A milling chatter monitoring method based on Grey Wolf optimized VMD and multi-entropy features

Through the method of optimizing variational modal decomposition and multi-entropy features, the problem of signal complexity and parameter optimization in milling flutter monitoring is solved, and the accurate identification and efficiency improvement of milling processing state is achieved, and the effective distinction between stable and mild and severe flutter.

CN114881067BActive Publication Date: 2024-11-19LIAONING UNIVERSITY OF PETROLEUM AND CHEMICAL TECHNOLOGY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210255231.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-03-15
Publication Date
2024-11-19
Estimated Expiration
2042-03-15

AI Technical Summary

Technical Problem

Existing milling flutter monitoring technology is difficult to effectively identify complex nonlinear signals, especially in modal aliasing and parameter optimization, which leads to insufficient decomposition efficiency and accuracy, and inconsistent efficiency and accuracy of different entropy characteristics, resulting in processing state misjudgment or misjudgment.

Method used

The variational modal decomposition and multi-entropy feature method based on gray wolf optimization are adopted. The periodic signal is filtered out through a comb filter. After EMD decomposition, the first signal reconstruction is carried out based on the correlation coefficient and energy characteristics, and the gray wolf optimization algorithm is used. Looking for the VMD parameters, further perform the second signal reconstruction, introduce multi-scale arrangement entropy and fuzzy entropy, and extract the milling flutter characteristics.

Benefits of technology

It improves the robustness and accuracy of VMD decomposition, effectively distinguishes the milling processing state, reduces the modal aliasing effect, improves the accuracy and efficiency of flutter recognition, and can clearly distinguish between stable, mild and severe flutter.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114881067B_ABST
    Figure CN114881067B_ABST
Patent Text Reader

Abstract

The invention relates to a milling chatter monitoring method based on grey wolf optimized VMD and multi-entropy features, comprising: acquiring a processing signal; filtering out the periodic signal frequency in the original signal based on the acquired processing signal; based on the filtered signal, using an empirical mode decomposition (EMD) algorithm to calculate the correlation coefficient and energy of each order of intrinsic mode function (IMF), and selecting the IMF component with the largest correlation coefficient and energy as the first reconstruction signal; based on the first reconstruction signal, using a GWO algorithm, taking the maximum energy entropy as the fitness function of GWO, and optimizing two parameter combinations of VMD; after optimizing, performing VMD decomposition, calculating the energy (ER) and energy entropy (EE) characteristics of each order of IMF components, and selecting the IMF component with larger ER and EE characteristics as the second reconstruction signal; based on the second reconstruction signal, calculating its multi-scale permutation entropy (MPE) and multi-scale fuzzy entropy (MFE); the invention realizes milling chatter monitoring and effectively distinguishes the milling processing state.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of mechanical processing state monitoring, and in particular relates to a milling chatter monitoring method based on multi-entropy characteristics of grey wolf optimized variational modal decomposition. Background Art

[0002] In metal milling, especially in the processing of low-rigidity workpieces, chatter is a key factor affecting many aspects such as surface quality, processing efficiency and tool life. Therefore, milling chatter online monitoring technology has been a hot topic in the manufacturing industry for many years. Milling chatter monitoring refers to obtaining the physical signal of milling processing through appropriate sensors, then preprocessing the signal, extracting signal features, replacing the original signal information with simplified features to reduce the signal dimension and improve monitoring efficiency, and finally identifying the processing state (stable, slight chatter, severe chatter) based on the extracted features. Chatter signals are nonlinear, non-stationary and multi-component. Therefore, in engineering practice, it is difficult to directly identify chatter based on the time domain waveform or spectrum of the sensor signal. For chatter signals with complex components, it is generally necessary to pre-process them and decompose them into a series of nonlinear components with clear physical meanings, and then extract feature quantities sensitive to chatter faults from these simple components, so as to realize online monitoring of the processing state.

[0003] EMD is an adaptive method suitable for analyzing nonlinear and non-stationary signals, with a high signal-to-noise ratio. At present, the EMD algorithm is increasingly used in milling chatter monitoring, bearing fault detection and other fields. However, due to the defects of the algorithm itself, modal aliasing and endpoint effects are very likely to occur in practical applications, resulting in invalid decomposition results. To solve this problem, chatter monitoring methods such as EEMD, VMD and data-driven have been proposed one after another, but there are still some limitations in their application, such as the long decomposition time of the EEMD algorithm, the VMD parameter optimization problem and the data volume requirements of the data-driven algorithm. The VMD algorithm has the characteristics of non-recursion, adaptive decomposition and solid theoretical foundation, but the choice of its penalty factor and decomposition level always restricts the decomposition efficiency and decomposition accuracy of VMD. Many optimization algorithms have been applied to seek the optimal parameter combination. In terms of chatter characteristics, energy entropy, sample entropy, approximate entropy, etc. have been applied one after another, which is to evaluate the degree of chaos of the waveform frequency, but when extracting its features, each type of entropy has different efficiency and accuracy, resulting in inconsistent chatter recognition results and misjudgment or omission of the machining state. Summary of the invention

[0004] In view of the problems existing in the above-mentioned prior art, the present invention proposes a milling chatter monitoring method based on Grey Wolf optimized variational mode decomposition and multi-entropy features. First, the periodic signal in the original signal is filtered out, and then the filtered signal is decomposed by utilizing the advantage of EMD decomposition with a higher signal-to-noise ratio, and then the noise is filtered out, and the first signal reconstruction is performed according to the correlation coefficient and energy characteristics. Based on the first reconstructed signal, the VMD parameters are globally optimized using the Grey Wolf optimization algorithm, and the IMF after the optimized VMD decomposition is reconstructed for the second time according to the energy and energy entropy characteristics. Based on the second reconstructed signal, the concept of multi-scale is introduced, and the multi-scale permutation entropy and multi-scale fuzzy entropy are extracted to realize milling chatter monitoring.

[0005] The technical solution of the present invention is as follows:

[0006] The present invention provides a milling chatter monitoring method based on grey wolf optimization variational mode decomposition and multi-entropy features, comprising:

[0007] S1. Obtain processing signals through a signal acquisition device;

[0008] S2. Based on the acquired processing signal, a comb filter is used to filter out the periodic signal frequency in the original signal; the periodic signal includes: the base frequency and multiple frequency of the spindle speed, the base frequency and multiple frequency of the tooth pass frequency;

[0009] S3, based on the filtered signal, the EMD algorithm is used to decompose it, the correlation coefficient and energy of each order of the intrinsic mode function (IMF) are calculated, and the IMF component with the largest correlation coefficient and energy is selected as the first reconstructed signal;

[0010] S4. Based on the first reconstruction signal, the Grey Wolf Optimization Algorithm (GWO) is used, and the maximum energy entropy is used as the fitness function of GWO to optimize the combination of VMD penalty factor and decomposition layer number;

[0011] S5. After optimizing the GWO parameters, VMD decomposition is performed to calculate the energy (ER) and energy entropy (EE) characteristics of each order IMF component, and the IMF component with larger ER and EE characteristics is selected as the second reconstruction signal;

[0012] S6. Calculate the MFE and MPE of the second reconstructed signal.

[0013] In the step S1, the signal acquisition device includes a microphone, a force sensor, a current sensor, and an acceleration sensor.

[0014] In the step S1, the acquired processing signals include sound signals, cutting force signals, current signals, and vibration signals.

[0015] In the step S1, the acquired processing signal will be synchronously uploaded to the PC.

[0016] The step S2 comprises:

[0017] S2a, obtaining the base frequency and multiple frequency of the spindle speed according to the spindle speed;

[0018] The fundamental frequency of the spindle speed = spindle speed / 60;

[0019] S2b, obtaining the base frequency and multiple frequency of the tooth-passing frequency according to the base frequency of the spindle speed and the number of teeth of the cutter;

[0020] The fundamental frequency of the tooth-through frequency = the number of teeth on the cutter * the fundamental frequency of the spindle speed.

[0021] The step S3 comprises:

[0022] S3a, calculating the correlation coefficient according to formula 1;

[0023] Formula 1:

[0024]

[0025]

[0026]

[0027] Among them, μ A and σ A are the mean and standard deviation of A, μ B and σ B are the mean and standard deviation of B respectively;

[0028] S3b, calculate the energy according to formula 2

[0029] Formula 2:

[0030]

[0031] where u i (t) represents the i-th order IMF component after EMD decomposition.

[0032] The step S4 comprises:

[0033] The basic steps of the GWO optimization algorithm are as follows:

[0034] 1. Initialize parameters and use maximum energy entropy as the fitness function;

[0035] 2. Initialize the gray wolf position;

[0036] 3. Calculate the value of the fitness function at different positions;

[0037] 10. Compare the values ​​of the fitness function and update the position of the gray wolf;

[0038] 11. Determine whether the maximum fitness function value or the maximum number of iterations has been reached;

[0039] 12. Output the optimal parameter combination.

[0040] The step S5 comprises:

[0041] S5a, calculate the IMFs energy according to formula 2;

[0042] S5b, calculate the IMFs energy entropy according to formula 3;

[0043] Formula 3:

[0044]

[0045] Among them, T i =R i / R represents the ratio of the energy of a certain order IMF to the energy of all IMFs.

[0046] The step S6 comprises:

[0047] S6a, calculate MPE,

[0048] (1) Calculate the coarse-grained time series according to Formula 4;

[0049] Formula 4:

[0050]

[0051] Among them, τ represents the scale factor, and the length of the time series after coarse-graining is N / τ;

[0052] (2) After coarse-graining the original sequence, the PE of the corresponding scale is calculated;

[0053] S6b, calculate MFE,

[0054] (1) Calculate the coarse-grained time series according to Formula 4;

[0055] (2) Calculate FE according to Formula 5:

[0056] Formula 5:

[0057] X(i)=[u(i),u(i+1),…,u(i+m-1)]-u0(i),i=1,2,…,N-m+1

[0058] in,

[0059] Introduce fuzzy membership function:

[0060]

[0061] Among them, r represents the similarity tolerance,

[0062]

[0063]

[0064] is the maximum absolute distance between window vectors X(i) and X(j).

[0065]

[0066] in,

[0067]

[0068] The beneficial effects of the present invention are as follows: milling chatter monitoring is realized by optimizing VMD and multi-entropy features by Gray Wolf. By introducing the concept of multi-scale, additional observation perspectives are provided at different time scales, and then the multi-scale permutation entropy and multi-scale fuzzy entropy features of the reconstructed signal are extracted, which effectively distinguishes the milling processing state (stable, slight chatter, severe chatter). Considering the influence of the penalty factor and the number of decomposition levels in the VMD algorithm on the decomposition efficiency and accuracy, the optimal parameter combination is found through the GWO optimization algorithm, making the VMD decomposition more robust. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 A flow chart of milling chatter monitoring based on gray wolf optimized VMD and multi-entropy features of the present invention;

[0070] Figure 2 It is the time domain and frequency domain waveform diagram of the simulation signal;

[0071] Figure 3 The time domain and frequency domain diagrams before and after filtering under different processing conditions;

[0072] Figure 4 Optimize the algorithm flow chart for the gray wolf;

[0073] Figure 5 Comparison of frequency distribution and energy entropy of IMFs;

[0074] Figure 6 The time domain and frequency domain diagrams are the second reconstructed signal;

[0075] Figure 7 are the MPE and MFE of the simulation signal.

[0076] Figure 8 Time domain and frequency domain diagrams of stable and chattering signals.

[0077] Fig. 9 This is a comparison of the time domain and frequency domain waveforms before and after filtering.

[0078] Fig.10 Optimize the VMD decomposition and spectrum of IMFs for GWO.

[0079] Fig.11 It is a comparison diagram of multi-scale arrangement entropy under stable and chattering processing states.

[0080] Fig.12 It is a comparison diagram of multi-scale fuzzy entropy under stable and chattering processing states. DETAILED DESCRIPTION

[0081] In order to better explain the present invention and facilitate understanding, the present invention is described in detail below through specific implementation methods.

[0082] like Figure 1 As shown, this embodiment provides a milling chatter monitoring method based on the fusion of the EMD algorithm and the optimized VMD algorithm, specifically, comprising the following steps:

[0083] S1, obtaining milling processing signal;

[0084] The sound signal, cutting force signal, current signal, and vibration signal of the machining process are obtained by using microphones, force sensors, current sensors, and acceleration sensors. The obtained signals are uploaded to the PC synchronously. This example verifies the effectiveness of this method by simulating the chatter signal of milling machining. The simulation signal is as follows:

[0085] s1=5sin(f SP ·2πt)+3sin(2f SP ·2πt)+2sin(3f SP ·2πt), 0<t≤1

[0086]

[0087] s3=25cos(f CH ·2πt+0.5sin(f SP ·2πt)), 1.5≤t≤2.5

[0088]

[0089] Among them, s1 is the periodic signal part, including the fundamental frequency and multiples of the spindle speed and the fundamental frequency and multiples of the tooth-through frequency; s2 is the slight vibration part; s3 is the severe vibration part. Figure 2 The time domain and frequency domain diagrams of the simulated signal.

[0090] S2. Based on the simulation signal, a comb filter is used to filter out periodic signals in the simulation signal, including the base frequency and multiples of the spindle speed and the base frequency and multiples of the tooth pass frequency.

[0091] Spindle speed base frequency = spindle speed / 60

[0092] Gear pass frequency base frequency = number of milling cutter teeth * spindle speed base frequency

[0093] Figure 3 The time domain and frequency domain waveforms before and after filtering of different processing states under simulation signals.

[0094] S3. Based on the filtered signal, EMD decomposition is used to obtain IMFs. In order for the selected IMFs to fully express the main information of the original signal, the energy ratio and correlation coefficient analysis of IMFs are shown in Table 1.

[0095] Table 1 Energy ratio and correlation coefficient of each order IMF

[0096]

[0097] After comparative analysis, it was found that the energy proportion and correlation coefficient of IMF1 were the highest, which means that it contains a lot of main information of the original signal, so IMF1 was selected as the first reconstructed signal. In addition, this method avoids the modal aliasing effect of EMD decomposition and retains valuable signals.

[0098] The correlation coefficient analysis is shown in Formula 1.

[0099] Formula 1:

[0100]

[0101]

[0102]

[0103] Among them, μ A and σ A are the mean and standard deviation of A, μ B and σ B are the mean and standard deviation of B respectively.

[0104] Energy analysis is shown in Formula 2.

[0105] Formula 2:

[0106]

[0107] where u i (t) represents the i-th order IMF component after EMD decomposition.

[0108] S4. Based on the first reconstructed signal, the Grey Wolf Algorithm (GWO) is used to globally optimize the penalty factor and the number of decomposition levels in VMD decomposition.

[0109] Assume that in a two-dimensional space, the position coordinates of the gray wolf are set to (K, α), and the gray wolf position is updated by comparing the size of the fitness function value. In the present invention, the maximum energy entropy criterion is used as the fitness function. Compare the size of the gray wolf fitness function value, and assign the optimal first three parameter combinations to α, β, and δ in turn. For specific implementation steps, see Figure 4 The initialization parameters of the Grey Wolf Optimization Algorithm are shown in Table 2.

[0110] Table 2 GWO parameter initialization

[0111]

[0112] Among them, M represents the size of the wolf pack, G max represents the maximum number of iterations, Dim is the search dimension, L1 and L2 are the search ranges of VMD parameters K and α respectively.

[0113] First, the maximum energy entropy criterion is used as the fitness function to perform GWO parameter optimization on the VMD parameters (decomposition level, penalty factor); then, VMD decomposition is performed, the energy entropy of different IMFs components is calculated, and the spectrum of their IMFs is compared. After the parameter optimization calculation, the optimal parameter combination of the simulation signal is obtained (4, 3500). For this parameter combination, the frequency distribution and energy entropy analysis of VMD decomposition are shown in Figure 5 .

[0114] S5. After optimizing the GWO parameters, VMD decomposition is performed to calculate the energy (ER) and energy entropy (EE) characteristics of each order IMF component, and the IMF component with larger ER and EE characteristics is selected as the second reconstructed signal.

[0115] In order to limit the influence of EMD modal aliasing and obtain rich flutter information, GWO-VMD is used to extract flutter frequency. The energy and energy entropy characteristics of each order IMF component are calculated using Formula 3 and Formula 4.

[0116] Formula 3:

[0117]

[0118] where u i (t) represents the i-th order IMF component after VMD decomposition.

[0119] Formula 4:

[0120]

[0121] Among them, T i =R i / R represents the ratio of the energy of a certain order IMF to the energy of all IMFs.

[0122] GWO-VMD further subdivides the frequency band of IMFs and separates modal aliasing. It can be seen from the simulation signal that the flutter frequencies of mild flutter and severe flutter are 108Hz, 133Hz and 158Hz respectively. It can be seen from Table 3 that the third-order IMF components with higher energy ratio and energy entropy values ​​are IMF2, IMF3 and IMF4 respectively. Therefore, the above third-order IMF components are used as the second reconstruction signal.

[0123] Table 3 Energy ratio and energy entropy of each order IMF

[0124]

[0125] from Figure 6 It can be seen that the reconstructed signal contains a lot of chatter information, and by analyzing the Hilbert frequency, it is found that the comb filter and EMD decomposition have filtered out periodic signals and useless information.

[0126] S6. Calculate MFE and MPE based on the reconstructed signal.

[0127] According to formula 5, the reconstructed signal is coarse-grained.

[0128] Formula 5:

[0129]

[0130] Among them, τ represents the scale factor, and N / τ is the length of the time series after coarse-graining.

[0131] After coarse-graining, the permutation entropy and fuzzy entropy at each time scale are calculated as the milling chatter feature. Figure 7 are the MPE and MFE of the simulation signal.

[0132] from Figure 7 It can be seen that MFE and MPE can clearly distinguish the three processing states (stable, slight chatter, and severe chatter). By introducing the concept of multi-scale, the characteristic indicators in different dimensions can be expressed more clearly.

[0133] experiment:

[0134] The milling experiment was carried out on a vertical CNC milling machine. The tool used was a Ti-Al-N alloy flat-end milling cutter with 4 teeth. The workpiece material was a Ti-6Al-4V aluminum alloy thin-walled part (100mm×200mm×10mm). The acceleration signal was obtained by the IEPE triaxial acceleration sensor B05Y31 with a sampling frequency of 10000Kz. This milling experiment adopted the side milling processing method, and the experimental processing parameters are shown in Table 4. The entire milling process was carried out under dry milling conditions.

[0135] Table 4 Milling experimental parameters

[0136]

[0137] Figure 8 These are the signals of Experiment 1 and Experiment 2 collected by acceleration sensors respectively. Figure 8 a and Figure 8 b is the time domain and frequency domain waveform during stable processing; Figure 8 c and Figure 8 d is the time domain and frequency domain waveform when chatter occurs. Based on the signals collected in Experiments 1 and 2, the base frequency and multiple frequencies of the spindle speed and tooth pass frequency are filtered using a comb filter. The calculation method is as follows:

[0138] Spindle speed base frequency = spindle speed / 60 = 10Hz

[0139] The fundamental frequency of tooth-passing frequency = number of milling cutter teeth * fundamental frequency of spindle speed = 40Hz.

[0140] Fig. 9 The time domain and frequency domain comparison diagrams of the two sets of experimental signals before and after filtering are shown in Figure 5. It can be seen from the figure that the comb filter effectively reduces the amplitude of the spindle rotation frequency and the tooth pass frequency. After filtering the periodic signal, the chatter frequency can be clearly seen from the frequency domain image. However, in actual production and processing, after filtering the periodic signal, the original signal still contains noise signals, which affects the characteristics of the signal. Therefore, the EMD decomposition algorithm is used for the filtered signal. This algorithm has the characteristics of a high signal-to-noise ratio. By selecting IMFs through correlation coefficients and energy ratios to complete signal reconstruction, a good noise reduction effect can be achieved. Tables 5 and 6 show the correlation coefficients and energy ratios of IMFs of each order for the two sets of experimental data.

[0141] Table 5 Correlation coefficients and energy ratios of IMFs of various orders in Experiment 1

[0142]

[0143] Table 6 Correlation coefficients and energy ratios of IMFs of various orders in Experiment 2

[0144]

[0145] According to the above table, it can be found that the energy ratio and correlation coefficient of the first-order IMF component are the highest, and it contains a lot of information of the original signal. Therefore, the first-order IMF is selected as the reconstructed signal.

[0146] In order to solve the problem of determining the penalty factor and the number of decomposition levels in VMD decomposition, the present invention adopts the gray wolf optimization algorithm to globally optimize its parameters. It is assumed that in a two-dimensional space, the position of the gray wolf is (K, α). The update of the gray wolf position is carried out by comparing the size of the fitness function. The present invention adopts the principle of the maximum energy entropy value of the fitness function. By comparing the fitness function value of the gray wolf's position, the gray wolf position is updated and iterated, and finally the maximum number of iterations or the maximum energy entropy value is reached, and the optimal parameter combination is output. The advantage of this algorithm is that the energy entropy calculation time is short, the efficiency of the optimization algorithm is greatly improved, and the robustness is strong. Table 7 shows the initial optimization parameters of the gray wolf optimization algorithm. After optimization by the gray wolf optimization algorithm, the number of decomposition layers and penalty factors of the two groups of experiments are (7,1163) and (7,4489), respectively.

[0147] Table 7 GWO initialization parameters

[0148]

[0149] The optimized parameters are brought into VMD decomposition to observe the distribution of the signal center frequency. The VMD decomposition results of the two groups of experimental signals and the spectrum of IMFs are shown in Fig.10 As shown. From the decomposition results, it can be seen that the VMD results determined by GWO optimization clearly extract the information of each frequency band. After verifying the effectiveness of GWO-VMO decomposition, according to the principle that energy will increase sharply when flutter occurs, the energy and energy entropy characteristics of each order of IMFs are calculated, and the two-order IMFs with relatively large EE and ER eigenvalues ​​are extracted as the second reconstruction signal. Table 8 shows the energy and energy entropy characteristics of each order of IMFs.

[0150] Table 8 EE and ER characteristics of IMFs

[0151]

[0152] According to Table 8, in Experiment 1, the third and fourth order IMF components have larger EE and ER features and are selected as reconstructed signals; in Experiment 2, the third and sixth order IMF components have larger EE and ER features and are selected as reconstructed signals; the multi-scale fuzzy entropy and multi-scale permutation entropy of the signal are calculated to discriminate the processing state. By introducing the concept of multi-scale, the change of entropy value can be observed from different time dimensions. Both permutation entropy and fuzzy entropy are indicators to measure the complexity or disorder of time series. The difference is that permutation entropy introduces the idea of ​​permutation when calculating the complexity between reconstructed subsequences, and fuzzy entropy introduces the idea of ​​fuzziness when calculating the complexity between reconstructed subsequences, making its division more detailed and more scientific.

[0153] Fig.11 and Fig.12 This is a comparison chart of multi-scale permutation entropy and multi-scale fuzzy entropy under two processing states: stable and flutter. In the comparison of multi-scale permutation entropy, the permutation entropy of time scales 1, 5, and 7 is more obvious, and other scales are more fuzzy. Based on the theorem of permutation entropy, it is found that the accuracy of multi-scale permutation entropy calculation depends on the size of the time series. The larger the time series, the more accurate it is. In the comparison of multi-scale fuzzy entropy, due to the advantages of fuzzy entropy itself, the idea of ​​fuzziness is added to it, making the distinction more accurate. Therefore, comprehensively judging, the milling flutter monitoring method based on gray wolf optimized VMD and multi-entropy features proposed in this invention patent can effectively distinguish milling processing states.

Claims

1. A milling chatter monitoring method based on gray wolf optimized VMD and multi-entropy features, characterized in that The steps include: S1. Obtain processing signals through a signal acquisition device; S2. Based on the acquired processing signal, a comb filter is used to filter out the periodic signal frequency in the original signal; the periodic signal includes: the base frequency and multiple frequency of the spindle speed, the base frequency and multiple frequency of the tooth pass frequency; S3, based on the filtered signal, the EMD algorithm is used to decompose it, the correlation coefficient and energy of each order of the intrinsic mode function IMF are calculated, and the IMF component with the largest correlation coefficient and energy is selected as the first reconstruction signal; S4, based on the first reconstructed signal, the Grey Wolf Optimization Algorithm GWO is used, and the maximum energy entropy is used as the fitness function of GWO to optimize the two parameter combinations of VMD; S5. After optimizing the GWO parameters, VMD decomposition is performed to calculate the energy ER and energy entropy EE characteristics of each order IMF component, and the two-order IMFs with the largest EE and ER eigenvalues ​​are extracted as the second reconstruction signal; S6. Calculate the MFE and MPE of the second reconstructed signal based on the second reconstructed signal; The step S3 comprises: S3a, calculating the correlation coefficient according to formula 1; Formula 1: Among them, μA and σA are the mean and standard deviation of A respectively, μB and σB are the mean and standard deviation of B respectively; S3b, calculate the energy according to formula 2, Formula 2: Where ui(t) represents the i-th order IMF component after EMD decomposition; The step S4 comprises: The basic steps of the GWO optimization algorithm are as follows:

1. Initialize parameters and use maximum energy entropy as the fitness function; 2. Initialize the gray wolf position; 3. Calculate the value of the fitness function at different positions; 4. Compare the values ​​of the fitness function and update the position of the gray wolf; 5. Determine whether the maximum fitness function value or the maximum number of iterations has been reached; 6. Output the optimal parameter combination.

2. A milling chatter monitoring method based on grey wolf optimized VMD and multi-entropy features according to claim 1, characterized in that In the step S1, the signal acquisition device includes a microphone, a force sensor, a current sensor, and an acceleration sensor.

3. The milling chatter monitoring method based on gray wolf optimized VMD and multi-entropy features according to claim 1 is characterized in that In the step S1, the acquired processing signals include sound signals, cutting force signals, current signals, and vibration signals.

4. The milling chatter monitoring method based on gray wolf optimized VMD and multi-entropy features according to claim 1 is characterized in that In the step S1, the acquired processing signal will be synchronously uploaded to the PC.

5. The milling chatter monitoring method based on gray wolf optimized VMD and multi-entropy features according to claim 1 is characterized in that The step S2 comprises: S2a, obtaining the base frequency and multiple frequency of the spindle speed according to the spindle speed; The fundamental frequency of the spindle speed = spindle speed / 60; S2b, obtaining the base frequency and multiple frequency of the tooth-passing frequency according to the base frequency of the spindle speed and the number of teeth of the cutter; The fundamental frequency of the tooth-through frequency = the number of teeth on the cutter * the fundamental frequency of the spindle speed.

6. The milling chatter monitoring method based on gray wolf optimized VMD and multi-entropy features according to claim 1 is characterized in that The step S5 comprises: S5a, calculate the IMFs energy according to formula 2; S5b, calculate the IMFs energy entropy according to formula 3; Formula 3: Where Ti = Ri / R represents the ratio of the energy of a certain order IMF to the energy of all IMFs.

7. The milling chatter monitoring method based on gray wolf optimized VMD and multi-entropy features according to claim 1 is characterized in that The step S6 comprises: S6a, calculate MPE, (1) Calculate the coarse-grained time series according to Formula 4; Formula 4: Among them, τ represents the scale factor, and the length of the time series after coarse-graining is N / τ; (2) After coarse-graining the original sequence, the PE of the corresponding scale is calculated; S6b, calculate MFE, (1) Calculate the coarse-grained time series according to Formula 4; (2) Calculate FE according to Formula 5: Formula 5: X(i)=[u(i),u(i+1),…,u(i+m-1)]-u0(i),i=1,2,…,N-m+1 in, Introduce fuzzy membership function: Among them, r represents the similarity tolerance, is the maximum absolute distance between window vectors X(i) and X(j); in,