Processing flutter signal feature sample extraction method, flutter detection method and device

Abstract

The invention provides a processing chatter signal feature sample extraction method and a chatter detection method and device, and belongs to the technical field of industrial signal processing, and the method comprises the steps: carrying out the optimization variational mode decomposition of an original processing chatter signal, and obtaining a plurality of IMF signal components; performing fusion index evaluation on the IMF signal components to obtain a quality score of each IMF signal component, and performing screening reconstruction on the IMF signal components according to the quality scores to obtain reconstructed signals; recursive hierarchical decomposition is carried out on the reconstructed signal based on a preset matrix operator to obtain a plurality of hierarchical node components, and singular spectrum entropy feature sample extraction is carried out on the hierarchical node components to obtain a feature sample data set; according to the method, weakening of short-time abrupt change features in signal segments is avoided through recursive hierarchical decomposition, local abrupt change features in signals are captured through singular spectrum entropy, sensitivity to short-time abrupt change or transient events is improved through combination of hierarchical decomposition and hierarchical singular spectrum entropy of singular spectrum entropy, and accurate feature samples are obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial signal processing, and in particular to a method for extracting characteristic samples of machining chatter signals, and a chatter detection method and device. Background Art

[0002] Machining chatter refers to the self-excited vibration phenomenon caused by the dynamic interaction between the tool and the workpiece during machining. This vibration not only degrades the quality of the machined surface, resulting in substandard dimensional accuracy and surface roughness, but also accelerates tool wear and can even cause tool breakage, severely impacting machining efficiency and equipment life. Furthermore, it can cause long-term damage to the machine tool structure, increasing maintenance costs and generating risks. Therefore, chatter detection is crucial during machining. By detecting and identifying chatter signals, machining parameters can be adjusted or suppression measures can be implemented to prevent further chatter exacerbation, thereby ensuring machining stability, improving product quality, and extending equipment life. However, due to the complex nature of actual machining conditions, the presence of ambient noise, and the nonstationary and nonlinear characteristics of regenerative chatter, acquired machining signals typically have a low signal-to-noise ratio, making them inadequate for direct use in machining chatter detection. Therefore, effective and robust machining chatter signal processing and chatter feature extraction methods are crucial.

[0003] Among industrial signal processing methods, a common approach currently uses variational mode decomposition to extract the IMF (Intrinsic Mode Function) of the raw signal data, followed by multi-scale entropy for feature extraction and sample data construction. However, when extracting signal feature samples for machining chatter detection, machining chatter exhibits short-term mutations, and its local signals or transient changes often contain important characteristic information. However, existing multi-scale entropy extraction methods require a coarse-graining process, where the signal is segmented and averaged at each scale. This coarse-graining process smooths local fluctuations in the signal and weakens key short-term mutation features in the signal, resulting in poor sensitivity to short-term mutations or transient events. Therefore, it is difficult to accurately extract feature samples from machining chatter signals.

[0004] Therefore, the existing technology has the technical problem of poor sensitivity to short-term mutations or transient events and difficulty in accurately extracting feature samples, which needs to be improved. Summary of the Invention

[0005] In view of this, it is necessary to provide a method for extracting characteristic samples of machining chatter signals, a chatter detection method and device, so as to solve the technical problems existing in the prior art of poor sensitivity to short-term mutations or transient events and difficulty in accurately extracting characteristic samples.

[0006] In a first aspect, the present invention provides a method for extracting characteristic samples of a machining chatter signal, comprising: The original machining chatter signal is subjected to optimized variational mode decomposition to obtain several IMF signal components; The IMF signal components are evaluated by fusion indicators to obtain the quality scores of each IMF signal component, and the IMF signal components are screened and reconstructed according to the quality scores to obtain the reconstructed signal; The reconstructed signal is recursively decomposed into several hierarchical node components based on a preset matrix operator, and the singular spectral entropy feature samples of the hierarchical node components are extracted to obtain a feature sample data set.

[0007] In one possible implementation, the original machining chatter signal is subjected to optimized variational mode decomposition to obtain several IMF signal components, including: Determine the optimal parameter combination for variational mode decomposition based on Newton-Raphson optimizer; Determine the initial parameters of variational mode decomposition according to the optimal parameter combination, and perform variational mode decomposition on the original machining chatter signal according to the initial parameters to obtain the IMF signal components; Among them, the Newton-Raphson optimizer takes minimizing the ratio of the permutation entropy and envelope entropy of the IMF signal component as the optimization objective.

[0008] In one possible implementation, the IMF signal components are evaluated by fusion indicators to obtain a quality score of each IMF signal component, including: Calculate the fusion index score of each IMF signal component; Determine the corresponding metric value vectors according to the scores of each fusion indicator; The quality score of each IMF signal component is determined according to each metric value vector.

[0009] In a possible implementation, the fusion indicators include: inverse spectral entropy, energy ratio, and Spearman rank correlation coefficient.

[0010] In one possible implementation, screening and reconstructing the IMF signal components according to the quality scores to obtain a reconstructed signal includes: Filtering the IMF signal components in descending order according to the quality scores to obtain filtered IMF signal components until the sum of the quality scores of the filtered IMF signal components reaches a preset score threshold; The reconstructed signal is obtained by superimposing the filtered IMF signal components.

[0011] In a possible implementation, the preset matrix operator includes a low-frequency average operator and a high-frequency difference operator.

[0012] In one possible implementation, the reconstructed signal is recursively and hierarchically decomposed based on a preset matrix operator to obtain several hierarchical node components, including: Determine the calling order of preset matrix operators based on the preset vector; Select the corresponding preset matrix operators in turn according to the calling order to perform signal decomposition on the reconstructed signal or the previous level node component to obtain a new level node component; Among them, the values of each dimension of the preset vector are randomly 0 or 1, 0 corresponds to the low-frequency average operator, and 1 corresponds to the high-frequency difference operator.

[0013] In a second aspect, the present invention provides a chatter detection method, comprising: A chatter detection sample set is generated based on a feature sample extraction method for machining chatter signals. An initial chatter detection network is trained based on the chatter detection sample set to obtain a fully trained chatter detection network. The chatter detection results are obtained by predicting and outputting the processing signal to be detected based on the chatter detection network.

[0014] The method for extracting characteristic samples of a machining chatter signal is any one of the above-mentioned methods for extracting characteristic samples of a machining chatter signal.

[0015] In a third aspect, the present invention provides a device for extracting characteristic samples of machining chatter signals, comprising: A signal decomposition unit is used to perform optimized variational mode decomposition on the original machining chatter signal to obtain a number of IMF signal components; a signal component screening unit, configured to perform fusion index evaluation on the IMF signal components to obtain a quality score of each IMF signal component, and to screen and reconstruct the IMF signal components according to the quality score to obtain a reconstructed signal; The feature sample extraction unit is used to recursively decompose the reconstructed signal into several hierarchical node components based on a preset matrix operator, and extract singular spectrum entropy feature samples from the hierarchical node components to obtain a feature sample data set.

[0016] In a fourth aspect, the present invention provides a computer-readable storage medium for storing a computer-readable program or instruction, which, when executed by a processor, can implement the steps of any of the above-mentioned machining chatter signal feature sample extraction methods and / or chatter detection methods.

[0017] The beneficial effects of the above-described embodiment are as follows: the method for extracting characteristic samples of machining chatter signals provided by the present invention can avoid weakening short-term mutation features during signal segmentation by recursively decomposing the reconstructed signal. The singular spectral entropy can effectively capture local mutation features in the signal during sample generation. The hierarchical singular spectral entropy, which combines hierarchical decomposition and singular spectral entropy, can adaptively capture key signal features from fine to coarse scales, improve sensitivity to short-term mutations or transient events, and obtain accurate characteristic samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0019] Figure 1 A schematic flow chart of an embodiment of a method for extracting characteristic samples of machining chatter signals provided by the present invention; Figure 2 Schematic diagram of the process of optimizing variational mode decomposition according to an embodiment of the present invention; Figure 3 A schematic diagram of a flow chart for evaluating quality scoring of fusion indicators according to an embodiment of the present invention; Figure 4 Schematic diagram of the process of screening and reconstructing IMF signal components according to an embodiment of the present invention; Figure 5 Schematic diagram of the process of recursive hierarchical decomposition according to an embodiment of the present invention; Figure 6 A schematic flow chart of an embodiment of a chatter detection method provided by the present invention; Figure 7 It is the time domain image of vibration acceleration data; Figure 8 This is the result diagram of time domain signal decomposition; Figure 9 The time domain comparison waveform of the original processing state signal and the reconstructed signal; Figure 10 It is a two-dimensional feature distribution diagram; Figure 11 A comparison chart of the test accuracy of chatter detection for different types of entropy processing; Figure 12 This is a structural schematic diagram of an embodiment of the device for extracting characteristic samples of machining chatter signals provided by the present invention. DETAILED DESCRIPTION

[0020] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0021] It should be understood that the schematic drawings are not drawn to scale. The flowcharts used in the present invention illustrate operations implemented according to some embodiments of the present invention. It should be understood that the operations of the flowcharts can be implemented out of sequence, and steps that have no logical contextual relationship can be reversed in order or implemented simultaneously. In addition, those skilled in the art, guided by the content of the present invention, can add one or more other operations to the flowcharts or remove one or more operations from the flowcharts. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in the form of software, or in one or more hardware modules or integrated circuits, or in different networks and / or processor systems and / or microcontroller systems.

[0022] The terms "first" and "second" in the embodiments of the present invention are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a technical feature designated as "first" or "second" may explicitly or implicitly include at least one such feature.

[0023] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present invention. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute a separate or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0024] The present invention provides a method for extracting characteristic samples of machining chatter signals, and a chatter detection method and device, which are respectively described below.

[0025] Figure 1 A flow chart of an embodiment of the method for extracting characteristic samples of machining chatter signals provided by the present invention is shown in FIG. Figure 1 As shown, the method for extracting characteristic samples of machining chatter signals includes: S101, performing optimized variational mode decomposition on the original processing chatter signal to obtain a number of IMF signal components; The raw machining chatter signals collected during machine tool processing generally have a low signal-to-noise ratio and cannot be directly used to generate samples. They require screening to remove noise and useless information. During this screening process, the raw machining chatter signals must first be decomposed into several IMF signal components. To ensure the quality of the decomposition, this embodiment uses a Newton-Raphson optimizer to optimize the parameters of the variational mode decomposition (VMD). Minimizing the ratio of permutation entropy to envelope entropy is the optimization objective. This generates the optimal initial parameters for the VMD, enabling a more comprehensive capture of the various features of the machining signal and improving the ability to decompose machining state signals.

[0026] S102, performing fusion index evaluation on the IMF signal components to obtain a quality score of each IMF signal component, and screening and reconstructing the IMF signal components according to the quality score to obtain a reconstructed signal; To ensure that the screening and reconstruction process effectively retains all valid information, the embodiment uses a fusion index to score the quality of each IMF signal component. This uses multiple metrics, including the inverse of spectral entropy, energy ratio, and Spearman rank correlation coefficient. The scores for each metric are then combined to perform screening and reconstruction to generate a reconstructed signal. By using different metrics to evaluate signal effectiveness from multiple perspectives, the embodiment effectively avoids overlooking useful information during the screening process.

[0027] S103 , recursively decomposing the reconstructed signal into layers based on a preset matrix operator to obtain a number of layer node components, and extracting singular spectral entropy feature samples from the layer node components to obtain a feature sample data set.

[0028] To capture and preserve short-term mutation features during sample generation, the embodiment first performs a hierarchical decomposition of the reconstructed signal and then uses singular spectral entropy to extract feature samples. During the hierarchical decomposition process, the embodiment decomposes the reconstructed signal into different scales layer by layer through a hierarchical analysis structure, avoiding the weakening of short-term mutation features caused by conventional multi-scale signal segmentation methods. Singular spectral entropy measures the complexity or uncertainty of the signal by analyzing the distribution of its singular values, which can enhance the focus on short-term mutation features. Through hierarchical decomposition and singular spectral entropy, the embodiment can effectively extract key features from the processing chatter signal and generate corresponding samples.

[0029] Compared with the prior art, the method for extracting characteristic samples of machining chatter signals provided by the present invention can avoid weakening short-term mutation features in signal segmentation by recursively decomposing the reconstructed signal. The singular spectral entropy can effectively capture local mutation features in the signal during sample generation. The hierarchical singular spectral entropy, which combines hierarchical decomposition and singular spectral entropy, can adaptively capture key features of the signal from fine to coarse scales, improve sensitivity to short-term mutations or transient events, and obtain accurate characteristic samples.

[0030] In some embodiments of the present invention, Figure 2 Schematic diagram of the process of optimizing variational mode decomposition according to an embodiment of the present invention, as shown in FIG. Figure 2 As shown in Figure 2, the optimized variational modal decomposition of the original machining chatter signal is performed to obtain several IMF signal components, including: S201. Determine an optimal parameter combination for variational mode decomposition based on a Newton-Raphson optimizer; S202, determining initial parameters of variational modal decomposition according to the optimal parameter combination, and performing variational modal decomposition on the original machining chatter signal according to the initial parameters to obtain IMF signal components; Among them, the Newton-Raphson optimizer takes minimizing the ratio of the permutation entropy and envelope entropy of the IMF signal component as the optimization objective.

[0031] Specifically, traditional optimization algorithms typically use a single feature as the fitness function when performing variational mode decomposition (VMD), such as signal capacity entropy, envelope entropy, or kurtosis. This makes it difficult for the optimization algorithm to fully capture the various features of the processing signal, thereby weakening the ability of VMD to decompose signals in different processing states.

[0032] In this regard, the embodiment takes minimizing the ratio of the permutation entropy and the envelope entropy of the IMF signal component as the optimization goal, and uses the Newton-Raphson optimizer to optimize the optimal number of modes for generating variational mode decomposition. and penalty factor The best parameter combination.

[0033] Optimization objective fitness function The formula is:

[0034] in, represents the permutation entropy, represents the envelope entropy.

[0035] The best embodiment is obtained by The parameter combination sets the initial parameters of variational modal decomposition, and then the original machining chatter signal is subjected to variational modal decomposition to obtain the various IMF signal components.

[0036] In some embodiments of the present invention, Figure 3 FIG. 1 is a flow chart of the quality scoring of the fusion indicator evaluation according to an embodiment of the present invention. Figure 3 As shown in Figure 1, the fusion index evaluation of the IMF signal components is performed to obtain the quality score of each IMF signal component, including: S301, calculating the fusion index score of each IMF signal component; S302, determining corresponding metric value vectors according to the scores of the fusion indicators; S303: Determine a quality score of each IMF signal component according to each metric value vector.

[0037] In some embodiments of the present invention, the fusion indicators include: inverse spectral entropy, energy ratio and Spearman rank correlation coefficient.

[0038] Specifically, when screening IMF signal components, existing research typically scores the IMF components using a single metric and then selects the top-ranked IMF components for signal reconstruction. However, relying on a single metric for IMF selection can overlook useful signal information from the machining process, reducing the robustness of the constructed machining chatter detection model in practical applications.

[0039] In this regard, the embodiment scores each IMF component by fusion index, which includes the inverse of spectrum entropy, energy ratio and Spearman rank correlation coefficient. The formula is:

[0040] Among them is Spectral entropy, Represents the smallest positive number.

[0041] Energy ratio The formula is:

[0042] in, is the energy of a single IMF, is the total energy of the signal.

[0043] Spearman rank correlation coefficient The formula is:

[0044] in, The value range is , For the The difference in variable ranks in samples, that is, ,in and Represent variables respectively and In the The rank of the samples, is the total number of samples.

[0045] Then in the process of fusing multiple scores, for the fusion index For each metric in , calculate the metric value vector of each IMF separately ,in To ensure all The value signs are consistent, all Perform non-negative processing, the formula is expressed as:

[0046] in, satisfy .

[0047] Finally, the weighted shares belonging to the same IMF are added together and averaged to obtain the final quality score of each IMF. The formula is expressed as:

[0048] In some embodiments of the present invention, Figure 4 FIG. 1 is a flow chart of IMF signal component screening and reconstruction according to an embodiment of the present invention. Figure 4 As shown, the IMF signal components are screened and reconstructed according to the quality score to obtain a reconstructed signal, including: S401, filtering IMF signal components in descending order according to their quality scores to obtain filtered IMF signal components, until the sum of the quality scores of the filtered IMF signal components reaches a preset score threshold; S402: Superimpose the filtered IMF signal components to obtain a reconstructed signal.

[0049] Specifically, after obtaining the quality score of each IMF, it can be expressed as The embodiment is first based on Arrange in descending order from large to small, the arrangement structure is expressed as The IMF components are then filtered out in descending order until the sum of the quality scores reaches a predetermined threshold. ( is a constant and satisfies ). The formula is:

[0050] in, , , the IMF corresponding to the range is identified as the selected optimal IMF signal component.

[0051] Then the selected optimal IMF signal components are superimposed to obtain the reconstructed signal.

[0052] In some embodiments of the present invention, the preset matrix operator includes a low-frequency average operator and a high-frequency difference operator.

[0053] In some embodiments of the present invention, Figure 5 FIG. 1 is a flow chart of recursive hierarchical decomposition according to an embodiment of the present invention, as shown in FIG. Figure 5As shown, the reconstructed signal is recursively decomposed hierarchically based on a preset matrix operator to obtain several hierarchical node components, including: S501, determining a calling order of preset matrix operators based on a preset vector; S502, selecting corresponding preset matrix operators in sequence according to the calling order to perform signal decomposition on the reconstructed signal or the node components of the previous level to obtain new node components of the first level; Among them, the values of each dimension of the preset vector are randomly 0 or 1, 0 corresponds to the low-frequency average operator, and 1 corresponds to the high-frequency difference operator.

[0054] Specifically, in order to avoid weakening the local fluctuation characteristics by conventional multi-scale entropy signal segmentation when constructing different scale features, the embodiment decomposes the reconstructed signal layer by layer through hierarchical decomposition.

[0055] First, for the time series of the input reconstruction signal ,in For the sequence length, the embodiment defines two matrix operators for time series reconstruction, where the low-frequency average operator The formula is:

[0056] The high-frequency difference operator formula is expressed as:

[0057] The low-frequency average operator and high-frequency difference operator can be expressed in matrices as follows:

[0058] in, , is the number of signals in the reconstructed signal, represents the low-frequency component of the time series, Represents the high-frequency component of the time series. Then the initial data sequence can be reconstructed as .

[0059] In order to continuously iterate the above operators to implement hierarchical analysis, the embodiment constructs an N-dimensional vector To determine the operator selected for each iteration. When the vector value is 0, it means the low-frequency average operator is selected, and when the vector value is 1, it means the high-frequency difference operator is selected. At the same time, based on the vector , determine an integer according to the following formula , used for Layers are decomposed hierarchically and assigned integer indices:

[0060] in, represents the number of levels of hierarchical decomposition, express Layer decomposition nodes.

[0061] By vector Pair Sequence Perform recursive hierarchical decomposition to obtain node components at different levels Expressed as:

[0062] Finally, for each hierarchical node, the embodiment calculates its singular spectrum entropy value to obtain the sample feature, which is expressed as follows:

[0063] Among them, represents the sample characteristics of the layered singular spectral entropy output, represents the singular spectral entropy, represents the embedding dimension, Indicates time lag.

[0064] Figure 6 A flow chart of an embodiment of the chatter detection method provided by the present invention is shown as follows: Figure 6 As shown, the chatter detection method includes: S601: Generate a chatter detection sample set based on a processing chatter signal feature sample extraction method, and train an initial chatter detection network based on the chatter detection sample set to obtain a fully trained chatter detection network. S602 : Predict and output the processing signal to be detected based on the chatter detection network to obtain a chatter detection result.

[0065] The method for extracting characteristic samples of a machining chatter signal is any one of the above-mentioned methods for extracting characteristic samples of a machining chatter signal.

[0066] Specifically, when implementing the chatter detection task, this embodiment first generates a chatter detection sample set using a method for extracting feature samples of processing chatter signals. This sample set is then divided into a training set and a test set in a ratio of 4:1. This embodiment then selects a support vector machine (SVM) as the classification model for training and learning, and uses the trained chatter detection network to distinguish the chatter state of the processing signal to be detected.

[0067] Furthermore, to verify the effectiveness of the embodiments of the present invention, the obtained sample data was classified into three sample types based on label information: stable state, mild chatter, and severe chatter. A support vector machine (SVM) was selected as the classification model for training and learning. A trained machining chatter state detection model was obtained, and the classification performance of the model was tested on a test set.

[0068] In the verification experiment, the processing parameters and tool holder overhang length of the experimental data are shown in Table 1 below: Table 1: Processing parameter settings

[0069] The time domain image of the vibration acceleration data collected in the embodiment is as follows: Figure 7 As shown, the embodiment first uses the Newton-Raphson optimizer to find the optimal decomposition parameters for each cutting state ,initial The upper and lower limit setting parameters are shown in Table 2 below. In Table 2, and for The upper and lower limits of and for The optimal decomposition parameters for each processing state obtained by the experiment are shown in Table 2 below. Then, based on the optimal decomposition parameters in Table 2, the embodiment uses the variational mode decomposition algorithm to decompose the signal data of the three processing states of stable state, slight chatter and severe chatter. The IMF time domain signal decomposition results of each processing state are as follows: Figure 8 shown.

[0070] Table 2: Initial The upper and lower limit range setting parameters and the optimal parameter

[0071] Then, the IMF is selected using the fusion index evaluation method given in the embodiment. Here, the threshold of the operator is set to 0.8. Table 3 shows the quality scores of each IMF calculation corresponding to each processing state. The bold values in the table represent the high-quality IMF signal components selected by the algorithm proposed in this invention for each processing state.

[0072] Table 3: Quality score of each IMF sub-signal corresponding to each processing state

[0073] Then the selected IMF components are reconstructed. Figure 9 The time domain comparison waveforms of the original processing state signal and the reconstructed signal are given. Figure 9 It can be seen that the reconstructed signal and the original signal maintain consistent trends and pulse characteristics in waveform characteristics, indicating that the IMF selection algorithm proposed in this invention can effectively extract the main information in the original signal and has good signal restoration capabilities.

[0074] Finally, the Hierarchical Singular Spectral Entropy (HSSE) method is used to extract feature samples. HSSE enhances sensitivity to signal chatter features by introducing multi-scale hierarchical analysis. The reconstructed signal for each processing state consists of 18,751 data points, divided into 190 samples, each containing 350 data points. To compare with HSSE, the present invention selects multi-scale sample entropy (MSE), multi-scale Renyi entropy (MRE), multi-scale spread entropy (MDE), multi-scale attention entropy (MAE), and multi-scale permutation entropy (MPE) to calculate the feature distribution of the reconstructed signal for each processing state under the same data conditions. All entropy calculations use the same parameter settings: embedding dimension is set to 4, delay is set to 1, and scale is set to 10. Figure 10 The two-dimensional feature distribution of each processing state extracted by six types of information entropy is shown. Figure 10 It can be seen that the hierarchical singular spectrum entropy HSSE feature extraction method used in the present invention has the clearest category distribution and the most obvious feature distribution boundary when distinguishing the three different processing states of stable state, slight chatter and severe chatter. In order to verify the chatter feature extraction effect of hierarchical singular spectrum entropy HSSE, the SVM machine learning model is used to identify and classify the feature signals extracted from the three processing states. The ratio of training samples to test samples is 4 to 1. In order to reduce the impact of data randomness and uneven distribution on the accuracy of processing chatter feature identification during the test process. The present invention conducted 5 feature recognition and classification experiments on processing chatter states for 6 different information entropies. Each experimental run will randomly re-divide the data set and calculate the average accuracy of the test set. The final test results are shown in Tables 4 and Figure 11 The average accuracy of the proposed method in 5 tests was 99.47%, which was the best result in machining chatter state identification and classification tests among the compared methods.

[0075] Table 4: Comparison of test accuracy of chatter detection using hierarchical singular spectrum entropy and different types of entropy

[0076] In summary, the method for extracting characteristic samples of machining chatter signals provided by the present invention can avoid weakening short-term mutation features during signal segmentation by recursively decomposing the reconstructed signal. Singular spectral entropy can effectively capture local mutation features in the signal during sample generation. Hierarchical singular spectral entropy, which combines hierarchical decomposition and singular spectral entropy, can adaptively capture key signal features from fine to coarse scales, improve sensitivity to short-term mutations or transient events, and obtain accurate characteristic samples.

[0077] In order to better implement the method for extracting characteristic samples of machining chatter signals in the embodiment of the present invention, based on the method for extracting characteristic samples of machining chatter signals, correspondingly, as shown in FIG. Figure 12 As shown, the present invention further provides a processing chatter signal feature sample extraction device, and the processing chatter signal feature sample extraction device 1200 includes: The signal decomposition unit 1201 is used to perform optimized variational mode decomposition on the original processing chatter signal to obtain a plurality of IMF signal components; A signal component screening unit 1202 is configured to perform fusion index evaluation on the IMF signal components to obtain a quality score of each IMF signal component, and screen and reconstruct the IMF signal components according to the quality score to obtain a reconstructed signal; The feature sample extraction unit 1203 is configured to perform recursive hierarchical decomposition on the reconstructed signal based on a preset matrix operator to obtain a plurality of hierarchical node components, and perform singular spectral entropy feature sample extraction on the hierarchical node components to obtain a feature sample data set.

[0078] The apparatus 1200 for extracting characteristic samples of machining chatter signals provided in the above embodiment can implement the technical solution described in the above embodiment of the method for extracting characteristic samples of machining chatter signals. The specific implementation principles of the above modules or units can be found in the corresponding contents of the above embodiment of the method for extracting characteristic samples of machining chatter signals, and will not be repeated here.

[0079] Accordingly, an embodiment of the present application further provides a computer-readable storage medium for storing a computer-readable program or instruction. When the program or instruction is executed by a processor, the steps or functions of the machining chatter signal feature sample extraction method and / or the chatter detection method provided in the above-mentioned method embodiments can be implemented.

[0080] Those skilled in the art will appreciate that all or part of the process flow of the above-described method embodiment can be implemented by instructing related hardware (such as a processor, controller, etc.) through a computer program, and the computer program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a magnetic disk, an optical disk, a read-only memory, or a random access memory.

[0081] The above describes in detail the method, apparatus, chatter detection device, and storage medium for extracting characteristic samples of machining chatter signals provided by the present invention. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is intended only to facilitate understanding of the method and core concepts of the present invention. Furthermore, those skilled in the art will appreciate that variations in the specific implementation methods and scope of application may occur based on the concepts of the present invention. Therefore, the contents of this specification should not be construed as limiting the present invention.

Claims

1. A method for extracting characteristic samples of machining chatter signals, characterized in that: include: The original machining chatter signal is subjected to optimized variational mode decomposition to obtain several IMF signal components; Performing a fusion index evaluation on the IMF signal components to obtain a quality score of each of the IMF signal components, and screening and reconstructing the IMF signal components according to the quality score to obtain a reconstructed signal; The reconstructed signal is recursively and hierarchically decomposed based on a preset matrix operator to obtain a plurality of hierarchical node components, and singular spectrum entropy feature samples are extracted from the hierarchical node components to obtain a feature sample data set.

2. The method for extracting characteristic samples of machining chatter signals according to claim 1, characterized in that: The optimized variational modal decomposition of the original machining chatter signal is performed to obtain several IMF signal components, including: Determine the optimal parameter combination for variational mode decomposition based on Newton-Raphson optimizer; determining initial parameters of variational modal decomposition according to the optimal parameter combination, and performing variational modal decomposition on the original machining chatter signal according to the initial parameters to obtain IMF signal components; The Newton-Raphson optimizer takes minimizing the ratio of the permutation entropy and the envelope entropy of the IMF signal component as an optimization goal.

3. The method for extracting characteristic samples of machining chatter signals according to claim 1, wherein: The performing fusion index evaluation on the IMF signal components to obtain a quality score of each of the IMF signal components includes: Calculating each fusion index score of each of the IMF signal components; Determine the corresponding metric value vectors according to the fusion index scores; A quality score of each of the IMF signal components is determined according to each of the metric value vectors.

4. The method for extracting characteristic samples of machining chatter signals according to claim 3, wherein: The fusion indicators include: inverse of spectrum entropy, energy ratio and Spearman rank correlation coefficient.

5. The method for extracting characteristic samples of machining chatter signals according to claim 1, wherein: The screening and reconstructing of the IMF signal components according to the quality score to obtain a reconstructed signal includes: Filtering the IMF signal components in descending order according to the quality scores to obtain filtered IMF signal components until the sum of the quality scores of the filtered IMF signal components reaches a preset score threshold; The filtered IMF signal components are superimposed to obtain a reconstructed signal.

6. The method for extracting characteristic samples of machining chatter signals according to claim 1, wherein: The preset matrix operators include a low-frequency average operator and a high-frequency difference operator.

7. The method for extracting characteristic samples of machining chatter signals according to claim 6, wherein: The recursive hierarchical decomposition of the reconstructed signal based on a preset matrix operator to obtain a plurality of hierarchical node components includes: Determining a calling order of the preset matrix operators based on a preset vector; Selecting the corresponding preset matrix operators in sequence according to the calling order to perform signal decomposition on the reconstructed signal or the previous level node component to obtain a new first level node component; The values of each dimension of the preset vector are randomly 0 or 1, 0 corresponds to the low-frequency average operator, and 1 corresponds to the high-frequency difference operator.

8. A chatter detection method, characterized in that: include: generating a chatter detection sample set based on a machining chatter signal feature sample extraction method, and training an initial chatter detection network based on the chatter detection sample set to obtain a fully trained chatter detection network; The chatter detection result is obtained by predicting and outputting the processing signal to be detected based on the chatter detection network. The method for extracting characteristic samples of machining chatter signals is the method for extracting characteristic samples of machining chatter signals according to any one of claims 1 to 7.

9. A device for extracting characteristic samples of machining chatter signals, characterized in that: include: A signal decomposition unit is used to perform optimized variational mode decomposition on the original machining chatter signal to obtain a number of IMF signal components; a signal component screening unit, configured to perform fusion index evaluation on the IMF signal components to obtain a quality score of each of the IMF signal components, and screen and reconstruct the IMF signal components according to the quality score to obtain a reconstructed signal; The feature sample extraction unit is used to perform recursive hierarchical decomposition on the reconstructed signal based on a preset matrix operator to obtain a plurality of hierarchical node components, and perform singular spectral entropy feature sample extraction on the hierarchical node components to obtain a feature sample data set.

10. A computer-readable storage medium, characterized in that Used to store computer-readable programs or instructions, which, when executed by a processor, can implement the steps of the method for extracting characteristic samples of machining chatter signals as described in any one of claims 1 to 7 and / or the method for detecting chatter as described in claim 8.

Images