Raman spectrum analysis method for rapidly identifying coating components

Abstract

The invention is suitable for the technical field of surface analysis and material characterization, and provides a Raman spectrum analysis method for rapidly identifying coating components, which comprises the following steps: step 1, obtaining a coating sample; 2, measuring the Raman spectrum of the coating; 3, element distribution of the coating is measured; step 4, determining components of the coating and preparing related standard substances; 5, measuring the Raman spectrum of a related standard substance; step 6, preprocessing the Raman spectrum of the standard substance and constructing a Raman spectrum database; 7, training the model; 8, rapidly analyzing the coating; according to the Raman spectrum analysis method for rapidly identifying the coating components, Raman spectrum data are rapidly analyzed and modeled by adopting algorithms such as principal component analysis-support vector machine and partial minimum discriminant analysis, and collected coating Raman spectrums are imported into a trained optimal model; and rapid and accurate identification of the coating components is realized.

Description

Technical Field

[0001] The invention belongs to the technical field of surface analysis and material characterization, and in particular relates to a Raman spectroscopy analysis method for rapidly identifying coating components. Background Art

[0002] In recent years, with the continuous development and application of new material technologies, functional coatings have played an increasingly important role in the fields of industry, aerospace, and electronic devices. In many occasions, in order to meet the requirements of wear resistance, corrosion resistance, high temperature resistance or other specific uses, various coating systems such as composite coatings and functional coatings are often prepared on the surface of the substrate material. Since different elements, compounds or phase compositions play a key role in the overall performance of the coating, how to accurately and efficiently obtain coating component information has become an important topic in the research and application of related industries.

[0003] The existing analysis and detection methods for coating components mainly include morphology observation, energy spectrum analysis, X-ray diffraction, infrared spectroscopy, etc. Although these methods can provide qualitative and semi-quantitative information on the elements or chemical structure in the coating to a certain extent, they often require more complicated sample preparation and operation processes, or require longer testing and analysis time. In addition, some conventional methods may damage the sample during the test process, or fail to accurately measure some micro-segments.

[0004] Raman spectroscopy has been increasingly widely used in component analysis in the field of materials due to its advantages such as being non-destructive, rapid, and insensitive to moisture interference. However, existing technologies still have shortcomings when applying Raman spectroscopy to coating detection. Conventional Raman spectroscopy detection can only provide qualitative or quasi-quantitative information, and how to accurately identify multi-component coatings in combination with other characterization methods remains a difficulty; the Raman signals of different coatings are relatively complex and are easily affected by factors such as background noise or baseline drift, and there is a lack of effective data preprocessing methods and rapid analysis models; there is a lack of rapid cross-sectional detection solutions for multi-layer structures such as composite coatings and functional coatings, especially in terms of high-throughput and automated identification.

[0005] Therefore, there is an urgent need to develop a method that can combine multiple characterization methods and utilize the advantages of Raman spectroscopy non-destructive testing to achieve rapid and accurate identification of coating components. Aiming at the difficulties and shortcomings of coating component analysis in practical applications, the present invention proposes a rapid identification method based on Raman spectroscopy. By establishing a standard database and introducing advanced data processing and model training methods, efficient identification and evaluation of coating components can be achieved. This technology is of great significance to accelerate the coating characterization process and improve the efficiency of material research and development. Summary of the invention

[0006] In view of the shortcomings of the prior art, an object of the embodiments of the present invention is to provide a Raman spectroscopy analysis method for quickly identifying coating components, so as to solve the problems in the above-mentioned background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A Raman spectroscopy analysis method for quickly identifying coating components comprises the following steps:

[0009] Step 1: Obtain coating samples: Clean or pre-treat the surface of the coating samples to be tested to ensure the accuracy of Raman testing;

[0010] Step 2: measuring the Raman spectrum of the coating: using a Raman spectrometer to perform a spectrum test on the coating sample and record the corresponding Raman spectrum data;

[0011] Step 3: Measure the element distribution of the coating: Characterize the element distribution of the coating sample by energy dispersive X-ray spectroscopy (EDS) to determine the content and distribution area of ​​each element in the coating;

[0012] Step 4: Identify coating components and prepare relevant standards: Based on the EDS test results, Raman spectrum results and literature information of the coating samples, screen the types of compounds or materials that may be contained in the coating, determine the main components of the coating, and prepare corresponding standard samples for these components;

[0013] Step 5, measuring the Raman spectra of relevant standard substances: performing Raman spectrum tests on the standard substances prepared in step 4 respectively, and obtaining Raman spectrum data corresponding to each standard substance;

[0014] Step 6: Preprocessing the Raman spectrum of the standard and building a Raman spectrum database: preprocessing the Raman spectrum data of the standard obtained in step 5, such as cosmic ray elimination, spectrum denoising, baseline correction, smoothing and standardization, and building and storing it in a Raman spectrum database;

[0015] Step 7: Training model: Rapidly analyze the coating Raman spectrum and conduct comparative analysis on its application effectiveness through principal component analysis-support vector machine and partial least squares discriminant analysis algorithm;

[0016] Step 8: Rapid analysis of coating: Input the pre-processed coating Raman spectrum into the trained model to achieve rapid identification of coating components;

[0017] As a further technical solution of the present invention, in step one, the sample to be tested is composed of a composite coating sample and a single coating sample. The samples are combined with epoxy resin, and their cross sections are ground and polished. According to actual needs, operations such as "ultrasonic cleaning" and "impurity removal" are performed, and Raman spectrum and element distribution measurements are performed on the cross sections of the coating samples.

[0018] As a further technical solution of the present invention, in step five, a surface scanning method is adopted when performing Raman spectrum test on the standard substance, and 1000 sets of Raman spectrum data are collected for each standard substance.

[0019] As a further technical solution of the present invention, in step six, cosmic ray elimination adopts a threshold-based peak replacement algorithm; baseline correction adopts a polynomial fitting algorithm combined with regularization processing; and smoothing processing adopts a Savitzky-Golay filtering method.

[0020] As a further technical solution of the present invention, in step seven, the principal component analysis-support vector machine algorithm includes two kernels: a linear kernel and an RBF kernel. The specific use steps are as follows:

[0021] Step 1: Standardize the coating Raman spectrum data set, perform dimensionality reduction on the Raman spectrum data set by principal component analysis and extract the main difference feature information, obtain the most significant feature variables principal component 1 and principal component 2, and input them into the SVM algorithm;

[0022] Step 2: By combining grid search and cross-validation, the optimal parameters of the linear kernel and the RBF kernel are determined respectively, and the parameters with the highest classification accuracy are used as the optimal parameter combination for constructing the principal component analysis-support vector machine model;

[0023] Step 3: Use the established principal component analysis-support vector machine model to verify the test set data, and then verify the classification performance of the model.

[0024] As a further technical solution of the present invention, in step seven, the Raman spectral data of the training set accounts for 80% of the total spectral data, and the Raman spectral data of the test set accounts for 20% of the total spectral data. The division ratio of the training set and the test set can be appropriately adjusted according to the volume of Raman spectral data.

[0025] As a further technical solution of the present invention, in step seven, the steps of using the principal component analysis-support vector machine model are as follows:

[0026] Step 1: Reduce the dimension of the Raman spectroscopy dataset by principal component analysis, and extract and identify the principal component information with significant differences by combining one-way ANOVA;

[0027] Step 2: Using the extracted principal component information with significant differences as the input variable of the support vector machine algorithm to generate a principal component analysis-support vector machine model recognition model;

[0028] Step 3: Use the principal component analysis-support vector machine model to classify and identify the Raman spectral characteristics of the coating components.

[0029] As a further technical solution of the present invention, in step seven, the steps of using the partial least squares discriminant analysis model are as follows:

[0030] Step 1: Standardize the coating Raman spectrum data set to obtain Raman spectrum data in a unified format;

[0031] Step 2: Construct the latent variable, select the direction with the highest correlation with the response variable as the first component, and then extract the direction with the second highest correlation as the second component;

[0032] Step 3: Based on the above components, a linear model is constructed to predict the response variable to achieve rapid classification and identification of coating components.

[0033] As a further technical solution of the present invention, in step seven, the steps of comparative analysis of the application effectiveness of the principal component analysis-support vector machine model and the partial least squares discriminant analysis model are as follows:

[0034] Step 1: Use the same test set to compare the performance differences between the principal component analysis-support vector machine model and the partial least squares discriminant analysis model in terms of accuracy, precision, recall, F1-score and other indicators;

[0035] Step 2: Comprehensively evaluate the usage cost and applicable scenarios of the two models based on aspects such as model training time, prediction time, and difficulty of hyperparameter tuning;

[0036] Step 3: Comprehensively consider various indicators and specific application requirements to determine the optimal model or model combination for rapid identification of actual coating Raman spectra.

[0037] As a further technical solution of the present invention, in step eight, the rapid analysis steps of the coating Raman spectrum are as follows:

[0038] Step 1: Perform pre-processing operations such as peak range selection, cosmic ray elimination, spectrum denoising, baseline correction, and smoothing on the collected coating Raman spectrum;

[0039] Step 2: Compare the classification performance of the principal component analysis-support vector machine model and the partial least squares discriminant analysis model. According to the model with the best performance, perform dimensionality reduction or standardization on the pre-processed Raman spectrum to meet the model import requirements;

[0040] Step 3: Use the model with the best performance to classify and identify the Raman spectral characteristics of the coating components.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] (1) The present invention uses Raman spectroscopy to detect coatings, without the need for destructive operations such as etching and dissolving the sample, which greatly reduces sample preparation and detection time;

[0043] (2) The present invention combines EDS results with Raman spectroscopy data, which can quickly identify the elements that may exist in the coating and their distribution, narrow the target range of subsequent Raman identification, and better eliminate noise and non-target signal interference, greatly improving the accuracy of the final identification;

[0044] (3) The present invention uses multiple processing methods such as cosmic ray elimination, baseline correction, and smoothing filtering to effectively remove spectral noise and baseline drift, retain key Raman feature information, and establish a Raman spectrum database of standard objects, and perform standardized preprocessing and storage, providing a high-quality data foundation for subsequent model training and fast matching;

[0045] (4) The present invention uses principal component analysis-support vector machine and partial least squares discriminant analysis algorithm to train and identify the preprocessed Raman spectroscopy data set, which can take into account both high accuracy and robustness. At the same time, by comparing the application effectiveness of multiple models, the best or combined model is selected for actual analysis, thereby improving data processing efficiency while ensuring accuracy;

[0046] (5) In the present invention, from preliminary sample preparation, standard material testing to spectral preprocessing and model training, each link is independent and process-oriented, which is convenient for batch application and upgrading. At the same time, this method can be promoted and used in different types of coatings, different industrial or scientific research environments, and can also be combined with an automated Raman test system and an intelligent control platform to achieve more advanced online real-time detection.

[0047] (6) The present invention reduces the operator's subjective judgment on the spectral peak shape and intensity, uses objective spectral data and machine learning models for analysis, avoids human errors, and can achieve rapid detection and automated data processing of large quantities of samples under the same or similar instrument conditions, reducing time and labor costs.

[0048] In order to more clearly illustrate the structural features and effects of the present invention, the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A flowchart of a Raman spectroscopy analysis method for rapidly identifying coating components provided by an embodiment of the present invention.

[0050] Figure 2 A schematic diagram of coating sample pretreatment provided in an embodiment of the present invention.

[0051] Figure 3 This is the Raman spectrum scan result of the coating provided by the embodiment of the present invention.

[0052] Figure 4 This is a Raman spectrum preprocessing process provided by an embodiment of the present invention.

[0053] Figure 5 A comparison chart of parameter selection for principal component analysis-support vector machine models based on two kernels, RBF kernel and linear kernel, provided in an embodiment of the present invention.

[0054] Figure 6 The main component scatter diagram and main component spectrum of the Raman spectral data after standardization and dimension reduction processing provided in the embodiment of the present invention.

[0055] Figure 7 The principal component analysis-support vector machine model linear kernel classification and RBF kernel classification training set and test set confusion matrix provided in the embodiment of the present invention.

[0056] Figure 8 The partial least squares discriminant analysis model provided in the embodiment of the present invention is used to analyze the main component scatter diagram and main component spectrum of the coating Raman spectrum. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0058] The specific implementation of the present invention is described in detail below in conjunction with specific embodiments.

[0059] Example 1

[0060] like Figure 1 As shown, a Raman spectroscopy analysis method for quickly identifying coating components provided by an embodiment of the present invention includes the following steps:

[0061] Step 1: Obtain coating samples: Clean or pre-treat the surface of the coating samples to be tested to ensure the accuracy of Raman testing;

[0062] Step 2: measuring the Raman spectrum of the coating: using a Raman spectrometer to perform a spectrum test on the coating sample and record the corresponding Raman spectrum data;

[0063] Step 3: Measure the element distribution of the coating: Characterize the element distribution of the coating sample by energy dispersive X-ray spectroscopy (EDS) to determine the content and distribution area of ​​each element in the coating;

[0064] Step 4: Identify coating components and prepare relevant standards: Based on the EDS test results, Raman spectrum results and literature information of the coating samples, screen the types of compounds or materials that may be contained in the coating, determine the main components of the coating, and prepare corresponding standard samples for these components;

[0065] Step 5, measuring the Raman spectra of relevant standard substances: performing Raman spectrum tests on the standard substances prepared in step 4 respectively, and obtaining Raman spectrum data corresponding to each standard substance;

[0066] Step 6: Preprocessing the Raman spectrum of the standard and building a Raman spectrum database: preprocessing the Raman spectrum data of the standard obtained in step 5, such as cosmic ray elimination, spectrum denoising, baseline correction, smoothing and standardization, and building and storing it in a Raman spectrum database;

[0067] Step 7: Training model: Rapidly analyze the coating Raman spectrum and conduct comparative analysis on its application effectiveness through principal component analysis-support vector machine and partial least squares discriminant analysis algorithm;

[0068] Step 8: Rapid analysis of coating: Input the pre-processed coating Raman spectrum into the trained model to achieve rapid identification of coating components;

[0069] In this embodiment, combined with a large number of previous studies on Raman spectra of coatings, a Raman spectroscopy analysis method of principal component analysis-support vector machine and partial least squares discriminant analysis is proposed, which is applied to the classification and identification of Raman spectroscopy data of different coating components, and the statistical spectral information contained in the coating components is mined to achieve rapid identification of coating components based on Raman spectroscopy detection technology.

[0070] like Figure 2 As shown, as a preferred embodiment of the present invention, in step 2, the sample to be tested is composed of a composite coating sample and a single coating sample, the samples are combined with epoxy resin, their cross sections are ground and polished, and operations such as "ultrasonic cleaning" and "impurity removal" are performed according to actual needs, and Raman spectrum and element distribution measurements are performed on the cross section of the coating sample.

[0071] like Figure 3 As shown, as a preferred embodiment of the present invention, in step five, a surface scanning method is used when performing Raman spectrum testing on the standard object, and 1000 sets of Raman spectrum data are collected for each standard object.

[0072] like Figure 4 As shown, as a preferred embodiment of the present invention, in step six, cosmic ray elimination adopts a threshold-based peak replacement algorithm; baseline correction adopts a polynomial fitting algorithm combined with regularization processing; and smoothing processing adopts Savitzky-Golay filtering method.

[0073] Example 2

[0074] As a preferred embodiment of the present invention, in step seven, the principal component analysis-support vector machine algorithm includes two kernels: a linear kernel and an RBF kernel. The specific use steps are as follows:

[0075] Step 1: Standardize the coating Raman spectrum data set, perform dimensionality reduction on the Raman spectrum data set by principal component analysis and extract the main difference feature information, obtain the most significant feature variables principal component 1 and principal component 2, and input them into the SVM algorithm;

[0076] Step 2: By combining grid search and cross-validation, the optimal parameters of the linear kernel and RBF kernel are determined respectively, and the parameters with the highest classification accuracy are used as the best parameter combination for constructing the principal component analysis-support vector machine model;

[0077] Step 3: Use the established principal component analysis-support vector machine model to verify the test set data, and then verify the classification performance of the model.

[0078] The number of training set samples is set to account for 80% of the total spectral data set. For the RBF kernel, when the optimal parameters C = 4.64 and γ = 0.028, the classification accuracy of the model reaches 98.75%. For the PCA-SVM model with a linear kernel, only the regularization parameter C needs to be optimized. When C = 2.15, the highest accuracy obtained by cross-validation is 99.29%, and the accuracy of the test set is 97.5%.

[0079] exist Figure 5 In (A), when the C value is low (close to 10 -3 ), the accuracy is low, which indicates that the model may be more tolerant to misclassification under these parameters, and there is not enough penalty to encourage the model to fit the data better; as the C value increases, the accuracy improves significantly, especially when the C value increases from 10 -3 to 10 0 This indicates that increasing the penalty for misclassification helps the model to classify data more accurately. When the C value is 2.15 and 10, the model accuracy reaches a peak of 0.9929, which is shown at the two red points. This is the optimal regularization degree of the model, which avoids overfitting and fully fits the data, indicating that the parameter settings of these two points have achieved the optimal balance on the given data set. When the C value exceeds 10 1 This may indicate that further increasing the C value will not improve the accuracy, and that too high a C value will lead to overfitting of the model.

[0080] exist Figure 5In (B), the best parameter combination is C = 4.64, γ = 0.028 through model calculation, with a cross-validation accuracy of 98.12% and a test set accuracy of 98.75%. When log10(C) is between 0.5 and 1.0 and log10(γ) is between -2.0 and -1.0, the model performs best, and the model stability and robustness are good within these parameter ranges, indicating that choosing the right parameters has a significant impact on the performance of the PCA-SVM model.

[0081] Example 3

[0082] As a preferred embodiment of the present invention, in step seven, the steps of using the principal component analysis-support vector machine model are as follows:

[0083] Step 1: Reduce the dimension of the Raman spectroscopy dataset by principal component analysis, and extract and identify the principal component information with significant differences by combining one-way ANOVA;

[0084] Step 2: Using the extracted principal component information with significant differences as the input variable of the support vector machine algorithm to generate a principal component analysis-support vector machine model recognition model;

[0085] Step 3: Use the principal component analysis-support vector machine model to classify and identify the Raman spectral characteristics of the coating components.

[0086] exist Figure 6 In (A), principal component 1 and principal component 2 describe 93.02% of the spectral information, so the first two principal components are selected as observations. From the principal component scatter plot, it can be seen that the samples of the FeS2 and FeO groups are mainly distributed on the negative semi-axis of principal component 1, while the Fe3O4 and Fe2O3 groups are concentrated on the positive semi-axis of principal component 1. The positive semi-axis of principal component 2 is all the FeS2 group, while the FeO and Fe2O3 groups are located on the negative semi-axis of principal component 2, and the Fe3O4 group is distributed on both sides of principal component 2. The spectral variables of principal component 1 account for 82.50% of the spectral variance. The FeO, FeS2 groups and the Fe2O3, Fe3O4 groups can be clearly distinguished based on the positive and negative semi-axis of principal component 1. Figure 6 The characteristic peaks in (B) mainly explain the differences in the spectral characteristics of FeO, FeS2, Fe2O3, and Fe3O4. The spectral variables of principal component 2 account for 10.52% of the variance and can further distinguish the FeO group from the FeS2 group.

[0087] By observation Figure 6(B) It is observed that the positive and negative characteristic changes of the main component loading match the Raman characteristic changes of the standard. The spectrum of main component 1 shows negative contributions from FeS2 (344 and 383 cm-1) and FeO (615 and 665 cm-1), and the positive contributions are mainly from Fe2O3 (221 and 289 cm-1) and Fe3O4 (289 and 554 cm-1). The difference between main component 2 and main component 1 is manifested by the negative contribution of Fe3O4 (554 cm-1) and the positive contribution of FeS2 (344 and 383 cm-1). In addition, the negative characteristic peaks of main component 1 are mainly the characteristic peaks of FeS2 and FeO, and the positive characteristic peaks show the spectral characteristic peaks of Fe2O3 and Fe3O4. The negative characteristic peaks of main component 2 come from the characteristic peaks of FeO and part of Fe3O4, while the positive characteristic peaks show the characteristic peaks of FeS2.

[0088] Figure 7 The confusion matrix of the training and test sets for the PCA-SVM model linear kernel classification and RBF kernel classification shows the high classification accuracy of the model. The linear kernel model accurately classified all category samples in the test set, indicating that it has strong generalization ability for this dataset.

[0089] Example 4

[0090] As a preferred embodiment of the present invention, in step seven, the steps of using the partial least squares discriminant analysis model are as follows:

[0091] Step 1: Standardize the coating Raman spectrum data set to obtain Raman spectrum data in a unified format;

[0092] Step 2: Construct the latent variable, select the direction with the highest correlation with the response variable as the first component, and then extract the direction with the second highest correlation as the second component;

[0093] Step 3: Based on the above components, a linear model is constructed to predict the response variable to achieve rapid classification and identification of coating components.

[0094] Figure 8 (a) shows the scatter distribution of the Raman spectroscopy dataset on latent variables 1 and 2. The FeO group is completely located on the positive semi-axis of latent variables 1 and 2, while the FeS2 group is mainly located on the negative semi-axis of latent variable 2. The distribution differences of Fe2O3 and Fe3O4 are not obvious, and they are mainly located on the negative semi-axis of latent variable 1 and the positive semi-axis of latent variable 2.

[0095] exist Figure 8(b), the spectral characteristics of latent variable 1 show the positive contribution of FeS2 and the negative contributions of Fe2O3 and Fe3O4, while the spectral characteristics of latent variable 2 show the negative contribution of FeS2, and the positive contributions come from Fe2O3, Fe3O4 and FeO. Figure 7 The positive contribution of latent variable 1 mainly comes from FeS2 and FeO groups, and the negative contribution comes from Fe2O3 and Fe3O4 groups; the positive contribution of latent variable 2 mainly comes from Fe2O3, Fe3O4 and FeO, and the negative contribution comes from FeS2.

[0096] Example 5

[0097] As a preferred embodiment of the present invention, in step seven, the steps of comparative analysis of the application effectiveness of the principal component analysis-support vector machine model and the partial least squares discriminant analysis model are as follows:

[0098] Step 1: Use the same test set to compare the performance differences between the principal component analysis-support vector machine model and the partial least squares discriminant analysis model in terms of accuracy, precision, recall, F1-score and other indicators;

[0099] Step 2: Comprehensively evaluate the usage cost and applicable scenarios of the two models based on aspects such as model training time, prediction time, and difficulty of hyperparameter tuning;

[0100] Step 3: Comprehensively consider various indicators and specific application requirements to determine the optimal model or model combination for rapid identification of actual coating Raman spectra.

[0101] In the process of studying coatings containing various iron compounds, the sensitivity, specificity, accuracy and overall classification accuracy of the coating components Fe3O4, FeO, FeS2 and Fe2O3 in the test set of the principal component analysis-support vector machine model linear check were all 100%. The sensitivity of the coating components Fe3O4, FeO, FeS2 and Fe2O3 in the test set of the principal component analysis-support vector machine model RBF check were 95%, 100%, 100%, 90%, specificity were 96.67%, 100%, 100%, 98.33%, accuracy were 96.25%, 100%, 100%, 96.25%, and the overall classification accuracy was 98.75%.

[0102] The sensitivity of the partial least squares discriminant analysis model for Fe3O4, FeO, FeS2 and Fe2O3 in the test set were 76.67%, 80.00%, 96.67% and 100%, respectively; the specificity was 93.33%, 98.89%, 100% and 92.22%, the accuracy was 89.17%, 94.17%, 99.17% and 94.17%, respectively; and the overall classification accuracy was 88.33%.

[0103] The principal component analysis-support vector machine linear kernel model performed best in classification accuracy, sensitivity, specificity and overall accuracy. In contrast, the partial least squares discriminant analysis model had the worst classification performance, and its overall accuracy was 11.67% lower than that of the PCA-SVM linear kernel model. Therefore, the principal component analysis-support vector machine model linear kernel model was selected for the rapid classification and identification of Fe3O4, FeO, FeS2 and Fe2O3 components in the coating.

[0104] In the above steps, the rapid analysis steps of coating Raman spectrum are as follows:

[0105] Step 1: Perform pre-processing operations such as peak range selection, cosmic ray elimination, spectrum denoising, baseline correction, and smoothing on the collected coating Raman spectrum;

[0106] Step 2: Compare the classification performance of the principal component analysis-support vector machine model and the partial least squares discriminant analysis model. According to the model with the best performance, perform dimensionality reduction or standardization on the pre-processed Raman spectrum to meet the model import requirements;

[0107] Step 3: Use the model with the best performance to classify and identify the Raman spectral characteristics of the coating components.

[0108] 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 and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A Raman spectroscopy analysis method for rapid identification of coating components, characterized in that: The steps include: Step 1: Obtain coating samples: Clean or pre-treat the surface of the coating samples to be tested to ensure the accuracy of Raman testing; Step 2: measuring the Raman spectrum of the coating: using a Raman spectrometer to perform a spectrum test on the coating sample and record the corresponding Raman spectrum data; Step 3: Measure the element distribution of the coating: Characterize the element distribution of the coating sample by energy dispersive X-ray spectroscopy (EDS) to determine the content and distribution area of ​​each element in the coating; Step 4: Identify coating components and prepare relevant standards: Based on the EDS test results, Raman spectrum results and literature information of the coating samples, screen the types of compounds or materials that may be contained in the coating, determine the main components of the coating, and prepare corresponding standard samples for these components; Step 5, measuring the Raman spectra of relevant standard substances: performing Raman spectrum tests on the standard substances prepared in step 4 respectively, and obtaining Raman spectrum data corresponding to each standard substance; Step 6: Preprocessing the Raman spectrum of the standard and building a Raman spectrum database: preprocessing the Raman spectrum data of the standard obtained in step 5, such as cosmic ray elimination, spectrum denoising, baseline correction, smoothing and standardization, and building and storing it in a Raman spectrum database; Step 7: Training model: Rapidly analyze the coating Raman spectrum and conduct comparative analysis on its application effectiveness through principal component analysis-support vector machine and partial least squares discriminant analysis algorithm; Step 8. Rapid analysis of coating: Input the pre-processed coating Raman spectrum into the trained model to achieve rapid identification of coating components.

2. A Raman spectroscopy analysis method for rapid identification of coating components according to claim 1, characterized in that: In the step 1, the sample to be tested is composed of a composite coating sample and a single coating sample. The samples are combined with epoxy resin, and their cross sections are ground and polished. According to actual needs, "ultrasonic cleaning" and "impurity removal" operations are performed, and Raman spectroscopy and element distribution measurements are performed on the cross sections of the coating samples.

3. A Raman spectroscopy analysis method for rapid identification of coating components according to claim 1, characterized in that: In the step 5, a surface scanning method is used when the Raman spectrum test is performed on the standard substance, and 1000 sets of Raman spectrum data are collected for each standard substance.

4. A Raman spectroscopy analysis method for rapid identification of coating components according to claim 1, characterized in that: In step six, cosmic ray elimination adopts a threshold-based spike replacement algorithm; baseline correction adopts a polynomial fitting algorithm combined with regularization processing; and smoothing processing adopts a Savitzky-Golay filtering method.

5. A Raman spectroscopy analysis method for rapid identification of coating components according to claim 1, characterized in that: In step 7, the principal component analysis-support vector machine algorithm includes two kernels: linear kernel and RBF kernel. The specific use steps are as follows: Step 1: Standardize the coating Raman spectrum data set, perform dimensionality reduction on the Raman spectrum data set by principal component analysis and extract the main difference feature information, obtain the most significant feature variables principal component 1 and principal component 2, and input them into the SVM algorithm; Step 2: By combining grid search and cross-validation, the optimal parameters of the linear kernel and the RBF kernel are determined respectively, and the parameters with the highest classification accuracy are used as the optimal parameter combination for constructing the principal component analysis-support vector machine model; Step 3: Use the established principal component analysis-support vector machine model to verify the test set data, and then verify the classification performance of the model.

6. A Raman spectroscopy analysis method for rapid identification of coating components according to claim 1, characterized in that: In step seven, the training set Raman spectral data accounts for 80% of the total spectral data, and the test set Raman spectral data accounts for 20% of the total spectral data. The division ratio of the training set and the test set can be appropriately adjusted according to the volume of Raman spectral data.

7. A Raman spectroscopy analysis method for rapid identification of coating components according to claim 1, characterized in that: In step 7, the steps for using the principal component analysis-support vector machine model are as follows: Step 1: Reduce the dimension of the Raman spectroscopy dataset by principal component analysis, and extract and identify the principal component information with significant differences by combining one-way ANOVA; Step 2: Using the extracted principal component information with significant differences as the input variable of the support vector machine algorithm to generate a principal component analysis-support vector machine model recognition model; Step 3: Use the principal component analysis-support vector machine model to classify and identify the Raman spectral characteristics of the coating components.

8. A Raman spectroscopy analysis method for rapid identification of coating components according to claim 1, characterized in that: In step 7, the steps for using the partial least squares discriminant analysis model are as follows: Step 1: Standardize the coating Raman spectrum data set to obtain Raman spectrum data in a unified format; Step 2: Construct the latent variable, select the direction with the highest correlation with the response variable as the first component, and then extract the direction with the second highest correlation as the second component; Step 3: Based on the above components, a linear model is constructed to predict the response variable to achieve rapid classification and identification of coating components.

9. A Raman spectroscopy analysis method for rapid identification of coating components according to claim 1, characterized in that: In step 7, the steps of comparative analysis of the application effectiveness of the principal component analysis-support vector machine model and the partial least squares discriminant analysis model are as follows: Step 1: Use the same test set to compare the performance differences between the principal component analysis-support vector machine model and the partial least squares discriminant analysis model in terms of accuracy, precision, recall, F1-score and other indicators; Step 2: Comprehensively evaluate the usage cost and applicable scenarios of the two models based on aspects such as model training time, prediction time, and difficulty of hyperparameter tuning; Step 3: Comprehensively consider various indicators and specific application requirements to determine the optimal model or model combination for rapid identification of actual coating Raman spectra.

10. A Raman spectroscopy analysis method for rapid identification of coating components according to claim 1, characterized in that: In step eight, the rapid analysis steps of the coating Raman spectrum are as follows: Step 1: Perform pre-processing operations such as peak range selection, cosmic ray elimination, spectrum denoising, baseline correction, and smoothing on the collected coating Raman spectrum; Step 2: Compare the classification performance of the principal component analysis-support vector machine model and the partial least squares discriminant analysis model. According to the model with the best performance, perform dimensionality reduction or standardization on the pre-processed Raman spectrum to meet the model import requirements; Step 3: Use the model with the best performance to classify and identify the Raman spectral characteristics of the coating components.

Images