents the index relationship of the descriptor; k Indicates the number of steps required to complete the description subsequence matching; r Indicates the first r step; d Indicates the peak height of the inorganic characteristic peak; t Indicates time.

8. The method for detecting impurity content in silicon carbide purification according to claim 1, characterized in that: The processing of organic impurity data and inorganic impurity data based on historical samples and construction of an impurity content prediction model in combination with a machine learning algorithm comprises the following steps: S31. Perform data cleaning, data coding and feature engineering on organic impurity data and inorganic impurity data based on historical samples; S32, dividing the processed organic impurity data and inorganic impurity data into a training set and a test set; S33. Construct an impurity content prediction model based on machine learning algorithm; S34, using the training set to train the impurity content prediction model, and using the test set to evaluate the trained impurity content prediction model.

9. The method for detecting impurity content in silicon carbide purification according to claim 1, characterized in that: The step of obtaining the real-time organic impurity data and the real-time inorganic impurity data after the purification process, inputting the real-time organic impurity data and the real-time inorganic impurity data into the impurity content prediction model, and obtaining the prediction results of the organic impurity content and the inorganic impurity content comprises the following steps: S41, collecting a preset number of real-time samples of purified silicon carbide for detection, and obtaining real-time mass spectrometry data by liquid chromatography-mass spectrometry; obtaining real-time spectrum data by X-ray fluorescence spectroscopy; S42, respectively processing the acquired real-time spectrum data and real-time mass spectrum data to obtain real-time organic impurity data and real-time inorganic impurity data based on the real-time sample; S43, processing the real-time organic impurity data and the real-time inorganic impurity data, and inputting the processed real-time organic impurity data and the real-time inorganic impurity data into the impurity content prediction model to obtain the prediction results of the organic impurity content and the inorganic impurity content.

10. An impurity content detection system for silicon carbide purification treatment, used to implement the impurity content detection method for silicon carbide purification treatment according to any one of claims 1 to 9, characterized in that: The impurity content detection system for silicon carbide purification treatment includes a sample collection module, an impurity data acquisition module, a model building module and an impurity content prediction module; The sample collection module, the impurity data acquisition module, the model building module and the impurity content prediction module are connected in sequence; The sample collection module is used to collect a preset number of historical samples of silicon carbide detection after purification from the silicon carbide database; The impurity data acquisition module is used to acquire mass spectrum data and spectral data of historical samples based on historical samples detected by silicon carbide, and process the mass spectrum data and spectral data respectively to obtain organic impurity data and inorganic impurity data based on historical samples; The model building module is used to process the organic impurity data and inorganic impurity data based on historical samples and build an impurity content prediction model in combination with a machine learning algorithm; The impurity content prediction module is used to obtain real-time organic impurity data and real-time inorganic impurity data after purification, input the real-time organic impurity data and real-time inorganic impurity data into the impurity content prediction model, and obtain the prediction results of the organic impurity content and the inorganic impurity content.

Images