Fabric wear resistance detection method based on data analysis

Abstract

The invention discloses a fabric wear resistance detection method based on data analysis, and the method comprises the steps: S1, collecting multi-dimensional fabric data of a fabric in a running-in period and a stable period, including force data, temperature data and deformation data; s2, optimizing the basic neural network based on association information between the microscopic features and the macroscopic features of the multi-dimensional fabric data to obtain an improved neural network model, and obtaining optimized association features based on the improved neural network model; and S3, converting the optimized correlation characteristics into an information entropy form, and building a quantitative relation model between the fabric structure change and the service life by combining a support vector machine to predict the wear-resistant service life of the fabric, the method comprehensively analyzes the characteristic relation, considers the random factor of the characteristic change, and makes the fabric wear resistance detection result deeper.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and in particular to a fabric wear resistance detection method based on data analysis. Background Art

[0002] The friction coefficient is a key parameter of the material surface properties, and its change directly reflects the strength of the interface interaction. At present, traditional detection methods only focus on the static friction coefficient and ignore the fluctuation characteristics of the friction coefficient in the dynamic process. When the fabric first rubs against other objects, it is in the running-in period. At this stage, the fiber structure on the surface of the fabric has not yet adapted to the friction, and the surface micro-roughness is relatively high. Taking cotton fiber fabric as an example, during the initial friction, the fiber ends are intertwined and connected, resulting in a relatively high friction coefficient. As the friction time increases, the fiber ends are gradually smoothed and straightened, and the friction coefficient will drop rapidly, with a drop of up to 30%-40%. The traditional detection method only measures an initial static friction coefficient, which cannot reflect this dynamic change process, resulting in deviations in the evaluation of the initial performance of the fabric.

[0003] Moreover, after the running-in period, the fabric enters a stable period. Although the friction coefficient is relatively stable at this time, it is not absolutely unchanged. Due to the influence of environmental factors (such as slight fluctuations in humidity and temperature) and tiny particles adsorbed on the surface of the fabric, the friction coefficient will still fluctuate within a certain range. For example, in an environment where the humidity increases from 40% to 60%, the friction coefficient of some synthetic fiber fabrics may increase by 10%-15%. The static friction coefficient of traditional testing cannot reflect the fluctuations caused by such environmental changes, resulting in inaccurate judgment of fabric performance in actual use scenarios. Therefore, a fabric wear resistance detection method based on data analysis is proposed herein. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention proposes the following technical solutions: A fabric wear resistance detection method based on data analysis, comprising: S1: Collect multi-dimensional fabric data of fabrics during the running-in period and the stabilization period, including force data, temperature data, and deformation data; S2: Optimize the basic neural network based on the correlation information between the micro-features and macro-features of multi-dimensional fabric data to obtain an improved neural network model, and obtain optimized correlation features based on the improved neural network model; S3: The optimized correlation features are converted into the form of information entropy, and combined with the support vector machine to construct a quantitative relationship model between fabric structure changes and life to predict the wear life of the fabric.

[0005] The force data includes the running-in period force data and stable period force data , the running-in period force data Including run-in positive pressure data and running-in tangential force data , the stable period force data Including stable positive pressure data and stable tangential force data ; The temperature data includes the running-in period temperature data and stable temperature data , the temperature data of the running-in period Including temperature rate and temperature maximum , the stable temperature data Including temperature fluctuation data and average temperature ; The deformation data includes deformation data during the running-in period and stable deformation data , the deformation data of the running-in period Including overall deformation data during running-in and local deformation data , the stable deformation data Including stable overall deformation data and small local deformation data .

[0006] The process of obtaining the correlation information between the microscopic features and macroscopic features of the multi-dimensional fabric data is as follows: Normalize the multi-dimensional fabric data into a comprehensive input tensor ; The synthesized input tensor Microscopic feature tensor and the macroscopic feature tensor ; Through the basic neural network model, the microscopic feature tensor and the macroscopic feature tensor Perform convolution normalization operation and obtain the flattened micro-feature tensor through the flattening function in the deep learning framework and the flattened macroscopic feature tensor ; Calculate the flattened microscopic feature tensor and the flattened macroscopic feature tensor The bilinear product of , the associated information is obtained through a multi-layer perceptron MLP containing two fully connected layers .

[0007] The microscopic feature tensor In the convolution normalization operation, a 3×3 size convolution kernel is used for convolution operation; The macroscopic feature tensor In the convolution normalization operation, a 5×5 size convolution kernel is used for convolution operation; The number of neurons in the first fully connected layer of the multilayer perceptron MLP with two fully connected layers , input the bilinear product result After that, output the first layer output result ; The number of neurons in the second fully connected layer of the multilayer perceptron MLP comprising two fully connected layers is , the output of the first layer As input, output related information .

[0008] The optimization correlation feature acquisition process is as follows: Associate information The micro-macro feature association matrix is ​​obtained through a Softmax function , based on the micro-macro feature correlation matrix Obtain an improved neural network model, and use the improved neural network model to analyze the microscopic feature tensor and the macroscopic feature tensor Perform convolution operation to obtain preliminary correlation features , for the preliminary correlation features Perform batch normalization and The activation function is processed to obtain the final optimized correlation features.

[0009] The improved neural network model is based on the micro-macro feature correlation matrix Get new weight , based on the new weights The original convolution kernel weights of the convolution operation in the basic neural network model are replaced to obtain an improved neural network model.

[0010] The form transformation process of optimizing the information entropy of the associated features is: The optimization correlation features are quantified, and the probability of different values ​​appearing in the quantized optimization correlation features is obtained as follows: , based on the occurrence probability of each value, the information entropy is obtained through the Shannon information entropy formula.

[0011] The process of constructing the quantitative relationship model between fabric structure change and lifespan is as follows: Combine the information entropy of optimized associated features into an information entropy vector , select the support vector machine model, and transform the information entropy vector As input feature, the wear life of the fabric As the output label, the radial basis kernel is selected as the kernel function. At the same time, the penalty factor of the support vector machine is determined, and the optimal penalty factor is selected through cross-validation. The support vector machine model is trained and the prediction results are verified after the training is completed. The actual wear life Compare and minimize the evaluation indicators to obtain the constructed quantitative relationship model.

[0012] The present invention has the following beneficial effects: In the present invention, firstly, the multi-dimensional fabric data is divided into microscopic and macroscopic feature tensors, and convolution normalization and flattening operations are performed using a basic neural network model, and then the correlation information between microscopic and macroscopic features is obtained through bilinear product and multi-layer perceptron (MLP) processing, and a correlation matrix is ​​generated. Based on this, the basic neural network model is optimized, so that the improved neural network model can more effectively capture the potential connection between complex features when processing fabric data; Secondly, by converting the optimized correlation features into information entropy, the uncertainty of different characteristics of the fabric during the wear process can be effectively quantified. For example, the randomness of wear caused by changes in factors such as temperature and force can be quantified through information entropy. A quantitative relationship model is constructed by combining support vector machine (SVM) and radial basis kernel function, and cross-validation is used to determine the optimal penalty factor to further optimize the model performance, making the model more capable of handling complex nonlinear problems and better adapting to the needs of fabric performance prediction in actual application scenarios. Finally, by combining the two, we can comprehensively analyze the characteristic relationship and take into account the random factors of characteristic changes, making the results of fabric wear resistance testing more in-depth. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 This is a method step diagram of a fabric wear resistance detection method based on data analysis proposed by the present invention. DETAILED DESCRIPTION

[0014] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0015] Embodiment 1 like Figure 1 As shown, the present invention proposes a fabric wear resistance detection method based on data analysis, comprising: S1: Collect multi-dimensional fabric data of fabrics during the running-in period and the stabilization period, including force data, temperature data, and deformation data; The pressure sensor is installed at the contact point between the loading device and the fabric of the friction tester to measure the force data of the fabric during the running-in period and the stable period. , For the running-in period force data, It is the force data of the stable period; Running-in period force data Including run-in positive pressure data and running-in tangential force data ; Specifically, the running-in positive pressure data during the running-in period The change in can reflect the vertical load that the fabric bears during the initial contact friction, and the running-in period tangential force data Can intuitively reflect the evolution of friction characteristics of the fabric surface during the running-in period; Stable period force data Including stable positive pressure data and stable tangential force data ; Specific, stable positive pressure data It can reflect the vertical load borne by the fabric under relatively stable friction conditions and stabilize the tangential force data. It can evaluate the wear degree of the fabric surface during the friction process and the influence of the fiber structure change on the friction force; Through the temperature sensor, the temperature data of the fabric surface is obtained non-contactly , is the temperature data during the running-in period, is the temperature data of the stable period; Temperature data during the running-in period Including temperature rate and temperature maximum ; Specifically, during the running-in period, the temperature usually rises rapidly, and the temperature rate Reflects the speed of frictional heat generation. During the running-in period, the maximum temperature Used to evaluate the thermal damage of fabrics in the initial friction stage. Excessive temperature may cause changes in the chemical structure of the fabric, such as melting and decomposition of fibers, thus affecting the wear resistance of the fabric; Temperature data during stable period Including temperature fluctuation data and average temperature ; Specifically, the temperature during the stable period is not absolutely stable, but will fluctuate within a certain range. It can be used to understand the thermal stability of the fabric during stable friction and its adaptability to environmental changes, and the average temperature value during the stable period It can comprehensively reflect the thermal state of the fabric during stable friction; Use deformation sensors to measure the deformation of fabrics during the running-in period and the stabilization period to obtain the deformation data of fabrics , Deformation data during the running-in period. is the deformation data in the stable period; Deformation data during the running-in period Including overall deformation data during running-in and local deformation data ; Specifically, the overall deformation data of the running-in period It reflects the macro deformation of the fabric in the initial friction stage and the local deformation data during the running-in period. It can locate the weak spots of fabric during the running-in period, which are usually the starting points of wear; Deformation data during stable period Including stable overall deformation data and small local deformation data ; Specific, overall deformation data during the stable period It can reflect the macro deformation of fabrics under relatively stable friction conditions, and the small local deformation data during the stable period. It can provide a more detailed understanding of the micromechanical behavior of fabrics during stable friction; Multidimensional data is represented as ; Specifically, by collecting and analyzing multi-dimensional data, the mechanical, thermal and deformation characteristics of the fabric during the friction process can be fully reflected; For example, force data can reflect the load and friction changes on the fabric, temperature data can reflect the heat generated by friction and its impact on the material, and deformation data shows the changes in the fabric structure. By combining this information, the wear resistance of the fabric can be more accurately evaluated, avoiding evaluation bias caused by single-dimensional data detection.

[0016] S2: Based on the correlation information between the micro-features and macro-features of multi-dimensional fabric data, the improved neural network is optimized to obtain an improved neural network model, and the optimized correlation features are obtained based on the improved neural network model; Normalize the force, temperature, and deformation data (multi-dimensional fabric data) during the running-in period and the stabilization period to form a comprehensive input tensor ,in is the combined time step of the running-in period data and the stable period data, is the comprehensive spatial dimension of the running-in period data and the stable period data, , for three channels of stress, temperature and deformation; Specifically, according to the requirements of fabric wear resistance testing and the frequency of data collection, the appropriate comprehensive time step can be determined. For example, if the data collection frequency is 10 times per second and it is hoped to analyze the changes in the friction characteristics of the fabric within 100 seconds, the time step can be set to 100; Comprehensive spatial dimension is the input tensor dimension of the running-in period data and the stabilization period data. Specifically, the force data contains 4 tensors, the temperature data contains 4 tensors, and the force-deformation data contains 4 tensors, namely ; The normalized force, temperature, and deformation data are taken as a channel respectively and combined into the channel dimension ,and , representing 3 different data; The synthesized input tensor Microscopic feature tensor and the macroscopic feature tensor ; Specifically, the microscopic characteristics include local deformation data during the running-in period , small local deformation data during stable period , Tangential force data during the running-in period , tangential force data during stable period , Temperature rate during running-in period ,These data reflect the changes in the fabric at the microscopic scale or in local areas; Macro characteristics: including overall deformation data during the running-in period , overall deformation data during stable period , positive pressure data during the running-in period , Stable period positive pressure data , Maximum temperature during the running-in period , Temperature fluctuation data during stable period and average temperature , reflecting the characteristics of the fabric at a larger scale or overall level; Through a basic neural network model, and through the basic neural network model of the microscopic feature tensor and the macroscopic feature tensor Perform feature flattening; Specifically: for the microscopic feature tensor The convolution operation is performed through a small-sized convolution kernel (3×3), and then processed by batch normalization (BN) and flattened. ,in, It is the flattening function in the deep learning framework, which converts it into a one-dimensional vector , that is, to expand the three-dimensional tensor into a long vector in a certain order to facilitate the subsequent bilinear product calculation; Macroscopic characteristics The convolution operation is performed through a large-size convolution kernel (5×5), and then the batch normalization (BN) is processed and flattened. After that, it becomes a one-dimensional vector ; Get the flattened microscopic feature tensor and the macroscopic feature tensor Then, calculate their bilinear product , based on the formula: ,in, represents the outer product operation, The representation is a matrix of size ; Specifically, the bilinear product can comprehensively calculate and express the relationship between the various dimensions of microscopic features and macroscopic features. Each element in The first The element of the macroscopic feature tensor The degree of correlation between the elements can be preliminarily obtained through bilinear product, which can facilitate further refinement through multi-layer perceptron (MLP); The associated information is obtained through a multi-layer perceptron MLP containing two fully connected layers ; Specifically: The number of neurons in the first fully connected layer is , input the bilinear product result Then, the formula is used to calculate: ; in, Represents the weight matrix of the first fully connected layer, and the bilinear product result Map to dimensional space, The function is used to introduce nonlinearity, so that the model can learn more complex feature relationships, filter out unimportant linear relationships, and highlight key association patterns; The number of neurons in the second fully connected layer is , taking the output of the first layer as input, through the formula ; in, Represents the weight matrix of the second fully connected layer, which is used to transform the output of the first fully connected layer Map to Dimensional space is used for calculation, and the nonlinear expression ability is enhanced again by functions, and finally the refined correlation information of micro-macro features is obtained. ; Specifically, in the actual process of fabric wear resistance analysis, the correlation between microscopic features and macroscopic features is crucial for accurately evaluating fabric performance. and the macroscopic feature tensor After the MLP refinement process, the potential connections between features can be captured more effectively, making the subsequent feature weight adjustment based on the correlation matrix more targeted, and improving the accuracy and effectiveness of the subsequent neural network model in processing fabric data and predicting fabric performance. Associate information The micro-macro feature association matrix is ​​obtained through a Softmax function; Specifically, related information is a In order to transform it into a matrix that can intuitively represent the correlation strength between micro-macro features, the conversion is performed based on the Softmax function. The formula is: , is the micro-macro feature correlation matrix; Specifically, the function of the Softmax function is to transform the input vector Each element of is converted into a probability value, so that the sum of the probabilities of all elements is 1, by specifying , which means that the last dimension of the MLP output is normalized, so the correlation matrix In , each row element represents the association probability between a certain dimension of micro features and each dimension of macro features, and each column element represents the association probability between a certain dimension of macro features and each dimension of micro features. In this way, the strength of the association between micro and macro features can be accurately characterized, providing a basis for the subsequent improvement of the feature weight adjustment of the neural network model. Based on the micro-macro feature correlation matrix Dynamically adjust the original convolution kernel weights of the convolution operation in the basic neural network model to obtain an improved neural network model; Specifically, let the original convolution kernel weight be , the new weight is calculated as , by changing the original convolution kernel weight Each element of the correlation matrix Multiply the corresponding elements in the correlation matrix The correlation strength information in the convolution kernel is integrated into the convolution kernel weight, so that the convolution kernel can perform targeted weighted processing on different features according to the correlation degree between micro-macro features when extracting features; Apply New Weights After replacing the original convolution kernel weights of the convolution operation in the basic neural network model, an improved neural network model is obtained. The microscopic feature tensor is improved by the improved neural network model. and the macroscopic feature tensor Perform convolution operation to obtain preliminary correlation features , the process is:

[0017] in, Represents the improved convolution operation in the neural network model. New weights used The micro-macro feature correlation matrix The correlation strength information in the convolution kernel is integrated into the convolution kernel weight, so that the convolution kernel can perform targeted weighted processing on different features according to the correlation degree between micro-macro features when extracting features. For example, for the closely related micro-fiber deformation features and macro-wear features, higher weights will be given during extraction to highlight these key features. Preliminary correlation features Perform batch normalization and The activation function is processed to obtain the final optimized correlation features. The process is expressed as:

[0018] in, represents the batch normalization operation, represents the activation function, Represents optimized associated features, the number of features is , ; Specifically, the convolution kernel weights are adjusted by integrating the micro-macro feature association matrix A into the traditional neural network model. The convolution kernel weight update of the traditional neural network model mainly relies on the back propagation of the overall data loss, while the improved neural network model uses the association matrix A (the association probability between microscopic fiber deformation and macroscopic wear depth) to make the convolution kernel Updated to , targetedly enhance the extraction weight of associated features, and first obtain the original association through feature flattening and bilinear product, then refine it through MLP, Softmax to generate the association matrix, and finally combine batch normalization and activation function to optimize feature extraction. The implementation steps are not simply superimposed, which comprehensively improves the association processing ability of the neural network model for complex fabric data.

[0019] S3: Convert the optimized correlation features into information entropy, and combine with support vector machine to build a quantitative relationship model between fabric structure change and life to predict the wear life of fabric; Optimize the associated features Perform quantification processing to optimize the associated features including continuous features and discrete features; For continuous features, they are divided into several intervals, each of which corresponds to a quantization level. For discrete features (such as the level of the number of fiber breaks), their discrete values ​​are used; For example, the temperature is divided into three intervals: low temperature, medium temperature, and high temperature, represented by 1, 2, and 3 respectively; the number of fiber breaks is divided into three levels: small, medium, and large, represented by 1, 2, and 3 respectively; According to the quantified optimized correlation features , let optimize the associated features have Different values, different values ​​in The number of occurrences in , No. The probability of a value occurring is ,in, Representation characteristics No. The number of times different values ​​appear in the sample data. For example, for the feature of fiber breakage number, it is quantified into three levels: small, medium, and large. When counting the sample data, the number of times the small level appears is , the level is , most of the levels are , In is from arrive variables, Representative features No. The number of times different values ​​appear in the sample data; Specifically, in actual data statistics, when the number of samples is large enough, the frequency will approach the probability. In the fabric wear resistance test, by statistically analyzing the optimized correlation features of a large number of different fabric samples, the probability of each eigenvalue occurring can be accurately obtained, which truly reflects the distribution of the eigenvalues ​​and provides a reliable basis for subsequent analysis. Based on the probability of occurrence of each value, the information entropy is obtained through the Shannon information entropy formula: ; For example, an optimized correlation feature has three quantized values, and the probabilities are , , , then its information entropy is: ; The process of constructing the quantitative relationship model between fabric structure change and lifespan is as follows: The information entropy of all optimized related features is combined into an information entropy vector , where m is the number of optimized correlation features; Specifically, during the wear process of fabrics, different optimized correlation features have different degrees of uncertainty. For example, changes in factors such as temperature and force will cause the wear of fabrics to have a certain degree of randomness. By obtaining the information entropy of each optimized correlation feature, this uncertainty can be quantified and the accuracy of the analysis can be improved. Select the support vector machine model and transform the information entropy vector As input features, the wear life of the fabric is used as the output label ; For the complex nonlinear problem of fabric wear life prediction, the radial basis kernel is selected as the kernel function, and its formula is: ,in, and are two different input feature vectors (information entropy vectors), , are the parameters of the kernel function; At the same time, the penalty factor of the support vector machine is determined. The penalty factor of the support vector machine is used to balance the tolerance and complexity of the model to the training error. The output labels are used as training data. The training data is divided into multiple subsets through the cross-validation method. Different penalty factors are traversed to calculate the performance of the model on the validation set under each parameter combination, and the penalty factor that makes the model perform best is selected. , using the selected optimal The support vector machine model is trained. After the training is completed, the optimized correlation features of the new samples are converted into information entropy vectors and input into the model. The model outputs the predicted wear life. , and verify the predicted results, and predict the wear life The actual wear life For comparison, minimize the evaluation index, expressed as:

[0020] in, To optimize the number of associated features, a quantitative relationship model between fabric structure change and life span is obtained; Specifically, the constructed quantitative relationship model between fabric structure change and lifespan has learned the mapping relationship between information entropy vector and fabric wear lifespan. When the information entropy vector of a new sample is input, the model can make predictions based on the learned relationship. At the same time, by calculating the evaluation index, the difference between the predicted result and the actual result can be quantified, making it more in line with actual needs. For example, suppose you need to predict the wear life of a new batch of cotton fabrics: Collected data on 100 different cotton fabric samples and obtained the optimized associated features and their actual wear life (the time recorded in hours under standard friction test conditions until the fabric reaches a specified level of wear); For new cotton fabric samples, measure their optimized associated feature data, such as temperature (quantized to 2), the force is (assuming the quantization is 2), the number of broken fibers is 8 (assuming the quantization is 2), and the three types of optimized associated feature information entropy are calculated according to the above method and combined into the information entropy vector , and input it into the constructed quantitative relationship model, and the model output predicts the wear life L=150 (hours).

[0021] In the application, several formulas involved are calculated by removing dimensions and taking their numerical values. The formulas are established by collecting a large amount of data and performing software simulation to obtain a formula for the most recent actual situation. Some coefficients or weights in the formulas are set by technicians in this field according to actual conditions, so they will not be elaborated here.

[0022] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product. A person of ordinary skill in the art may appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein may be implemented in electronic hardware, or in a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0023] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A fabric wear resistance detection method based on data analysis, characterized in that: The method steps include: S1: Collect multi-dimensional fabric data of fabrics during the running-in period and the stabilization period, including force data, temperature data, and deformation data; S2: Optimize the basic neural network based on the correlation information between the micro-features and macro-features of multi-dimensional fabric data to obtain an improved neural network model, and obtain optimized correlation features based on the improved neural network model; S3: The optimized correlation features are converted into the form of information entropy, and combined with the support vector machine to construct a quantitative relationship model between fabric structure changes and life to predict the wear life of the fabric.

2. A fabric wear resistance detection method based on data analysis according to claim 1, characterized in that: The force data includes the running-in period force data and stable period force data , the running-in period force data Including run-in positive pressure data and running-in tangential force data , the stable period force data Including stable positive pressure data and stable tangential force data ; The temperature data includes the running-in period temperature data and stable temperature data , the temperature data of the running-in period Including temperature rate and temperature maximum , the stable temperature data Including temperature fluctuation data and average temperature ; The deformation data includes deformation data during the running-in period and stable deformation data , the deformation data of the running-in period Including overall deformation data during running-in and local deformation data , the stable deformation data Including stable overall deformation data and small local deformation data .

3. The method for detecting fabric wear resistance based on data analysis according to claim 1 is characterized in that: The process of obtaining the correlation information between the microscopic features and macroscopic features of the multi-dimensional fabric data is as follows: Normalize the multi-dimensional fabric data into a comprehensive input tensor ; The synthesized input tensor Microscopic feature tensor and the macroscopic feature tensor ; Through the basic neural network model, the microscopic feature tensor and the macroscopic feature tensor Perform convolution normalization operation and obtain the flattened micro-feature tensor through the flattening function in the deep learning framework and the flattened macroscopic feature tensor ; Calculate the flattened microscopic feature tensor and the flattened macroscopic feature tensor The bilinear product of , the associated information is obtained through a multi-layer perceptron MLP containing two fully connected layers .

4. The method for detecting fabric wear resistance based on data analysis according to claim 3 is characterized in that: The microscopic feature tensor In the convolution normalization operation, a 3×3 size convolution kernel is used for convolution operation; The macroscopic feature tensor In the convolution normalization operation, a 5×5 size convolution kernel is used for convolution operation; The number of neurons in the first fully connected layer of the multilayer perceptron MLP with two fully connected layers , input the bilinear product result After that, output the first layer output result ; The number of neurons in the second fully connected layer of the multilayer perceptron MLP comprising two fully connected layers is , the output of the first layer As input, output related information .

5. The method for detecting fabric wear resistance based on data analysis according to claim 1 is characterized in that: The optimization correlation feature acquisition process is as follows: Associate information The micro-macro feature association matrix is ​​obtained through a Softmax function , based on the micro-macro feature correlation matrix Obtain an improved neural network model, and use the improved neural network model to analyze the microscopic feature tensor and the macroscopic feature tensor Perform convolution operation to obtain preliminary correlation features , for the preliminary correlation features Perform batch normalization and The activation function is processed to obtain the final optimized correlation features.

6. The method for detecting fabric wear resistance based on data analysis according to claim 1 is characterized in that: The improved neural network model is based on the micro-macro feature correlation matrix Get new weight , based on the new weights The improved neural network model is obtained by replacing the original convolution kernel weights of the convolution operation in the basic neural network model.

7. The method for detecting fabric wear resistance based on data analysis according to claim 1, characterized in that: The form transformation process of optimizing the information entropy of the associated features is: The optimization correlation features are quantified, and the probability of different values ​​appearing in the quantized optimization correlation features is obtained as follows: , based on the occurrence probability of each value, the information entropy is obtained through the Shannon information entropy formula.

8. The method for detecting fabric wear resistance based on data analysis according to claim 1, characterized in that: The process of constructing the quantitative relationship model between fabric structure change and lifespan is as follows: Combine the information entropy of optimized associated features into an information entropy vector , select the support vector machine model, and transform the information entropy vector As input feature, the wear life of the fabric As the output label, the radial basis kernel is selected as the kernel function. At the same time, the penalty factor of the support vector machine is determined, and the optimal penalty factor is selected through cross-validation. The support vector machine model is trained and the prediction results are verified after the training. The actual wear life Compare and minimize the evaluation indicators to obtain the constructed quantitative relationship model.

Images