High-speed dry milling flutter monitoring method based on real-time cutting force signal

Abstract

The invention discloses a high-speed dry milling flutter monitoring method based on a real-time cutting force signal, which comprises the following steps of: intercepting a three-way cutting force signal in a high-speed dry milling process by using a reverse exponential attenuation window function, and decomposing the intercepted signal by using a successive variational mode decomposition method; according to the normalized frequency deviation degree and the normalized spectral entropy, a flutter component containing flutter information is extracted to reconstruct a flutter signal, the energy ratio of the flutter signal to an original signal is calculated, cutting parameters are combined to form a feature vector, and the machining stability state of high-speed dry milling is judged based on a Gaussian Bayes classifier to achieve flutter monitoring.

Description

Technical Field

[0001] The present invention relates to the technical field of mechanical manufacturing processing chatter monitoring, and in particular to a high-speed dry milling processing chatter monitoring method based on real-time cutting force signals. Background Art

[0002] High-speed dry milling is an advanced machining technology with great development prospects. It can complete rapid machining without the use of cutting fluid, which not only helps reduce machining costs but also avoids environmental pollution caused by cutting fluid emissions. However, due to its extremely high cutting speed, high-speed dry milling can make the process system more susceptible to machining chatter due to the regeneration effect under unreasonable process parameters. Chatter can lead to deterioration of the quality of the workpiece's machined surface and increased tool wear. Long-term chattering can cause irreversible damage to the machine tool's spindle system. Therefore, how to monitor chatter during high-speed dry milling and promptly identify whether chatter is occurring in the process system is one of the key technologies to ensure machining quality, extend tool life, and protect the machine tool's spindle system. Summary of the Invention

[0003] In view of the above-mentioned deficiencies in the prior art, the technical problem to be solved by the present invention is: to provide a high-speed dry milling chatter monitoring method based on real-time cutting force signals, use a sliding window strategy to intercept the real-time signal, then extract a relatively pure chatter signal by decomposing and reconstructing the cutting force signal, calculate the energy ratio of the chatter signal to the original signal, and combine the process parameters as the input features of the machine learning model to realize the discrimination of the chatter state.

[0004] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0005] A chatter monitoring method for high-speed dry milling based on real-time cutting force signals comprises the following steps:

[0006] Step 1: Collect three-dimensional cutting force signals during high-speed dry milling;

[0007] Step 2: intercepting the three-directional cutting force signal;

[0008] Step 3: Decompose the three-dimensional cutting force signal described in step 2 to obtain the intrinsic mode function;

[0009] Step 4: Identify the intrinsic mode functions containing the chatter information in step 3;

[0010] Step 5: Reconstruct the intrinsic mode function containing the chatter information in step 4 to obtain a reconstructed chatter signal;

[0011] Step 6: Extracting the chatter energy ratio of the reconstructed chatter signal described in step 5;

[0012] Step 7: Construct a feature vector by combining the chatter energy ratio and cutting parameters described in step 6;

[0013] Step 8: Train a Gaussian Bayes classifier based on the feature vectors and corresponding labels extracted from the historical data;

[0014] Step 9: Based on the feature vector in the real-time signal, the Gaussian Bayesian classifier trained in step 8 is used to realize the chatter monitoring of high-speed dry milling.

[0015] Compared with the prior art, this application has the following beneficial effects:

[0016] The present invention,

[0017] 1. The present invention uses an inverse exponential decay window function to intercept the cutting force signal. The left-side attenuation and right-side enhancement characteristics of the inverse exponential decay window function can highlight the real-time signal while ensuring sufficient frequency resolution. At the same time, its continuous and differentiable, no sudden change characteristics are suitable for time-frequency analysis.

[0018] 2. The successive variational modal decomposition method used in the present invention has excellent decomposition performance and can accurately and clearly extract the various modes of the cutting force signal, and its reconstruction error is controllable during signal reconstruction;

[0019] 3. The normalized frequency deviation and normalized spectral entropy used in the present invention can effectively identify the components containing chatter information among the components obtained after successive variational modal decomposition of the signal within the window;

[0020] 4. The chatter energy ratio index proposed in this invention can be used as one of the input features of the Gaussian Bayesian classifier and can effectively reflect the intensity of machining chatter.

[0021] 5. This invention addresses the phenomenon of abnormally high chatter energy ratio caused by transient impact effects caused by drastic changes in cutting parameters during cutting-in and cutting-out states. Cutting parameters are used as auxiliary input features to prevent the classifier from misjudging the chatter state.

[0022] 6. The methods and indicators used in this invention, such as successive variational modal decomposition, Gaussian Bayesian classifier, normalized frequency deviation, normalized spectral entropy, and flutter energy ratio, require minimal computational effort and can ensure timely flutter monitoring.

[0023] 7. The chatter monitoring method proposed in the present invention is reliable and timely, and can achieve real-time monitoring of the machining stability state during high-speed dry milling;

[0024] 8. The indicators used in the present invention are based on the time domain and frequency domain characteristics of the cutting force signal. There is no need to identify the modal characteristics of the machining system through hammer modal experiments, and it can be quickly and effectively extended to other multi-blade intermittent cutting machining scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 The present invention proposes a method flow for monitoring chatter in high-speed dry milling based on real-time cutting force signals;

[0026] Figure 2 Schematic diagram of a machining chatter monitoring process in an embodiment;

[0027] Figure 3 are tool parameters in the embodiment;

[0028] Figure 4 are the vibration test parameters in the embodiment;

[0029] Figure 5 is the three-directional cutting force signal collected in the embodiment;

[0030] Figure 6 The cutting force signal segment obtained by intercepting the window function in the embodiment;

[0031] Figure 7 is the intrinsic mode function after the cutting force signal segment is decomposed in the embodiment;

[0032] Figure 8 is the normalized frequency deviation and normalized spectral entropy of each eigenmode function of the stable signal in the embodiment;

[0033] Figure 9 are the normalized frequency deviation and normalized spectral entropy of each intrinsic mode function of the chatter signal in the embodiment;

[0034] Figure 10 Reconstructing a chatter signal in an embodiment;

[0035] Figure 11 is the vibration energy ratio of Experiment 1 in the embodiment;

[0036] Figure 12 is the vibration energy ratio of Experiment 2 in the embodiment;

[0037] In the figure, 1 is a cutting force sensor, 2 is a workbench, 3 is a fixture, 4 is a workpiece, 5 is a three-axis dynamometer, 6 is a charge amplifier, and 7 is an industrial computer. DETAILED DESCRIPTION

[0038] The present invention will be described in further detail below with reference to the accompanying drawings.

[0039] Specific implementation: see Figures 1-12 ,

[0040] In the embodiment, the present invention uses an inverse exponential decay window function to intercept the three-dimensional cutting force signal during high-speed dry milling, decomposes the intercepted signal through the successive variational modal decomposition method, extracts the chatter component containing chatter information according to the normalized frequency offset and the normalized spectral entropy to reconstruct the chatter signal, calculates the energy ratio of the chatter signal to the original signal, combines the cutting parameters to form a feature vector, and discriminates the machining stability state of high-speed dry milling based on the Gaussian Bayes classifier to realize chatter monitoring.

[0041] The content of the method proposed in the present invention will be described in detail below with reference to the accompanying drawings.

[0042] like Figure 1 As shown,

[0043] Step 1: Collect three-dimensional cutting force signals during high-speed dry milling.

[0044] Specifically, such as Figure 2 As shown in the figure, the cutting force sensor 1 is installed on the workbench 2, and the workpiece 4 is clamped on the sensor by the fixture 3. The sensor transmits the cutting force signal to the industrial computer 7 through the three-axis dynamometer 5 and the charge amplifier 6 to realize the cutting force signal S in the X, Y and Z directions. x (n), S y (n), S z (n) collection.

[0045] Step 2: Intercept the three-dimensional cutting force signal collected in step 1.

[0046] The three-dimensional cutting force signal S collected in step 1 is processed using the reverse exponential decay window function. x (n), S y (n), S z (n) is intercepted, and the exponential decay window function is calculated according to the following formula:

[0047]

[0048] Where N is the window length, k is the attenuation coefficient that controls the attenuation speed, and n is the sample index. The present invention requires weakening the signal on the left side of the window and retaining the signal strength on the right side. The window function needs to be flipped, and the reverse exponential attenuation function is

[0049] w f (n) = w(N-1-n) (2)

[0050] Use w f (n) The three-dimensional cutting forces after cutting are

[0051] S x,w (n) = w f (n)·S x (n)

[0052] S y,w (n) = w f (n)·S y (n)

[0053] S z,w (n) = w f (n)·S z (n)

[0054] Step 3: Decompose the three-dimensional cutting force signal in step 2 to obtain the intrinsic mode function.

[0055] Specifically, the cutting force signal in each direction after interception is decomposed using successive variational mode decomposition. For the cutting force signal x(t) in one direction, the decomposition process is:

[0056] Step 301: Set the initial residual signal and the first eigenmode function Initialize the regularization parameter α and randomly initialize the center frequency ω L ;

[0057] Step 302: Set constraints

[0058] Each eigenmode component u L (t) should be tightly packed around its center frequency ω L :

[0059]

[0060] Wherein, j is a complex unit.

[0061] Residual component r L (t) The energy in u L (t) corresponds to the smallest bandwidth:

[0062]

[0063] Where ω is the instantaneous frequency, and The residual components r L (t) and the extracted L-1 intrinsic mode functions u i The Fourier transform of (t).

[0064] To avoid u L (t) overlaps with the obtained eigenmode components, and we have:

[0065]

[0066] Step 303: For the L-1th eigenmode function u L-1(t) Solve the minimization problem (7) to obtain the Lth eigenmode function u L (t), repeat the iteration until the residual component r L The ratio of the energy of (t) to the energy of the original signal x(t) is less than the set threshold ε, and all eigenmode functions are obtained;

[0067]

[0068] Among them, α is obtained by using the Lagrange multiplier method.

[0069]

[0070] Where λ is the Lagrange multiplier.

[0071] Step 4: Identify the intrinsic mode function containing the chatter information described in step 3.

[0072] Calculate the normalized frequency deviation δ of all components except the DC component u1(t) f and normalized spectral entropy H s , by setting a reasonable threshold Th δ and Th H Identify the component u with more chatter information c (t)

[0073]

[0074] Among them, f d =ω / 2π is the center frequency of the eigenmode function, is the main shaft harmonic frequency sequence, including the main shaft fundamental frequency and harmonic frequency, z=1,2,…,n; f tp is the tooth passing frequency.

[0075]

[0076] Among them, p i is the relative power spectrum of IMFi, that is

[0077]

[0078] Step 5: Reconstruct the intrinsic mode function described in step 4 to obtain a reconstructed chatter signal.

[0079] The vibration components identified in step 4 are summed and reconstructed to obtain the vibration signal s c (t)

[0080]

[0081] Among them, u c,i(t) is the i-th intrinsic mode function containing chatter information.

[0082] Step 6: Extract the dither energy ratio of the reconstructed dither signal described in step 5.

[0083] Calculate the vibration components s in three directions respectively c The energy ratio of the cutting force signal in three directions is obtained by calculating the energy ratio of the cutting force signal in three directions, r x 、r y 、r z .

[0084]

[0085] Step 7: Construct a feature vector by combining the chatter energy ratio and cutting parameters described in step 6;

[0086] Combined flutter energy ratio r x 、r y 、r z and cutting parameters (spindle speed Ω, feed per tooth f t , radial cutting width a e , axial cutting depth a p ) constitutes the eigenvector X=[Ω,f t ,a e ,a p ,r x ,r y ,r z ] as input features of Gaussian Bayes classifier;

[0087] Step 8: Train a Gaussian Bayes classifier based on the feature vectors and corresponding labels extracted from the historical data in step 7.

[0088] Each feature x i In Category C k The following Gaussian distribution

[0089]

[0090] Among them, μ k,i Category C k The mean of the next i-th feature, Category C k The variance of the next i-th feature

[0091]

[0092] Among them, N k Category C k The number of samples.

[0093] Calculate μ of the sample k,i and And the prior probability P(C k ) Complete training

[0094]

[0095] Where N is the total number of samples.

[0096] Step 9: Based on the feature vector described in step 6 in the real-time signal, the Gaussian Bayesian classifier described in step 8 is used to realize chatter monitoring of high-speed dry milling.

[0097] For the new sample X extracted from the real-time signal, calculate the posterior probability P(C k |X), that is, given feature x i Category C k According to Bayes’ theorem, the posterior probability can be expressed as

[0098]

[0099] Select the category with the largest posterior probability as the predicted category

[0100]

[0101] After completing the classification of sample X, the classification result is output to achieve chatter monitoring.

[0102] The following is an explanation based on actual working conditions:

[0103] The machine tool used in this experiment is Baoji BV8H vertical machining center, and the milling cutter used is ATN coated carbide integral flat bottom spiral milling cutter. The tool parameters are as follows: Figure 3 As shown in the figure, the material being machined is 30CrMnSiNi2a high-strength steel. The cutting force acquisition system used consists of a 5070 charge amplifier, a 5697A triaxial dynamometer, a 9257B piezoelectric force sensor, and an industrial computer equipped with an Intel Core i3-9100F CPU, 8.00GB of RAM, and Windows 10 x64. To ensure sufficient data accuracy, the acquisition frequency was set to 50,000 Hz according to the sampling theorem.

[0104] Among them, tool parameters such as Figure 3 As shown;

[0105] The experimental schemes of continuously changing cutting depth (experiment 1) and cutting width (experiment 2) were designed respectively. This scheme combines the continuously changing cutting depth and cutting width conditions to simulate the dynamic cutting process in actual machining, so as to efficiently identify the critical conditions for the occurrence of chatter and its correlation with cutting parameters. The specific experimental parameters are as follows: Figure 4 shown.

[0106] When implementing,

[0107] (1) Cutting force signal acquisition

[0108] Perform step 1 for Experiment 1 and Experiment 2 respectively to collect the three-dimensional cutting force signals, such as Figure 5 shown.

[0109] (2) Chatter feature extraction

[0110] Execute step 2 and use the sliding window strategy to intercept the signal. The window function w f (n) The length N is 10000, the attenuation factor λ is 0.001, and the overlap rate is 90%. The cutting force signal fragments obtained after interception in the stable state and the cutting force signal in the chatter state are as follows: Figure 6 As shown;

[0111] Execute step 3 and use successive variational mode decomposition to decompose the cutting force signals in the three directions within the window respectively to obtain the eigenmode functions, such as Figure 7 As shown;

[0112] Execute step 4 and calculate the delta of each eigenmode function except the DC component u1(t) for each eigenmode function in each direction. f and H s , according to the threshold Th δ and Th H Filter out the chatter components with chatter information, and the δ of each component f and H s like Figure 8 and Figure 9 As shown;

[0113] Execute steps 5 and 6 to sum all the dither components to reconstruct the dither signal S c (t), such as Figure 10 As shown, the vibration energy ratio r between the vibration signal and the original signal is calculated x 、r y 、r z , the vibration energy of each intercepted segment is as follows Figure 11 and Figure 12 As shown;

[0114] Execute step 3, combined with the flutter energy ratio r x 、r y 、r z and cutting parameters Ω, f t 、a e 、a p Construct the feature vector X=[Ω,f t ,a e ,a p ,r x,r y ,r z ], the feature vector is input into the trained Gaussian Bayesian classifier to classify the machining stability state and realize chatter monitoring.

[0115] In the high-speed dry milling scenario of this embodiment, the chatter monitoring method for high-speed dry milling based on real-time cutting force signals proposed by the present invention achieved an accuracy rate of 99.02%. This accuracy rate is the ratio of the number of test samples that correctly predicted the machining stability state to the total number of test samples. When run on an industrial computer, the calculation time for each signal segment was approximately 18 milliseconds. This shows that the chatter monitoring method for high-speed dry milling based on real-time cutting force signals provided by the present invention can quickly and accurately identify chatter states during high-speed dry milling, demonstrating excellent real-time performance and practical value.

[0116] In summary, the present invention collects three-dimensional cutting force signals during high-speed dry milling; uses a sliding window strategy and an inverse exponential decay window function to intercept the cutting force signals; uses the successive variational modal decomposition method to decompose the signal segments within the window to obtain a series of intrinsic mode functions; calculates the normalized frequency offset and normalized spectral entropy of the remaining intrinsic mode functions excluding the DC component to identify whether the intrinsic mode functions contain chatter information; sums and reconstructs the intrinsic mode functions containing chatter information to obtain a chatter signal; calculates the chatter energy ratio of the chatter signal to the original signal, and combines it with the cutting parameters to form a feature vector as the input feature of the Gaussian Bayes classifier; trains the Gaussian Bayes classifier based on historical examples to obtain a prediction model; and uses the Gaussian Bayes classifier to determine whether chatter occurs in the machining process corresponding to the intercepted signal segment. The present invention can timely and effectively reflect the machining stability state by monitoring the machining chatter of high-speed dry milling, protect the machining quality and the spindle system, and extend the tool life.

[0117] Although the embodiments of the present invention have been shown and described, it is apparent to those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and basis of the present invention. The scope of the present invention is defined by the appended claims and their equivalents. Therefore, the embodiments of the present invention are merely illustrative examples of the present invention. No matter from which point of view, the embodiments of the present invention do not constitute a limitation on the present invention.

Claims

1. A chatter monitoring method for high-speed dry milling based on real-time cutting force signals, characterized in that: The following steps are involved: Step 1: Collect three-dimensional cutting force signals during high-speed dry milling; Step 2: intercepting the three-directional cutting force signal; Step 3: Decompose the three-dimensional cutting force signal described in step 2 to obtain the intrinsic mode function; Step 4: Identify the intrinsic mode functions containing the chatter information in step 3; Step 5: Reconstruct the intrinsic mode function containing the chatter information in step 4 to obtain a reconstructed chatter signal; Step 6: Extracting the chatter energy ratio of the reconstructed chatter signal described in step 5; Step 7: Construct a feature vector by combining the chatter energy ratio and cutting parameters described in step 6; Step 8: Train a Gaussian Bayes classifier based on the feature vectors and corresponding labels extracted from the historical data; Step 9: Based on the feature vector in the real-time signal, the Gaussian Bayesian classifier trained in step 8 is used to realize the chatter monitoring of high-speed dry milling.

2. The method for monitoring chatter in high-speed dry milling based on real-time cutting force signals according to claim 1, characterized in that: The three-axis cutting force signal is collected by deploying a force sensor, a three-axis dynamometer, a charge amplifier and an industrial computer.

3. The method for monitoring chatter in high-speed dry milling based on real-time cutting force signals according to claim 1, characterized in that: In step 2, based on the inverse exponential decay function w f (n) The sliding window strategy intercepts the three-dimensional cutting force signal, w f (n) = w(N-1-n) (2) where N is the window length, k is the attenuation coefficient that controls the attenuation speed, and n is the sample index.

4. The method for monitoring chatter in high-speed dry milling based on real-time cutting force signals according to claim 1, characterized in that: In step 3, the energy ratio of the residual signal to the original signal is set as the cutoff condition, and the cutting force signals in each direction are decomposed using successive variational mode decomposition to obtain the intrinsic mode function clusters of the decomposed cutting force signals in each direction. Specifically, For the cutting force signal x(t) in one direction, t is a continuous time variable, and its decomposition process is: 301) Set the initial residual signal and the first eigenmode function Initialize the regularization parameter α and randomly initialize the center frequency ω L ; 302) Setting constraints; Each eigenmode component u L (t) should be tightly packed around its center frequency ω L : Wherein, j is a complex unit. Residual component r L (t) The energy in u L (t) corresponds to the smallest bandwidth: Where ω is the instantaneous frequency, and The residual components r L (t) and the extracted L-1 intrinsic mode functions u i Fourier transform of (t); To avoid u L (t) overlaps with the obtained eigenmode components, and we have: 303) For the L-1th eigenmode function u L-1 (t) Solve the minimization problem to obtain the Lth eigenmode function u L (t), repeat the iteration until the residual component r L The ratio of the energy of (t) to the energy of the original signal x(t) is less than the set threshold ε, and all eigenmode functions are obtained. Among them, α is obtained by using the Lagrange multiplier method. Where λ is the Lagrange multiplier.

5. The method for monitoring chatter in high-speed dry milling based on real-time cutting force signals according to claim 1, characterized in that: In step 4, the normalized frequency deviation δ of each eigenmode function in each direction is calculated after removing the DC component. f and normalized spectral entropy H s , the threshold value is set by the calculation results to identify the intrinsic mode function containing the chatter information; Among them, f d =ω / 2π is the center frequency of the eigenmode function, is the main shaft harmonic frequency sequence, including the main shaft fundamental frequency and harmonic frequency, z=1,2,…,n; f tp is the tooth passing frequency; Among them, p i is the ith eigenmode function u i The relative power spectrum of (t), that is, 6. The method for monitoring chatter in high-speed dry milling based on real-time cutting force signals according to claim 1, characterized in that: In step 5, the m intrinsic mode functions u containing the vibration information in each direction are c (t) are added and reconstructed respectively to obtain the reconstructed chatter signals in three directions, Among them, u c,i (t) is the i-th intrinsic mode function containing chatter information.

7. The method for monitoring chatter in high-speed dry milling based on real-time cutting force signals according to claim 6, characterized in that: In step 6, the energy ratios of the reconstructed chatter signals in the three directions and the cutting force signals intercepted in step 2 are calculated to obtain the chatter energy ratios r in each direction. x 、r y 、r z As a flutter feature; Among them, s c (t) is the reconstructed chatter signal, and x(t) is the original signal.

8. The method for monitoring chatter in high-speed dry milling based on real-time cutting force signals according to claim 7, characterized in that: In step 7, the intercepted cutting force signal segment is combined with the vibration energy ratio r in the three directions. x 、r y 、r z and cutting parameters constitute the input feature vector X=[Ω,f t ,a e ,a p ,r x ,r y ,r z ], where Ω is the spindle speed, f t is the feed per tooth, a e is the radial cutting width, a p is the axial cutting depth.

9. The method for monitoring chatter in high-speed dry milling based on real-time cutting force signals according to claim 1, characterized in that: In step 8, based on the historical data, the chatter energy ratio and cutting parameters are extracted according to steps 2 to 7 to form an input feature vector, and the Gaussian Bayesian classifier is trained with the machining stability state as a label, wherein the machining stability state includes a stable state and a chatter state; Specifically, each feature x i In Category C k The following Gaussian distribution Among them, μ k,i Category C k The mean of the next i-th feature, Category C k The variance of the next i-th feature; Among them, N k Category C k The number of samples; Calculate μ of the sample k,i and And the prior probability P(C k ) Complete training; Where N is the total number of samples.

10. The method for monitoring chatter in high-speed dry milling based on real-time cutting force signals according to claim 9, characterized in that: In step 9, the feature vector of the real-time signal is input into the trained Gaussian Bayesian classifier for discrimination; Specifically, For the new sample X extracted from the real-time signal, calculate the posterior probability P(C k |X), that is, given feature x i Category C k According to Bayes' theorem, the posterior probability can be expressed as Select the category with the largest posterior probability as the predicted category After completing the classification of sample X, the classification result is output to achieve chatter monitoring.

Images