Method and system for quantitative evaluation of milling tool wear based on multi-domain feature fusion of cutting force

Abstract

This invention provides a method and system for quantitatively assessing milling cutter wear based on multi-domain feature fusion of cutting force. The method includes: conducting a cutting experiment on the milling cutter to obtain an original cutting force signal containing milling cutter wear characteristics; performing smoothing and collaborative noise reduction processing on the original cutting force signal to obtain a smoothed output signal; extracting time-domain, frequency-domain, and time-frequency-domain features from the output signal and constructing a multi-domain feature matrix; based on the correlation between features and wear, processing the multi-domain feature matrix using a feature optimization algorithm to obtain a sensitive feature matrix; and converting the sensitive feature matrix into a fused feature vector based on a weighted fusion strategy. This invention can comprehensively capture the changes in cutting force caused by milling cutter wear, improve the accuracy of wear prediction and the robustness of the model, and achieve accurate quantitative assessment and anomaly warning of milling cutter wear status.

Description

Technical Field

[0001] This invention relates to the field of end mill wear assessment technology, specifically to a quantitative assessment method and system for end mill wear based on the fusion of multi-domain characteristics of cutting force. Background Technology

[0002] Currently, methods for monitoring the wear condition of end mills are mainly divided into direct and indirect methods. Direct methods typically involve directly observing the width of the wear band on the flank face of the tool using techniques such as optical microscopy and computer vision. While offering high accuracy, these methods are limited by factors such as cutting fluid and chip obstruction, as well as the sensor's installation location, making real-time online monitoring difficult. Indirect methods involve collecting sensor signals closely related to the tool's wear condition (such as cutting force, vibration, acoustic emission, and spindle power), and then using signal processing and machine learning algorithms to establish a mapping relationship to assess the amount of wear. Among these, the cutting force signal is the most sensitive to tool wear and can directly reflect changes in the physical state during the cutting process, thus becoming the most widely used monitoring method.

[0003] However, existing tool wear assessment methods based on cutting forces still have the following technical problems and shortcomings in practical applications:

[0004] Weak signal anti-interference capability: The milling process is an intermittent cutting process, accompanied by strong mechanical vibration, random noise, and high-frequency chatter. Traditional signal denoising methods (such as low-pass filtering, wavelet thresholding, etc.) often smooth out the abrupt changes in the signal that reflect the micro-chipping or damage of the tool while filtering out high-frequency noise, resulting in the loss of key information and affecting the accuracy of the assessment.

[0005] Feature extraction has limitations: existing technologies mostly focus on extracting features in the single time or frequency domain. While time-domain features (such as mean and root mean square) can reflect energy changes, they are not sensitive to early wear; while frequency-domain features (such as power spectrum) can reflect frequency structure changes, they are insufficient when processing non-stationary and nonlinear cutting signals. A single feature domain is insufficient to comprehensively characterize the complex evolution process of a milling cutter from initial wear, normal wear to rapid wear.

[0006] Feature redundancy and poor model robustness: When constructing evaluation models, a large number of extracted high-dimensional features are often directly input into the model without fully considering the correlation between features and wear amount, as well as the collinearity between features. This not only leads to large computational load and model complexity, but also easily introduces redundant information and noise interference, causing model overfitting and resulting in poor generalization ability and robustness of the model under different cutting conditions.

[0007] Lack of in-depth analysis of wear trends: Existing methods mostly focus on static prediction of current wear levels, lacking effective identification of wear rate changes and inflection points in wear stages (such as the transition from normal wear to rapid wear). This makes it difficult to accurately predict the remaining tool life and optimal tool change timing, failing to meet the high requirements of predictive maintenance in intelligent manufacturing.

[0008] In summary, how to effectively extract highly wear-sensitive fusion features from noisy cutting force signals and establish a high-precision, robust quantitative evaluation model remains a key technical problem that urgently needs to be solved in the field of end mill wear monitoring. Summary of the Invention

[0009] To address the shortcomings of existing technologies, this invention provides a method and system for quantitatively evaluating milling cutter wear based on the fusion of multi-domain features of cutting force, thereby solving the problems mentioned in the background art.

[0010] To achieve the above objectives, the present invention provides the following technical solution: a method for quantitatively evaluating milling cutter wear based on multi-domain feature fusion of cutting force, comprising the following steps:

[0011] S1: Perform a cutting experiment on the milling cutter to be evaluated to obtain the original cutting force signal containing the wear characteristics of the milling cutter;

[0012] Among them, the original cutting force signal matrix in the original cutting force signal , where n is the number of samples.

[0013] S2: Perform smoothing and collaborative noise reduction processing on the original cutting force signal to obtain a smoothed output signal;

[0014] S3: Extract time-domain features, frequency-domain features, and time-frequency-domain features based on the output signal, and construct a multi-domain feature matrix;

[0015] S4: Based on the correlation between features and wear amount, the multi-domain feature matrix is ​​processed by a feature optimization algorithm to obtain a sensitive feature matrix;

[0016] S5: Convert the sensitive feature matrix into a fused feature vector based on a weighted fusion strategy;

[0017] S6: Input the fused feature vector into the wear prediction model to obtain the predicted wear amount sequence;

[0018] S7: Perform a moving average filter on the predicted wear amount sequence to obtain a smoothed wear amount curve.

[0019] S8: The slope of the smooth wear curve is detected and a threshold is determined. After the threshold is determined, the abnormal wear area of ​​the end mill to be evaluated is obtained.

[0020] As a preferred embodiment, the specific process for performing smoothing and collaborative noise reduction processing on the original cutting force signal to obtain a smoothed output signal includes:

[0021] Set the moving window to 2a+1, where a is the order of the polynomial fitting and is a positive integer greater than or equal to 2.

[0022] In the moving window, the original cutting force signal matrix For each data point, select 2m+1 data points within a window centered on that data point to form a subsequence. , where i represents the index of the current data point, m represents the half-window width as a positive integer, and m>a;

[0023] Perform a k-th order polynomial least squares fit on the subsequence to obtain the fitted polynomial. ;

[0024] Where k represents the order of the polynomial fitting, Indicates the summation index. The coefficients of the polynomial are represented by 't', and the relative position coordinates within the window are represented by 't'. ;

[0025] Calculate the function value of the fitted polynomial at the center of the window, use it as the filtered output value of the data point, slide the window to the next data point, and repeat the above process until all data points have been traversed to obtain the smoothed output signal. .

[0026] As a preferred embodiment, the specific process for constructing the multi-domain feature matrix includes:

[0027] Extract the characteristics of the output signal in the time domain, frequency domain, and time-frequency domain;

[0028] The process of extracting the time-domain feature vector T includes: calculating the mean, root mean square (RMS), kurtosis, and impulse factor of the signal;

[0029] The frequency domain feature vector P extraction process includes: performing a fast Fourier transform on the signal to obtain the power spectral density, and extracting the main frequency amplitude, spectral centroid, and spectral variance;

[0030] The time-frequency domain feature vector E extraction process includes: using wavelet packet transform to decompose the signal to the third level, extracting the energy features of 8 frequency bands, and calculating the wavelet packet entropy;

[0031] Based on the obtained time-domain, frequency-domain, and time-frequency-domain feature vectors, a multi-domain feature matrix is ​​constructed.

[0032] As a preferred embodiment, the specific process of obtaining the sensitive feature matrix by processing the multi-domain feature matrix through a feature optimization algorithm based on the correlation between features and wear amount includes:

[0033] Calculate the correlation coefficient between each feature in the multi-domain feature matrix and the actual wear amount, and take the absolute value as the evaluation score. , Representing the multi-domain characteristic matrix The i-th feature in This indicates the actual wear on the flank face of the milling cutter. Represents the Pearson correlation coefficient function;

[0034] A predetermined number of sensitive features are selected based on the evaluation scores;

[0035] Principal component analysis or kernel principal component analysis is performed on the screened sensitive features to reduce dimensionality and eliminate feature collinearity, thus obtaining the sensitive feature matrix.

[0036] As a preferred embodiment, the calculation formula for the weighted fusion strategy is as follows:

[0037] ;

[0038] Where k represents the number of sensitive features selected. This represents the weighting coefficient of the i-th sensitive feature. This represents the i-th sensitive feature in the sensitive feature matrix.

[0039] in, The contribution of each feature to wear prediction is determined by calculating it using a random forest regression model.

[0040] As a preferred embodiment, the formula for calculating the smooth wear curve is: ;

[0041] Where i represents the index of the current data point. This represents the total width of the sliding window. Indicates the relative offset within the window. This represents the original predicted wear value at the (i+j)th position within the window range.

[0042] As a preferred embodiment, the smooth wear curve The process of slope detection is as follows:

[0043] Extracting smooth wear curves Wear rate in ;

[0044] in, This represents the wear rate at time i. This indicates the amount of smooth wear at the current moment. This indicates the amount of smooth wear at the previous moment.

[0045] As a preferred embodiment, the specific process for determining the threshold of the smooth wear curve includes:

[0046] Calculate the smooth wear curve The difference between the value and the preset wear threshold ;

[0047] When the difference When the value is greater than 0, the corresponding area is identified as an abnormal area, i.e., the tool fails.

[0048] As a preferred embodiment, a milling cutter wear quantitative assessment system based on multi-domain feature fusion of cutting force is provided to implement the above-mentioned milling cutter wear quantitative assessment method based on multi-domain feature fusion of cutting force, including:

[0049] The signal acquisition module is used to perform cutting experiments on the milling cutter to be evaluated and acquire the original cutting force signal containing the wear characteristics of the milling cutter.

[0050] The signal acquisition module is connected to a force measuring instrument, which has 3 channels and is used to acquire cutting forces in the X, Y, and Z directions.

[0051] The signal preprocessing module is used to perform smoothing and collaborative noise reduction processing on the original cutting force signal to obtain a smoothed output signal.

[0052] The feature extraction module is used to extract time-domain features, frequency-domain features, and time-frequency-domain features based on the output signal, and to construct a multi-domain feature matrix;

[0053] The feature extraction module includes:

[0054] The time-domain feature extraction unit is used to calculate the mean, root mean square (RMS), kurtosis, and impulse factor of the signal to obtain the time-domain feature vector T.

[0055] The frequency domain feature extraction unit is used to perform a fast Fourier transform on the signal to obtain the power spectral density, extract the main frequency amplitude, spectral centroid and spectral variance, and obtain the frequency domain feature vector P.

[0056] The time-frequency domain feature extraction unit is used to decompose the signal to the third level using wavelet packet transform, extract the energy features of 8 frequency bands, and calculate the wavelet packet entropy to obtain the time-frequency domain feature vector E;

[0057] The feature optimization module is used to process the multi-domain feature matrix to obtain a sensitive feature matrix based on the correlation between features and wear amount;

[0058] The feature fusion module is used to convert the sensitive feature matrix into a fused feature vector based on a weighted fusion strategy;

[0059] The wear prediction module is used to input the fused feature vector into the wear prediction model to obtain a predicted wear amount sequence.

[0060] The result smoothing module is used to perform a moving average filter on the predicted wear amount sequence to obtain a smoothed wear amount curve.

[0061] The anomaly detection module is used to detect the slope and determine the threshold of the smooth wear curve. After the threshold determination, the abnormal wear area of ​​the milling cutter to be evaluated is obtained.

[0062] The anomaly detection module includes:

[0063] The slope detection unit is used to extract the smooth wear curve. Wear rate in ;

[0064] in, This represents the wear rate at time i. This indicates the amount of smooth wear at the current moment. This indicates the amount of smooth wear at the previous moment;

[0065] The trend analysis unit is used to fit the smooth wear curve using a cubic spline interpolation algorithm to generate a wear trend curve.

[0066] Threshold determination unit, used to calculate smooth wear curve The difference between the value and the preset wear threshold ;

[0067] When the difference When the value is greater than 0, the corresponding area is identified as an abnormal area, i.e., the tool fails.

[0068] This invention provides a method and system for quantitatively evaluating milling cutter wear based on multi-domain feature fusion of cutting force, which has the following beneficial effects:

[0069] 1. This invention employs smoothing and collaborative noise reduction processing. By performing polynomial least squares fitting within a moving window, it can effectively filter out machine tool vibration noise and occasional chatter interference during the cutting process. Compared with traditional low-pass filtering or simple mean filtering, this method can retain the local abrupt change characteristics and impact components of the cutting force signal to the greatest extent while filtering out high-frequency noise, avoiding the loss of tool breakage or micro-chipping characteristics caused by global smoothing, and providing a high-quality signal foundation for subsequent accurate evaluation;

[0070] 2. This invention constructs a multi-domain feature matrix that includes time domain, frequency domain, and time-frequency domain. The time domain features reflect the amplitude change of the cutting force, the frequency domain features reveal the shift in the signal frequency structure, and the time-frequency domain features capture the non-stationary and nonlinear characteristics of the cutting process. This multi-dimensional fusion method overcomes the limitations of a single feature domain in describing complex wear states and can comprehensively and sensitively respond to the changes of the milling cutter from initial wear to rapid wear.

[0071] 3. This invention introduces a feature optimization algorithm based on Pearson correlation coefficient and a weighted fusion strategy based on random forest contribution. By eliminating redundant and irrelevant features and using principal component analysis to eliminate collinearity, the data dimensionality and computational complexity are significantly reduced. At the same time, by weighting the contribution of each feature to the wear amount, the fused feature vector is made more representative. This not only improves the input quality of the wear prediction model, but also effectively eliminates the instability of a single feature under specific working conditions, and enhances the robustness of the evaluation system under different processing environments.

[0072] 4. This invention not only achieves quantitative prediction of wear, but also identifies changes in wear rate and inflection points in wear stages through slope detection and cubic spline interpolation. This method can accurately capture the moment when the tool transitions from a normal wear stage to a rapid wear stage. Combined with preset threshold judgment, it can provide on-site operators with precise tool change timing suggestions and structural health monitoring basis, thereby maximizing tool life and reducing production costs while ensuring machining quality. Attached Figure Description

[0073] Figure 1 The flowchart of the method for quantitative evaluation of milling cutter wear based on multi-domain feature fusion of cutting force according to the present invention is shown below.

[0074] Figure 2 This is a schematic diagram of the process for quantitatively evaluating milling cutter wear based on multi-domain feature fusion of cutting force according to the present invention.

[0075] Figure 3 This is a block diagram of the milling cutter wear quantitative assessment system based on multi-domain feature fusion of cutting force according to the present invention. Detailed Implementation

[0076] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0077] like Figure 1 and 2As shown, this embodiment of the invention provides a method for quantitatively evaluating milling cutter wear based on multi-domain feature fusion of cutting force, including the following steps:

[0078] S1: Perform a cutting experiment on the end mill to be evaluated to obtain the original cutting force signal containing the wear characteristics of the end mill;

[0079] Among them, the original cutting force signal matrix in the original cutting force signal , where n is the number of samples.

[0080] Specifically, This represents the cutting force signal in the X direction. This represents the cutting force signal in the Y direction. This represents the cutting force signal in the Z direction.

[0081] In this embodiment, the milling cutter to be evaluated is machined using a CNC machine tool. The feed direction of the milling cutter is the X direction, the direction perpendicular to the X direction and parallel to the surface of the workpiece to be evaluated is the Y direction, and the direction perpendicular to the surface of the area of ​​the workpiece to be evaluated is the Z direction. The milling machine is started to perform cutting, and the original cutting force signal in the X direction is obtained by a force measuring instrument. The force gauge has 3 channels, indicating that it collects forces in the X, Y, and Z axes.

[0082] S2: Perform smoothing and collaborative noise reduction processing on the original cutting force signal to obtain a smoothed output signal;

[0083] In this embodiment, the specific process for smoothing and coordinating noise reduction of the original cutting force signal includes:

[0084] Set the moving window to 2a+1, where a is the order of the polynomial fitting and is a positive integer greater than or equal to 2.

[0085] In the moving window, the original cutting force signal matrix For each data point, select 2m+1 data points within a window centered on that data point to form a subsequence. , where i represents the index of the current data point, m represents the half-window width as a positive integer, and m>a.

[0086] For example, the width of the moving window is set to 50 (a is 25), for the original cutting force signal matrix For each data point, a subsequence is formed by selecting 51 data points within a window centered on that data point.

[0087] Perform a k-th order polynomial least squares fit on the subsequence to obtain the fitted polynomial. ;

[0088] Where k represents the order of the polynomial fitting, Indicates the summation index. The coefficients of the polynomial are represented by 't', and the relative position coordinates within the window are represented by 't'. ;

[0089] It should be noted that polynomial least squares fitting is chosen here to extract low-frequency trend components from the subsequence and filter out random noise and high-frequency vibration interference during the cutting process. Compared with traditional low-pass filtering methods, polynomial fitting can preserve the local abrupt changes in the cutting force signal and avoids the loss of impact signals caused by global smoothing by fitting within the window.

[0090] Calculate the function value of the fitted polynomial at the center of the window. This value is used as the filtered output value of the data point. The sliding window moves to the next data point, and the above process is repeated until all data points have been traversed to obtain the smoothed output signal. .

[0091] Calculate the output signal By filtering through the aforementioned moving window, machine tool vibration noise is significantly reduced, abnormal peak noise signals introduced by occasional chatter during the cutting process are eliminated, and a smooth cutting force response signal output is finally obtained.

[0092] in, This represents the specific signal components after smoothing in the X, Y, and Z directions.

[0093] S3: Extract time-domain features, frequency-domain features, and time-frequency-domain features based on the output signal, and construct a multi-domain feature matrix;

[0094] Specifically, the characteristics of the output signal in the time domain, frequency domain, and time-frequency domain are extracted;

[0095] The process of extracting the time-domain feature vector T includes: calculating the mean, root mean square (RMS), kurtosis, and impulse factor of the signal;

[0096] The frequency domain feature vector P extraction process includes: performing a fast Fourier transform on the signal to obtain the power spectral density, and extracting the main frequency amplitude, spectral centroid, and spectral variance;

[0097] The time-frequency domain feature vector E extraction process includes: using wavelet packet transform to decompose the signal to the third level, extracting the energy features of 8 frequency bands, and calculating the wavelet packet entropy;

[0098] Based on the obtained time-domain, frequency-domain, and time-frequency-domain feature vectors, a multi-domain feature matrix is ​​constructed. ;

[0099] Among them, for the output signal The purpose of multi-domain feature extraction is to comprehensively capture the changes in cutting force amplitude, frequency structure shift, and nonlinear non-stationary features caused by milling cutter wear, thereby improving the ability to characterize the wear state. The calculated multi-domain feature matrix... Satisfying N is the number of samples, and D is the total dimension of the features.

[0100] S4: Based on the correlation between features and wear amount, the multi-domain feature matrix is ​​processed by a feature optimization algorithm to obtain the sensitive feature matrix;

[0101] Specifically, by performing feature sensitivity analysis on the historical cutting data of the milling cutter to be evaluated, the correlation coefficients between different features and the wear amount VB value can be obtained;

[0102] Based on a feature optimization strategy, the importance of features is ranked and filtered. One feature optimization evaluation criterion is calculated in this process. , Representing the multi-domain characteristic matrix The i-th feature in This indicates the actual wear on the flank face of the milling cutter. This represents the Pearson correlation coefficient function; redundant and irrelevant features can be further eliminated through the feature optimization evaluation criteria.

[0103] Principal component analysis or kernel principal component analysis is used to reduce the dimensionality of the screened features, eliminate collinearity among features, and obtain the sensitive feature matrix. This provides high-quality data for subsequent model input.

[0104] S5: Convert the sensitive feature matrix into a fused feature vector based on a weighted fusion strategy;

[0105] Specifically, the sensitive feature matrix By converting the weighted fusion strategy into a fusion feature vector, the fusion feature vector can be further calculated. ;

[0106] ;

[0107] Where k represents the number of sensitive features selected. This represents the weighting coefficient of the i-th sensitive feature. This represents the i-th sensitive feature in the sensitive feature matrix.

[0108] In this embodiment, the above weighting coefficients The weights are determined based on the contribution of each feature to the single-feature prediction model. Specifically, the Gini impurity or the decrease in out-of-bag error for each feature is calculated using a random forest regression model and used as the weights. This eliminates the instability of a single feature under specific working conditions, thus laying the foundation for accurate quantification of milling cutter wear.

[0109] S6: Input the fused feature vector into the wear prediction model to obtain the predicted wear amount sequence;

[0110] Specifically, by transforming the collected multi-domain features of cutting force into corresponding wear data, the processed data can more intuitively reflect the tool life. The purpose of this is to achieve a reliable conversion from features to wear, thus realizing an accurate mapping of physical quantities.

[0111] In this embodiment, the wear prediction model uses support vector regression or extreme learning machine. The model is trained with fused feature vectors as input and actual wear amount VB as label.

[0112] It should be noted that the calculated predicted wear sequence The wear prediction model uses support vector regression or extreme learning machine. The specific model construction adopts grid search method for cross-validation to optimize hyperparameters. The specific verification process is based on existing technology and will not be described in detail here.

[0113] S7: Perform a moving average filter on the predicted wear amount sequence to obtain a smooth wear amount curve;

[0114] The formula for calculating the smooth wear curve is: ;

[0115] Where i represents the index of the current data point. This represents the total width of the sliding window. Indicates the relative offset within the window. This represents the original predicted wear value at the (i+j)th position within the window range.

[0116] Specifically, for the wear quantity sequence output by the prediction model, there may be individual points where the prediction jumps. By using a moving average to calculate the mean, the influence of random prediction errors can be eliminated by smoothing out data fluctuations. This can effectively avoid misinterpreting individual abnormal prediction points as severe wear.

[0117] S8: Perform slope detection and threshold determination on the smooth wear curve. After threshold determination, the abnormal wear area of ​​the end mill to be evaluated is obtained.

[0118] In this embodiment, the wear curve is smoothed. The process of slope detection is as follows:

[0119] Extracting smooth wear curves Wear rate in ;

[0120] in, This represents the wear rate at time i. This indicates the amount of smooth wear at the current moment. This indicates the amount of smooth wear at the previous moment.

[0121] It should be noted that the smooth wear curve is extracted first. Wear rate in This is because it can be used to identify the inflection point where tool wear changes from the normal wear stage to the rapid wear stage, thus providing specific quantitative basis for the timing of tool changes and structural health monitoring of milling cutters;

[0122] A cubic spline interpolation algorithm is used to smooth the wear curve. The wear trend curve is obtained by fitting the curve, and the second derivative of the trend curve is calculated to find the inflection point.

[0123] For example, a cubic spline interpolation algorithm is used to smooth the wear curve. Interpolation is performed, and further processing is conducted based on the smoothed wear curve. Point set, construct system of equations; where, Represents the coefficient matrix. Let represent the unknown vector, and let be the value of the second derivative to be solved. Let be the right-hand vector, and be the difference term calculated from the function value and the node spacing. Solve the above three bending moment equations: Natural splines are used as boundary conditions to generate wear trend curves. Similarly, the moment when wear intensifies can be determined by analyzing the extreme points of the second derivative.

[0124] The specific process for determining the threshold of the smooth wear curve includes:

[0125] Calculate the smooth wear curve The difference between the value and the preset wear threshold ;

[0126] When the difference When the value is greater than 0, the corresponding area is identified as an abnormal area, i.e., the tool fails.

[0127] This application presents a quantitative assessment method for milling cutter wear based on multi-domain feature fusion of cutting force. Through smoothing and collaborative noise reduction, it effectively filters out machine tool vibration noise and cutting chatter interference while preserving the local abrupt changes and impact components of the cutting force signal to the greatest extent. This avoids the loss of tool breakage features caused by global smoothing, providing a high-quality signal foundation for subsequent assessment. A multi-domain feature matrix encompassing time, frequency, and time-frequency domains is constructed, comprehensively reflecting changes in cutting force amplitude, frequency structure shifts, and nonlinear non-stationary characteristics, overcoming the limitations of a single feature domain in describing complex wear states. The method introduces feature optimization based on Pearson correlation coefficient and a weighted fusion strategy based on random forest contribution, eliminating redundant and irrelevant features, eliminating feature collinearity, significantly reducing data dimensionality, and enhancing the robustness of the assessment system under different working conditions. Furthermore, through slope detection and threshold determination, it not only achieves accurate quantitative prediction of wear but also keenly captures the inflection point of tool transition from normal wear to rapid wear, providing a reliable basis for determining the optimal tool change timing and tool health monitoring. This helps to maximize tool life and reduce production costs while ensuring machining quality.

[0128] like Figure 2 As shown, this embodiment also provides a quantitative assessment system for milling cutter wear based on multi-domain feature fusion of cutting force, used to implement the above-mentioned quantitative assessment method for milling cutter wear based on multi-domain feature fusion of cutting force, including:

[0129] The signal acquisition module is used to perform cutting experiments on the milling cutter to be evaluated and acquire the original cutting force signal containing the wear characteristics of the milling cutter.

[0130] The signal acquisition module is connected to the force gauge, which has 3 channels and is used to acquire cutting forces in the X, Y, and Z directions.

[0131] The signal preprocessing module is used to perform smoothing and collaborative noise reduction on the original cutting force signal to obtain a smoothed output signal.

[0132] The feature extraction module is used to extract time-domain features, frequency-domain features, and time-frequency-domain features based on the output signal, and to construct a multi-domain feature matrix;

[0133] The feature extraction module includes:

[0134] The time-domain feature extraction unit is used to calculate the mean, root mean square (RMS), kurtosis, and impulse factor of the signal to obtain the time-domain feature vector T.

[0135] The frequency domain feature extraction unit is used to perform a fast Fourier transform on the signal to obtain the power spectral density, extract the main frequency amplitude, spectral centroid and spectral variance, and obtain the frequency domain feature vector P.

[0136] The time-frequency domain feature extraction unit is used to decompose the signal to the third level using wavelet packet transform, extract the energy features of 8 frequency bands, and calculate the wavelet packet entropy to obtain the time-frequency domain feature vector E;

[0137] The feature optimization module is used to process the multi-domain feature matrix to obtain the sensitive feature matrix based on the correlation between features and wear amount;

[0138] The feature fusion module is used to convert the sensitive feature matrix into a fused feature vector based on a weighted fusion strategy.

[0139] The wear prediction module is used to input the fused feature vector into the wear prediction model to obtain the predicted wear amount sequence;

[0140] The result smoothing module is used to perform moving average filtering on the predicted wear amount sequence to obtain a smoothed wear amount curve.

[0141] The anomaly detection module is used to detect the slope of the smooth wear curve and determine the threshold. After the threshold determination, the abnormal wear area of ​​the milling cutter to be evaluated is obtained.

[0142] The anomaly detection module includes:

[0143] The slope detection unit is used to extract the smooth wear curve. Wear rate in ;

[0144] in, This represents the wear rate at time i. This indicates the amount of smooth wear at the current moment. This indicates the amount of smooth wear at the previous moment;

[0145] The trend analysis unit is used to fit the smooth wear curve using a cubic spline interpolation algorithm to generate a wear trend curve.

[0146] Threshold determination unit, used to calculate smooth wear curve The difference between the value and the preset wear threshold ;

[0147] When the difference When the value is greater than 0, the corresponding area is identified as an abnormal area, i.e., the tool fails.

[0148] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for quantitative evaluation of milling cutter wear based on multi-domain feature fusion of cutting force, characterized in that, Includes the following steps: S1: Perform a cutting experiment on the end mill to be evaluated to obtain the original cutting force signal containing the wear characteristics of the end mill; Wherein, the original cutting force signal matrix in the original cutting force signal n is the number of sampling samples; S2: Perform smoothing and collaborative noise reduction processing on the original cutting force signal to obtain a smoothed output signal; S3: Extract time-domain features, frequency-domain features, and time-frequency-domain features based on the output signal, and construct a multi-domain feature matrix; S4: Based on the correlation between features and wear amount, the multi-domain feature matrix is ​​processed by a feature optimization algorithm to obtain a sensitive feature matrix; S5: Convert the sensitive feature matrix into a fused feature vector based on a weighted fusion strategy; S6: Input the fused feature vector into the wear prediction model to obtain the predicted wear amount sequence; S7: Perform a moving average filter on the predicted wear amount sequence to obtain a smoothed wear amount curve. S8: Perform slope detection and threshold determination on the smooth wear curve. After threshold determination, the abnormal wear area of ​​the end mill to be evaluated is obtained.

2. The method according to claim 1, wherein, The specific process for performing smoothing and collaborative noise reduction on the original cutting force signal to obtain a smoothed output signal includes: Set the moving window to 2a+1, where a is the order of the polynomial fitting and is a positive integer greater than or equal to 2; In the moving window, the original cutting force signal matrix For each data point, select 2m+1 data points within a window centered on that data point to form a subsequence. , where i represents the index of the current data point, m represents the half-window width as a positive integer, and m>a; Perform a k-th order polynomial least squares fit on the subsequence to obtain the fitted polynomial. ; Where k represents the order of the polynomial fitting, Indicates the summation index. The coefficients of the polynomial are represented by 't', and the relative position coordinates within the window are represented by 't'. ; Calculate the function value of the fitted polynomial at the center of the window, use it as the filtered output value of the data point, slide the window to the next data point, and repeat the above process until all data points have been traversed to obtain the smoothed output signal. .

3. The method for quantitatively evaluating milling cutter wear based on multi-domain feature fusion of cutting force according to claim 1, characterized in that, The specific process for constructing the multi-domain feature matrix includes: Extract the characteristics of the output signal in the time domain, frequency domain, and time-frequency domain; The process of extracting the time-domain feature vector T includes: calculating the mean, root mean square (RMS), kurtosis, and impulse factor of the signal; The frequency domain feature vector P extraction process includes: performing a fast Fourier transform on the signal to obtain the power spectral density, and extracting the main frequency amplitude, spectral centroid, and spectral variance; The time-frequency domain feature vector E extraction process includes: using wavelet packet transform to decompose the signal to the third level, extracting the energy features of 8 frequency bands, and calculating the wavelet packet entropy; Based on the obtained time-domain, frequency-domain, and time-frequency-domain feature vectors, a multi-domain feature matrix is ​​constructed.

4. The method for quantitatively evaluating milling cutter wear based on multi-domain feature fusion of cutting force according to claim 1, characterized in that, The specific process of obtaining the sensitive feature matrix by processing the multi-domain feature matrix through a feature optimization algorithm based on the correlation between features and wear amount includes: Calculate the correlation coefficient between each feature in the multi-domain feature matrix and the actual wear amount, and take the absolute value as the evaluation score. , Representing the multi-domain characteristic matrix The i-th feature in This indicates the actual wear on the flank face of the milling cutter. Represents the Pearson correlation coefficient function; A predetermined number of sensitive features are selected based on the evaluation scores; Principal component analysis or kernel principal component analysis is performed on the screened sensitive features to reduce dimensionality and eliminate feature collinearity, thus obtaining the sensitive feature matrix.

5. The method for quantitatively evaluating milling cutter wear based on multi-domain feature fusion of cutting force according to claim 1, characterized in that, The calculation formula for the weighted fusion strategy is as follows: ； Where k represents the number of sensitive features selected. This represents the weighting coefficient of the i-th sensitive feature. This represents the i-th sensitive feature in the sensitive feature matrix; in, The contribution of each feature to wear prediction is determined by calculating it using a random forest regression model.

6. The method for quantitatively evaluating milling cutter wear based on multi-domain feature fusion of cutting force according to claim 1, characterized in that, The formula for calculating the smooth wear curve is: ; Where i represents the index of the current data point. This represents the total width of the sliding window. Indicates the relative offset within the window. This represents the original predicted wear value at the (i+j)th position within the window range.

7. The method for quantitatively evaluating milling cutter wear based on multi-domain feature fusion of cutting force according to claim 1, characterized in that, The smooth wear curve The process of slope detection is as follows: Extracting smooth wear curves Wear rate in ; in, This represents the wear rate at time i. This indicates the amount of smooth wear at the current moment. This indicates the amount of smooth wear at the previous moment.

8. The method for quantitatively evaluating milling cutter wear based on multi-domain feature fusion of cutting force according to claim 1, characterized in that, The specific process for determining the threshold of the smooth wear curve includes: Calculate the smooth wear curve The difference between the value and the preset wear threshold ; When the difference When the value is greater than 0, the corresponding area is identified as an abnormal area, i.e., the tool fails.

9. A quantitative assessment system for milling cutter wear based on multi-domain feature fusion of cutting force, used to implement the quantitative assessment method for milling cutter wear based on multi-domain feature fusion of cutting force as described in any one of claims 1-8, characterized in that, include: The signal acquisition module is used to perform cutting experiments on the milling cutter to be evaluated and acquire the original cutting force signal containing the wear characteristics of the milling cutter. The signal acquisition module is connected to a force measuring instrument, which has 3 channels and is used to acquire cutting forces in the X, Y, and Z directions. The signal preprocessing module is used to perform smoothing and collaborative noise reduction processing on the original cutting force signal to obtain a smoothed output signal. The feature extraction module is used to extract time-domain features, frequency-domain features, and time-frequency-domain features based on the output signal, and to construct a multi-domain feature matrix; The feature extraction module includes: The time-domain feature extraction unit is used to calculate the mean, root mean square (RMS), kurtosis, and impulse factor of the signal to obtain the time-domain feature vector T. The frequency domain feature extraction unit is used to perform a fast Fourier transform on the signal to obtain the power spectral density, extract the main frequency amplitude, spectral centroid and spectral variance, and obtain the frequency domain feature vector P. The time-frequency domain feature extraction unit is used to decompose the signal to the third level using wavelet packet transform, extract the energy features of 8 frequency bands, and calculate the wavelet packet entropy to obtain the time-frequency domain feature vector E; The feature optimization module is used to process the multi-domain feature matrix to obtain a sensitive feature matrix based on the correlation between features and wear amount; The feature fusion module is used to convert the sensitive feature matrix into a fused feature vector based on a weighted fusion strategy; The wear prediction module is used to input the fused feature vector into the wear prediction model to obtain a predicted wear amount sequence. The result smoothing module is used to perform a moving average filter on the predicted wear amount sequence to obtain a smoothed wear amount curve. The anomaly detection module is used to detect the slope and determine the threshold of the smooth wear curve. After the threshold determination, the abnormal wear area of ​​the milling cutter to be evaluated is obtained. The anomaly detection module includes: The slope detection unit is used to extract the smooth wear curve. Wear rate in ; in, This represents the wear rate at time i. This indicates the amount of smooth wear at the current moment. This indicates the amount of smooth wear at the previous moment; The trend analysis unit is used to fit the smooth wear curve using a cubic spline interpolation algorithm to generate a wear trend curve. Threshold determination unit, used to calculate smooth wear curve The difference between the value and the preset wear threshold ; When the difference When the value is greater than 0, the corresponding area is identified as an abnormal area, i.e., the tool fails.

Images