A tool wear prediction method based on a multi-element fast iterative filter decomposition method

Abstract

The application discloses a tool wear prediction method based on a multi-element fast iterative filtering decomposition method and relates to the tool wear prediction field. The tool wear prediction method comprises the following steps: S1, establishing a wear prediction model based on a support vector machine model; S2, collecting original machining signals of a milling process of a machine tool through three-way vibration sensors arranged on a machine tool spindle; and S3, decomposing the original machining signals by using a multi-element fast iterative filtering decomposition method to obtain a plurality of multi-channel intrinsic modal components. The tool wear prediction method can predict the wear condition of a tool during a tool mechanical milling process based on an indirect monitoring method, does not need to stop and restart the equipment, reduces the milling precision error caused by frequent shutdown and disassembly of the tool, can reduce the detection time of the tool, does not affect the production efficiency, can predict the wear condition of the tool in time, and can replace the tool before the tool reaches the maximum wear value.

Description

Technical Field

[0001] This invention relates to the field of tool wear prediction, and specifically to a tool wear prediction method based on multivariate fast iterative filtering decomposition. Background Technology

[0002] In actual daily machining, cutting tools are subjected to continuous impact and extremely high temperatures, inevitably leading to tool wear and even breakage. Tool wear affects machine tool efficiency and machining accuracy, making tool wear status crucial in actual machining. Previously, tool condition assessment relied on operator experience, who judged tool condition by observing chips and listening to machining noise. This method is experience-dependent, prone to error, and premature tool replacement leads to resource waste, while delayed replacement reduces workpiece surface finish.

[0003] To address this issue, direct and indirect detection methods for tool wear have been proposed. 1) Direct detection involves directly measuring the wear area on the tool's flank face after each pass using instruments. This method requires stopping production and removing the tool, resulting in low production efficiency. 2) Indirect detection infers tool wear by detecting changes in physical quantities caused by tool wear. These indirect observation signals are typically acoustic emission signals, spindle power signals, etc. When linking observed signals with wear amounts, physical modeling and artificial intelligence methods are commonly used. Physical modeling establishes a model of the observed data and wear amount, using the observed data as input and the wear amount as output. However, the observed data is easily affected by noise, leading to large errors in the output wear amount. Artificial intelligence methods establish machine learning algorithms to fit a model of the observed physical quantities and wear values. During production, the real-time measured physical quantities are used as the output model to obtain the wear amount. However, this method requires a large amount of experimental data, thus consuming significant amounts of materials and tools, resulting in high costs. Furthermore, when the tool or workpiece material changes, the model needs to be rebuilt, which is time-consuming and labor-intensive. Summary of the Invention

[0004] To address the aforementioned shortcomings of existing technologies, this invention provides a simple, convenient, low-cost, fast, and highly specific tool wear prediction method based on multivariate fast iterative filtering decomposition.

[0005] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows:

[0006] A tool wear prediction method based on multivariate fast iterative filtering decomposition is provided, which includes the following steps:

[0007] S1: Establish a wear prediction model based on the support vector machine model;

[0008] S2: The original machining signals of the milling process are acquired by the three-dimensional vibration sensor installed on the machine tool spindle;

[0009] S3: The original processing signal is decomposed using the multivariate fast iterative filtering decomposition method to obtain several multi-channel intrinsic mode components;

[0010] S4: Calculate the weighted coefficient kurtosis index value of each multi-channel intrinsic mode component, and reconstruct the signal of the multi-channel intrinsic mode component with the largest weighted coefficient kurtosis index value to obtain the reconstructed signal;

[0011] S5: Extract the time-domain and frequency-domain features of the reconstructed signal, and perform principal component analysis on the time-domain and frequency-domain features of the reconstructed signal to obtain the dimension-reduced feature vector;

[0012] S6: Collect the wear area of ​​the tool to be predicted, and input the wear area and the dimensionality-reduced feature vector into the wear prediction model to obtain the tool wear prediction result.

[0013] Furthermore, the original machining signals include X-axis, Y-axis, and Z-axis signals; the multivariate fast iterative filtering decomposition method includes the following steps:

[0014] D1: Integrate the X-axis signal, Y-axis signal and Z-axis signal to obtain the integrated signal;

[0015] D2: The integrated signal is fed into the pre-filter to obtain the single-channel intrinsic mode component u;

[0016] The pre-filter includes the following:

[0017]

[0018]

[0019]

[0020] Where V(t) is the integrated signal at time t; N is the number of sample points; k is the number of extreme points; L is the filter length; and x is the delay time.

[0021] D3: Input the single-channel intrinsic mode component u into the fast iterative filter decomposition model to obtain m multi-channel intrinsic mode components IMFi, i = (1,2,3,...,m).

[0022] Furthermore, the fast iterative filtering decomposition model includes the following:

[0023]

[0024] Where i is the number of decomposition sequences, i = (1, 2, 3, ..., m); D is the diagonal matrix associated with the filter vectors; I is the identity matrix; N0 is the number of iterations in the decomposition process; Let be the transpose matrix of the single-channel intrinsic mode component u at the i-th decomposition.

[0025] Furthermore, the stopping criterion for the fast iterative filtering decomposition method is:

[0026]

[0027] in, For the single-channel intrinsic mode component at the (k+1)th step of the i-th decomposition in the inner loop; δ represents the single-channel intrinsic mode component at step k during the i-th decomposition of the inner loop; δ is the adjustment parameter.

[0028] Furthermore, the weighted coefficient kurtosis index value WSK = S·K ur ·|ρ|, where S is the sparsity of the multi-channel intrinsic mode components; K ur ρ is the kurtosis value of the multi-channel intrinsic mode components; ρ is the Pearson correlation coefficient of the multi-channel intrinsic mode components.

[0029]

[0030] m×n is the matrix size of the multi-channel intrinsic mode components, and τ is the number of non-zero elements;

[0031]

[0032] Where, x i denoted as the vibration amplitude corresponding to the discrete sequence points of the time-domain waveform of the multi-channel intrinsic modal components, where n is the number of discrete sequence points of the multi-channel intrinsic modal components;

[0033]

[0034] Where X is the original processing signal; Y is any multi-channel intrinsic mode component; cov(X,Y) is the covariance between X and Y; Let X be the standard deviation; The standard deviation of Y; μ X Let μ be the mean of X. Y Let Y be the mean of Y.

[0035] Furthermore, the extracted time-domain features include: signal average value, signal maximum value, signal minimum value, signal peak and valley values, signal root mean square value, signal standard deviation value, and signal kurtosis value;

[0036] Frequency domain features are extracted by performing a Fourier transform on time domain features; the Fourier transform includes the following formula:

[0037]

[0038] Where F(ω) is the frequency domain characteristic; ω is the frequency, t is the time, and e -iωt It is a complex function, and f(t) is a time-domain characteristic.

[0039] Furthermore, the process of establishing a wear prediction model includes the following steps:

[0040] A1: Randomly divide the sample set into a training set and a test set in a 2:1 ratio;

[0041] A2: Set the range of values ​​for the penalty coefficient c to [x1, x2] and the range of values ​​for the Gaussian kernel function g to [x3, x4]; substitute the training set into the Grey Wolf Optimized Support Vector Machine algorithm to iteratively optimize the model and obtain the optimal penalty coefficient c. i and the optimal Gaussian kernel function g i ;

[0042] A3: Based on the optimal penalty coefficient c i and the optimal Gaussian kernel function g i Establish a wear prediction model.

[0043] Furthermore, when the Grey Wolf Optimization Algorithm performs initial optimization on the Support Vector Machine model, the number of wolves is set to 20, the maximum number of iterations is 30, the optimization parameter is 2, and the parameter value range is [0.001, 100].

[0044] The beneficial effects of this invention are as follows:

[0045] 1. This invention is based on an indirect monitoring method, which can predict the wear of cutting tools during the mechanical milling process without stopping and restarting the equipment. This reduces the errors in milling accuracy caused by frequent machine stops and tool disassembly, and also reduces tool inspection time, without affecting production efficiency. At the same time, it can predict the wear of cutting tools in a timely manner, and replace the tools before they reach their maximum wear value, avoiding workpiece accuracy defects caused by tool wear, improving the yield rate, and thus effectively reducing the production cost of enterprises.

[0046] 2. The three-dimensional vibration signal of the tool is decomposed using a multivariate fast iterative filtering decomposition method, eliminating the need to pre-set the decomposition values ​​before calculation, thus accelerating the processing speed of the vibration signal and reducing the error caused by noise. The support vector machine is optimized using the Grey Wolf Optimized Support Vector Machine algorithm, which accelerates the convergence speed of the model and reduces the number of parameter adjustments during training. The combination of the multivariate fast iterative filtering decomposition method and the Grey Wolf optimization method enables faster and more accurate prediction of tool wear. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the process of the present invention;

[0048] Figure 2 This is the envelope spectrum of the original processing signal;

[0049] Figure 3 Screening chart for weighted sparsity kurtosis index;

[0050] Figure 4 This is a reduced-dimensional eigenvector envelope spectrum.

[0051] Figure 5 This is a schematic diagram comparing the decomposition time of the present invention with that of NA-MEMD and FA-MVEMD;

[0052] Figure 6 This is a schematic diagram comparing the decomposition time of the present invention with that of FA-MVEMD. Detailed Implementation

[0053] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0054] like Figure 1 As shown, a tool wear prediction method based on multivariate fast iterative filtering decomposition includes the following steps:

[0055] S1: Establish a wear prediction model based on the support vector machine model;

[0056] The process of establishing a wear prediction model includes the following steps:

[0057] A1: Randomly divide the sample set into a training set and a test set in a 2:1 ratio;

[0058] A2: Set the range of values ​​for the penalty coefficient c to [x1, x2] and the range of values ​​for the Gaussian kernel function g to [x3, x4]; input the training set into the Grey Wolf Optimized Support Vector Machine algorithm in MATLAB, and iteratively optimize the model to obtain the optimal penalty coefficient c. i and the optimal Gaussian kernel function g i ;

[0059] A3: Based on the optimal penalty coefficient c i and the optimal Gaussian kernel function g i Establish a wear prediction model.

[0060] Furthermore, when the Grey Wolf Optimization Algorithm performs initial optimization on the Support Vector Machine model, the number of wolves is set to 20, the maximum number of iterations is 30, the optimization parameter is 2, and the parameter value range is [0.001, 100].

[0061] S2: The original machining signals of the milling process are acquired by the three-axis vibration sensor installed on the machine tool spindle; the original machining signals include X-axis signals, Y-axis signals and Z-axis signals;

[0062] S3: The original processed signal is decomposed using a multivariate fast iterative filtering decomposition method to obtain several multi-channel intrinsic mode components; the multivariate fast iterative filtering decomposition method includes the following steps:

[0063] D1: Integrate the X-axis signal, Y-axis signal and Z-axis signal to obtain the integrated signal;

[0064] D2: The integrated signal is fed into the pre-filter to obtain the single-channel intrinsic mode component u;

[0065] The pre-filter includes the following:

[0066]

[0067]

[0068]

[0069] Where V(t) is the integrated signal at time t; N is the number of sample points, which is determined according to the actual situation. When collecting the original processed signal, the collection time period and time interval are set. The longer the time period and the smaller the interval, the larger the number of sample points; otherwise, the smaller the number of sample points; k is the number of extreme points; L is the filter length; and x is the delay time.

[0070] D3: Input the single-channel intrinsic mode component u into the fast iterative filter decomposition model to obtain m multi-channel intrinsic mode components IMFi, i = (1,2,3,...,m).

[0071] Fast iterative filtering decomposition models include the following:

[0072]

[0073] Where i is the number of decomposition sequences, i = (1, 2, 3, ..., m); D is the diagonal matrix associated with the filter vectors; I is the identity matrix; N0 is the number of iterations in the decomposition process; Let be the transpose matrix of the single-channel intrinsic mode component u at the i-th decomposition.

[0074] The stopping criterion for the fast iterative filtering decomposition method is:

[0075]

[0076] in, For the single-channel intrinsic mode component at the (k+1)th step of the i-th decomposition in the inner loop; δ represents the single-channel intrinsic mode component at step k during the i-th decomposition of the inner loop; δ is the adjustment parameter.

[0077] S4: Calculate the weighted coefficient kurtosis index value of each multi-channel intrinsic mode component, and reconstruct the signal of the multi-channel intrinsic mode component with the largest weighted coefficient kurtosis index value to obtain the reconstructed signal;

[0078] Weighted coefficient kurtosis index value WSK = S·K ur ·|ρ|, where S is the sparsity of the multi-channel intrinsic mode components; K ur ρ is the kurtosis value of the multi-channel intrinsic mode components; ρ is the Pearson correlation coefficient of the multi-channel intrinsic mode components.

[0079]

[0080] m×n is the matrix size of the multi-channel intrinsic mode components, and τ is the number of non-zero elements;

[0081]

[0082] Where, x i denoted as the vibration amplitude corresponding to the discrete sequence points of the time-domain waveform of the multi-channel intrinsic modal components, where n is the number of discrete sequence points of the multi-channel intrinsic modal components;

[0083]

[0084] Where X is the original processing signal; Y is any multi-channel intrinsic mode component; cov(X,Y) is the covariance between X and Y; Let X be the standard deviation; The standard deviation of Y; μ X Let μ be the mean of X. Y Let Y be the mean of Y.

[0085] Signal reconstruction is performed using Matlab, and the reconstruction formula includes the following:

[0086] IMFnew=Z(:,1)+Z(:,2)+X(:,1)+X(:,2)+Y(:,1)+Y(:,2);

[0087] Where Z(:,1), Z(:,2), X(:,1), X(:,2), Y(:,1) and Y(:,2) are the intrinsic modal components in the X-axis, Y-axis and Z-axis directions obtained by multivariate decomposition of the original processing signal, respectively.

[0088] S5: Extract the time-domain and frequency-domain features of the reconstructed signal, and perform principal component analysis on the time-domain and frequency-domain features of the reconstructed signal to obtain the dimension-reduced feature vector;

[0089] The extracted time-domain features include: signal average value, signal maximum value, signal minimum value, signal peak and valley values, signal root mean square value, signal standard deviation value, and signal kurtosis value;

[0090] Frequency domain features are extracted by performing a Fourier transform on time domain features; the Fourier transform includes the following formula:

[0091]

[0092] Where F(ω) is the frequency domain characteristic; ω is the frequency, t is the time, and e -iωt It is a complex function, and f(t) is a time-domain characteristic.

[0093] S6: Collect the wear area of ​​the tool to be predicted, and input the wear area and the dimensionality-reduced feature vector into the wear prediction model to obtain the tool wear prediction result.

[0094] Performance Comparison

[0095] In this embodiment, a sampling frequency of 10kHz is used, and the tool wear test conditions are shown in Table 1.

[0096] Table 1

[0097]

[0098] The vibration sensor collects the raw machining signal from the cutting tool and transmits it to the computer. The raw machining signal is as follows: Figure 2 As shown, the original processed signal is then decomposed using a multivariate fast iterative filtering algorithm to obtain 10 solid-state moduli. The weighted coefficient kurtosis index value for each solid-state modulus is then calculated, resulting in the following... Figure 3 The weighted sparsity kurtosis index screening chart shown is composed of... Figure 3 It can be seen that the weighting coefficient kurtosis index of the second mode vector IMF2 is the largest. Signal reconstruction is performed on the second mode vector IMF2, and time-domain and frequency-domain features are extracted from the reconstructed signal. Principal component analysis is then performed on the time-domain and frequency-domain features to obtain the following results: Figure 4 The reduced eigenvector envelope spectrum shown in the figure clearly highlights the features, which contrasts sharply with the messy envelope spectrum of the original processed signal.

[0099] Simultaneously, existing technologies, such as Noise-Assisted Multivariate Empirical Mode Decomposition (NA-MEMD) and Fast Adaptive Multivariate Empirical Mode Decomposition (FA-MVEMD), were used to record the original processing signals, resulting in the following data: Figure 5-6 The diagram showing the decomposition time comparison is composed of... Figure 5 It can be seen that the decomposition time required by the method MvFIF provided by this invention is significantly less than that required by the prior art NA-MEMD; the decomposition time required by MvFIF is similar to that required by the prior art FA-MVEMD, but from... Figure 6 It is evident that the decomposition time required by MvFIF is significantly less than that required by FA-MVEMD. MvFIF has a clear advantage in decomposition speed compared to NA-MEMD and FA-MVEMD, and can decompose the original signal more quickly.

[0100] The wear prediction model established by the dimensionality reduction feature vector is used to collect the wear area of ​​the tool and input it into the wear prediction model (hereinafter referred to as the MvFIF prediction model) and the wear prediction model established by NA-MEMD decomposition (hereinafter referred to as the NA-MEMD prediction model) to obtain the prediction results. The prediction result evaluation comparison table is shown in Table 2.

[0101] Table 2

[0102]

[0103] As shown in Table 2, the mean absolute error (MAE), mean square error (MSE), root mean square error (RMSEP), and mean absolute percentage error (MAPE) of the prediction results obtained by the MvFIF prediction model are all lower than those of the NA-MEMD prediction model. Therefore, the accuracy of the MvFIF prediction model is significantly higher than that of the NA-MEMD prediction model.

[0104] In summary, the solution provided by this invention represents a significant advancement over existing technologies and is inventive.

Claims

1. A tool wear prediction method based on multivariate fast iterative filtering decomposition, characterized in that, Includes the following steps: S1: Establish a wear prediction model based on the support vector machine model; S2: The original machining signals of the milling process are acquired by the three-dimensional vibration sensor installed on the machine tool spindle; S3: The original processing signal is decomposed using the multivariate fast iterative filtering decomposition method to obtain several multi-channel intrinsic mode components; S4: Calculate the weighted coefficient kurtosis index value of each multi-channel intrinsic mode component, and reconstruct the signal of the multi-channel intrinsic mode component with the largest weighted coefficient kurtosis index value to obtain the reconstructed signal; S5: Extract the time-domain and frequency-domain features of the reconstructed signal, and perform principal component analysis on the time-domain and frequency-domain features of the reconstructed signal to obtain the dimension-reduced feature vector; S6: Collect the wear area of ​​the tool to be predicted, and input the wear area and the dimensionality-reduced feature vector into the wear prediction model to obtain the tool wear prediction result.

2. The tool wear prediction method based on multivariate fast iterative filtering decomposition as described in claim 1, characterized in that, The original machining signals include X-axis signals, Y-axis signals, and Z-axis signals; the multivariate fast iterative filtering decomposition method includes the following steps: D1: Integrate the X-axis signal, Y-axis signal and Z-axis signal to obtain the integrated signal; D2: Substituting the integrated signal into the pre-filter yields the single-channel intrinsic mode components. u ; The pre-filter includes the following: ； ； ； Wherein, V ( t ) for time t The integrated signal below; N Number of sample points; k The number of extreme points; L This is the filter length; x For delay time; D3: Single-channel intrinsic mode components u The input to the fast iterative filter decomposition model yields m multi-channel intrinsic mode components (IMFs). i , i = (1,2,3...,m).

3. The tool wear prediction method based on multivariate fast iterative filtering decomposition as described in claim 2, characterized in that, The fast iterative filtering decomposition model includes the following: ； in, i The number of decomposed sequences, i = (1, 2, 3, ..., m); D This is the diagonal matrix associated with the filter vectors; I It is the identity matrix; N 0 represents the number of iterations in the decomposition process; For the single-channel intrinsic mode components at the i-th decomposition u The transpose of .

4. The tool wear prediction method based on multivariate fast iterative filtering decomposition as described in claim 3, characterized in that, The stopping criterion for the fast iterative filtering decomposition method is: ； in, For the i-th channel of the inner loop k+ The IMF in one step; For the i-th channel of the inner loop k The IMF's steps; To adjust the parameters.

5. The tool wear prediction method based on multivariate fast iterative filtering decomposition as described in claim 1, characterized in that, The weighted coefficient kurtosis index value ,in, S The sparsity of the intrinsic modal components; K ur The kurtosis value of the intrinsic modal component; ρ The Pearson correlation coefficient is the intrinsic modal component. ； m×n Let be the matrix size of the intrinsic modal components. τ The number of non-zero elements; ； in, x i The vibration amplitude corresponding to the discrete points of the time-domain waveform of the inherent modal components. n The number of discrete sequence points for the intrinsic modal components; ； in, X This is the original processing signal; Y For any intrinsic modal component; cov( X , Y )for X and Y Covariance between them; б X for X Standard deviation; б Y for Y Standard deviation; μ X for X The mean, μ Y for Y The mean.

6. The tool wear prediction method based on multivariate fast iterative filtering decomposition as described in claim 1, characterized in that, The extracted time-domain features include: signal average value, signal maximum value, signal minimum value, signal peak and valley values, signal root mean square value, signal standard deviation value, and signal kurtosis value; The frequency domain features are extracted by performing a Fourier transform on the time domain features; the Fourier transform includes the following formula: ； in, F ( ω () represents frequency domain characteristics; ω For frequency, t For time, It is a complex function. f ( t () represents the time-domain feature.

7. The tool wear prediction method based on multivariate fast iterative filtering decomposition as described in claim 1, characterized in that, The process of establishing the wear prediction model includes the following steps: A1: Randomly divide the sample set into a training set and a test set in a 2:1 ratio; A2: Set the penalty coefficients separately. c The range of values ​​[ x 1, x 2] and Gaussian kernel function g The range of values ​​[ x 3, x 4] The training set is fed into the Grey Wolf Optimized Support Vector Machine algorithm to iteratively optimize the model and obtain the optimal penalty coefficient. c i and the optimal Gaussian kernel function g i ; A3: Based on the optimal penalty coefficient c i and the optimal Gaussian kernel function g i Establish a wear prediction model.

8. The tool wear prediction method based on multivariate fast iterative filtering decomposition as described in claim 7, characterized in that, When the gray wolf optimized support vector machine algorithm performs initial optimization on the support vector machine model, the wolf pack size is set to 20, the maximum number of iterations is 30, the optimization parameter is 2, and the parameter value range is [0.001, 100].

Images