Non-destructive testing method for flame retardant performance of non-metallic shell of electric energy metering box

Abstract

This invention relates to the field of material performance testing technology, specifically disclosing a non-destructive testing method for the flame retardant performance of non-metallic casings of electricity metering boxes. The method combines spectroscopic techniques with machine learning and prediction techniques. It acquires flame retardant performance data from multiple samples through flame combustion tests as labels, and uses a spectrometer for non-destructive in-situ detection to obtain spectral data of the material surface. After preprocessing and feature extraction, various machine learning algorithms are trained and optimized, ultimately selecting a Bayesian-optimized BO-XGBoost model as the prediction algorithm. This invention achieves non-destructive testing of the flame retardant performance of non-metallic casings of electricity metering boxes.

Description

Technical Field

[0001] This invention relates to the field of material performance testing technology, and more specifically, to a non-destructive testing method for the flame retardant properties of the non-metallic casing of an electricity metering box. Background Technology

[0002] Combustion performance testing of polymer materials is typically a destructive testing process. Existing combustion performance testing equipment, such as cone calorimeters, horizontal and vertical combustion test apparatuses, and various combustion testing machines, all operate on the principle of placing the test sample in a flame environment and observing phenomena such as the combustion state and flame exit extinguishing rate as the basis for judging the material's combustion performance. Because the test sample needs to be shaped appropriately for the combustion test before being placed in a flame, after the flame retardancy test, the sample is unusable regardless of whether it burns or not.

[0003] As a terminal electrical equipment directly serving users, the flame-retardant performance of the non-metallic casing of electricity metering boxes is crucial for ensuring the safety of users' lives and property. Currently, testing of the flame-retardant performance of the non-metallic casing of electricity metering boxes can only be done through sampling, making it impossible to test each box individually. This poses a certain safety hazard to electrical equipment.

[0004] According to the procurement requirements of power companies, the chemical composition of the non-metallic casing of electricity metering boxes is relatively fixed. Its main component is a composite of polycarbonate (PC) and ABS resin, with each manufacturer adding flame retardants, plasticizers, and reinforcing materials based on their own production processes. Therefore, the flame-retardant performance of the non-metallic casing mainly depends on the type and content of the added flame retardants. By measuring the relationship between the composition and content of flame retardants in the polymer composite system and their flame-retardant performance, the flame-retardant performance of the casing sample can be determined and analyzed. However, due to the large number of manufacturers of electricity metering boxes, differences in processes and raw materials, and variations in the stability of process control, even products from the same batch may have different flame-retardant performance. Summary of the Invention

[0005] To overcome the aforementioned deficiencies in the prior art, this invention provides a non-destructive testing method for the flame retardant performance of non-metallic casings of electricity metering boxes. This method combines spectral detection technology with machine learning prediction technology. Accurate flame retardant performance data is obtained by collecting samples and conducting flame combustion tests as labels. The sample spectral data is acquired using in-situ non-destructive spectral detection technology. After preprocessing and feature extraction, various machine learning algorithms are trained and optimized. Finally, a Bayesian-optimized BO-XGBoost model is selected for predictive analysis, thus achieving non-destructive testing of the flame retardant performance of non-metallic casings of electricity metering boxes and solving the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A non-destructive testing method for the flame retardant performance of non-metallic casings of energy metering boxes combines spectroscopic techniques with machine learning and prediction techniques. It utilizes in-situ spectral measurement techniques that do not damage the sample to obtain the chemical composition spectrum of the test object. Through large-scale experimental data learning and training of a machine learning model, it achieves prediction of the flame retardant performance of materials based on in-situ spectral characteristics. The method includes the following steps:

[0008] Step S1: Collect non-metallic shell samples of the power metering box, and use a portable infrared spectrometer, a portable Raman spectrometer, or an infrared / Raman spectrometer equipped with a movable fiber optic probe to perform in-situ spectroscopic detection on the collected non-metallic shell samples and acquire spectral data.

[0009] Step S2: Conduct a flame combustion test on the collected non-metallic shell sample of the power metering box;

[0010] Step S3: Preprocessing spectral data and establishing a machine learning model;

[0011] Step S4: Use Bayesian optimization to search for the hyperparameters of the XGBoost model, including the maximum tree depth (max_depth), learning rate (learning_rate), minimum split loss (gamma), and subsample ratio (subsample). Evaluate the model performance using five-fold cross-validation.

[0012] Step S5: Use the optimized BO-XGBoost model to predict and identify the flame retardant properties of the non-metallic shell material of the power metering box.

[0013] As a further aspect of the present invention, step S1 includes:

[0014] Step S11: Collect a wide range of samples of electricity metering boxes from various brands and manufacturers on the market, and obtain non-metallic shells to ensure the diversity and representativeness of the test data;

[0015] Step S12: Use clean water and anhydrous ethanol alternately to clean the sample to remove dust, oil and other contaminants from the surface, ensuring the accuracy of the flame combustion test.

[0016] As a further aspect of this invention, in step S2, the flame retardant performance test is conducted according to the national standard GB / T2408-2021. Specifically, the cleaned and dried non-metallic shell sample is cut into strips with a length of 125.0 mm ± 5.0 mm, a width of 13.0 mm ± 0.5 mm, and a thickness of < 13.0 mm, ensuring that its dimensions meet the standard requirements. The test uses horizontal and vertical combustion testing equipment, and combustion tests are conducted under specified flame height, ignition time (10 seconds or 30 seconds), and test environment (23 ± 2℃, relative humidity 50 ± 10%). During the test, the flame propagation speed after combustion, self-extinguishing time after flame removal, dripping phenomenon, and whether cotton is ignited are observed. Based on the comprehensive performance of the combustion test, the flame retardant level is divided into V0 (the highest level, indicating short burning time, no dripping, and strong self-extinguishing ability), V1 (slightly longer self-extinguishing time, but no burning dripping), V2 (burning dripping and igniting cotton), and other levels (unqualified samples). Test data serves as a key label for model training, providing a foundation for building machine learning models, while repeated experiments with different samples improve data reliability.

[0017] As a further aspect of the present invention, step S3 includes:

[0018] Step S31: The acquired spectral data are sequentially subjected to Savitzky-Golay filtering, baseline correction, and normalization.

[0019] Step S32: Extract key band features from the spectral data, including information such as peak position and intensity that are directly related to the flame retardant components and content;

[0020] Step S33: Initial model training is performed using Support Vector Machine (SVM), Inverse Neural Network, CatBoost, BO-CatBoost (Bayesian optimized CatBoost), XGBoost, and BO-XGBoost (Bayesian optimized XGBoost), and the BO-XGBoost model is selected as the final prediction model.

[0021] As a further aspect of the present invention, step S5 includes:

[0022] Step S51: Use a portable infrared or Raman spectrometer, or an infrared or Raman spectrometer with a movable fiber optic probe, to test the non-metallic casing of the power metering box in situ to obtain the spectral data of the test sample.

[0023] Step S52: Use the optimized BO-XGBoost model to read in the spectral data of the test sample, predict the flame retardant performance of the test sample, and output the result as "qualified sample" or "unqualified sample".

[0024] In step S53, the sample predicted by the optimized BO-XGBoost model is further verified by the combustion test in step S13.

[0025] As a further aspect of the present invention, in step S31, Savitzky-Golay filtering is used to smooth the acquired spectral data. By setting a window width of 7 points and a second-order polynomial order, random noise in the spectral signal is removed while retaining the shape of key spectral characteristic peaks. A least squares baseline fitting algorithm is used for baseline correction to eliminate spectral baseline drift caused by test environment or instrument errors, ensuring that the spectral data is based on an accurate zero point. Max-min normalization is used to normalize the spectral data, mapping the spectral intensity values ​​to the range of [0,1] to eliminate the influence of external factors such as sample thickness and light source intensity on the characteristic intensity.

[0026] As a further aspect of the present invention, in step S32, key waveband features are extracted from the original spectrum. Since flame retardants typically have unique characteristic absorption peaks in infrared and Raman spectra, key wavebands related to these chemical components and their contents need to be extracted from the spectrum. In infrared spectra, different chemical groups will produce absorption peaks in specific wavenumber ranges. For example, phosphate groups typically have significant characteristic absorption peaks in the 1200-1300 cm⁻¹ range, bromides will show characteristic peaks in the 500-600 cm⁻¹ range, and chlorides will show characteristic peaks in the 600-800 cm⁻¹ range. -1 The spectrum contains relatively weak characteristic absorption peaks. By analyzing these peaks in the sample spectrum, information related to the type and content of flame retardants can be extracted. In Raman spectroscopy, the vibrational modes of chemical bonds produce unique scattering peaks in specific wavelength regions. For example, nitrogen-containing or phosphorus-containing chemical bonds typically exhibit different frequency shifts in Raman spectra. This information can be used to further identify and quantify the content of flame retardants. Extracting peak positions and intensities directly related to content includes the following specific steps:

[0027] Step Y1: Use a peak detection algorithm to identify significant peaks in the spectrum. The peak detection algorithm is based on the first derivative to find the inflection point of the spectrum and determine its wavelength or wave value.

[0028] Step Y2: The intensity of the peak reflects the strength of the light absorbed in that band. It can be obtained by calculating the maximum absorption value of each peak in the spectrum. The intensity of the peak is proportional to the content of the substance. Therefore, by measuring the intensity of these peaks, the concentration or content of the flame retardant can be inferred.

[0029] As a further aspect of the present invention, in step S33, the combustion experiment results of step S2 are used as labels, and the spectral data collected in step S1 are used as feature inputs. In order to obtain the best prediction model, a variety of classic machine learning algorithms are used for preliminary training and comparison. The machine learning algorithms include support vector machine, inverse neural network, CatBoost, BO-CatBoost, XGBoost and BO-XGBoost. The BO-XGBoost model with the highest accuracy, precision, recall and F1 score is selected as the final machine prediction model.

[0030] As a further aspect of the present invention, in step S4, the hyperparameters of the XGBoost model are searched using the Bayesian optimization method. The initial hyperparameter settings are: max_depth between 3 and 10, learning_rate between 0.01 and 0.3, gamma between 0 and 5, and subsample between 0.5 and 1.0. The optimal hyperparameter results are: max_depth = 7, learning_rate = 0.15, gamma = 1.2, and subsample = 0.8.

[0031] The technical effects and advantages of this invention, a non-destructive testing method for the flame retardant performance of non-metallic casings of energy metering boxes, are as follows: This invention combines spectroscopic technology with machine learning and prediction techniques to achieve non-destructive testing of the flame retardant performance of non-metallic casings of energy metering boxes. Accurate data is obtained by collecting multiple samples for flame combustion tests to acquire labels. Non-destructive spectral data is obtained using in-situ spectral detection. After preprocessing and feature extraction, various machine learning algorithms are trained and optimized, and finally, a Bayesian-optimized BO-XGBoost model is selected for prediction. This method overcomes the limitations of existing destructive testing and sampling inspections, enabling product-by-product testing to ensure the safety of power equipment. Furthermore, the testing process is non-destructive and does not affect the subsequent use of the samples, providing an efficient, accurate, non-destructive, and widely applicable solution for testing the flame retardant performance of non-metallic casings of energy metering boxes. Attached Figure Description

[0032] Figure 1 This is a flowchart of a non-destructive testing method for the flame retardant performance of a non-metallic casing of an energy metering box, according to the present invention.

[0033] Figure 2 This is a diagram showing the combustion test results of the first sample using a horizontal and vertical combustion apparatus according to the present invention.

[0034] Figure 3 This is a diagram showing the combustion test results of the second sample using a horizontal and vertical combustion apparatus according to the present invention.

[0035] Figure 4This is a diagram showing the combustion test results of the third sample using a horizontal and vertical combustion apparatus according to the present invention. Detailed Implementation

[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0037] Example 1

[0038] See Figure 1 The flowchart shown illustrates a non-destructive testing method for the flame retardant performance of a non-metallic casing of an energy metering box. This invention combines spectroscopic techniques with machine learning and prediction techniques. It utilizes in-situ spectroscopic measurement technology, which does not damage the sample, to obtain the chemical composition spectrum of the test object. Through large-scale experimental data learning and training of a machine learning model, it achieves prediction of the material's flame retardant performance based on in-situ spectral characteristics. The method includes the following steps:

[0039] Step S1: Collect a sample of the non-metallic shell of the power metering box. Use a portable infrared spectrometer, a portable Raman spectrometer, or an infrared / Raman spectrometer equipped with a movable fiber optic probe to perform in-situ spectroscopic detection on the collected non-metallic shell sample and collect spectral data.

[0040] Step S11: Collect a wide range of samples of electricity metering boxes from various brands and manufacturers on the market to ensure the diversity and representativeness of the test data and obtain non-metallic casings;

[0041] Step S12: Use clean water and anhydrous ethanol alternately to clean the sample to remove dust, oil and other contaminants from the surface and ensure the accuracy of the flame combustion test.

[0042] Step S2: Conduct a flame combustion test on the collected non-metallic shell sample of the power metering box;

[0043] Step S3: Preprocessing spectral data and establishing a machine learning model;

[0044] Step S31: The acquired spectral data are sequentially subjected to Savitzky-Golay filtering, baseline correction, and normalization.

[0045] Step S32: Extract key band features from the original spectrum, including information such as peak position and intensity that are directly related to the flame retardant components and content;

[0046] Step S33: Initial model training is performed using Support Vector Machine (SVM), Inverse Neural Network, CatBoost, BO-CatBoost, XGBoost, and BO-XGBoost (Bayesian-optimized XGBoost), and the BO-XGBoost model is selected as the final prediction model.

[0047] Step S4: Use Bayesian optimization to search for the hyperparameters of the XGBoost model, including the maximum tree depth (max_depth), learning rate (learning_rate), minimum split loss (gamma), and subsample ratio (subsample). Evaluate the model performance using five-fold cross-validation.

[0048] Step S5: Use the optimized BO-XGBoost model to predict and identify the flame retardant properties of the non-metallic shell material of the power metering box.

[0049] Step S51: Use an infrared or Raman spectrometer with a movable fiber optic probe to test the non-metallic shell of the power metering box in situ to obtain the spectral data of the test sample.

[0050] Step S52: Use the optimized BO-XGBoost model to read in the spectral data of the test sample, predict the flame retardant performance of the test sample, and output the result as "qualified sample" or "unqualified sample".

[0051] In step S53, the sample predicted by the optimized BO-XGBoost model is further verified by the combustion test in step S13.

[0052] Further, in step S2, the flame retardant performance test is conducted according to the national standard GB / T 2408-2021. Specifically, the cleaned and dried non-metallic shell sample is cut into strips with a length of 125.0mm ± 5.0mm, a width of 13.0mm ± 0.5mm, and a thickness of < 13.0mm, ensuring that its dimensions meet the standard requirements. The test uses horizontal and vertical combustion testing equipment, and combustion tests are conducted under specified flame height, ignition time (10 seconds or 30 seconds), and test environment (23±2℃, relative humidity 50±10%). During the test, the flame propagation speed after combustion, self-extinguishing time after flame removal, dripping phenomenon, and whether cotton is ignited are observed. Based on the comprehensive performance of the combustion test, the flame retardant level is divided into V0 (the highest level, indicating short burning time, no dripping, and strong self-extinguishing ability), V1 (slightly longer self-extinguishing time, but no burning dripping), V2 (burning dripping and ignition of cotton), and other levels (unqualified samples). Test data serves as a key label for model training, providing a foundation for building machine learning models, while repeated experiments with different samples improve data reliability.

[0053] Furthermore, in step S31, Savitzky-Golay filtering is used to smooth the acquired spectral data. By setting a window width of 7 points and a second-order polynomial order, random noise in the spectral signal is removed while retaining the shape of key spectral characteristic peaks. The least squares baseline fitting algorithm is used for baseline correction to eliminate spectral baseline drift caused by test environment or instrument errors, ensuring that the spectral data is based on an accurate zero point. Max-min normalization is used to normalize the spectral data, mapping the spectral intensity values ​​to the range of [0,1] to eliminate the influence of external factors such as sample thickness and light source intensity on the characteristic intensity.

[0054] Further, in step S32, key waveband features are extracted from the original spectrum. Since flame retardants typically have unique characteristic absorption peaks in infrared and Raman spectra, key wavebands related to these chemical components and their contents need to be extracted from the spectrum. In infrared spectra, different chemical groups will produce absorption peaks within specific wavenumber ranges. For example, phosphate groups typically have significant characteristic absorption peaks in the 1200-1300 cm⁻¹ infrared spectral range, bromides will show characteristic peaks in the 500-600 cm⁻¹ infrared spectral range, while chlorides show characteristic peaks in the 600-800 cm⁻¹ range. -1 The infrared spectral region exhibits weak characteristic absorption peaks. By analyzing these peaks in the sample spectrum, information related to the type and content of flame retardants can be extracted. In Raman spectroscopy, the vibrational modes of chemical bonds produce unique scattering peaks in specific wavelength regions. For example, nitrogen-containing or phosphorus-containing chemical bonds typically exhibit different frequency shifts in Raman spectra. This information can be used to further identify and quantify the content of flame retardants. Extracting peak positions and intensities directly related to content includes the following specific steps:

[0055] Step Y1: Use a peak detection algorithm to identify significant peaks in the spectrum. The peak detection algorithm is based on the first derivative to find the inflection point of the spectrum and determine its wavelength or wave value.

[0056] Step Y2: The intensity of the peak reflects the strength of the light absorbed in that band. It can be obtained by calculating the maximum absorption value of each peak in the spectrum. The intensity of the peak is proportional to the content of the substance. Therefore, by measuring the intensity of these peaks, the concentration or content of the flame retardant can be inferred.

[0057] Furthermore, in step S33, the combustion experiment results of step S2 are used as labels, and the spectral data collected in step S1 are used as feature inputs. In order to obtain the best prediction model, a variety of classic machine learning algorithms are used for preliminary training and comparison. The machine learning algorithms include support vector machine, inverse neural network, CatBoost, BO-CatBoost, XGBoost and BO-XGBoost. The table below shows the accuracy, precision, recall and F1 score of support vector machine, inverse neural network, CatBoost, BO-CatBoost, XGBoost and BO-XGBoost on the training set.

[0058] Table 1. Accuracy, precision, recall, and F1 score of Support Vector Machine, Reverse Neural Network, CatBoost, XGBoost, and BO-XGBoost on the training set.

[0059] Model Accuracy (%) Accuracy (%) Recall rate (%) F1 Score (%) Support Vector Machine 92.5 91.8 90.2 91.0 Inverse Neural Network 89.8 88.7 87.4 88.0 CatBoost 94.2 93.5 92.8 93.1 BO-CatBoost 95.0 94.5 93.7 94.1 XGBoost 93.6 92.3 91.7 92.0 BO-XGBoost 96.1 95.4 94.8 95.1

[0060] Based on the accuracy, precision, recall, and F1 score of different models on the training set, the BO-XGBoost model was ultimately selected as the final machine prediction model.

[0061] Furthermore, in step S4, the hyperparameters of the XGBoost model are searched using the Bayesian optimization method. The initial hyperparameter settings are: max_depth between 3 and 10, learning_rate between 0.01 and 0.3, gamma between 0 and 5, and subsample between 0.5 and 1.0. The optimal hyperparameter results are: max_depth = 7, learning_rate = 0.15, gamma = 1.2, and subsample = 0.8.

[0062] In this embodiment, 239 non-metallic shell samples of electricity metering boxes from different manufacturers and models on the market were collected. The flame retardant performance of the shell samples was tested, yielding 206 qualified samples (flame retardant level V0) and 33 unqualified samples (flame retardant level below V0). Spectral data were obtained from the 239 collected samples using a portable Raman spectrometer. The obtained flame retardant performance test results (qualified - flame retardant level V0, unqualified - flame retardant level below V0) were used to train supervised machine learning on the obtained Raman spectral data. Following steps S3, S4, and S5 of the present invention, a BO-XGBoost model was used for training to obtain a prediction model. The trained prediction model was used to evaluate and predict the flame retardant performance of three untested flame retardant performance samples using in-situ Raman spectral data. A horizontal and vertical combustion tester was used to test the flame retardant performance of the three samples, and the results were obtained. The combustion test results of the three samples are as follows: Figure 2 , Figure 3 , Figure 4 The photographs of the flame combustion test are shown. The degree of agreement between the model prediction results and the combustion test results for the three samples—the identification accuracy rate—is shown in Table 2.

[0063] Example 2

[0064] The BO-XGBoost model trained in Example 1 was used to predict the Raman spectral data of eight non-metallic energy metering box shell samples; the flame retardant performance of the samples was tested using a horizontal and vertical combustion tester. The degree of agreement between the model prediction results and the sample combustion test results—the identification accuracy—is shown in Table 2.

[0065] Table 2. Degree of agreement between model predictions and combustion test results - identification accuracy

[0066]

[0067]

[0068] This invention combines spectral technology with machine learning and prediction techniques to achieve non-destructive testing of the flame-retardant performance of non-metallic casings for electricity metering boxes. Flame-retardant performance data is obtained by collecting multiple samples for flame combustion tests and using this data as labels. Non-destructive spectral data is obtained through in-situ spectral analysis of the samples. After preprocessing and feature extraction, various machine learning algorithms are trained and optimized, and finally, a Bayesian-optimized BO-XGBoost model is selected for prediction. This method overcomes the limitations of existing destructive testing and sampling inspections, enabling product-by-product testing to ensure the safety of electrical equipment. Furthermore, the testing process is non-destructive and does not affect the subsequent use of the samples, providing an efficient, accurate, non-destructive, and widely applicable solution for testing the flame-retardant performance of non-metallic casings for electricity metering boxes.

[0069] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0070] In conclusion, 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 principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A non-destructive testing method for the flame retardant performance of a non-metallic casing of an electricity metering box, utilizing in-situ spectroscopic measurement technology that does not damage the sample to obtain the chemical composition spectrum of the test object, thereby achieving prediction of the flame retardant performance of the material based on in-situ spectral characteristics, characterized in that... Includes the following steps: Step S1: Collect a sample of the non-metallic shell of the power metering box. Use a portable infrared spectrometer, a portable Raman spectrometer, or an infrared / Raman spectrometer equipped with a movable fiber optic probe to perform in-situ spectroscopic detection on the collected non-metallic shell sample and collect spectral data. Step S2: Conduct a flame combustion test on the collected non-metallic shell sample of the power metering box; Step S3: Preprocessing spectral data and establishing a machine learning model; Step S4: Use Bayesian optimization to search for hyperparameters of the XGBoost model, including maximum tree depth, learning rate, minimum split loss, and subsample ratio. Evaluate model performance using five-fold cross-validation. Step S5: Use the optimized BO-XGBoost model to predict and identify the flame retardant properties of the non-metallic shell material of the power metering box.

2. The non-destructive testing method for the flame retardant performance of the non-metallic casing of an energy metering box according to claim 1, characterized in that, In step S1, collecting samples of the non-metallic casing of the power metering box specifically includes: extensively collecting samples of power metering boxes from various brands and manufacturers on the market to ensure the diversity and representativeness of the test data, and obtaining the non-metallic casing; using clean water and anhydrous ethanol alternately to clean the sample to remove surface dust, oil and other contaminants to ensure the accuracy of the flame combustion test.

3. The non-destructive testing method for the flame retardant performance of the non-metallic casing of an energy metering box according to claim 1, characterized in that, In step S2, the flame combustion test specifically includes: according to the GB / T2408-2021 standard, cutting the sample into standard test strips and conducting a combustion test to obtain flame retardant performance data, providing labels for subsequent model training.

4. The non-destructive testing method for the flame retardant performance of the non-metallic casing of an energy metering box according to claim 1, characterized in that, In step S3, Savitzky-Golay filtering is used to smooth the acquired spectral data. By setting a window width of 7 points and a second-order polynomial order, random noise in the spectral signal is removed while retaining the shape of key spectral characteristic peaks. The least squares baseline fitting algorithm is used for baseline correction to eliminate spectral baseline drift caused by test environment or instrument errors, ensuring that the spectral data is based on an accurate zero point. Max-min normalization is used to normalize the spectral data, mapping the spectral intensity values ​​to the range of [0,1] to eliminate the influence of external factors such as sample surface finish and scattered light intensity on the characteristic intensity.

5. The non-destructive testing method for the flame retardant performance of the non-metallic casing of an energy metering box according to claim 1, characterized in that, In step S3, key band features are extracted from the original spectrum, including peak positions and intensity information directly related to the flame retardant components and content. Extracting peak positions and intensities directly related to content includes the following specific steps: Step Y1: Use a peak detection algorithm to identify significant peaks in the spectrum. The peak detection algorithm is based on the first derivative to find the inflection point of the spectrum and determine its wavelength or wave value. Step Y2: The intensity of the peak reflects the strength of the absorption ability of the test substance in this band of spectrum. It is obtained by calculating the maximum absorption value of the spectral peak. The intensity of the peak reflects the content of the substance. Therefore, by measuring the intensity of these peaks, the concentration or content of a specific flame retardant in the test sample can be calculated.

6. The non-destructive testing method for the flame retardant performance of the non-metallic casing of an energy metering box according to claim 1, characterized in that, In step S3, support vector machines, inverse neural networks, CatBoost, BO-CatBoost, XGBoost and BO-XGBoost are used for preliminary model training. Finally, the BO-XGBoost model with the highest accuracy, precision, recall and F1 score on the training set is selected as the final machine prediction model.

7. The non-destructive testing method for the flame retardant performance of the non-metallic casing of an energy metering box according to claim 1, characterized in that, In step S4, the Bayesian optimization method is used to search for the hyperparameters of the XGBoost model. The initial hyperparameter settings are: max_depth between 3 and 10, learning_rate between 0.01 and 0.3, gamma between 0 and 5, and subsample between 0.5 and 1.

0. The optimal hyperparameter results are: max_depth = 7, learning_rate = 0.15, gamma = 1.2, and subsample = 0.

8.

8. The non-destructive testing method for the flame retardant performance of the non-metallic casing of an energy metering box according to claim 1, characterized in that, In step S5, a portable infrared or Raman spectrometer, or an infrared or Raman spectrometer with a movable fiber optic probe, is used to test the non-metallic shell of the power metering box in situ to obtain the spectral data of the test sample. The optimized BO-XGBoost model is used to read the spectral data of the test sample, predict the flame retardant performance of the test sample, and output the result as "qualified sample" or "unqualified sample". The sample predicted by the optimized BO-XGBoost model is further verified by the combustion test in step S1.

Images