A spectrum recognition method and system based on vector database

Abstract

The present invention discloses a spectral identification method and system based on a vector database, and the method comprises the following steps: screening substances of a specified chemical system from a known database, including all simple substances or compounds composed of specified chemical elements, and superimposing the simulated XRDs of different simple substances or compounds according to different mass fractions to generate a vector database; searching the vector database according to the XRD of a mixture to be tested, determining the closest matching mixture combination ratio, and determining the material parameters corresponding to the XRD of the mixture to be tested. The spectral identification method and system based on a vector database provided by the present invention solves the problem that deep learning methods are difficult to expand the research system and that training is time-consuming. By using mature vector database technology, the post-processing analysis process is greatly simplified, tedious analysis and a large amount of manual intervention are eliminated, and a more accurate concentration estimate of each component in the mixture can be given.

Description

Technical Field

[0001] The present invention relates to the technical field of material spectrum analysis, and in particular to a spectrum recognition method and system based on a vector database. Background Art

[0002] The measurement and characterization methods of material science involve a large number of spectral measurements. Spectral analysis of materials can obtain parameters such as the composition and proportion of the materials. At present, the measurement methods in the industry are mainly aimed at the measurement and resolution of a single spectrum. When different materials are mixed and multiple spectra are superimposed, the identification of mixed spectral lines is a difficult problem.

[0003] X-ray diffraction (XRD) is a very important characterization technology in the field of materials science, with the characteristics of rapidity, high precision, strong penetration, non-contact detection and portable instruments. XRD can accurately measure the crystal structure and phase change phenomenon of materials, and is often used for non-destructive testing and analysis of product samples. It is a common method for technicians in this field to identify whether the synthetic materials meet expectations.

[0004] However, analyzing XRD is a tedious task that requires rich experience. In addition, in the actual operating environment, due to the different experimental synthesis technology routes, the experimental synthesis products are often not pure single-component materials, but mixtures of multiple compounds. Therefore, the measured XRD is the superposition of the XRD spectra of multiple single materials or known compounds. At the same time, due to the different instrument measurement parameters, the measured XRD will also deviate from the standard diffraction card. At present, researchers use computer software such as JADE and standard diffraction databases to assist in XRD identification, which requires a lot of manual work, which greatly increases the difficulty of spectral analysis and reduces work efficiency.

[0005] In order to improve efficiency and reduce manual intervention, the current measurement methods in the industry are usually carried out by researchers to measure and resolve a single spectrum, build a library in advance, use computer software such as JADE and standard diffraction databases to assist in XRD identification, or use deep learning models to perform automatic XRD analysis, use trained convolutional neural networks to automatically analyze the spectrum, and give the probability of the existence of a certain phase.

[0006] Both existing methods have defects to a certain extent. For auxiliary analysis, due to the difference in instrument measurement parameters, the XRD obtained by auxiliary analysis of a single pure substance deviates from the standard diffraction card, and still requires a lot of manual post-processing.

[0007] Convolutional neural networks are widely used in deep learning to process grid-structured data. They can automatically learn and extract high-level features without manual intervention, improve efficiency, and greatly reduce manual workload. However, there are still the following shortcomings:

[0008] (1) The training data is limited and it is difficult to cover a variety of chemical systems. It is difficult to expand the material systems of interest to different research fields, especially for experimental materials in the research and development process, which require faster and more efficient analysis and measurement.

[0009] (2) The training and post-processing process is rather complicated.

[0010] (3) It is difficult to give a more accurate estimate of the mass fraction of each component in the mixture.

[0011] Therefore, the prior art needs to be further improved and developed. Summary of the invention

[0012] In order to solve the above-mentioned problems in the prior art, the present invention proposes a spectral recognition method and system based on a vector database, which uses vector database technology and relies on a large-scale inorganic crystal material database. Preliminary analysis results are provided through a vector similarity algorithm integrated in the vector database, forming a software implementation system and software implementation method for spectral analysis technology that can be performed quickly and efficiently.

[0013] The technical solution of the present invention is as follows:

[0014] A spectral recognition method based on a vector database, comprising the following steps:

[0015] S1. Data preprocessing: screening substances of a specified chemical system from a known database, including all simple substances or compounds composed of specified chemical elements, superimposing the simulated XRD of different simple substances or compounds according to different mass fractions, and generating a vector database of the specified chemical system;

[0016] S2. Search the vector database according to the XRD of the mixture to be tested, determine the closest matching mixture combination ratio, and determine the material parameters corresponding to the XRD of the mixture to be tested.

[0017] The spectral identification method based on vector database, wherein the known database is an inorganic crystal database.

[0018] The spectral recognition method based on vector database, wherein the screening process in step S1 includes:

[0019] S12. Based on the material stability information contained in the inorganic crystal database, a subset of chemical substances that can be stably synthesized is screened out.

[0020] The spectral recognition method based on vector database, wherein the screening process in step S1 includes:

[0021] S13. Screening thermodynamically stable chemical substances in the chemical substance subset according to the substances included in the inorganic crystal database, wherein a thermodynamic stability criterion of the chemical substance is that the decomposition energy (Energy Above Hull) in the thermodynamic phase diagram is less than 0.5 eV / atom.

[0022] The spectral recognition method based on the vector database, wherein, after the step S12 and / or the step S13, further comprises the following steps:

[0023] S14. Remove inorganic crystalline substances with repetitive structures.

[0024] The spectral recognition method based on vector database, wherein the step S1 further comprises:

[0025] S15, generating a simulated XRD for the single chemical substance selected according to the Bragg diffraction formula to obtain a simulated XRD data set for the single chemical substance;

[0026] S16, for a simulated XRD data set of a single chemical substance, traverse all mixture combinations that can be obtained from the chemical substance set, perform XRD superposition on all possible multi-mixture combinations using different mass fractions, and obtain a simulated XRD data set of the mixture;

[0027] S17, extracting features from the superimposed mixture XRD data set, and performing an embedding process, by which the mixture simulation XRD is converted into a vector of a given length to form the vector database.

[0028] The spectral recognition method based on the vector database, wherein, before the step S17, further includes the following processing:

[0029] S161. Apply data enhancement technology to the simulated XRD data, use different half-peak widths to broaden the diffraction peaks formed by the simulation, and add random background noise that conforms to the Gaussian distribution.

[0030] Any of the above-mentioned spectral recognition methods based on vector database, wherein the step S2 further comprises:

[0031] S21, processing the XRD of the mixture to be detected according to the same simulation processing process as the vector database to match the vector in the vector database.

[0032] The spectral recognition method based on vector database, wherein the step S21 further comprises:

[0033] S211, performing an average half-peak width broadening of the XRD spectrum peak of the mixture to be detected to embed and generate an XRD embedding vector to be detected, and matching the simulation data embedding vector corresponding to the half-peak width in the vector database;

[0034] S212, retrieve at least one group of most similar XRD in the vector database, and directly give which compounds the XRD to be tested is composed of, the preset mass fraction, and the offset value of each main diffraction peak in the XRD to be tested compared with the simulated XRD;

[0035] S213. Post-process the search results to provide a reasonable estimate of the mass fraction of each component.

[0036] A system for implementing any of the above-mentioned spectral recognition methods based on a vector database, comprising the following modules:

[0037] A data preprocessing module is used to screen substances of a specified chemical system from a known database, including all simple substances or compounds composed of specified chemical elements, and generate a vector database based on the simulated XRD superposition of different components;

[0038] A module for detecting the XRD of the mixture to be detected is used to search the vector database according to the XRD of the mixture to be detected, determine the closest matching mixture combination ratio, and determine the material parameters corresponding to the XRD of the mixture to be detected.

[0039] The present invention provides a spectral recognition method and system based on a vector database, which has the beneficial effect of solving the problem that deep learning methods are difficult to expand the research system and are time-consuming to train. By using mature vector database technology and relying on a large database of known inorganic crystal materials, the crystal materials are classified according to the chemical systems widely studied in the field, a large number of simulated XRDs are generated and converted into vectors and stored in the vector database, and the XRD to be analyzed is searched through the vector similarity algorithm integrated in the vector database to find a group of vectors with the most similar shapes, which correspond to a series of possible components respectively, and simple post-processing is performed to provide preliminary analysis results. The present invention provides a complete vector database construction workflow.

[0040] The present invention further provides a spectral identification method and system based on a vector database, which can directly perform vector similarity comparison from massive mixture XRDs, greatly simplifying the post-processing analysis process, eliminating tedious analysis and a large amount of manual intervention, and can provide a more accurate concentration estimate of each component in the mixture. Compared with traditional JADE-assisted analysis and automated analysis based on deep learning, the present invention provides a better XRD automated analysis solution, and is easy to expand to a variety of chemical composition systems, and even to other spectral analyses commonly used in materials science, such as infrared absorption spectroscopy. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 The figure is a flowchart of the overall workflow of a preferred embodiment of the spectral recognition method and system based on a vector database of the present invention.

[0042] Figure 2 A flowchart of database creation for a preferred embodiment of the spectral recognition method and system based on a vector database of the present invention.

[0043] Figure 3 This is an example diagram of the embedding generation program of the preferred embodiment of the spectral recognition method and system based on the vector database of the present invention.

[0044] Figure 4 The figure is a flowchart of the retrieval work of a preferred embodiment of the spectral recognition method and system based on a vector database of the present invention.

[0045] Figure 5 This is an example diagram of quality score estimation of a preferred embodiment of the spectral recognition method and system based on a vector database of the present invention. DETAILED DESCRIPTION

[0046] The preferred embodiments of the present invention are described in detail below.

[0047] like Figure 1 As shown, in a preferred embodiment of a spectral identification method and system based on a vector database provided by the present invention, the overall workflow defined includes two parts, one part is the workflow for creating a vector database, and the other part is the workflow for querying the vector database and performing analysis. In the process of generating the vector database, diffraction spectra of materials of different proportions are screened from existing known databases, such as an inorganic crystal database, for example, 60% Cu2O, 20% MgO, and 20% TiO2, and the diffraction spectra are mixed to form a mixed diffraction spectrum, which is then broadened and noisy to generate an embedding, which is inserted into the database to form a vector database. Parameters such as different compound compositions and grain size are repeatedly adjusted to perform mixed calculations of the above-mentioned material diffraction spectra.

[0048] In actual inspection and measurement, after obtaining the XRD of the experimental material of unknown component, the peak width is checked and an embedding is generated. A search is performed in the vector database to find the closest and matching one or more diffraction spectra, and after a simple analysis, further analysis results such as the diffraction peak offset data, grain size, and composition concentration can be obtained.

[0049] The database creation process of the preferred embodiment of the method and system of the present invention establishes a complete workflow, which can quickly establish a vector database for a specified chemical system. Figure 2 As shown, the workflow shown first screens compounds of a specified chemical system from a pre-established inorganic crystal database, based on chemical composition, such as the Ca-Ti-O system, which includes all single substances or compounds composed of these three chemical elements.

[0050] Then, according to the stability information contained in the inorganic crystal database, according to the energy stability screening, the compounds that can be stably synthesized are screened out (the decomposition energy Energy Above Hull in the thermodynamic phase diagram is required to be less than 0.5eV / atom), and the compounds with repeated structures are removed; then, for the screened compounds, according to the Bragg diffraction formula, a simulated XRD is generated, specifically but not limited to using the diffraction analysis module provided by the open source software package Pymatgen to generate a simulated diffraction spectrum, which has integrated the relevant implementation methods of the diffraction formula and can be called through the software interface, but Pymatgen can only perform single-phase data processing; similar to the software function of Pymatgen, in the preferred embodiment of the present invention, the processing software is set to mix and superimpose according to the composition ratio of the screened compounds and the diffraction spectrum data of different components in proportion, and the mixed XRD is calculated after the component ratio is adjusted. After mixing, the corresponding diffraction peak offset, grain size, composition concentration and other parameters can be obtained, and a vector database is formed.

[0051] After obtaining the simulated XRD of many single compounds from the known inorganic crystal database, in the preferred embodiment of the present invention, multiple compounds are mixed and calculated in advance according to different mass fractions through calculation processing to generate all combinations of multiple substances. Specifically, all binary mixtures and ternary mixtures, or more complex mixtures (but the corresponding calculation amount will increase significantly) can be generated. Different mass fractions are assigned to different components of each element or compound in each mixture, and their XRDs are superimposed respectively. Different mass fractions are assigned to each substance in each combination, and their XRDs are superimposed to obtain the XRD of the simulated mixture. In actual application scenarios, the mixture to be detected may be composed of elements or compounds in various proportions. Therefore, when forming a vector database, component simulation calculations of various proportions of elements or compounds are also required.

[0052] For example, to generate XRD of a binary mixture containing two compounds, we first need to traverse all binary combinations. If there are 20 compounds, there are 190 combinations. Then, for each combination, different mass fractions are used for XRD superposition. For example, component 1 can be 0%, 10%, 20% to 100%, and the corresponding component 2 is from 100%, 90%, 80% to 0%. In this way, each binary combination has ten sets of mass fractions, so the total database entries are 1900. The reason why different mass fractions can be used to superimpose the simulated diffraction spectra is that the diffraction peak intensity in XRD is proportional to the mass fraction of the component.

[0053] The superimposed mixture XRD also needs to be feature extracted before it can be inserted into the database. The method and system of the preferred embodiment of the present invention define this process as embedding, which is to convert the XRD of the mixture into a feature vector of a given length. These feature vectors contain a large amount of information in the corresponding XRD, which is sufficient to identify mixtures of different components and particle sizes.

[0054] The embedding process is shown in Figure 3 As shown. The simulated XRD generated in an example of the present invention is a vector with a length of 800. Before the embedding is generated, in order to ensure that the data is more consistent with the experimental situation, the simulated XRD data is also processed by applying data enhancement technology in a preferred embodiment of the present invention: Figure 3 As shown in (a), different half-peak widths are applied in advance to perform Gaussian broadening on the diffraction peaks formed by the simulation, and random background noise that conforms to the Gaussian distribution is added to meet the noise influence of normal situations. The Gaussian peak function used for broadening is shown below, where X is the function argument, f(x) has the form of a Gaussian function, P is the peak intensity, e is the base of the natural logarithm, and the maximum value of the random background noise in the example does not exceed 5% of the intensity of the strongest diffraction peak.

[0055] .

[0056] In the preferred embodiment of the method and system of the present invention, two types of embedding are also proposed. Figure 3 As shown in (b), one embedding method is to use a simple convolution technique to reduce the dimension of the original XRD vector, move along the vector through a sliding window of a given length, collect the distribution of the diffraction intensity in the sliding window, and use the mean, variance, and skewness statistics to measure the distribution of the diffraction intensity in the window. In this way, the convolution vector is obtained. Unlike the convolutional neural network in deep learning, this technology mainly uses statistics to extract the statistical distribution characteristics of the diffraction intensity, and does not require a training step to determine the convolution kernel.

[0057] Furthermore, in a preferred embodiment of the present invention, an average pooling technique is applied to the vectors reduced in dimension by convolution, and the values ​​in a window of a given size are averaged to further reduce the dimension. The use of reduced-dimensional vectors can improve the processing speed and ensure the accuracy and effectiveness of the processing.

[0058] Another example of embedding is to extract the top eight diffraction peaks in the original vector of XRD, record the peak positions and peak intensities of these diffraction peaks as an embedded vector with a length of 8, such as Figure 3 (b) is shown. Note that the first eight are selected for example only. According to the actual computing power and efficiency balance, more or fewer dimensions can be set to adapt to different situation requirements and perform embedding processing that meets actual needs.

[0059] In a preferred embodiment of the spectral recognition method and system based on a vector database of the present invention, the retrieval process of the vector database is as follows: Figure 4 As shown, before searching, for the XRD to be analyzed, the program of the method and system embodiment of the present invention will first detect the average half-peak width of the peak in the XRD to be analyzed, and then Figure 3 The method is to generate an embedded vector for the XRD to be tested to match the embedded vector of the simulated data corresponding to the half-peak width stored in the database. Through the search, a group of the most similar XRDs in the vector database can be directly obtained; in another example, multiple groups of relatively similar XRDs can also be output and sorted according to the degree of similarity. The number of groups output can be controlled as much as possible to facilitate subsequent manual identification, and the similarity of each group of XRDs can be clearly determined and displayed, so that subsequent manual further judgment can be made to determine the validity of the simulated data XRD.

[0060] In the preferred embodiment of the method and system of the present invention, by comparing with the aforementioned vector database search, it is directly given which compounds these XRDs are composed of, what is the preset mass fraction of each compound, and the offset values ​​of each main diffraction peak in the XRD to be measured compared with the simulated XRD. The values ​​of these peak offsets often contain information on lattice distortion.

[0061] like Figure 4 As shown in the figure, it is illustrated that the XRD to be tested is a sample of experimental sintered Cu-Mg-Ti-O. The retrieval program of the embodiment of the method and system of the present invention directly gives the first six most likely results by comparing the similarity of each vector in the vector database. The two most similar XRDs contain 60% Cu2O, 20% TiO2, 20% MgCuO2, and 60% Cu2O, 20% TiO2 and 20% MgO, respectively. In the result comparison on the right, in each group of XRD comparisons, the top graph represents the simulated XRD of the mixture, the middle graph represents the simulated XRD after the average peak width of the XRD to be tested is widened, and the bottom one is the experimental XRD to be tested. The offset value of the main diffraction peak is marked in small black letters below the middle graph.

[0062] In a preferred embodiment of the spectral recognition method and system based on a vector database described in the present invention, after the search results are given, the program will continue to perform simple post-processing to give a reasonable estimate of the mass fraction of each component. The cosine similarity between the sample to be tested and the vector of a given concentration calculated by the vector database can be used to make a more accurate mass fraction estimate.

[0063] The entries retrieved from the database already contain a rough mass score of each component. Since the mass score of the component mainly affects the intensity of the diffraction peak, and the intensity of the diffraction peak is reflected in the eight-strong peak embedding vector of the aforementioned example, the cosine similarity between the eight-strong peak embedding vector of the sample to be tested and the eight-strong peak embedding vector of the database entry with known mass score should be correlated with the difference between the mass score of each component in the sample to be tested and the mass score of the known database entry.

[0064] ,

[0065] Among them, taking the ternary mixture as an example, d i It is the cosine similarity of the eight peaks embedded in the XRD of samples with mass fractions x, y, z and a, b, c, respectively. The cosine similarity has a functional relationship f with the difference in mass fractions of the components.

[0066] This correlation can be simply expressed as a function f, where the left side of the equation is the cosine similarity and the right side is the difference in mass fractions of the components (taking a ternary mixture as an example). The cosine similarity used here is a comprehensive similarity, which is composed of Figure 3 (c) shows the peak intensity embedding of the eight strongest peaks and the peak position embedding of the eight strongest peaks in a weighted manner. The peak intensity has a higher weight.

[0067] ,

[0068] Among them, d peak yes Figure 3 (c) Cosine similarity calculated based on the peak intensity embedding shown in (d) theta It is the cosine similarity calculated based on the embedding of the peak position. For the calculation of mass score, when the components are the same, the peak intensity has a greater weight than the peak position.

[0069] In the embodiment of the present invention, it is not necessary to find the specific form of the function f. For a mixture of a given component, all mass fraction combinations can be traversed to generate a superposition spectrum and obtain eight peaks, and then the cosine similarity of all these combinations with a specified mass fraction combination can be calculated. In this way, the present invention can draw a curve of the function f and directly obtain the mass fraction of the sample to be tested from the curve.

[0070] like Figure 5 The figure shows an example of a Cu-Mg-Ti-O ternary mixture system in a preferred embodiment of the spectral identification method and system based on a vector database of the present invention. One of the closest possible compositions directly retrieved from the database is 60% cuprous oxide, 20% titanium dioxide, and 20% MgCuO2. The exemplary software of the present invention calculates all combinations of mass fractions of the three components, accurate to 1%, and the cosine similarity of these combinations with the combination of 6:2:2, and lists Figure 5 The table in the figure is the curve of function f. The cosine similarity of the sample to be tested and the combination of 6:2:2 is calculated, and the three quality scores closest to the cosine similarity can be found by looking up the table, thus obtaining a more accurate cosine similarity estimation result.

[0071] In a preferred embodiment of the spectral identification method and system based on the vector database of the present invention, a software implementation method and system are implemented, which realizes that in the preprocessing process, a single substance and multiple compounds are set according to the known inorganic crystal database data according to the experimental purpose, such as the Cu-Mg-Ti-O material spectrum, mixed according to different mass fractions, and their XRDs are superimposed to obtain the simulated XRD of various mixtures, thereby forming a vector database. For example, to generate a simulated XRD of a binary mixture containing two compounds, first traverse all combinations of two elements or compounds. For example, when there are 20 compounds, the total number of combinations is C2 20 According to different data processing capabilities and actual experimental requirements, the more components there are in the mixture, the higher the amount of data processing required.

[0072] In a preferred embodiment of the vector database-based spectroscopy identification method and system of the present invention, different mass fractions are used for each combination to simulate XRD superposition.

[0073] like Figure 4 The figure shows a preferred embodiment of the spectral identification method and system based on the vector database of the present invention. The XRD to be tested is a sample of the experimental sintering Cu-Mg-Ti-O series. During the search process, the software program of the present invention directly gives the first six most likely results by comparing the similarity of each vector in the vector database. The two most similar simulated XRDs contain 60% Cu2O, 20% TiO2, 20% MgCuO2, and 60% Cu2O, 20% TiO2 and 20% MgO respectively. Figure 4 In the result comparison on the right, the top graph shows the simulated XRD of the mixture, the middle graph shows the simulated XRD after broadening according to the average peak width of the XRD to be tested, and the lower graph shows the XRD to be tested, where the offset value of the main diffraction peak is marked in small black letters below the middle graph. In these two and other four mixture search results, manual further judgment is required to determine the rationality of the test results of the XRD to be tested.

[0074] like Figure 5 The figure shows a preferred embodiment of the spectral identification method and system based on the vector database of the present invention. The search result belongs to the Cu-Mg-Ti-O system. One of the possible compositions retrieved by the software program is 60% cuprous oxide, 20% titanium dioxide, and 20% MgCuO2. The software program will traverse all the XRDs of the mixture of these three compounds by mass fraction, calculate the function curve of mass fraction and cosine similarity, and directly look up the table from the curve to obtain the estimated values ​​of the mass fractions of the three components in the XRD to be tested, which are 66%, 28% and 6% respectively.

[0075] In a preferred embodiment of the vector database-based spectral identification method and system described in the present invention, mature vector database technology is used, relying on a large-scale inorganic crystal material database, and the crystal materials are classified according to the chemical system widely studied in the field. A large number of simulated XRDs are generated and converted into vectors and stored in the vector database. Through the vector similarity algorithm integrated in the vector database, the vector database is searched according to the XRD to be tested to find at least one group of the most similar vectors, which correspond to a series of possible components respectively, and simple post-processing can be performed to provide preliminary analysis results.

[0076] The preferred embodiment of the present invention directly performs vector similarity comparison from the simulated XRD of a large number of mixtures, fully utilizes the computing and processing capabilities of existing hardware equipment to perform data preprocessing to form a vector database of simulated XRD, and performs retrieval and matching comparison based on the XRD to be tested, which greatly simplifies the post-processing analysis process, eliminates tedious analysis and a large amount of manual intervention, and provides a more accurate concentration estimate of each component in the mixture.

[0077] The key difference between the preferred embodiments of the spectral identification method and system based on the vector database of the present invention and the prior art is that when analyzing the XRD of the mixture to be tested, the tedious manual analysis workload or the complicated discrimination algorithm process is eliminated. Instead, a pre-formed vector database is used to directly obtain the closest component and mass fraction corresponding to the XRD of the mixture through vector comparison retrieval and matching. The speed is very fast and the efficiency is extremely high. It can also be easily expanded to a variety of chemical composition systems, and even to other spectral analyses commonly used in materials science, such as infrared absorption spectroscopy.

[0078] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all these improvements and changes should fall within the scope of protection of the appended claims of the present invention.

Claims

1. A spectral recognition method based on vector database, It is characterized in that The following steps are involved: S1. Screening substances of a specified chemical system from a known database, including all simple substances or compounds composed of specified chemical elements, superimposing the simulated XRDs of different simple substances or compounds according to different mass fractions, and generating a vector database of the specified chemical system; S2, searching the vector database according to the XRD of the mixture to be tested, determining the closest matching mixture combination ratio, and determining the material parameters corresponding to the XRD of the mixture to be tested; Wherein, the step S1 further includes: S15, generating a simulated XRD for the single chemical substance selected according to the Bragg diffraction formula to obtain a simulated XRD data set for the single chemical substance; S16, for a simulated XRD data set of a single chemical substance, traverse all mixture combinations that can be obtained from the simulated XRD data set of the single chemical substance, perform XRD superposition on all possible multi-component mixture combinations using different mass fractions, and obtain a simulated XRD data set of the mixture; S17, extracting features from the superimposed mixture XRD data set, and performing an embedding process, by which the mixture simulation XRD is converted into a vector of a given length to form the vector database.

2. The spectral recognition method based on the vector database according to claim 1, It is characterized in that The known database is an inorganic crystal database.

3. The spectral recognition method based on vector database according to claim 2, It is characterized in that The screening process in step S1 includes: S12. Based on the material stability information contained in the inorganic crystal database, a subset of chemical substances that can be stably synthesized is screened out.

4. The spectral recognition method based on vector database according to claim 3, It is characterized in that The screening process in step S1 includes: S13. Screening thermodynamically stable chemical substances in the chemical substance subset according to the substances included in the inorganic crystal database, wherein a thermodynamic stability criterion of the chemical substance is that the decomposition energy in the thermodynamic phase diagram is less than 0.5 eV / atom.

5. The spectral recognition method based on vector database according to claim 4, It is characterized in that After step S12 and / or step S13, the following steps are also included: S14. Remove inorganic crystalline substances with repetitive structures.

6. The spectral recognition method based on vector database according to claim 5, It is characterized in that Before step S17, the following processing is also included: S161. Apply data enhancement technology to the simulated XRD data, use different half-peak widths to broaden the diffraction peaks formed by the simulation, and add random background noise that conforms to the Gaussian distribution.

7. The spectral recognition method based on a vector database according to any one of claims 1 to 6, It is characterized in that The step S2 further comprises: S21, processing the XRD of the mixture to be detected according to the same simulation processing process as the vector database to match the vector in the vector database.

8. The spectral recognition method based on vector database according to claim 7, It is characterized in that The step S21 further includes: S211, performing an average half-peak width broadening of the XRD spectrum peak of the mixture to be detected to embed and generate an XRD embedding vector to be detected, and matching the simulation data embedding vector corresponding to the half-peak width in the vector database; S212, retrieve at least one group of most similar XRD in the vector database, and directly give which compounds the XRD to be tested is composed of, the preset mass fraction, and the offset value of each main diffraction peak in the XRD to be tested compared with the simulated XRD; S213. Post-process the search results to provide a reasonable estimate of the mass fraction of each component.

9. A system for implementing the spectral recognition method based on a vector database as claimed in any one of claims 1 to 8, It is characterized in that Includes the following modules: A data preprocessing module is used to screen substances of a specified chemical system from a known database, including all simple substances or compounds composed of specified chemical elements, and generate a vector database based on the simulated XRD superposition of different components; A module for detecting the XRD of the mixture to be detected is used to search the vector database according to the XRD of the mixture to be detected, determine the closest matching mixture combination ratio, and determine the material parameters corresponding to the XRD of the mixture to be detected.

Images