Micro-milling flutter detection method based on stack integration

Abstract

The invention particularly relates to a micro-milling flutter detection method based on stack integration. The method comprises the following steps: forming a training sample by using flutter characteristics and labels representing machining states; the training samples are sent into a plurality of base classifiers at the same time, and corresponding flutter detection base models are obtained; taking the output of each flutter detection base model as the input and sending the input into a meta classifier to obtain a final flutter detection model; and for a to-be-tested sample, the corresponding flutter characteristics of the to-be-tested sample are sent to the flutter detection model, and the processing state corresponding to the current to-be-tested sample is obtained. According to the method, the idea of stack integration is adopted, a classification and regression tree, a k-nearest neighbor method, a feedforward neural network and a gating circulation unit are used as base classifiers, a support vector machine is used as an element classifier, complementarity of different base classifiers and the correction effect of the element classifier are integrated, and the robustness of the algorithm is improved. Therefore, the feature capture capability and generalization capability of the model are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of micro-milling chatter detection, and particularly relates to a micro-milling chatter detection method based on stacked integration. Background Art

[0002] Existing chatter detection methods mostly classify the machining state based on a single model. The commonly used single models can be mainly divided into two categories. One is to establish a corresponding chatter detection model by setting appropriate thresholds, and the other is to obtain a corresponding chatter detection model based on machine learning algorithms. The threshold method is relatively simple to use, but it is often difficult to select appropriate thresholds. In addition, when the machining conditions change, the thresholds also need to be adjusted accordingly. In contrast, the chatter detection model obtained based on machine learning algorithms has better adaptability. However, the chatter detection model obtained based on a single machine learning algorithm has insufficient feature capture ability and is often ineffective when facing some difficult-to-classify samples, thus reducing the accuracy of the chatter detection model. Therefore, developing a micro-milling chatter detection method based on stacked integration has important practical application value.

[0003] Document 1 "X. Li, S. Wan, X. Huang, J. Hong, Milling chatter detection based on VMD and difference of power spectral entropy, The International Journal of Advanced Manufacturing Technology 111(2020)2051–2063" discloses a chatter detection method based on the threshold method. This method first processes the original signal using the variational mode decomposition method to obtain the chatter component, and then filters the chatter component using the least mean square algorithm. Then, the power spectral entropy corresponding to the chatter component and the filtered signal are calculated respectively, and the difference obtained by subtracting the two is used as the chatter feature. By comparing it with the set threshold 0, the detection of the machining state is realized. If the difference in power spectral entropy is less than 0, it is considered that the machining process is stable; otherwise, it is considered that chatter occurs during the machining process. However, certain prior knowledge is often required when selecting appropriate thresholds. In addition, when the samples are complex, it is difficult to correctly distinguish different machining states relying on the change of a single feature.

[0004] The literature 2 "F. Shi, H. Cao, X. Zhang, X. Chen, A reinforced k-nearest neighbors method with application to chatter identification in high-speed milling, IEEE Transactions on Industrial Electronics 67(12)(2020)10844–10855." discloses a chatter detection method based on the k-nearest neighbor method. This method first processes the obtained acceleration signal and uses the root mean square value, etc. as chatter features to obtain a series of samples. Then, after training the obtained samples using the k-nearest neighbor method, a corresponding chatter detection model is obtained. When a sample to be detected is input, first calculate its distance from each sample point in the training set, and according to the distance, select the k training samples closest to the sample to be detected, and count the machining states corresponding to these k closest training samples. The machining state with the most occurrences is considered to be the machining state corresponding to the sample to be tested. However, the size of the k value will greatly affect the accuracy of the detection model. In addition, the chatter detection model obtained only relying on the k-nearest neighbor method has insufficient feature capture ability and will be relatively ineffective when facing some difficult-to-classify samples.

[0005] The typical characteristics of the above literature are: The chatter detection model obtained based on the threshold method requires certain prior knowledge when determining a suitable threshold, and the detection accuracy is greatly affected by the threshold. In addition, the chatter detection model obtained based on a single machine learning algorithm has insufficient feature capture ability and is relatively ineffective when facing some difficult-to-classify samples. Therefore, it is necessary to develop a chatter detection method applicable to the micro-milling process with strong feature capture ability.

[0006] It should be noted that the information disclosed in the above background art section is only used to strengthen the understanding of the background of the present invention, and therefore may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0007] Aiming at the problems of difficult selection of a suitable threshold and insufficient feature capture ability of the chatter detection model obtained based on a single machine learning algorithm, the present invention provides a micro-milling chatter detection method based on stacking integration.

[0008] Other features and advantages of the present invention will become apparent through the following detailed description, or be learned in part through the practice of the present invention.

[0009] According to a first aspect of the present invention, there is provided a micro-milling chatter detection method based on stacking integration, the method comprising:

[0010] Combine the flutter characteristics and the labels representing the processing status to form training samples;

[0011] Send the training samples into multiple base classifiers simultaneously to obtain the corresponding flutter detection base models;

[0012] Take the outputs of the respective flutter detection base models as inputs and send them into the meta-classifier to obtain the final flutter detection model;

[0013] For the sample to be tested, send its corresponding flutter characteristics into the flutter detection model to obtain the processing status corresponding to the current sample to be tested.

[0014] In some exemplary embodiments, the method further includes obtaining flutter characteristics, specifically:

[0015] Collect the acceleration signal at the workpiece in the micro-milling cutting experiment and perform filtering processing on the acceleration signal at the workpiece;

[0016] Perform variational mode extraction on the filtered acceleration signal at the workpiece to obtain the flutter component;

[0017] Perform time-frequency analysis on the flutter component to obtain its corresponding marginal spectrum;

[0018] Calculate the flutter characteristics based on the acceleration signal at the workpiece, the flutter component, and the marginal spectrum.

[0019] In some exemplary embodiments, the time-frequency analysis is specifically: perform Hilbert-Huang transform on the flutter component.

[0020] In some exemplary embodiments, the labels representing the processing status include 1 and 2, where 1 represents stable and 2 represents flutter.

[0021] In some exemplary embodiments, the base classifiers include: classification and regression tree, k-nearest neighbor method, feedforward neural network, and gated recurrent unit.

[0022] In some exemplary embodiments, the meta-classifier is a support vector machine.

[0023] In some exemplary embodiments, the step of taking the outputs of the respective flutter detection base models as inputs and sending them into the meta-classifier to obtain the final flutter detection model is specifically:

[0024] H o =ω s ·f M +b m

[0025] Among them, H o represents the obtained hyperplane, f M is a feature vector composed of the outputs of four base models, ωs is the weight vector corresponding to the model, and b m is the bias term. After solving the above formula, the final flutter detection model is obtained.

[0026] According to the second aspect of the present invention, there is provided a storage medium having stored thereon a computer program, which when executed by a processor implements the method for detecting micro-milling flutter based on stacked integration described in the first aspect above.

[0027] According to the third aspect of the present invention, there is provided a computer program product having stored thereon a computer program, which when executed by a processor implements the method for detecting micro-milling flutter based on stacked integration described in the first aspect above.

[0028] According to the fourth aspect of the present invention, there is provided an electronic device, comprising:

[0029] a processor; and

[0030] a memory for storing executable instructions of the processor;

[0031] wherein the processor is configured to implement the method for detecting micro-milling flutter based on stacked integration described in the first aspect above when executing the executable instructions.

[0032] The method for detecting micro-milling flutter based on stacked integration provided by the embodiments of the present invention adopts the idea of stacked integration, and at the same time uses classification and regression trees, k-nearest neighbor method, feed-forward neural network and gated recurrent unit as base classifiers, and uses support vector machine as a meta-classifier, integrating the complementarity of different base classifiers and the correction effect of the meta-classifier, thereby improving the feature capture ability and generalization ability of the model. The method first processes the collected acceleration signal using an adaptive filter, and extracts the required flutter components therefrom using the variational mode extraction method; then performs time-frequency analysis on the obtained flutter components to calculate different flutter features, and obtains the corresponding sample set; then uses different base classifiers to train the samples to obtain the corresponding flutter detection base models, and then the outputs of each base model are sent as new inputs into the meta-classifier to obtain the final flutter detection model. Compared with Document 1, the present invention does not require prior knowledge to set appropriate thresholds, so it has better versatility; compared with Document 2, the present invention integrates the complementarity of different base classifiers and the correction effect of the meta-classifier when establishing the flutter detection model, so it has strong feature capture ability.

[0033] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present invention. Description of the Drawings

[0034] The accompanying drawings here are incorporated into the specification and form a part of this specification, showing embodiments consistent with the present invention, and are used together with the specification to explain the principles of the present invention. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0035] Figure 1 It is a flowchart of the micro-milling chatter detection method based on stack integration according to an embodiment of the present invention;

[0036] Figure 2 It is the workpiece acceleration signal collected in an embodiment of the present invention when the spindle speed is 19,500 revolutions per minute and the axial cutting depth is 50 microns.

[0037] Figure 3 It is the machine tool acceleration signal collected in an embodiment of the present invention when the spindle speed is 19,500 revolutions per minute and the axial cutting depth is 50 microns.

[0038] Figure 4 It is the filtered signal obtained by processing the acceleration signal collected in an embodiment of the present invention when the spindle speed is 19,500 revolutions per minute and the axial cutting depth is 50 microns using an adaptive filter.

[0039] Figure 5 It is the chatter component obtained by performing variational mode extraction on the filtered signal when the spindle speed is 19,500 revolutions per minute and the axial cutting depth is 50 microns in an embodiment of the present invention.

[0040] Figure 6 It is the marginal spectrum obtained by performing time-frequency analysis on the chatter component when the spindle speed is 19,500 revolutions per minute and the axial cutting depth is 50 microns in an embodiment of the present invention.

[0041] Figure 7 It is the workpiece acceleration signal collected in an embodiment of the present invention when the spindle speed is 19,500 revolutions per minute and the axial cutting depth is 310 microns.

[0042] Figure 8 It is the machine tool acceleration signal collected in an embodiment of the present invention when the spindle speed is 19,500 revolutions per minute and the axial cutting depth is 310 microns.

[0043] Figure 9 It is the filtered signal obtained by processing the acceleration signal collected in an embodiment of the present invention when the spindle speed is 19,500 revolutions per minute and the axial cutting depth is 310 microns using an adaptive filter.

[0044] Figure 10This is the flutter component obtained by performing variational mode extraction on the filtered signal when the spindle speed is 19,500 revolutions per minute and the axial cutting depth is 310 microns in the embodiments of the present invention.

[0045] Figure 11 This is the marginal spectrum obtained by performing time-frequency analysis on the flutter component when the spindle speed is 19,500 revolutions per minute and the axial cutting depth is 310 microns in the embodiments of the present invention. Detailed implementation manners

[0046] Example embodiments will now be described more fully with reference to the accompanying drawings. However, the example embodiments can be implemented in various forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this invention will be more complete and comprehensive, and will fully convey the concept of the example embodiments to those skilled in the art. The described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.

[0047] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and thus repeated descriptions thereof will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0048] In view of the drawbacks and deficiencies of the prior art, in this example embodiment, a micro-milling flutter detection method based on stacked integration is provided. This method first performs adaptive filtering on the acquired acceleration signal, then uses the variational mode extraction method to extract the corresponding flutter component from the filtered signal, and then further processes the obtained flutter component to calculate different flutter characteristics, thereby obtaining a series of samples corresponding to different machining states. Next, the obtained samples are simultaneously fed into the base classifiers classification and regression tree, k-nearest neighbor method, feedforward neural network, and gated recurrent unit to obtain the corresponding flutter detection base models. Finally, the outputs of each base model are used as inputs and fed into the meta-classifier support vector machine to obtain the final flutter detection model. For the samples to be tested, by feeding the corresponding characteristic values thereof into the flutter detection model, the machining state corresponding to the current sample to be tested can be obtained.

[0049] Expected technical effects: The micro-milling flutter detection method provided by the present invention combines the advantages of different classifiers and is a micro-milling flutter detection method with stronger feature capture ability.

[0050] Reference Figure 1As shown in the figure, the micro-milling chatter detection method based on stacked integration can specifically include the following steps:

[0051] Form training samples by combining chatter features and labels representing the machining state;

[0052] Send the training samples into multiple base classifiers simultaneously to obtain corresponding chatter detection base models;

[0053] Take the outputs of each chatter detection base model as inputs and send them into a meta-classifier to obtain the final chatter detection model;

[0054] For the sample to be tested, send its corresponding chatter features into the chatter detection model to obtain the machining state corresponding to the current sample to be tested.

[0055] The above steps are specifically as follows:

[0056] Step 1: Conduct micro-milling cutting experiments and simultaneously collect the acceleration signal S w (t) at the workpiece and the acceleration signal S m (t) at the machine tool, where t represents the time index.

[0057] Step 2: Use an adaptive filter to process the acquired acceleration signals to obtain the filtered signal S f (t). The specific calculation process can be expressed as follows:

[0058] S f (t) = S w (t) - h T (t - 1)S x (t)

[0059]

[0060] h(t) = h(t - 1) + k(t)S f (t)

[0061] In the formula, h T (t - 1) is the weight vector, S x (t) is a real-valued vector composed of the machine tool acceleration signals. The superscript 'T' represents transpose, k(t) is the Kalman gain vector, Q(t) is an intermediate variable, and λ(t) is a forgetting factor that changes with time.

[0062] Step 3: Perform variational mode extraction on the filtered signal S f (t) to obtain the chatter component S' c (t) contained therein:

[0063]

[0064] where α is the penalty factor, δ(t) is the Dirac function, S′ c (t) and S r (t) represent the flutter component and the residual component contained in the filtered signal, and ‘*’ represent the partial derivative and convolution symbols, ω d is the central frequency of the desired intrinsic mode function, e is the natural constant, j is the imaginary unit, β(t) represents the impulse response function, and γ(t) is the Lagrange multiplier.

[0065] Step 4: Perform the Hilbert-Huang transform on the obtained flutter component S′ c (t) to obtain its corresponding marginal spectrum H(f):

[0066] H(f) = ∫H(t,f)dt

[0067] where:

[0068]

[0069] where f represents the frequency index, S h (t) is the Hilbert transform of the flutter component S′ c (t), and H(t,f) is an intermediate variable.

[0070] Step 5: Calculate the following different flutter characteristics:

[0071] F e2 = max(S f (t))

[0072]

[0073]

[0074] where represents the average value of S w (t), represents the average value of H(f), and N is the number of data points.

[0075] Step 6: Combine the flutter characteristics and the labels representing the processing states into a series of samples, where stable is labeled as ‘1’ and flutter is labeled as ‘2’.

[0076] Step 7: Feed the series of samples obtained in Step 6 into the base classifiers decision tree, k-nearest neighbor method, feedforward neural network, and gated recurrent unit simultaneously to obtain the corresponding flutter detection base models M CART , M KNN , M FNN and M GRU . The calculation process is as follows:

[0077] 1) Classification and regression tree: The Gini index is used to measure the purity of the sample set, and thus for node partitioning of the classification and regression tree

[0078]

[0079] where D is the sample set, N D is the total number of samples in the sample set, C g is the label indicating the sample processing status, N g is the number of samples belonging to the processing status C g (g = 1, 2,..., G), and G is the total number of categories corresponding to the samples. Assume that the feature F sl (l = 1, 2,..., 6) has V different values Feature F sl The corresponding Gini index is calculated as follows:

[0080]

[0081] where V is the number of possible values of feature F sl v represents in the index, N v is the feature F sl with the value of the number of. is the sample subset obtained by partitioning the sample set D based on the feature F sl b v represents the index corresponding to the sample subset . The feature corresponding to the minimum Gini index is selected as the better feature, and the current sample set will be split into different sample subsets accordingly. Then, the above process is repeated for different sample subsets to obtain the flutter detection base model M based on the classification and regression tree CART .

[0082] 2) k-nearest neighbor method: For the k-nearest neighbor method, its training process is implicit. Denote the flutter detection base model obtained after training based on the k-nearest neighbor method as M KNN . When a new test sample is input, first calculate its distance from each sample point in the training set, and according to the magnitude of the distance value, select the k training samples closest to the test sample, and count their corresponding processing statuses. The processing status with the most occurrences is considered the processing status corresponding to the test sample

[0083]

[0084] where f N is the feature vector corresponding to the test sample, i represents the sample index, is the feature vector corresponding to the i-th sample in the training set, fNl and are the values of the feature vector under each feature.

[0085] 3) Feedforward neural network: The feedforward neural network can effectively capture the non-linear relationship between features and processing states, thereby realizing the detection of processing states. Its calculation process is as follows:

[0086]

[0087] Among them, H p and H g are intermediate variables, p is a positive integer index, ω pl and are the weights between the input layer and the hidden layer, ω gp is the weight between the hidden layer and the output layer, L p is the number of hidden neurons, is the output of the hidden neuron, is the output of the output neuron, σ s is the sigmoid activation function, is the softmax activation function. After solving the above process, the flutter detection base model M FNN can be obtained.

[0088] 4) Gated recurrent unit: The gated recurrent unit is a type of recurrent neural network, and it avoids the long-term dependence problem of the recurrent neural network. Its calculation process is as follows:

[0089]

[0090] Among them, t c represents the current time step, is the update gate, is the reset gate, and are the candidate hidden state and the current hidden state, is the output of the gated recurrent unit at the current time step, is the feature vector input at the current time step, W z , W r , W h and W y are weight matrices, b z , b r , b h and b y are the corresponding biases, ⊙ represents the Hadamard product, and 'tanh' represents the hyperbolic tangent function. After solving, the flutter detection base model M GRU can be obtained.

[0091] Step 8: Use the outputs of each base model obtained in Step 7 as inputs and feed them into the meta-classifier support vector machine, and the final flutter detection model M can be obtained. SVM :

[0092] H o = ω s ·f M + b m

[0093] where H o represents the obtained hyperplane, f M is the feature vector composed of the outputs of the base models M CART , M KNN , M FNN and M GRU , ω s is the weight vector corresponding to the model, b m is the bias term. After solving the above formula, the final flutter detection model M SVM is obtained.

[0094] The following uses two specific embodiments to illustrate the method of the present invention. Among them, Embodiment 1 corresponds to a stable micro-milling process, and Embodiment 2 corresponds to a micro-milling process with flutter.

[0095] Embodiment 1:

[0096] The parameters of the milling cutter used in the experiment are: the number of teeth N = 2 teeth, the diameter D = 1 mm; the experimental equipment is a Jingdiao GR50 five-axis CNC machining center; the cutting method is down milling; the workpiece material is aluminum alloy; the cutting parameters are the radial depth of cut a e = 0.5 mm, the feed per tooth c = 0.005 mm / tooth, the spindle speed is 19,500 r / min, and the axial depth of cut a p = 50 μm.

[0097] Step 1: Refer to Appendix Figure 2 and Figure 3 , conduct a micro-milling cutting experiment, and simultaneously collect the acceleration signal S w (t) at the workpiece and the acceleration signal S m (t) at the machine tool, where t = 1, 2,..., 256000.

[0098] Step 2: Use an adaptive filter to process the acquired acceleration signal to obtain the filtered signal S f (t). Refer to Appendix Figure 4 , and the specific calculation process can be expressed as follows:

[0099] S f (t) = S w (t) - h T (t - 1)S x(t)

[0100]

[0101] h(t) = h(t - 1) + k(t)S f (t)

[0102] where h T (t - 1) is the weight vector, S x (t) is a real-valued vector composed of the machine tool acceleration signal, the superscript 'T' represents transpose, k(t) is the Kalman gain vector, Q(t) is an intermediate variable, and λ(t) is a forgetting factor that varies with time.

[0103] Step 3. Perform variational mode extraction on the filtered signal S f (t) to obtain the chatter component S c ′(t) contained therein, referring to Appendix Figure 5 :

[0104]

[0105] where α = 2000, δ(t) is the Dirac function, S′ c (t) and S r (t) represent the chatter component and the residual component contained in the filtered signal, and '*' represent the partial derivative and convolution symbols, ω d is the central frequency of the desired intrinsic mode function, e = 2.7183, j is the imaginary unit, β(t) represents the impulse response function, and γ(t) is the Lagrange multiplier.

[0106] Step 4. Perform time-frequency analysis on the obtained chatter component S′ c (t) to obtain its corresponding marginal spectrum H(f), referring to Appendix Figure 6 :

[0107] H(f) = ∫H(t, f)dt

[0108] where:

[0109]

[0110] where f represents the frequency index, S h (t) is the Hilbert transform of the chatter component S′ c (t), and H(t, f) is an intermediate variable.

[0111] Step 5. Calculate the following different chatter characteristics:

[0112] F e2 = max(S f(t)) = 0.53

[0113]

[0114] Step Six: Feed the result calculated in Step Five, i.e., the sample vector [3.14, 0.53, 0.05, 0.13, 5.73, 7.75], into the flutter detection base model M CART , M KNN , M FNN and M GRU to obtain the corresponding output [0, 0, 1, 0].

[0115] Step Seven: Feed the output of the base model obtained in Step Six into the final flutter detection model M SVM , and the model output is '1', indicating that the machining process is in a stable state.

[0116] Example 2:

[0117] The parameters of the milling cutter used in the experiment are: the number of teeth N = 2 teeth, the diameter D = 1 mm; the experimental equipment is the Jingdiao GR50 five-axis CNC machining center; the cutting method is down milling; the workpiece material is aluminum alloy; the cutting parameters are the radial depth of cut a e = 0.5 mm, the feed per tooth c = 0.005 mm / tooth, the spindle speed is 19,500 revolutions per minute, and the axial depth of cut a p = 310 microns.

[0118] Step One: Refer to Attachment Figure 7 and Figure 8 to conduct a micro-milling cutting experiment, and simultaneously collect the acceleration signal S w (t) at the workpiece and the acceleration signal S m (t) at the machine tool, where t = 1, 2,..., 256000.

[0119] Step Two: Use an adaptive filter to process the acquired acceleration signal to obtain the filtered signal S f (t). Refer to Attachment Figure 9 , and the specific calculation process can be expressed as follows:

[0120] S f (t) = S w (t) - h T (t - 1)S x (t)

[0121]

[0122] h(t) = h(t - 1) + k(t)S f (t)

[0123] where h T(t - 1) is the weight vector, S x (t) is a real - valued vector composed of the machine tool acceleration signal. The superscript 'T' represents transpose, k(t) is the Kalman gain vector, Q(t) is an intermediate variable, and λ(t) is a forgetting factor that changes with time.

[0124] Step 3: Perform variational mode extraction on the filtered signal S f (t) to obtain the chatter component S′ c (t) contained therein. Refer to Appendix Figure 10 :

[0125]

[0126] where α = 2000, δ(t) is the Dirac function, S′ c (t) and S r (t) represent the chatter component and the residual component contained in the filtered signal, and '*' represent the partial derivative and convolution symbols, ω d is the central frequency of the desired intrinsic mode function, e = 2.7183, j is the imaginary unit, β(t) represents the impulse response function, and γ(t) is the Lagrange multiplier.

[0127] Step 4: Perform time - frequency analysis on the obtained chatter component S c ′(t) to obtain its corresponding marginal spectrum H(f). Refer to Appendix Figure 11 :

[0128] H(f)=∫H(t,f)dt

[0129] where:

[0130]

[0131] where f represents the frequency index, S h (t) is the Hilbert transform of the chatter component S′ c (t), and H(t,f) is an intermediate variable.

[0132] Step 5: Calculate the following different chatter characteristics:

[0133] F e2 =max(S f (t)) = 13.01

[0134]

[0135] Step 6: Form the sample vector [5.42, 13.01, 0.36, 0.31, 6.68, 8.55] from the results calculated in Step 5 and send it into the chatter detection base model MCART , M KNN , M FNN and M GRU , the corresponding output [0, 1, 1, 1] is obtained.

[0136] Step Seven: Feed the output of the base model obtained in Step Six into the final chatter detection model M SVM , and the model output is '2', indicating that chatter occurs during the machining process.

[0137] As can be seen from the above examples, the method proposed by the present invention does not require manual intervention and has good adaptability. The proposed method uses multiple different base models and takes the output results of each base model as inputs and feeds them into the meta-model. Even if one of the base models makes a wrong judgment, the finally obtained chatter detection model can still integrate the complementarity of different base models and the correction effect of the meta-model, so as to achieve accurate detection of the machining state.

[0138] It should be noted that, on the other hand, the present application also provides a storage medium, which may be included in an electronic device; or may exist alone without being assembled into the electronic device. The above storage medium carries one or more programs, and when the above one or more programs are executed by an electronic device, the electronic device is caused to implement the method described in the following embodiments. For example, the electronic device may implement the various steps of the method as Figure 1 shown.

[0139] In one embodiment, the present application provides a computer program product, including a computer program, which when executed by a processor implements the steps in the above method embodiments.

[0140] In addition, the above drawings are only schematic illustrations of the processes included in the method according to the exemplary embodiments of the present invention, rather than for limiting purposes. It is easy to understand that the processes shown in the above drawings do not indicate or limit the time sequence of these processes. Additionally, it is also easy to understand that these processes may be executed synchronously or asynchronously in, for example, multiple modules.

[0141] Those skilled in the art will readily think of other embodiments of the present invention after considering the specification and practicing the invention herein. The present application aims to cover any variations, uses, or adaptations of the present invention, which follow the general principles of the present invention and include the common general knowledge or conventional technical means in the technical field not disclosed by the present invention. The specification and embodiments are only regarded as exemplary, and the true scope and spirit of the present invention are pointed out by the claims.

[0142] It should be understood that the present invention is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present invention is only defined by the appended claims.

Claims

1. A micro-milling chatter detection method based on stacking integration, characterized in that: The method comprises: The chatter features and labels representing the processing status are combined into training samples; The training samples are simultaneously fed into multiple base classifiers to obtain the corresponding chatter detection base model; The outputs of each chatter detection base model are fed into the meta-classifier as input to obtain the final chatter detection model; For the sample to be tested, its corresponding chatter feature is sent to the chatter detection model to obtain the processing state corresponding to the current sample to be tested.

2. The method according to claim 1, characterized in that The method further includes obtaining a vibration feature, specifically: Collect the acceleration signal of the workpiece in the micro-milling experiment, and filter the acceleration signal of the workpiece; Perform variational mode extraction on the filtered acceleration signal at the workpiece to obtain the chatter component; Perform time-frequency analysis on the chatter component to obtain its corresponding marginal spectrum; The chatter characteristics are calculated based on the acceleration signal, chatter components and marginal spectra at the workpiece.

3. The method according to claim 2, characterized in that The time-frequency analysis specifically includes: performing Hilbert-Huang transform on the chatter component.

4. The method according to claim 1, characterized in that: The labels indicating the processing status include 1 and 2, wherein 1 indicates stability and 2 indicates vibration.

5. The method according to claim 1, characterized in that: The base classifier includes: classification and regression tree, k-nearest neighbor method, feedforward neural network and gated recurrent unit.

6. The method according to claim 5, characterized in that The meta-classifier is a support vector machine.

7. The method according to claim 6, characterized in that The output of each flutter detection base model is sent as input to the meta-classifier to obtain the final flutter detection model, which is specifically: H o =ω s ·f M +b m Among them, H o represents the hyperplane, f M is a feature vector composed of the outputs of the four basis models, ω s is the weight vector corresponding to the model, b m is the bias term. After solving the above equation, the final flutter detection model is obtained.

8. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the micro-milling chatter detection method based on stacking integration according to any one of claims 1 to 7 is implemented.

9. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the micro-milling chatter detection method based on stacking integration according to any one of claims 1 to 7 is implemented.

10. An electronic device, characterized in that: include: processor; as well as A memory, configured to store executable instructions of the processor; The processor is configured to perform the micro-milling chatter detection method based on stacking integration according to any one of claims 1 to 7 by executing the executable instructions.

Images