High-speed grinding wheel state recognition detection method based on INOA-SVM model

Abstract

The present application relates to the technical field of grinding wheel state detection, and particularly relates to a high-speed grinding wheel state recognition detection method based on an INOA-SVM model, comprising the following steps: S1, obtaining and wear dividing of original vibration signals; S2, data processing of the original vibration signals; S3, constructing and verifying the INOA-SVM model; S4, recognition detection of the high-speed grinding wheel wear state. The present application can decompose complex original vibration data into simple signals for iteration and eliminate non-grinding vibration by processing high-speed grinding wheel vibration data, and can also accelerate the recognition detection efficiency of the grinding state recognition detection model, optimize the SVM model through the INOA model, so that the grinding state recognition detection model jumps out of local convergence, and has a faster convergence speed and higher stability, so as to improve the accuracy and stability of high-speed grinding wheel state recognition detection.

Description

Technical Field

[0001] This invention relates to the field of grinding wheel condition detection technology, and in particular to a method for identifying and detecting the condition of high-speed grinding wheels based on the INOA-SVM model. Background Technology

[0002] Grinding machines, as high-precision machining equipment, are widely used in surface treatment and finishing of metal, glass, and ceramic surfaces. The grinding process refers to the process in which the grinding wheel head drives the high-speed rotating grinding wheel to grind the surface of the workpiece at a certain feed speed, thereby removing the material from the surface of the workpiece. Therefore, the degree of wear of the grinding wheel during the grinding process has a very important impact on the surface processing quality.

[0003] Currently, in grinding processes, to ensure machining quality, the dressing of grinding wheels mostly depends on the experience of the machining personnel and is performed periodically. However, this dressing method leads to low machining efficiency and a certain degree of waste. Therefore, the detection of grinding wheel wear is of great significance.

[0004] Currently, the detection of grinding wheel wear includes:

[0005] 1. Direct detection method:

[0006] Based on whether there is contact, it can be divided into:

[0007] A. Contact inspection: Contact inspection usually involves contact with the surface of the grinding wheel to obtain the physical properties of the grinding wheel surface (e.g., surface morphology, hardness, roughness, etc.). It has high inspection accuracy, but its operation process is complicated, time-consuming and labor-intensive, and there is a potential risk of wear on the surface of the grinding wheel.

[0008] B. Non-contact inspection: This method mainly relies on high-precision inspection equipment such as cameras, microscopes, white light interferometers, and laser rangefinders to detect grinding wheel wear. It has high detection accuracy, but it requires disassembling the grinding wheel and then using high-precision inspection equipment for inspection. Obviously, non-contact inspection is not suitable for actual processing. In addition, although this method can achieve online inspection, the inspection equipment such as cameras will be affected by splashed cutting fluid and flying grinding debris, which will lead to a decrease in the measurement environment.

[0009] 2. Indirect detection method: Its core is to use various sensors to analyze the changes in signals during the grinding process of the grinding wheel. These signals can indirectly reflect the wear degree of the grinding wheel. Indirect detection can achieve online detection without interfering with normal grinding operations.

[0010] Currently, vibration signals are the most commonly used indirect detection. As the wear of the grinding wheel increases, the field of view of the vibration signal tends to increase. Therefore, vibration signals are often used in indirect detection. However, due to the large amount of data collected, they cannot be directly applied to the detection of grinding wheel wear. The vibration signal must be processed before feature extraction. The extracted features are used to establish the relationship between the vibration signal and the wear of the high-speed grinding wheel. However, the traditional method of detecting grinding wheel wear by extracting vibration signal features cannot fully express the grinding wheel wear state corresponding to the vibration signal, thus affecting the accuracy of grinding wheel wear state identification and detection. Summary of the Invention

[0011] To address the shortcomings of existing production technologies, the applicant provides a method for identifying and detecting the condition of high-speed grinding wheels based on the INOA-SVM model. By improving the method for identifying and detecting the condition of high-speed grinding wheels, the accuracy of identifying and detecting the condition of high-speed grinding wheels is enhanced.

[0012] The technical solution adopted in this invention is as follows:

[0013] A method for identifying and detecting the condition of a high-speed grinding wheel based on the INOA-SVM model includes the following steps:

[0014] S1. Acquisition of raw vibration signals and wear classification: Acquire the raw vibration signal dataset of the high-speed grinding wheel in the X-axis, Y-axis, and Z-axis directions. The original vibration signal dataset was analyzed according to the wear stages of the high-speed grinding wheel. Wear stages are divided to obtain a vibration signal dataset after the wear stages are divided. ;

[0015] S2. Data processing of the original vibration signal dataset in S1: The original vibration signal dataset in S1 is processed sequentially. Hard thresholding fast iterative filtering, reconstruction, and time-domain and frequency-domain feature extraction are performed to obtain the processed vibration signal dataset. ;

[0016] S3. Construct and validate the INOA-SVM model: Analyze the regularization coefficients in the SVM model using the INOA model. Kernel function Optimization was performed to obtain the INOA-SVM model, which uses the vibration signal dataset after dividing S1 into wear stages. The processed vibration signal dataset in S2 Input the data into the INOA-SVM model and validate the model. If the validation meets the requirements, the INOA-SVM model is the optimal INOA-SVM model, and S4 is executed. If the validation does not meet the requirements, the regularization coefficients are re-optimized. Kernel function The process continues until the verification meets the requirements; once this INOA-SVM model is confirmed, it is the optimal INOA-SVM model.

[0017] S4. Identification and detection of wear condition of high-speed grinding wheel: Real-time detection of vibration signals of the high-speed grinding wheel in the X, Y, and Z axes. and the vibration signal The input is fed into the optimal INOA-SVM model in S3 to output the vibration signal. The wear condition corresponding to high-speed wear grinding wheels;

[0018] in: Indicates time, Indicates the number of times the original vibration signal was collected. This indicates the wear stage of a high-speed grinding wheel.

[0019] Therefore, by performing hard-threshold fast iterative filtering and reconstruction processing on high-speed grinding wheel vibration data, the complex original vibration data (i.e., linear signals, nonlinear signals, and non-stationary signals) can be decomposed into simple signals for iterative processing, and non-grinding vibrations can be eliminated. Simultaneously, it can accelerate the recognition and detection efficiency of the grinding condition identification and detection model, and improve the regularization coefficients in the SVM model through the INOA model. Kernel function Optimization is performed to enable the grinding condition identification and detection model to escape local convergence and to achieve faster convergence speed and higher stability, thereby improving the accuracy and stability of high-speed grinding wheel condition identification and detection.

[0020] Further, S3 includes the following steps:

[0021] S3-1. SVM model parameter initialization, including: star-crested jayfinch population size, first iteration number, and regularization coefficient. Kernel function ;

[0022] S3-2. Refraction back learning is performed based on the Tent-Sine-Logistic composite chaotic mapping to optimize the initial star gargoyle population in S3-1;

[0023] S3-3. Input all star-shaped parrotbill populations into the SVM model, and calculate the fitness of each star-shaped parrotbill individual through triple cross-validation. ;

[0024] S3-4. Randomly generate two random numbers. and If random number If yes, then execute S3-5; otherwise, execute S3-6.

[0025] S3-5, Foraging and Storage Strategies: Introducing Adaptive Variable Inertia Weight Strategy Parameters And set a nonlinear weighting factor. ;

[0026] S3-6. Storage Search and Retrieval Strategy Based on INOA Model: Introducing Adaptive Variable Inertia Weight Strategy Parameters into the storage search and retrieval strategy. And set a nonlinear weighting factor. ;

[0027] S3-7, Update the fitness of the Star Parrotbill population ;

[0028] S3-8, will Elite-based dimensional reverse learning and dynamic local optimal escape are introduced into the INOA algorithm;

[0029] S3-9. If the current first iteration count reaches the maximum first iteration count, then execute S3-10; otherwise, return to S3-4. Continue until the current first iteration count reaches the maximum first iteration count, then execute S3-10.

[0030] S3-10, to obtain the globally optimal regularization coefficient Kernel function And obtain the INOA-SVM model;

[0031] S3-11. Divide the vibration signal dataset in S1 into wear stages. The processed vibration signal dataset in S2 The input is fed into the INOA-SVM model in S3-10, and the INOA-SVM model is validated. If the validation meets the requirements, the INOA-SVM model is the optimal INOA-SVM model, and S4 is executed. If the validation does not meet the requirements, S3-4 is returned, and this process continues until the validation meets the requirements, at which point the INOA-SVM model is considered the optimal INOA-SVM model. Therefore, the regularization coefficient of the SVM model is... Kernel function The INOA model is used to optimize the high-speed grinding wheel condition recognition and detection results within a given range. The INOA model enhances the richness of the initial star-shaped magpie population through composite chaotic mapping and refraction back learning, and accelerates the optimization speed through an adaptive variable inertia weight strategy. Elite-based dimensional reverse learning and a dynamic local optimum escape mechanism enable the INOA model to escape local optima, thereby allowing the INOA model to find the optimal regularization coefficient. Kernel function This capability enables the SVM model to accurately identify and detect the wear state of high-speed grinding wheels.

[0032] Furthermore, in S3-2, the expression for the composite chaotic mapping is:

[0033] ;

[0034] in: Indicates a Tent mapping. Represents a Sine mapping. Represents the Logistic mapping;

[0035] By using refraction-backward learning, the original solution and the refraction solution are merged to form a hybrid star parrotbill population of size 2NP. The individuals are then sorted according to their fitness. The 2NP hybrid star parrotbill populations are arranged in descending order based on fitness, and the top NP individuals are selected as the final initial population.

[0036] The expression for the individual Star Crowfinch after refraction-back learning is:

[0037] ;

[0038] in: Indicates the lower bound. Indicates the upper bound. Indicates the ratio of refraction distances. Indicates refractive index, This indicates a pre-star crow-sparrow individual that has undergone reverse learning.

[0039] In S3-3, a radial basis kernel function is used. In the SVM model, wear state The expression is:

[0040] ;

[0041] Lagrange factor The calculation formula is:

[0042] ;

[0043] Radial basis kernel function The calculation formula is:

[0044] 0 ;

[0045] in: Indicates the number of times the original vibration signal was collected. This represents the output of the training samples. Moisturizing wear amount, Represents the first in the dataset 1 eigenvector Indicates the bias amount. Represents the regularization coefficient. Indicates the error variable. Represents the kernel function;

[0046] fitness The calculation formula is:

[0047] ;

[0048] in: Indicates the number of correct categories. This indicates the number of classification errors. This refers to the correct result of triple cross-validation in S3-3. This refers to the result of the triple cross-validation error in S3-3. Therefore, in S3-2, the Tent-Sine-Logistic composite chaotic mapping can generate more uniformly distributed initial solutions in the solution space, thus significantly improving the diversity of the star parrotbill population; refraction-backward learning can preserve high-quality solutions and improve the overall quality of the star parrotbill population; by changing the refraction distance ratio... Refractive index It can generate refracted solutions at different distances and in different directions to increase the diversity of the star parrotbill population, thereby improving the optimization efficiency of the entire star parrotbill population.

[0049] Furthermore, in S3-5, an adaptive variable inertia weight strategy parameter is introduced. Nonlinear weighting factor Then, the expression for foraging and storage strategies is:

[0050] ;

[0051] in: Indicates the first The generation The new position of the star crows Indicates the current generation The new position of the star crows This represents numbers randomly generated based on a normal distribution, Lévy flight, and a range of 0-1. Indicates the current number The optimal individual position of the generation. This represents the random number generated by Levi's flight. , , , A random number between 0 and 1 , Two random individuals in the population. This represents a linear decreasing factor between 0 and 1;

[0052] Adaptive variable inertia weight strategy parameters The calculation formula is:

[0053] ;

[0054] in: This represents the initial value of the inertia weight. The inertia weight corresponding to the maximum number of iterations. This indicates the current iteration number. Indicates the maximum number of first iterations;

[0055] Nonlinear weighting factor The calculation formula is:

[0056] ;

[0057] in: This represents the lower bound of the learning factor. This represents a random generation function. This indicates the current iteration number. Indicates the maximum number of first iterations. This represents the upper bound of the learning factor. Therefore, in S3-5, the adaptive variable inertia weight strategy parameters are adjusted using an iterative process. This allows the INOA model to perform a fast search in the early stages and reduce the step size in the later stages, thus searching for the optimal solution more accurately through nonlinear weighting factors. It can enhance the ability of individual star parrotbills to learn the best position in the star parrotbill population.

[0058] Furthermore, in S3-6, the first The generation Line number The new position of the Starry Night Crows The expression is:

[0059] ;

[0060] in: Representing the contemporary first Line number The new position of the Starry Night Crows , , , , , , A random number between 0 and 1 Indicates the current generation The first reference point for each star crow to find its location. Indicates the current generation Each star crow and sparrow seeks a second reference point for its location;

[0061] When a group of elite individuals gather at point C The elite mechanism detected that the gap between these elite individuals and the global optimum was less than a preset value. The system will perform dimension-wise reverse learning on these elite individuals to generate a new batch of solutions distributed in the opposite region of point C. If one of these solutions finds a new region E, it will jump out of point C.

[0062] The expression for the elite mechanism detection is:

[0063]

[0064] in: Indicates the first The individual in the first The position of the dimension Indicates the first The upper bound of the dimension, Indicates the first The lower bound of dimensionality This represents a function that generates randomly. This represents the threshold for the probability of elite reverse learning. Functions for generating random numbers, fitness function Indicates the position of elite individuals;

[0065] If the attraction at point C is strong, the solution generated by reverse learning will be pulled back and stagnate at point C. At this point, a dynamic local optimum escape mechanism is triggered, randomly resetting half of the population or applying a perturbation to the global optimum.

[0066] The expression for the dynamic local optimal escape mechanism is:

[0067] ;

[0068] Adaptive step size The calculation formula is:

[0069] ;

[0070] in: This indicates the position of the crow-lark after the disturbance. Indicates Cauchy mutation, This indicates the position of the perturbed star, the Crow-sparrow. Indicates the escape distance parameter. This represents a random individual in the Starry Night Parrotbill population. This represents the average fitness value of the current generation. Indicates the first The average fitness value of the generation, This indicates the current iteration number. Indicates the maximum number of first iterations. Indicates the first The position of the optimal individual within the entire Star Parrotbill population. Therefore, in S3-6, this is determined through a non-linear weighting factor. This can enhance the learning ability of individual Star Parrotbills to the optimal position within the Star Parrotbill population; in S3-8, Elite-based dimensional reverse learning and dynamic local optimum escape are introduced into the INOA algorithm to enable it to escape local optima; triggering the dynamic local optimum escape mechanism ensures that the INOA algorithm will not get stuck at point C.

[0071] Further, S2 includes the following steps:

[0072] S2-1, Applying fast iterative filtering to the original signal using a hard threshold. Decompose to obtain eigenmode functions ;

[0073] S2-2, to eigenmode functions Perform reconstruction processing;

[0074] S2-3, Regarding the reconstructed... eigenmode functions Perform time-domain and frequency-domain feature extraction processing, and then process the time-domain features... eigenmode functions Dimensionality reduction is performed. Therefore, data processing (i.e., hard threshold fast iterative filtering, reconstruction, and extraction of time and frequency domain features) is applied to the original vibration signal to remove noise and outliers, making it usable as input data (i.e., highly accurate data) for the INOA-SVM model. Furthermore, based on the characteristics of the high-speed grinding wheel vibration signal in initial wear, normal wear, early rapid wear, and rapid wear, weighted kurtosis indices are extracted for each component of the vibration signal. Components with high weighted kurtosis indices are selected for signal reconstruction to further improve the accuracy of subsequent high-speed grinding wheel condition identification and detection.

[0075] Further, S2-1 includes the following steps:

[0076] S2-1-1, Set the number of second iterations Threshold parameters ;

[0077] S2-1-2, Calculate the filter length and filter ;

[0078] S2-1-3, Transform the diagonal matrix The value is less than the threshold parameter All eigenvalues ​​are set to zero to obtain the matrix. to convert the original signal Decomposed into Each intrinsic mode function and residual term Therefore, in S2-1, the number of the second iteration... This enables fast iterative filtering with hard thresholding to directly calculate the intrinsic mode function. This reduces the iteration time of the entire hard threshold fast iterative filtering, thereby accelerating the recognition and detection efficiency of the grinding state recognition and detection model.

[0079] Furthermore, in S2-1-1, the number of iterations... The calculation formula is:

[0080] ;

[0081] in: Indicates a constant. Indicates sorted in descending order Eigenvalues;

[0082] In S2-1-2, the filter is a non-negative, symmetric Fokker-Planck compactly supported low-pass filter with an area of ​​1. The size of this low-pass filter is... ;

[0083] Filter length The calculation formula is:

[0084] ;

[0085] in: Indicates signal length. To set parameters, This indicates the number of extreme points in the decomposed signal;

[0086] filter The calculation formula is:

[0087] ;

[0088] low-pass filter The calculation formula is:

[0089] ;

[0090] in: The first term of the low-pass filter Line number Column elements, , Indicates the core function;

[0091] Intrinsic mode functions The calculation formula is:

[0092] ;

[0093] in: Represents a unit vector. For matrix eigenvectors, Represented by matrix A diagonal matrix whose eigenvalues ​​are diagonal elements. express Discrete Fourier Transform , express Inverse Discrete Fourier Transform ;

[0094] Intrinsic mode functions Represented as:

[0095] ;

[0096] Hard-threshold fast iterative filtering (htFIF) for calculating intrinsic mode functions The calculation formula is:

[0097] ;

[0098] The intrinsic mode functions are calculated using direct fast iterative filtering. The calculation formula yields the first eigenmode function. and the intrinsic mode function From the original signal Separation from the middle, making Determine the residual term Is it below the threshold parameter? If it is lower, the original signal is output. If the decomposition result is not found, then return to S2-1-2, and so on until the residual term is found. Below the threshold parameter Then, output the original signal. The decomposition results;

[0099] ;

[0100] in: Indicates the first There are 10 intrinsic mode functions. Therefore, the entire hard-threshold fast iterative filtering process consists of only one matrix. Adjusting one parameter can speed up the calculation and improve the recognition and detection efficiency of the grinding condition identification and detection model.

[0101] Furthermore, in S2-2, the effective weighted sparsity kurtosis The calculation formula is:

[0102] ;

[0103] Weighted sparsity kurtosis criterion The calculation formula is:

[0104] ;

[0105] in: This indicates the number of modal functions in the original proof after reconstruction. For the sparsity of the signal, The kurtosis of the signal, Represents the original vibration signal and the intrinsic mode function The correlation coefficient;

[0106] Intrinsic mode functions Comprehensive indicators The calculation formula is:

[0107] ;

[0108] Permutation Entropy The expression is:

[0109] ;

[0110] signal energy The expression is:

[0111] ；

[0112] in: Indicates the first eigenmode functions Therefore, in S2-2, through reconstruction processing, multiple intrinsic mode functions can be selected. The sub-signal containing the most wear signals from high-speed grinding wheels is used to further accelerate the recognition and detection efficiency of the grinding condition identification and detection model.

[0113] Furthermore, in S2-3, the time-domain features include: root mean square value. ,average value Absolute average amplitude Root amplitude Maximum value kurtosis ,variance kurtosis factor Margin factor Pulse factor Peak factor skewness Distortion Frequency domain characteristics include: mean square frequency. Frequency variance Center frequency Therefore, in S2-3, by extracting the feature values ​​of the time-domain and frequency-domain characteristics of the vibration signal, the wear state of the high-speed grinding wheel can be reflected to the greatest extent.

[0114] Furthermore, in S1, the wear state of the high-speed grinding wheel includes: initial wear, normal wear, early rapid wear, and rapid wear.

[0115] The beneficial effects of this invention are as follows:

[0116] This invention utilizes hard-threshold fast iterative filtering and reconstruction processing on high-speed grinding wheel vibration data. This decomposes complex raw vibration data (i.e., linear, nonlinear, and non-stationary signals) into simpler signals for iterative processing, eliminating non-grinding vibrations. Simultaneously, it accelerates the recognition and detection efficiency of the grinding condition identification and detection model by applying the INOA model to the regularization coefficients in the SVM model. Kernel function Optimization is performed to enable the grinding condition identification and detection model to escape local convergence and to achieve faster convergence speed and higher stability, thereby improving the accuracy and stability of high-speed grinding wheel condition identification and detection.

[0117] The present invention also includes the following advantages:

[0118] 1. Regularization coefficients of the SVM model in this invention Kernel function The INOA model is used to optimize the high-speed grinding wheel condition recognition and detection results within a given range. The INOA model enhances the richness of the initial star-shaped magpie population through composite chaotic mapping and refraction back learning, and accelerates the optimization speed through an adaptive variable inertia weight strategy. Elite-based dimensional reverse learning and a dynamic local optimum escape mechanism enable the INOA model to escape local optima, thereby allowing the INOA model to find the optimal regularization coefficient. Kernel function This capability enables the SVM model to accurately identify and detect the wear state of high-speed grinding wheels; in S3-2, the Tent-Sine-Logistic composite chaotic mapping can generate more uniformly distributed initial solutions in the solution space, thus significantly improving the diversity of the star parrotbill population; refraction back learning can retain high-quality solutions and improve the overall quality of the star parrotbill population; by changing the refraction distance ratio Refractive index It can generate refraction solutions at different distances and in different directions to increase the diversity of the star parrotbill population, thereby improving the optimization efficiency of the entire star parrotbill population; in S3-5, the adaptive variable inertia weight strategy parameters are adjusted according to the iterative process. This allows the INOA model to perform a fast search in the early stages and reduce the step size in the later stages, thus searching for the optimal solution more accurately through nonlinear weighting factors. This can enhance the learning ability of individual star parrotbills to the optimal position within the star parrotbill population; in S3-6, through nonlinear weighting factors... This can enhance the learning ability of individual Star Parrotbills to the optimal position within the Star Parrotbill population; in S3-8, Elite-based dimensional reverse learning and dynamic local optimum escape are introduced into the INOA algorithm to enable it to escape local optima; triggering the dynamic local optimum escape mechanism ensures that the INOA algorithm will not get stuck at point C.

[0119] 2. This invention performs data processing on the original vibration signal (i.e., hard threshold fast iterative filtering, reconstruction, and extraction of time-domain and frequency-domain features), which can remove noise and outliers from the original vibration signal, thus making it usable as input data (i.e., data with high accuracy) for the INOA-SVM model. Furthermore, based on the characteristics of the high-speed grinding wheel vibration signal in initial wear, normal wear, early rapid wear, and rapid wear, the weighted kurtosis index of each component of the vibration signal is extracted. Components with high weighted kurtosis index values ​​are selected for signal reconstruction to further improve the accuracy of subsequent high-speed grinding wheel state identification and detection; in S2-1, the second iteration number... This enables fast iterative filtering with hard thresholding to directly calculate the intrinsic mode function. This reduces the iteration time of the entire hard threshold fast iterative filtering, thereby accelerating the recognition and detection efficiency of the grinding condition identification and detection model; the entire hard threshold fast iterative filtering only involves a matrix. Adjusting one parameter can speed up calculations and improve the recognition and detection efficiency of the grinding condition identification and detection model; in S2-2, through reconstruction processing, multiple intrinsic mode functions can be selected. The sub-signal containing the most wear signals of high-speed grinding wheels is used to further accelerate the recognition and detection efficiency of the grinding condition identification and detection model. In S2-3, by extracting feature values ​​from the time-domain and frequency-domain features of the vibration signal, the wear condition of the high-speed grinding wheel can be reflected to the greatest extent. Attached Figure Description

[0120] Figure 1 This is a flowchart of the high-speed grinding wheel condition identification and detection method based on the INOA-SVM model of the present invention;

[0121] Figure 2 This is a flowchart of S3 of the present invention;

[0122] Figure 3 This is a flowchart of S2-1 of the present invention. Detailed Implementation

[0123] The specific embodiments of the present invention will now be described with reference to the accompanying drawings.

[0124] like Figures 1 to 3 The diagram shows the preferred embodiment of the present invention. This embodiment describes a high-speed grinding wheel condition identification and detection method based on the INOA (Improved Star-Crow-Sparrow)-SVM model, which includes the following steps:

[0125] S1. Acquisition of raw vibration signals and wear classification: Acquire the raw vibration signal dataset of the high-speed grinding wheel in the X-axis, Y-axis, and Z-axis directions. The original vibration signal dataset was analyzed according to the wear stages of the high-speed grinding wheel. Wear stages are divided to obtain a vibration signal dataset after the wear stages are divided. ;

[0126] S2. Data processing of the original vibration signal dataset in S1: The original vibration signal dataset in S1 is processed sequentially. Hard thresholding fast iterative filtering, reconstruction, and time-domain and frequency-domain feature extraction are performed to obtain the processed vibration signal dataset. ;

[0127] S3. Construct and validate the INOA (Improved Star Crowbird)-SVM model: Adjust the regularization coefficients in the SVM model using the INOA model. Kernel function Optimization was performed to obtain the INOA-SVM model, which uses the vibration signal dataset after dividing S1 into wear stages. The processed vibration signal dataset in S2 Input the data into the INOA-SVM model and validate the model. If the validation meets the requirements, the INOA-SVM model is the optimal INOA-SVM model, and S4 is executed. If the validation does not meet the requirements, the regularization coefficients are re-optimized. Kernel function The process continues until the verification meets the requirements; once this INOA-SVM model is confirmed, it is the optimal INOA-SVM model.

[0128] S4. Identification and detection of wear condition of high-speed grinding wheel: Real-time detection of vibration signals of the high-speed grinding wheel in the X, Y, and Z axes. and the vibration signal The input is fed into the optimal INOA-SVM model in S3 to output the vibration signal. The wear condition corresponding to high-speed wear grinding wheels;

[0129] in: Indicates time, Indicates the number of times the original vibration signal was collected. This represents the wear stage of a high-speed grinding wheel. Therefore, by performing hard-threshold fast iterative filtering and reconstruction processing on the vibration data of the high-speed grinding wheel, the complex original vibration data (i.e., linear, nonlinear, and non-stationary signals) can be decomposed into simpler signals for iterative processing. Non-grinding vibrations are eliminated (for example, during the operation of a high-speed grinding wheel, factors such as the installation and working environment can cause noise, outliers, and other interference factors in the collected vibration signals, making them unsuitable for direct use in high-speed grinding wheel condition identification and detection. Therefore, data processing of the original vibration signals is necessary. However, during the wear process of a high-speed grinding wheel, the vibration signals are non-stationary, and processing them using methods for stationary signals cannot accurately identify and detect the wear state of the high-speed grinding wheel). Simultaneously, it can accelerate the identification and detection efficiency of the grinding condition identification and detection model by using the INOA model to adjust the regularization coefficients in the SVM model. Kernel function Optimization is performed to enable the grinding condition identification and detection model to escape local convergence and to achieve faster convergence speed and higher stability, thereby improving the accuracy and stability of high-speed grinding wheel condition identification and detection.

[0130] It should be noted that in the SVM model, if the regularization coefficient... High, kernel function A high regularization coefficient carries a significant risk of overfitting, excessively increasing the sensitivity to noise, leading to a collapse in generalization ability and resulting in unreliable results. This negatively impacts the accuracy of high-speed grinding wheel condition detection and identification. High, kernel function If the value is low, there is a risk of extreme underfitting, which means that the system cannot learn the effective features of the vibration signal of the high-speed grinding wheel, resulting in poor classification ability and unreliable results, thus affecting the accuracy of high-speed grinding wheel condition detection and recognition.

[0131] Specifically, in the INOV model, each individual star gargoyle represents a set of regularization coefficients. Kernel function .

[0132] In this embodiment, in S1, the wear state of the high-speed grinding wheel includes: initial wear, normal wear, early rapid wear, and rapid wear.

[0133] In this embodiment, step S2 includes the following steps:

[0134] S2-1, Applying fast iterative filtering to the original signal using a hard threshold. Decompose to obtain eigenmode functions ;

[0135] S2-2, to eigenmode functions Perform reconstruction processing;

[0136] S2-3, Regarding the reconstructed... eigenmode functions Perform time-domain and frequency-domain feature extraction processing, and then process the time-domain features... eigenmode functions Perform dimensionality reduction processing;

[0137] S2-1 includes the following steps:

[0138] S2-1-1, Set the number of second iterations Threshold parameters ;

[0139] S2-1-2, Calculate the filter length and filter ;

[0140] S2-1-3, Transform the diagonal matrix The value is less than the threshold parameter All eigenvalues ​​are set to zero to obtain the matrix. to convert the original signal Decomposed into Each intrinsic mode function and residual term ;

[0141] In this embodiment, in S2-1-1, the number of iterations The calculation formula is:

[0142] ;

[0143] in: Indicates a constant. Indicates sorted in descending order Eigenvalues;

[0144] In S2-1-2, the filter is a non-negative, symmetric Fokker-Planck compactly supported low-pass filter with an area of ​​1. The size of this low-pass filter is... ;

[0145] Filter length The calculation formula is:

[0146] ;

[0147] in: Indicates signal length. To set parameters, This indicates the number of extreme points in the decomposed signal;

[0148] filter The calculation formula is:

[0149] ;

[0150] low-pass filter The calculation formula is:

[0151] ;

[0152] in: The first term of the low-pass filter Line number Column elements, , Indicates the core function;

[0153] Intrinsic mode functions The calculation formula is:

[0154] ;

[0155] in: Represents a unit vector. For matrix eigenvectors, Represented by matrix A diagonal matrix whose eigenvalues ​​are diagonal elements. express Discrete Fourier Transform , express Inverse Discrete Fourier Transform ;

[0156] Intrinsic mode functions Represented as:

[0157] ;

[0158] Hard-threshold fast iterative filtering (htFIF) for calculating intrinsic mode functions The calculation formula is:

[0159] ;

[0160] The intrinsic mode functions are calculated using direct fast iterative filtering. The calculation formula yields the first eigenmode function. and the intrinsic mode function From the original signal Separation from the middle, making Determine the residual term Is it below the threshold parameter? If it is lower, the original signal is output. If the decomposition result is not found, then return to S2-1-2, and so on until the residual term is found. Below the threshold parameter Then, output the original signal. The decomposition results;

[0161] ;

[0162] in: Indicates the first One intrinsic mode function;

[0163] In S2-2, effective weighted sparsity kurtosis The calculation formula is:

[0164] ;

[0165] Weighted sparsity kurtosis criterion The calculation formula is:

[0166] ;

[0167] in: This indicates the number of modal functions in the original proof after reconstruction. For the sparsity of the signal, The kurtosis of the signal, Represents the original vibration signal and the intrinsic mode function The correlation coefficient;

[0168] Intrinsic mode functions Comprehensive indicators The calculation formula is:

[0169] ;

[0170] Permutation Entropy The expression is:

[0171] ;

[0172] signal energy The expression is:

[0173] ；

[0174] in: Indicates the first eigenmode functions ;

[0175] In S2-3, the time-domain features include: root mean square value. ,average value Absolute average amplitude Root amplitude Maximum value kurtosis ,variance kurtosis factor Margin factor Pulse factor Peak factor skewness Distortion Frequency domain characteristics include: mean square frequency. (Reflects changes in the position of the frequency band, and is a weighted average of the squares of the signal frequencies), frequency variance (The square of the frequency standard deviation is another measure of the dispersion of power spectrum energy; the dimension reflects the distribution of spectral energy.) Center frequency (This reflects the change in the position of the frequency center in the spectrum. When the wear of the high-speed grinding wheel increases, the amplitude corresponding to some frequency components will change, and the position of its frequency center will also change accordingly.) Therefore, the wear state of a high-speed grinding wheel during operation directly alters the dynamic characteristics of the machine tool processing system. These changes are manifested through vibration signals of specific patterns. Therefore, detecting these vibration signals is an effective way to identify and detect the wear state of high-speed grinding wheels. However, various noises and outliers exist during processing. Therefore, data processing of the original vibration signal (i.e., hard threshold fast iterative filtering, reconstruction, and extraction of time and frequency domain features) can remove noise and outliers, making it suitable as input data (i.e., highly accurate data) for the INOA-SVM model. Furthermore, based on the characteristics of high-speed grinding wheel vibration signals in initial wear, normal wear, early rapid wear, and rapid wear, weighted kurtosis indices are extracted for each component of the vibration signal. Components with high weighted kurtosis indices are selected for signal reconstruction to further improve the accuracy of subsequent high-speed grinding wheel state identification and detection. In S2-1, the second iteration number... This enables fast iterative filtering with hard thresholding to directly calculate the intrinsic mode function. This reduces the iteration time of the entire hard threshold fast iterative filtering, thereby accelerating the recognition and detection efficiency of the grinding state identification and detection model; the low-pass filter effectively reduces noise interference during iterative signal decomposition; the entire hard threshold fast iterative filtering only requires a matrix... Adjusting one parameter can speed up calculations and improve the recognition and detection efficiency of the grinding condition identification and detection model; in S2-2, through reconstruction processing, multiple intrinsic mode functions can be selected. The sub-signal containing the most wear signals of high-speed grinding wheels is used to further accelerate the recognition and detection efficiency of the grinding condition identification and detection model. In S2-3, by extracting feature values ​​from the time-domain and frequency-domain features of the vibration signal, the wear condition of the high-speed grinding wheel can be reflected to the greatest extent.

[0176] Specifically, in S2-1, the hard threshold fast iterative filter (htFIF) has the characteristics of fast decomposition speed and few adjustment parameters, which can accelerate the recognition and detection efficiency of the grinding state recognition and detection model.

[0177] Specifically, in S2-2, the effective weighted sparsity kurtosis Can consider intrinsic mode functions Sparsity, kurtosis, and correlation.

[0178] In this embodiment, step S3 includes the following steps:

[0179] S3-1. SVM model parameter initialization, including: star-crested jayfinch population size, first iteration number, and regularization coefficient. Kernel function ;

[0180] S3-2. Refraction back learning is performed based on the Tent-Sine-Logistic composite chaotic mapping to optimize the initial star gargoyle population in S3-1;

[0181] S3-3. Input all star-shaped parrotbill populations into the SVM model, and calculate the fitness of each star-shaped parrotbill individual through triple cross-validation. ;

[0182] S3-4. Randomly generate two random numbers. and If random number If yes, then execute S3-5; otherwise, execute S3-6.

[0183] S3-5, Foraging and Storage Strategies: Introducing Adaptive Variable Inertia Weight Strategy Parameters And set a nonlinear weighting factor. ;

[0184] S3-6. Storage Search and Retrieval Strategy Based on INOA Model: Introducing Adaptive Variable Inertia Weight Strategy Parameters into the storage search and retrieval strategy. And set a nonlinear weighting factor. ;

[0185] S3-7, Update the fitness of the Star Parrotbill population ;

[0186] S3-8, will Elite-based dimensional reverse learning and dynamic local optimal escape are introduced into the INOA algorithm;

[0187] S3-9. If the current first iteration count reaches the maximum first iteration count, then execute S3-10; otherwise, return to S3-4. Continue until the current first iteration count reaches the maximum first iteration count, then execute S3-10.

[0188] S3-10, to obtain the globally optimal regularization coefficient Kernel function And obtain the INOA-SVM model;

[0189] S3-11. Divide the vibration signal dataset in S1 into wear stages. The processed vibration signal dataset in S2 Input the INOA-SVM model into S3-10 and validate the INOA-SVM model. If the validation meets the requirements, the INOA-SVM model is the optimal INOA-SVM model, and then execute S4. If the validation does not meet the requirements, return to S3-4, until the validation meets the requirements, and the INOA-SVM model is the optimal INOA-SVM model.

[0190] In S3-2, the expression for the composite chaotic mapping is:

[0191] ;

[0192] in: Indicates a Tent mapping. Represents a Sine mapping. Represents the Logistic mapping;

[0193] By using refraction-backward learning, the original solution and the refraction solution are merged to form a hybrid star parrotbill population of size 2NP. The individuals are then sorted according to their fitness. The 2NP hybrid star parrotbill populations are arranged in descending order based on fitness, and the top NP individuals are selected as the final initial population.

[0194] The expression for the individual Star Crowfinch after refraction-back learning is:

[0195] ;

[0196] in: Indicates the lower bound. Indicates the upper bound. Indicates the ratio of refraction distances. Indicates refractive index, This indicates a pre-star crow-sparrow individual that has undergone reverse learning.

[0197] In S3-3, a radial basis kernel function is used. In the SVM model, wear state The expression is:

[0198] ;

[0199] Lagrange factor The calculation formula is:

[0200] ;

[0201] Radial basis kernel function The calculation formula is:

[0202] 0 ;

[0203] in: Indicates the number of times the original vibration signal was collected. This represents the output of the training samples. Moisturizing wear amount, Represents the first in the dataset 1 eigenvector Indicates the bias amount. Represents the regularization coefficient. Indicates the error variable. Represents the kernel function;

[0204] fitness The calculation formula is:

[0205] ;

[0206] in: Indicates the number of correct categories. This indicates the number of classification errors. This refers to the correct result of triple cross-validation in S3-3. This refers to the result of a triple cross-validation error in S3-3;

[0207] In S3-5, an adaptive variable inertia weight strategy parameter is introduced. Nonlinear weighting factor Then, the expression for foraging and storage strategies is:

[0208] ;

[0209] in: Indicates the first The generation The new position of the star crows Indicates the current generation The new position of the star crows This represents numbers randomly generated based on a normal distribution, Lévy flight, and a range of 0-1. Indicates the current number The optimal individual position of the generation. This represents the random number generated by Levi's flight. , , , A random number between 0 and 1 , Two random individuals in the population. This represents a linear decreasing factor between 0 and 1;

[0210] Adaptive variable inertia weight strategy parameters The calculation formula is:

[0211] ;

[0212] in: This represents the initial value of the inertia weight. The inertia weight corresponding to the maximum number of iterations. This indicates the current iteration number. Indicates the maximum number of first iterations;

[0213] Nonlinear weighting factor The calculation formula is:

[0214] ;

[0215] in: This represents the lower bound of the learning factor. This represents a random generation function. This indicates the current iteration number. Indicates the maximum number of first iterations. This represents the upper bound of the learning factor;

[0216] In S3-6, the first The generation Line number The new position of the Starry Night Crows The expression is:

[0217] ;

[0218] in: Representing the contemporary first Line number The new position of the Starry Night Crows , , , , , , A random number between 0 and 1 Indicates the current generation The first reference point for each star crow to find its location. Indicates the current generation Each star crow and sparrow seeks a second reference point for its location;

[0219] When a group of elite individuals gather at point C The elite mechanism detected that the gap between these elite individuals and the global optimum was less than a preset value. (For example, these elite individuals are relatively close to the global optimum), so we will perform dimension-wise backward learning on these elite individuals to generate a new batch of solutions distributed in the opposite region of point C. If one of these solutions finds a new region E (i.e., a better region than point C), then we will jump out of point C.

[0220] The expression for the elite mechanism detection is:

[0221]

[0222] in: Indicates the first The individual in the first The position of the dimension Indicates the first The upper bound of the dimension, Indicates the first The lower bound of dimensionality This represents a function that generates randomly. This represents the threshold for the probability of elite reverse learning. Functions for generating random numbers, fitness function Indicates the position of elite individuals;

[0223] If the attraction at point C is strong, the solution generated by reverse learning will be pulled back and stagnate at point C. At this point, a dynamic local optimum escape mechanism is triggered, randomly resetting half of the population or applying a perturbation to the global optimum.

[0224] The expression for the dynamic local optimal escape mechanism is:

[0225] ;

[0226] Adaptive step size The calculation formula is:

[0227] ;

[0228] in: This indicates the position of the crow-lark after the disturbance. Indicates Cauchy mutation, This indicates the position of the perturbed star, the Crow-sparrow. Indicates the escape distance parameter. This represents a random individual in the Star Parrotbill population. This represents the average fitness value of the current generation. Indicates the first The average fitness value of the generation, This indicates the current iteration number. Indicates the maximum number of first iterations. Indicates the first The position of the optimal individual within the entire Star Parrotbill population. Therefore, the performance of the SVM model is highly dependent on the regularization coefficient. Kernel function These two parameters, if optimized manually or using traditional methods (such as grid search), have inherent limitations and are difficult to find the optimal regularization coefficients. Kernel function This leads to the inability to establish an accurate high-speed grinding wheel condition recognition and detection model, affecting the accuracy of high-speed grinding wheel condition recognition and detection. Therefore, the regularization coefficient of the SVM model... Kernel function The INOA model is used to optimize the high-speed grinding wheel condition recognition and detection results within a given range. The INOA model enhances the richness of the initial star-shaped magpie population through composite chaotic mapping and refraction back learning, and accelerates the optimization speed through an adaptive variable inertia weight strategy. Elite-based dimensional reverse learning and a dynamic local optimum escape mechanism enable the INOA model to escape local optima, thereby allowing the INOA model to find the optimal regularization coefficient. Kernel function Traditional optimization methods (such as genetic algorithms and particle swarm optimization) suffer from slow optimization speed and are prone to getting trapped in local optima, thus failing to find the optimal regularization coefficient. Kernel function These two parameters enable the SVM model to accurately identify and detect the wear state of high-speed grinding wheels; in S3-2, the Tent-Sine-Logistic composite chaotic mapping can generate more uniformly distributed initial solutions in the solution space, thus significantly improving the diversity of the star parrotbill population; refraction inverse learning can retain high-quality solutions and improve the overall quality of the star parrotbill population; by changing the refraction distance ratio Refractive index It can generate refraction solutions at different distances and in different directions to increase the diversity of the star parrotbill population, thereby improving the optimization efficiency of the entire star parrotbill population; in S3-5, the adaptive variable inertia weight strategy parameters are adjusted according to the iterative process. This allows the INOA model to perform a fast search in the early stages and reduce the step size in the later stages, thus searching for the optimal solution more accurately through nonlinear weighting factors. This can enhance the learning ability of individual star parrotbills to the optimal position within the star parrotbill population; in S3-6, through nonlinear weighting factors... This can enhance the learning ability of individual Star Parrotbills to the optimal position within the Star Parrotbill population; in S3-8, Elite-based dimensional reverse learning and dynamic local optimum escape are introduced into the INOA algorithm to enable it to escape local optima; triggering the dynamic local optimum escape mechanism ensures that the INOA algorithm will not get stuck at point C.

[0229] Specifically, in S3-2, back-learning can improve the optimization ability of the INOA model in the early stage of iteration, but it is very easy to get trapped in local optima in the later stage, resulting in poor convergence accuracy and reduced convergence speed. Therefore, the principle of light refraction (i.e., refraction back-learning) is introduced on the basis of directional learning to find the optimal solution, so as to retain high-quality solutions and improve the overall quality of the star gargoyle population.

[0230] Specifically, in S3-3, the fitness of each individual Star Crowfinch This refers to the accuracy of the SVM model.

[0231] In summary, this invention, through hard-threshold fast iterative filtering and reconstruction processing of high-speed grinding wheel vibration data, can decompose complex original vibration data (i.e., linear signals, nonlinear signals, and non-stationary signals) into simple signals for iterative processing and eliminate non-grinding vibrations. Simultaneously, it can accelerate the identification and detection efficiency of the grinding condition recognition and detection model by using the INOA model to adjust the regularization coefficients in the SVM model. Kernel function Optimization is performed to enable the grinding condition identification and detection model to escape local convergence and to achieve faster convergence speed and higher stability, thereby improving the accuracy and stability of high-speed grinding wheel condition identification and detection.

[0232] The above description is an explanation of the present invention and not a limitation thereof. The scope of the present invention is defined by the claims. Within the scope of protection of the present invention, any form of modification may be made.

Claims

1. A method for identifying and detecting the condition of a high-speed grinding wheel based on an INOA-SVM model, characterized in that, Includes the following steps: S1. Acquisition of raw vibration signals and wear classification: Acquire the raw vibration signal dataset of the high-speed grinding wheel in the X-axis, Y-axis, and Z-axis directions. The original vibration signal dataset was analyzed according to the wear stages of the high-speed grinding wheel. Wear stages are divided to obtain a vibration signal dataset after the wear stages are divided. ; S2. Data processing of the original vibration signal dataset in S1: The original vibration signal dataset in S1 is processed sequentially. Hard thresholding fast iterative filtering, reconstruction, and time-domain and frequency-domain feature extraction are performed to obtain the processed vibration signal dataset. ; S3. Construct and validate the INOA-SVM model: Analyze the regularization coefficients in the SVM model using the INOA model. Kernel function Optimization was performed to obtain the INOA-SVM model, which uses the vibration signal dataset after dividing S1 into wear stages. The processed vibration signal dataset in S2 Input the data into the INOA-SVM model and validate the model. If the validation meets the requirements, the INOA-SVM model is the optimal INOA-SVM model, and S4 is executed. If the validation does not meet the requirements, the regularization coefficients are re-optimized. Kernel function The process continues until the verification meets the requirements; once this INOA-SVM model is confirmed, it is the optimal INOA-SVM model. S4. Identification and detection of wear condition of high-speed grinding wheel: Real-time detection of vibration signals of the high-speed grinding wheel in the X, Y, and Z axes. and the vibration signal The input is fed into the optimal INOA-SVM model in S3 to output the vibration signal. The wear condition corresponding to high-speed wear grinding wheels; in: Indicates time, Indicates the number of times the original vibration signal was collected. This indicates the wear stage of a high-speed grinding wheel; S3 includes the following steps: S3-1. SVM model parameter initialization, including: star-crested jayfinch population size, first iteration number, and regularization coefficient. Kernel function ; S3-2. Refraction back learning is performed based on the Tent-Sine-Logistic composite chaotic mapping to optimize the initial star gargoyle population in S3-1; S3-3. Input all star-shaped parrotbill populations into the SVM model, and calculate the fitness of each star-shaped parrotbill individual through triple cross-validation. ; S3-4. Randomly generate two random numbers. and If random number If yes, then execute S3-5; otherwise, execute S3-6. S3-5, Foraging and Storage Strategies: Introducing Adaptive Variable Inertia Weight Strategy Parameters And set a nonlinear weighting factor. ; S3-6. Storage Search and Retrieval Strategy Based on INOA Model: Introducing Adaptive Variable Inertia Weight Strategy Parameters into the storage search and retrieval strategy. And set a nonlinear weighting factor. ; S3-7, Update the fitness of the Star Parrotbill population ; S3-8, will Elite-based dimensional reverse learning and dynamic local optimal escape are introduced into the INOA algorithm; S3-9. If the current first iteration count reaches the maximum first iteration count, then execute S3-10; otherwise, return to S3-4. Continue until the current first iteration count reaches the maximum first iteration count, then execute S3-10. S3-10, to obtain the globally optimal regularization coefficient Kernel function And obtain the INOA-SVM model; S3-11. Divide the vibration signal dataset in S1 into wear stages. The processed vibration signal dataset in S2 Input the INOA-SVM model into S3-10 and validate the INOA-SVM model. If the validation meets the requirements, the INOA-SVM model is the optimal INOA-SVM model, and then execute S4. If the validation does not meet the requirements, return to S3-4, until the validation meets the requirements, and the INOA-SVM model is the optimal INOA-SVM model. In S3-2, the expression for the composite chaotic mapping is: ； in: Indicates a Tent mapping. Represents a Sine mapping. Represents the Logistic mapping; By using refraction-backward learning, the original solution and the refraction solution are merged to form a hybrid star parrotbill population of size 2NP. The individuals are then sorted according to their fitness. The 2NP hybrid star parrotbill populations are arranged in descending order based on fitness, and the top NP individuals are selected as the final initial population. The expression for the individual Star Crowfinch after refraction-back learning is: ； in: Indicates the lower bound. Indicates the upper bound. Indicates the ratio of refraction distances. Indicates refractive index, This indicates a pre-star crow-sparrow individual that has undergone reverse learning. In S3-3, a radial basis kernel function is used. In the SVM model, wear state The expression is: ； Lagrange factor The calculation formula is: ； Radial basis kernel function The calculation formula is: 0 ； in: Indicates the number of times the original vibration signal was collected. This represents the output of the training samples. Moisturizing wear amount, Represents the first in the dataset 1 eigenvector Indicates the bias amount. Represents the regularization coefficient. Indicates the error variable. Represents the kernel function; fitness The calculation formula is: ； in: Indicates the number of correct categories. This indicates the number of classification errors. This refers to the correct result of triple cross-validation in S3-3. This refers to the result of a triple cross-validation error in S3-3.

2. The method for identifying and detecting the condition of a high-speed grinding wheel based on the INOA-SVM model as described in claim 1, characterized in that: In S3-5, an adaptive variable inertia weight strategy parameter is introduced. Nonlinear weighting factor Then, the expression for foraging and storage strategies is: ； in: Indicates the first The generation The new position of the star crows Indicates the current generation The new position of the star crows This represents numbers randomly generated based on a normal distribution, Lévy flight, and a range of 0-1. Indicates the current number The optimal individual position of the generation. This represents the random number generated by Levi's flight. , , , A random number between 0 and 1 , Two random individuals in the population. This represents a linear decreasing factor between 0 and 1; Adaptive variable inertia weight strategy parameters The calculation formula is: ； in: This represents the initial value of the inertia weight. The inertia weight corresponding to the maximum number of iterations. This indicates the current iteration number. Indicates the maximum number of first iterations; Nonlinear weighting factor The calculation formula is: ； in: This represents the lower bound of the learning factor. This represents a random generation function. This indicates the current iteration number. Indicates the maximum number of first iterations. This represents the upper bound of the learning factor.

3. The method for identifying and detecting the condition of a high-speed grinding wheel based on the INOA-SVM model as described in claim 1, characterized in that: S2 includes the following steps: S2-1, Applying fast iterative filtering to the original signal using a hard threshold. Decompose to obtain eigenmode functions ; S2-2, to eigenmode functions Perform reconstruction processing; S2-3, Regarding the reconstructed... eigenmode functions Perform time-domain and frequency-domain feature extraction processing, and then process the time-domain features... eigenmode functions Dimensionality reduction processing is performed.

4. The method for identifying and detecting the condition of a high-speed grinding wheel based on the INOA-SVM model as described in claim 3, characterized in that: S2-1 includes the following steps: S2-1-1, Set the number of second iterations Threshold parameters ; S2-1-2, Calculate the filter length and filter ; S2-1-3, Transform the diagonal matrix The value is less than the threshold parameter All eigenvalues ​​are set to zero to obtain the matrix. to convert the original signal Decomposed into Each intrinsic mode function and residual term .

5. The method for identifying and detecting the condition of a high-speed grinding wheel based on the INOA-SVM model as described in claim 4, characterized in that: In S2-1-1, the calculation formula for the number of iterations is as follows: ； in: Indicates a constant. Indicates sorted in descending order Eigenvalues; In S2-1-2, the filter is a non-negative, symmetric Fokker-Planck compactly supported low-pass filter with an area of ​​1. The size of this low-pass filter is... ; Filter length The calculation formula is: ； in: Indicates signal length. To set parameters, This indicates the number of extreme points in the decomposed signal; filter The calculation formula is: ； low-pass filter The calculation formula is: ； in: The first term of the low-pass filter Line number Column elements, , Indicates the core function; Intrinsic mode functions The calculation formula is: ； in: Represents a unit vector. For matrix eigenvectors, Represented by matrix A diagonal matrix whose eigenvalues ​​are diagonal elements. express Discrete Fourier Transform , express Inverse Discrete Fourier Transform ; Intrinsic mode functions Represented as: ； Hard-threshold fast iterative filtering (htFIF) for calculating intrinsic mode functions The calculation formula is: ； The intrinsic mode functions are calculated using direct fast iterative filtering. The calculation formula yields the first eigenmode function. and the intrinsic mode function From the original signal Separation from the middle, making Determine the residual term Is it below the threshold parameter? If it is lower, the original signal is output. If the decomposition result is not found, then return to S2-1-2, and so on until the residual term is found. Below the threshold parameter Then, output the original signal. The decomposition results; ； in: Indicates the first One intrinsic mode function.

6. The method for identifying and detecting the condition of a high-speed grinding wheel based on the INOA-SVM model as described in claim 3, characterized in that: In S2-2, effective weighted sparsity kurtosis The calculation formula is: ； Weighted sparsity kurtosis criterion The calculation formula is: ； in: This indicates the number of modal functions in the original proof after reconstruction. For the sparsity of the signal, The kurtosis of the signal, Represents the original vibration signal and the intrinsic mode function The correlation coefficient; Intrinsic mode functions Comprehensive indicators The calculation formula is: ； Permutation Entropy The expression is: ; signal energy The expression is: ； in: Indicates the first eigenmode functions .

7. The method for identifying and detecting the condition of a high-speed grinding wheel based on the INOA-SVM model as described in claim 3, characterized in that: In S2-3, the time-domain features include: Root mean square value ,average value Absolute average amplitude Root amplitude Maximum value kurtosis ,variance kurtosis factor Margin factor Pulse factor Peak factor skewness Distortion ; Frequency domain characteristics include: Mean square frequency Frequency variance Center frequency .

8. The method for identifying and detecting the condition of a high-speed grinding wheel based on the INOA-SVM model as described in claim 1, characterized in that: In S1, the wear conditions of high-speed grinding wheels include: Initial wear, normal wear, early rapid wear, and rapid wear.

Images