Time-frequency characterization method for friction and vibration behavior of cemented carbide with metal powder sintered composite layer

Abstract

A time-frequency characterization method for the friction and vibration behavior of a sintered composite layer cemented carbide with metal powder, which relates to a time-frequency characterization method for friction and vibration behavior. The present invention aims to solve the problems that the existing time-frequency methods are not deep enough in extracting the image feature textures during analysis, the extracted features are not extensive enough and lack strong persuasiveness, and it is necessary to combine time-domain and frequency-domain analysis, while the improved time-frequency analysis methods often require data adaptation and lack generality. The steps of the present invention include preprocessing the collected friction and vibration data files; performing noise reduction processing on the collected vibration signals; obtaining the time-frequency analysis image of the vibration signals; performing grayscale processing on the time-frequency image; extracting features from the time-frequency image of the vibration signals; performing dimensionality reduction and screening on the vibration behavior features based on the MIC maximum mutual information coefficient - BP neural network method; and performing curve fitting on the stability characterization of the test system based on the vibration behavior. The present invention belongs to the technical field of friction and vibration behavior characterization.

Description

Technical Field

[0001] The present invention relates to a time-frequency characterization method for friction vibration behavior, belonging to the technical field of friction vibration behavior characterization. Background Art

[0002] With the continuous application of difficult-to-machine materials such as titanium alloys in the engineering field, it poses a great challenge to the wear resistance and service life of cemented carbide tools for their cutting processing. And the strengthening means on the surface of cemented carbide are also constantly improving. Among them, the sintered metal powder composite layer is widely used in the field of alloy surface strengthening. The strengthening effects of composite layers made of different metal powders and different sintering parameters are very different. Each adjustment of parameters, elements and ratios requires friction and wear detection of surface anti-wear and stability. And a large amount of test data will be generated during the detection process. Among them, vibration behavior data and friction force data are two important performance detection and evaluation indicators. Generally, the acquisition cost of vibration behavior data is lower than that of friction force data.

[0003] Currently, most of the vibration behavior analyses for friction and wear tests stay at the analysis in the time domain, frequency domain and single time-frequency domain. Time domain and frequency domain analyses need to process the original data, which requires a large amount of space for operation and has poor anti-interference ability; the existing time-frequency methods do not extract the image feature textures deeply enough during analysis, and the extracted features are not extensive enough and lack persuasive power, and need to combine time domain and frequency domain analyses; while the improved time-frequency analysis methods often require data adaptation and lack universality.

[0004] Therefore, improving the persuasive power of time-frequency analysis feature extraction, improving the universality of the analysis method, reducing the amount of computation, and reducing the feature storage space play an important role in analyzing multi-batch and multi-feature vibration behavior data generated by the detection of the strengthening performance of sintered metal powder composite layers, and improving the stability of the detection system and the data prediction and characterization ability. Summary of the Invention

[0005] In order to solve the problems that the existing time-frequency methods do not extract the image feature textures deeply enough during analysis, the extracted features are not extensive enough and lack persuasive power, and need to combine time domain and frequency domain analyses, while the improved time-frequency analysis methods often require data adaptation and lack universality, the present invention further provides a time-frequency characterization method for the friction vibration behavior of cemented carbide with a sintered metal powder composite layer.

[0006] The technical solutions adopted by the present invention to solve the above problems are as follows: The steps of the present invention include:

[0007] Step 1, preprocess the collected friction vibration data file;

[0008] Step 2, perform noise reduction processing on the collected vibration signal;

[0009] Step 3, obtain the time-frequency analysis image of the vibration signal;

[0010] Step 4: Perform grayscale processing on the time-frequency image;

[0011] Step 5: Extract features from the time-frequency image of the vibration signal;

[0012] Step 6: Perform dimensionality reduction and screening on the vibration behavior features based on the MIC maximum mutual information coefficient - BP neural network method;

[0013] Step 7: Verify the regression prediction accuracy of feature selection based on the random forest RF;

[0014] Step 8: Fit the stability characterization curve of the test system based on the vibration behavior.

[0015] Furthermore, the preprocessing of the collected friction vibration data file in Step 1 refers to using LabVIEW to convert other types of collected files into '.xlsx' format files of Excel tables.

[0016] Furthermore, in Step 2, the variational mode decomposition (VMD) is used to remove the obvious interference parts in the vibration and perform noise reduction processing.

[0017] Furthermore, in Step 3, the short-time Fourier transform and the continuous wavelet transform are respectively used to perform time-frequency analysis on the friction signal to obtain the time-frequency image.

[0018] Furthermore, the grayscale processing of the time-frequency image in Step 4 specifically includes:

[0019] Remove the color of the image, simplify the complexity of the image, and shorten the calculation time;

[0020] Adopt the grayscale algorithm to perform grayscale processing on the image according to 256 grayscale levels. The feature information such as the color and shape changes of the image will be saved in the grayscale image in the form of grayscale pixel points, and the feature transformation is performed through the pixel distribution and the number of pixel points of each grayscale level.

[0021] Furthermore, the feature extraction from the time-frequency image of the vibration signal in Step 5 specifically includes:

[0022] Compare and adapt the eigenvalue extraction methods based on the image features obtained from the time-frequency analysis;

[0023] For the STFT image features, select 7 Hu invariant moment features A total of 8 features including the grayscale mean are extracted as the image feature values;

[0024] For the CWT image features, select the mean and variance of energy, contrast, homogeneity, and entropy of the gray-level co-occurrence matrix of the CWT image in four directions of 0°, 45°, 90°, and 135° as a total of 8 features as the image feature values for extraction; finally, a total of 16 time-frequency features of friction vibration are obtained.

[0025] Further, the steps of dimensionality reduction and screening of vibration behavior features based on the MIC maximum mutual information coefficient - BP neural network method in step 6 are specifically as follows:

[0026] Step 601: Obtain 16 eigenvalue of the friction and wear characteristics vibration of the sintered composite layer of metal powder reinforced cemented carbide through step 5;

[0027] Step 602: Use the friction force, which is the characterization parameter of the stability of the friction and wear test system, as the output of feature screening, and use the 16 eigenvalue of STFT and CWT as the input for feature selection based on the MIC maximum mutual information coefficient. The specific implementation is as follows:

[0028] By discretizing the input feature X and the output feature Y in a two-dimensional space, dividing the data into grids according to the distribution of the eigenvalue, calculating the mutual information between the feature and the target variable within each grid, and finally traversing all grids to find the grid with the maximum mutual information, so as to obtain the MIC value of each feature as the contribution value of the feature;

[0029] Based on the feature selection experience, when the gap between the maximum feature contribution and the minimum feature contribution is large, take half of the difference between the maximum MIC value and the minimum MIC value among the features as the threshold for initially screening the feature importance, that is, initially screen and retain features with an importance greater than 0.3;

[0030] Step 602: Use the RMSE value of the BP neural network friction force prediction to further screen the features, and calculate the mean value of the RMSE values of 30 predictions each time a feature is reduced.

[0031] Further, the specific steps of curve fitting for characterizing the stability of the test system based on vibration behavior in step 8 include:

[0032] Step 801: Based on the selected gray mean, entropy mean, contrast mean, entropy variance, and energy mean features as the numerical basis, based on the vector normalization formula:

[0033]

[0034] In formula (1), c represents a coefficient used to increase the data magnitude, set to 100; n represents the number of samples; a i represents the actual measured data; y iIt represents the normalized data. The normalization is achieved by using the norm(x) function in MATLAB to perform vector normalization on the data, eliminating the influence of dimensions.

[0035] Step 802: Based on the normalized data, perform feature weight division. Specifically, based on the importance of the contribution value detected by MIC, set the sum of the selected feature contributions as C, and the contribution value of each feature as x i , where i represents different features. Take the percentage of the contribution value of each feature in the total contribution value of the selected features, that is: where y i is the weight divided for each retained feature;

[0036] Step 803: Multiply each normalized feature by its respective weight and then sum to obtain the six-feature fusion parameter. Based on the fusion parameter, fit the time-frequency feature law curve of the vibration behavior;

[0037] Step 804: Through the vector normalization, weight division, and weighted fitting of the gray mean, entropy mean, contrast mean, entropy variance, and energy mean, obtain the time-frequency feature law curve of the vibration behavior, thereby characterizing the stability change curve of the sintered metal powder composite layer alloy strengthening performance detection system based on the vibration behavior.

[0038] The beneficial effects of the present invention are as follows:

[0039] 1. The analysis method proposed by the present invention is highly adaptable to the vibration behavior analysis of the surface properties of sintered metal powder composite layer cemented carbide and highly adaptable to the detection of the stability of the test system, which can significantly improve the accuracy of the analysis and the accuracy of the prediction of relevant parameters;

[0040] 2. Compared with the traditional time-domain and frequency-domain analysis of vibration signals, the method of the present invention can effectively reduce the amount of data from the source, convert the complex original data into images for storage, save storage and computing space, and extract a large number of image features to dilute the interference of impurity signals, thereby improving the anti-interference ability of the analysis method;

[0041] 3. Compared with the traditional single time-frequency analysis method and the improved time-frequency analysis method, the method of the present invention extracts the time-frequency image features more comprehensively and completely, has stronger versatility, and well balances the accuracy and versatility of the vibration behavior time-frequency processing method;

[0042] 4. The machine learning algorithm used in the present invention performs feature fusion at the feature layer level to achieve feature selection and data prediction with multiple inputs and single output, and has good robustness during operation. Description of the Drawings

[0043] Figure 1It is a schematic structural diagram of a surface performance detection system for sintered metal powder composite layer cemented carbide;

[0044] Figure 2 It is a schematic diagram of a signal data file format conversion program based on LabVIEW;

[0045] Figure 3 It is a schematic diagram of vibration signal noise reduction processing based on VMD;

[0046] Figure 4 It is a schematic diagram of obtaining the time-frequency image of the vibration signal based on STFT and CWT;

[0047] Figure 5 It is a schematic diagram of graying the time-frequency images of STFT and CWT based on the gray scale algorithm;

[0048] Figure 6 It is a schematic diagram of the time-frequency image eigenvalue extraction program;

[0049] Figure 7 It is a schematic diagram of the MIC maximum mutual information coefficient feature selection program and the importance image;

[0050] Figure 8 It is a schematic diagram of the random forest RF prediction verification result;

[0051] Figure 9 It is a schematic diagram of the test system stability characterization curve;

[0052] Figure 10 It is a schematic diagram of file format conversion;

[0053] Figure 11 It is a schematic diagram of some noise reduction images;

[0054] Figure 12 It is a schematic diagram of the two-dimensional, three-dimensional and gray-scale images of STFT and CWT time-frequency processing;

[0055] Figure 13 It is a flow chart of the present invention. Specific implementation mode

[0056] Specific implementation mode 1: As Figures 1 to 13 shown, a time-frequency characterization method for the friction and vibration behavior of a metal powder sintered composite layer cemented carbide, the specific steps include:

[0057] Step 1, preprocess the collected friction and vibration data file;

[0058] Step 2, perform noise reduction processing on the collected vibration signal;

[0059] Step 3, obtain the time-frequency analysis image of the vibration signal;

[0060] Step 4: Perform grayscale processing on the time-frequency image;

[0061] Step 5: Extract features from the time-frequency image of the vibration signal;

[0062] Step 6: Perform dimensionality reduction and screening on the vibration behavior features based on the MIC maximum mutual information coefficient - BP neural network method;

[0063] Step 7: Verify the regression prediction accuracy of feature selection based on the random forest RF;

[0064] Step 8: Fit the stability characterization curve of the test system based on the vibration behavior.

[0065] Specific Embodiment 2: As Figures 1 to 13 shown, the preprocessing of the collected friction vibration data file in Step 1 refers to using LabVIEW to convert the collected other types of files into '.xlsx' format files of Excel tables.

[0066] Specific Embodiment 3: As Figures 1 to 13 shown, in Step 2, the variational mode decomposition (VMD) is used to remove the obvious interference parts in the vibration and perform noise reduction processing.

[0067] Among them, if the vibration information of the overall detection system is comprehensively considered, noise reduction can be not carried out when the interference of external impurity signals is small.

[0068] The [imf, residual] = vmd(x, 'NumIMF', a) program in MATLAB is used to perform noise reduction on the image. A total of 9 layers of decomposition are performed, and the IMF2 - IMF8 components are reconstructed.

[0069] Specific Embodiment 4: As Figures 1 to 13 shown, in Step 3, the short-time Fourier transform and the continuous wavelet transform are respectively used to perform time-frequency analysis on the friction signal to obtain the time-frequency image.

[0070] Among them, the STFT uses a Hanning window to set the window length to 127,

[0071] The CWT needs to adjust the wavelet function according to the range of the cone of influence curve in the image so that the fuzzy features outside the cone of influence curve do not affect the overall time-frequency features. Wavelet functions such as 'haar', 'db','sym', 'cmor','mexh', 'gaus', 'bior','morse', 'amor', 'bump' can be called according to the signal features to reduce the influence of the fuzzy features inside the cone of influence curve. Using the 'amor' wavelet function for CWT time-frequency analysis of the vibration signal obtained in the friction and wear test meets the needs of feature extraction.

[0072] The short-time Fourier transform toolbox of MATLAB and [wt,f]=cwt(x,'wavelet function',fs) are used for STFT and CWT time-frequency analysis of vibration signals, and the obtained time-frequency images and influence cone curves are verified as Figure 4 shown below.

[0073] Specific implementation method 5: As Figures 1 to 13 shown below, the grayscale processing of the time-frequency image in step 4 specifically includes:

[0074] Remove the color of the image, simplify the complexity of the image, and shorten the calculation time;

[0075] Use the grayscale algorithm to process the image according to 256 grayscale levels. The characteristic information such as the color and shape changes of the image will be saved in the grayscale image in the form of grayscale pixel points, and the characteristic conversion is performed through the pixel distribution and the number of pixel points of each grayscale level as the characteristic information.

[0076] Specific implementation method 6: As Figures 1 to 13 shown below, the feature extraction of the vibration signal time-frequency image in step 5 specifically includes:

[0077] Compare and adapt the eigenvalue extraction methods according to the image features obtained from the time-frequency analysis;

[0078] For the STFT image features, select 7 Hu invariant moment features A total of 8 features including the grayscale mean value are extracted as the image feature values;

[0079] For the CWT image features, select the mean and variance of the energy, contrast, uniformity, and entropy of the gray-level co-occurrence matrix of the CWT image in four directions of 0°, 45°, 90°, and 135° as a total of 8 features as the image feature values for extraction; finally, a total of 16 time-frequency features of friction vibration are obtained.

[0080] Among them, Figure 7 The features are gradually removed according to the ranking in to observe the change of the RSME value until the value is the smallest. The specific implementation is as follows: First, obtain the predicted RMSE values of the system friction force with all input features retained and the input features initially screened through the BP neural network, and verify the result of the initial screening of the features; then, based on the 10 features initially screened out, reduce one feature with the smallest contribution each time according to the ranking of the feature MIC contribution value, and perform a prediction calculation of the output feature friction force to detect the change of the RMSE value until the RMSE value of the predicted system friction force with the retained features reaches the smallest. Thus, the advanced screening of the input features is realized, and the final features are obtained.

[0081] The prediction calculation is shown in Table 1. The finally retained features are: grayscale mean value, entropy mean value, contrast mean value, entropy variance, energy mean value.

[0082] Table 1 RMSE values of friction force predicted by BP neural network

[0083]

[0084] Specific implementation method seven: As Figures 1 to 13 shown, the steps of dimensionality reduction and screening of vibration behavior characteristics based on the MIC maximum mutual information coefficient - BP neural network method in step 6 are specifically as follows:

[0085] Step 601: Obtain 16 characteristic values of the friction and wear test vibration of the metal powder sintered composite layer strengthened cemented carbide through step 5;

[0086] Step 602: Taking the friction force, the characterization parameter of the stability of the friction and wear test system, as the output of feature screening, and taking the 16 characteristic values of STFT and CWT as the input, perform feature selection based on the MIC maximum mutual information coefficient. The specific implementation is as follows:

[0087] By discretizing the input feature X and the output feature Y in a two-dimensional space, dividing the data into grids according to the distribution of the characteristic values, calculating the mutual information between the feature and the target variable within each grid, and finally traversing all grids to find the grid with the maximum mutual information, thereby obtaining the MIC value of each feature as the contribution value of the feature;

[0088] Based on feature selection experience, when the gap between the maximum feature contribution and the minimum feature contribution is large, take half of the difference between the maximum MIC value and the minimum MIC value among the features as the threshold for initially screening the importance of features, that is, initially screen and retain features with an importance greater than 0.3;

[0089] Step 602: Use the RMSE value of the friction force predicted by the BP neural network to perform the next step of screening on the features, and calculate the average value of the RMSE values of 30 predictions each time a feature is reduced.

[0090] Specific implementation method eight: As Figures 1 to 13 shown, the specific steps of curve fitting for characterizing the stability of the test system based on vibration behavior in step 8 include:

[0091] Step 801: Based on the selected grayscale mean, entropy mean, contrast mean, entropy variance, and energy mean features as the numerical basis, based on the vector normalization formula:

[0092]

[0093] In formula (1), c represents a coefficient used to increase the data magnitude, set to 100; n represents the number of samples; a i represents the actual measurement data; y iRepresents the normalized data; the normalization of the data is achieved by using the norm(x) function in MATLAB to perform vector normalization on the data, eliminating the influence of dimensions;

[0094] Step 802: Based on the normalized data, perform feature weight division. The specific implementation is as follows: Based on the importance of the contribution value detected by MIC, set the sum of the selected feature contributions to C, and the contribution value of each feature to be x i , where i represents different features, and the percentage of the contribution value of each feature to the total contribution value of the selected features, that is: where y i is the weight assigned to each retained feature;

[0095] Step 803: Multiply each normalized feature by its respective weight and then sum to obtain the six-feature fusion parameter. Based on the fusion parameter, fit the time-frequency feature law curve of the vibration behavior;

[0096] Step 804: Through the vector normalization processing, weight division, and weighted fitting of the gray mean, entropy mean, contrast mean, entropy variance, and energy mean, obtain the time-frequency feature law curve of the vibration behavior, thereby characterizing the stability change curve of the sintered metal powder composite layer alloy strengthening performance detection system based on the vibration behavior.

[0097] Specific Embodiment Nine: As Figures 1 to 13 shown, the verification of the regression prediction accuracy of feature selection based on the random forest RF in Step Seven specifically includes:

[0098] Respectively use the 4th, 5th, and 6th features gradually selected in Table 1 of Step 6 as inputs,

[0099] with the friction force as the output, perform feature selection and prediction accuracy verification. To adapt to the data, prevent overfitting in prediction, and improve the generalization ability, select the number of decision trees to be 150 and the number of leaves to be 1.

[0100] The verification results of the random forest prediction are as Figure 8 shown. The selected gray mean, entropy mean, contrast mean, entropy variance, and energy mean are already the best features. In this way, the vibration behavior features that best describe the stability of the sintered metal powder composite layer strengthening performance detection system and have the greatest prediction contribution are obtained.

[0101] Example

[0102] A method for time-frequency characterization of the friction vibration behavior of a metal powder sintered composite layer cemented carbide, specifically including:

[0103] Step One: Use LabVIEW to convert data files in other formats into '.xlsx' format files;

[0104] Step 2: According to the signal characteristics and component preservation requirements, perform VMD noise reduction processing on the original signal data;

[0105] Step 3: Perform STFT and CWT time-frequency analysis on the vibration signal data after noise reduction processing to obtain a time-frequency image, and perform grayscale processing on the time-frequency image using a grayscale algorithm;

[0106] Step 4: Extract features based on the characteristics of the time-frequency image. For the STFT image, select the Hu invariant moment Gray mean value; for the CWT image, select the GLCM gray-level co-occurrence matrix: the mean and variance of energy, contrast, uniformity, and entropy in four directions of 0°, 45°, 90°, and 135°. A total of 16 features are extracted;

[0107] Step 5: Using the system stability friction force as the input and the 16 features as the output, initially perform feature selection using the MIC maximum mutual information coefficient method to exclude features with a contribution level to the output lower than 0.3;

[0108] Step 6: Use the BP neural network in combination with the MIC maximum mutual information coefficient method to gradually screen the features;

[0109] Step 7: Use the random forest RF regression prediction algorithm to verify the selected feature information and predict the system stability friction force again;

[0110] Step 8: Perform curve fitting on the obtained selected features; use the vector normalization method to normalize the features; divide the weights of the features to be fitted based on the feature contribution values obtained by the MIC maximum mutual information coefficient method, and finally fit the detection system stability change curve based on vibration behavior according to the weights.

[0111] The above is only a preferred embodiment of the present invention, and does not impose any form of limitation on the present invention. Although the present invention has been disclosed above with a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can make some changes or modifications to the above-disclosed technical content to make equivalent embodiments with equivalent changes within the scope of the technical solution of the present invention. However, as long as it does not depart from the content of the technical solution of the present invention, any simple modification, equivalent replacement, and improvement made to the above embodiments within the spirit and principle of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A time-frequency characterization method for friction vibration behavior of metal powder sintered composite layer cemented carbide, characterized in that: The specific steps include: Step 1: preprocessing the collected friction vibration data file; Step 2: performing noise reduction processing on the collected vibration signal; Step 3, obtaining a time-frequency analysis image of the vibration signal; Step 4: grayscale the time-frequency image; Step 5: Extract features from the vibration signal time-frequency image; specifically including: Compare and adapt the feature value extraction methods based on the image features obtained by time-frequency analysis; For STFT image features, select 7 Hu invariant moment features A total of 8 features including grayscale mean are extracted as image feature values; For CWT image features, 8 features including energy, contrast, uniformity, mean and variance of entropy of CWT image gray-level co-occurrence matrix at four directions of 0°, 45°, 90° and 135° were selected as image feature values ​​for extraction; finally, 16 friction vibration time-frequency features were obtained; Step 6: Perform dimension reduction screening on vibration behavior characteristics based on the MIC maximum mutual information coefficient-BP neural network method; the specific steps include: Step 601, obtaining 16 eigenvalues ​​of friction and wear vibration of the metal powder sintered composite layer reinforced cemented carbide through step 5; Step 602: Taking the friction force, a parameter representing the stability of the friction and wear test system, as the output of feature screening, and taking the 16 eigenvalues ​​of STFT and CWT as input, feature selection based on the maximum mutual information coefficient of MIC is performed, which is specifically implemented as follows: By discretizing the input feature X and the output feature Y in two-dimensional space, the data is divided into grids according to the distribution of the feature values, and the mutual information between the feature and the target variable is calculated in each grid. Finally, all grids are traversed to find the grid with the maximum mutual information, so as to obtain the MIC value of each feature as the contribution value of the feature; Based on the experience of feature selection, when the difference between the maximum feature contribution and the minimum feature contribution is large, half of the difference between the maximum MIC value and the minimum MIC value between the features is used as the threshold for preliminary screening of feature importance, that is, the features with importance greater than 0.3 are retained in the preliminary screening; Step 602: Use the RMSE value of friction force prediction of the BP neural network to perform the next step of feature screening. Each time a feature is cut, the RMSE values ​​of 30 predictions are obtained to calculate the average. Step 7: Verify the prediction accuracy of feature selection regression based on random forest RF; Step 8: Fit the stability characterization curve of the test system based on the vibration behavior.

2. The time-frequency characterization method for friction vibration behavior of metal powder sintered composite layer cemented carbide according to claim 1 is characterized in that: The preprocessing of the collected friction vibration data file in step 1 refers to using LabVIEW to convert other types of collected files into an Excel spreadsheet's '.xlsx' format file.

3. The time-frequency characterization method for friction vibration behavior of metal powder sintered composite layer cemented carbide according to claim 1, characterized in that: In step 2, VMD variational mode decomposition is used to remove the obvious interference parts in the vibration and perform noise reduction.

4. The method for time-frequency characterization of friction vibration behavior of metal powder sintered composite layer cemented carbide according to claim 1, characterized in that: In step 3, short-time Fourier transform and continuous wavelet transform are used to perform time-frequency analysis on the friction signal to obtain a time-frequency image.

5. The time-frequency characterization method for friction vibration behavior of metal powder sintered composite layer cemented carbide according to claim 1, characterized in that: The grayscale processing of the time-frequency image in step 4 specifically includes: Decolorize the image, simplify the complexity of the image, and shorten the calculation time; The grayscale algorithm is used to process the image according to 256 grayscale levels. The characteristic information such as image color and shape changes will be saved in the grayscale image in the form of grayscale pixels, and feature conversion is performed using the pixel distribution and pixel number of each grayscale level as characteristic information.

6. The method for time-frequency characterization of friction vibration behavior of metal powder sintered composite layer cemented carbide according to claim 1, characterized in that: The specific steps of fitting the stability characterization curve of the test system based on vibration behavior in step 8 include: Step 801: take the grayscale mean, entropy mean, contrast mean, The entropy variance and energy mean characteristics are numerically based on the vector normalization formula: In formula (1), c represents the coefficient, which is used to increase the data level and is set to 100; n represents the number of samples; a i Represents the actual measurement data; y i Represents the normalized data; the data is vector-normalized by using the norm(x) function in MATLAB to achieve normalization and eliminate the dimension effect; Step 802: Based on the normalized data, feature weights are divided; the specific implementation is: based on the importance of the MIC detection contribution value, the total contribution of the screening features is set to C, and the contribution value of each feature is x i , i represents different features, and the contribution value of each feature accounts for the percentage of the total contribution value of the screening features, that is: where y i That is, the weight assigned to each retained feature; Step 803: multiply each normalized feature by its own weight and then sum them up to obtain six-feature fusion parameters, and fit the vibration behavior time-frequency characteristic law curve based on the fusion parameters; Step 804: by gray level mean, entropy mean, contrast mean, The vector normalization processing, weight division and weighted fitting of entropy variance and energy mean are performed to obtain the time-frequency characteristic law curve of vibration behavior, thereby characterizing the stability change curve of the sintered metal powder composite layer alloy strengthening performance detection system based on vibration behavior.

Images