Early warning method for chatter in floor boring based on vibration signal feature clustering

Abstract

This invention belongs to the field of fault early warning technology, specifically relating to an early warning method for chatter in floor-mounted boring machining based on vibration signal feature clustering. The method includes: acquiring a spindle vibration signal sequence; performing variational mode decomposition on the signal to obtain the frequency band energy entropy based on energy proportion, standard deviation of high-frequency noise components, and mean effective amplitude, and constructing a machining state feature vector by combining weighted spectral kurtosis; determining the cluster centers of stable cutting states based on the local density, relative distance, and sample size confidence of historical samples; obtaining a chatter early warning index based on the Euclidean distance between the current feature vector and the cluster centers and the cluster radius, and issuing an early warning accordingly. This invention, through the chatter early warning index, reduces the masking of weak chatter features by strong background noise, achieving sensitive detection of early chatter risks.

Description

Technical Field

[0001] This invention relates to the field of fault early warning technology. More specifically, this invention relates to an early warning method for chatter in floor boring machining based on vibration signal feature clustering. Background Technology

[0002] As a key piece of equipment in the field of heavy machinery machining, the floor-type boring machine plays a vital role in the boring and milling of large workpieces. During the heavy cutting process of the floor-type boring machine, self-excited vibration (i.e., chatter) often occurs between the tool and the workpiece. This chatter not only reduces the surface finish of the workpiece and causes tool chipping, but in severe cases, it can even damage the machine tool spindle. Therefore, real-time monitoring of the machining status of the floor-type boring machine spindle and timely early warning of chatter are crucial for ensuring the manufacturing quality and production safety of heavy equipment.

[0003] In related technologies, a threshold-based vibration amplitude monitoring method is commonly used. This method includes: installing a vibration sensor on the machine tool spindle, collecting the vibration signal of the spindle in real time, and calculating the root mean square value or peak value of the vibration signal; comparing the calculated vibration amplitude with a preset fixed alarm threshold; and when the vibration amplitude exceeds the threshold, determining that the machine tool is chattering and triggering an alarm shutdown.

[0004] However, the aforementioned technologies have certain limitations when dealing with complex factory environments. Due to the presence of significant electromagnetic interference and background noise in factory environments, subtle early chatter characteristics are easily masked by noise, rendering simple amplitude-based monitoring methods ineffective. Furthermore, vibration fluctuations caused by normal machining parameter adjustments are often difficult to distinguish in amplitude from vibration deviations caused by early chatter. Fixed threshold methods are prone to false alarms or missed alarms, thus affecting the machining efficiency of floor-type boring machines and the precise protection of the equipment. Summary of the Invention

[0005] To address the technical problems of existing technologies where weak chatter characteristics are easily masked in strong background noise environments and it is difficult to distinguish between fluctuations caused by normal machining parameter adjustments and early chatter symptoms, resulting in low early warning accuracy, this invention provides an early warning method for chatter in floor-type boring machines based on vibration signal feature clustering. The method includes: acquiring a floor-type boring machine spindle vibration signal sequence; performing variational mode decomposition on the spindle vibration signal sequence to obtain intrinsic mode function components; obtaining the frequency band energy entropy of the spindle vibration signal sequence based on the energy proportion of the intrinsic mode function components, the standard deviation of the high-frequency noise components, and the mean effective amplitude; and obtaining the acceleration characteristic of the spindle vibration signal sequence based on the energy proportion and spectral kurtosis of the intrinsic mode function components. The system employs the following methods: spectral kurtosis; constructing a state feature vector based on the frequency band energy entropy and weighted spectral kurtosis; using the number of other vectors contained within a preset distance range as the local density; obtaining the relative distance of the state feature vector based on the local density; obtaining the center index of the state feature vector based on the local density, relative distance, and total number of samples; determining the cluster center of the stable cutting state based on the center index; obtaining the flutter early warning index based on the Euclidean distance between the current state feature vector and the cluster center, and the cluster radius of the stable cutting state cluster center; the cluster radius is obtained based on the distance between the historical state feature vector and the cluster center; and generating an early warning control command based on the flutter early warning index.

[0006] This invention extracts the energy concentration features and nonlinear impact features of early chatter in a strong background noise environment by performing variational mode decomposition on the spindle vibration signal sequence and calculating the frequency band energy entropy and weighted spectral kurtosis. This solves the problem that a single indicator is difficult to distinguish between background noise and weak chatter signals, and is beneficial to improving the anti-interference capability of chatter detection. This invention uses the number of other vectors contained in the state feature vector of the historical feature sample set within a preset distance range as the local density, and determines the cluster center of the stable cutting state accordingly. It can automatically identify the normal working mode of the machine tool without manually labeling a large number of samples, which is beneficial to improving the adaptability of the monitoring system to different processing conditions. This invention obtains the early chatter warning index based on Euclidean distance and cluster radius. The early chatter warning index is kept at a low level within the safety boundary to suppress false alarms, and rises rapidly after the boundary is exceeded to indicate risk. This is beneficial to achieve accurate early warning of chatter in the spindle of a floor boring machine.

[0007] Preferably, the step of obtaining the spindle vibration signal sequence of the floor-type boring machine includes: obtaining the analog voltage signal output by the piezoelectric accelerometer installed on the end face of the spindle box of the floor-type boring machine; converting the analog voltage signal into a standard voltage signal through a charge amplifier; filtering out interference signals with excessively high frequencies in the standard voltage signal using a filter; and acquiring the processed signal at fixed time intervals through an analog-to-digital conversion module to obtain the spindle vibration signal sequence.

[0008] Preferably, the method for obtaining the standard deviation of the high-frequency noise component and the mean effective amplitude includes: selecting the intrinsic mode function component with the highest center frequency, using the standard deviation of this component as the standard deviation of the high-frequency noise component, and using the average of the absolute values ​​of all data points in the main shaft vibration signal sequence as the mean effective amplitude.

[0009] Preferably, the frequency band energy entropy satisfies the following relationship: In the formula, The frequency band energy entropy of the main shaft vibration signal sequence, For the first The energy percentage of each eigenmode function component The standard deviation of the high-frequency noise component in the spindle vibration signal sequence is the standard deviation of the high-frequency noise component. The effective amplitude mean of the spindle vibration signal sequence The total number of intrinsic mode function components. It is an exponential function with the natural constant as the base.

[0010] This invention obtains the frequency band energy entropy by calculating the ratio of the standard deviation of the high-frequency noise component to the mean of the effective amplitude. When the standard deviation of the background noise component is large relative to the mean of the effective amplitude, the value of the frequency band energy entropy is amplified accordingly, which cancels the masking effect of strong background noise on weak flutter characteristics. This is beneficial for expressing flutter symptoms in the state feature vector under different noise levels.

[0011] Preferably, the weighted spectral kurtosis satisfies the following relationship: In the formula, The weighted spectral kurtosis characteristics of the main shaft vibration signal sequence, The total number of intrinsic mode function components. For the first The energy percentage of each eigenmode function component For the first Spectral kurtosis of each intrinsic mode function component.

[0012] This invention weights the spectral kurtosis of each intrinsic mode function component with its energy proportion to obtain a weighted spectral kurtosis. This weighted spectral kurtosis mainly reflects the impact characteristics in the frequency bands that carry the main cutting energy, effectively suppressing the pollution of the overall features by false high spectral kurtosis values ​​generated by random interference in the low-energy frequency band, and helping to improve the directivity of the feature vector to the real cutting chatter impact.

[0013] Preferably, obtaining the relative distance of the state feature vectors based on local density includes: for the state feature vector with the highest local density, taking the maximum value of all distance values ​​in the historical feature sample set as its relative distance; for each remaining state feature vector, obtaining the Euclidean distance between the state feature vector and all state feature vectors with local densities higher than that state feature vector, and taking the minimum value of the Euclidean distance as the relative distance of that state feature vector.

[0014] Preferably, the central index satisfies the following relationship: In the formula, For the first The central index of each state feature vector For the first Local density of a state feature vector, For the first The relative distance between the state feature vectors The total number of samples in the historical feature sample set. Set a preset confidence threshold for the number of samples. For adjustment coefficients, It is an exponential function with the natural constant as the base.

[0015] This invention constructs a centrality index by calculating the difference between the total number of samples and a preset confidence threshold for the number of samples. This forces the clustering centrality index of the state feature vector to be reduced during the cold start phase of the monitoring process. This prevents accidental clustering caused by data sparsity from being mistakenly identified as stable cutting centers, which helps to ensure the robustness of the model learning process and makes fault warnings more accurate.

[0016] Preferably, before obtaining the cluster radius, the method further includes: marking the cluster centers of stable cutting states as stable classes; for all state feature vectors other than the cluster centers of stable cutting states, traversing each state feature vector in descending order of local density; for the currently traversed state feature vector, finding the vector with the closest Euclidean distance to the vector among all state feature vectors with a local density higher than the vector, using it as a guiding vector, and marking the category of the currently traversed state feature vector as the same category as the guiding vector.

[0017] Preferably, the method for obtaining the cluster radius further includes: filtering out all state feature vectors that are ultimately marked as stable classes; calculating the average Euclidean distance between all the filtered state feature vectors and the cluster centers of the stable cutting states, and using the average value as the cluster radius of the cluster centers of the stable cutting states.

[0018] This invention uses the average Euclidean distance between all selected state feature vectors and the cluster centers of stable cutting states as the cluster radius, and adaptively generates the boundary of the stable region based on the actual distribution of historical processing data, providing an objective reference benchmark for chatter early warning.

[0019] Preferably, the flutter early warning index satisfies the following relationship: In the formula, The flutter early warning index is the feature vector of the current state. Let Euclidean distance be the feature vector of the current state and the cluster center of the stable cutting state. The cluster radius, To activate the gain coefficient, For the boundary tolerance factor, It is the hyperbolic tangent function.

[0020] The beneficial effects of this invention are as follows: By combining the evaluation of signal complexity using frequency band energy entropy and the screening of impact components using weighted spectral kurtosis, and by dynamically correcting the features using environmental noise levels, this invention enables the separation of weak early signs of chatter from masked vibration signals in the complex electromagnetic and mechanical background noise environment of a floor-type boring machine. This solves the problem of traditional methods failing to accurately identify chatter features in low signal-to-noise ratio environments, making data analysis in harsh industrial settings more reliable. Furthermore, this invention determines the cluster center of stable cutting states through the local density distribution of statistical state feature vectors and automatically constructs a stable working reference benchmark in the absence of fault samples. This eliminates the need for pre-training with large amounts of fault data during the monitoring process, reducing the debugging cost of the system when switching between different machining tasks and improving its adaptability to different working conditions. Finally, this invention constructs a dynamic tolerance boundary based on the cluster radius, allowing the fault warning system to remain tolerant of normal vibration amplitude fluctuations caused by adjustments in cutting depth or feed rate, only alarming for abnormal trends deviating from stable modes. This reduces unnecessary downtime due to false alarms while ensuring machining safety, thus guaranteeing the efficiency of continuous production. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the early warning method for chatter in floor boring based on vibration signal feature clustering in this invention;

[0022] Figure 2 This is a schematic diagram illustrating the comparison of the early warning effects in this invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0025] This invention discloses an early warning method for chatter in floor boring based on vibration signal feature clustering, referring to... Figure 1 This includes steps S1 to S4:

[0026] S1. Obtain the vibration signal sequence of the floor-type boring machine spindle containing timing information through a piezoelectric accelerometer.

[0027] It should be noted that chatter during floor-type boring is a vibration between the tool and the workpiece, which is transmitted to the spindle head. Due to electromagnetic interference and background noise from machine operation in the factory environment, the directly acquired signal will contain a lot of interference noise. Therefore, this invention installs a sensor at the front end of the spindle head to acquire a sequence of vibration signals from the floor-type boring machine spindle containing timing information.

[0028] Specifically, the analog voltage signal output from the piezoelectric accelerometer mounted on the end face of the spindle box of the floor-type boring machine is acquired. The analog voltage signal is converted into a standard voltage signal using a charge amplifier. A filter is used to remove excessively high-frequency interference signals from the standard voltage signal. The processed signal is then acquired at fixed time intervals using an analog-to-digital converter to obtain a spindle vibration signal sequence.

[0029] S2. Based on the sub-mode characteristics of variational mode decomposition and combined with the environmental noise level, construct the state feature vector of the principal shaft vibration signal sequence.

[0030] It should be noted that when chatter occurs, the signal energy concentrates at a specific frequency, and the waveform exhibits sharp spikes. A single feature is difficult to accurately describe this change and is easily affected by noise. Therefore, this invention decomposes the signal into components of different frequencies, calculates their energy distribution and waveform spikes separately, and combines them into a feature vector that comprehensively reflects the processing state.

[0031] Specifically, variational mode decomposition is performed on any spindle vibration signal sequence to obtain several intrinsic mode function (EMF) components with different center frequencies. The energy value of each EMF component is calculated, and the proportion of each EMF component's energy value to the total energy of all components is also calculated. The EMF component with the highest center frequency is selected, and its standard deviation is used as the standard deviation of the high-frequency noise component. The average of the absolute values ​​of all data points in the spindle vibration signal sequence is used as the mean effective amplitude. Combining the energy proportion with the standard deviation and mean effective amplitude of the high-frequency noise component, the frequency band energy entropy of the spindle vibration signal sequence is calculated.

[0032] Specifically, the frequency band energy entropy satisfies the following relationship:

[0033] ;

[0034] In the formula, The frequency band energy entropy of the main shaft vibration signal sequence, For the first The energy percentage of each eigenmode function component The standard deviation of the high-frequency noise component in the spindle vibration signal sequence is the standard deviation of the high-frequency noise component. The effective amplitude mean of the spindle vibration signal sequence The total number of intrinsic mode function components. It is an exponential function with the natural constant as the base.

[0035] in, This reflects the uniformity of energy distribution across different frequencies. A larger value indicates a more uniform energy distribution, representing a more stable cutting process and resulting in a higher frequency band energy entropy; a smaller value indicates that energy is concentrated at a few frequencies, suggesting possible flutter and resulting in a lower frequency band energy entropy. Used to eliminate the influence of background noise on the calculation results. The higher the background noise ratio, the better. A value greater than 1 amplifies the frequency band energy entropy, preventing misjudgment due to noise masking signal characteristics; conversely, a smaller background noise ratio... The closer it is to 1, the more it maintains the original magnitude of the frequency band energy entropy.

[0036] Furthermore, the ratio of the fourth central moment to the square of the second central moment of each intrinsic mode function component is calculated to obtain the spectral kurtosis of each component. The spectral kurtosis of each component is weighted and aggregated with its energy proportion to obtain the weighted spectral kurtosis of the principal shaft vibration signal sequence; and the frequency band energy entropy of each principal shaft vibration signal sequence is combined with the weighted spectral kurtosis to form a state feature vector.

[0037] Specifically, the weighted spectral kurtosis satisfies the following relation:

[0038] ;

[0039] In the formula, The weighted spectral kurtosis characteristics of the main shaft vibration signal sequence, The total number of intrinsic mode function components. For the first The energy percentage of each eigenmode function component For the first Spectral kurtosis of each intrinsic mode function component.

[0040] in, This represents the impact characteristics of a single frequency band component. The larger the number, the more likely it is to be the first. The sharper the signal waveform within each intrinsic mode function component, the stronger the nonlinear impulse component it contains; The smaller the value, the more stable the overall Gaussian distribution of the signal, corresponding to a stable cutting state; through Weighting is applied so that the final weighted spectral kurtosis features mainly reflect the impact characteristics of the high-energy frequency band.

[0041] S3. Based on the local density distribution of the state feature vector set and the sample size confidence level, obtain the cluster centers of the stable cutting state.

[0042] It should be noted that in actual machining, the process is mostly in a stable cutting state. The feature vectors corresponding to these states cluster together in the data space, forming a dense region. In contrast, the feature vectors during chatter are usually more dispersed. To find the data center representing the stable state, it is necessary to analyze the density of the data. Furthermore, at the beginning of machining, the data is scarce, and the calculated results may be inaccurate, requiring limitations. Therefore, this invention calculates the local density of each data point and adjusts the score according to the amount of data, thereby selecting reliable centers of stable cutting states.

[0043] Specifically, state feature vectors obtained from multiple processing steps are collected to establish a historical feature sample set. The Euclidean distance between any two state feature vectors in the sample set is calculated. For each state feature vector, the number of other vectors contained within a preset distance range is counted, which is used as the local density of that vector.

[0044] Furthermore, all state feature vectors are sorted in descending order of local density. For the vector with the highest local density, the maximum value of all distance values ​​in the historical feature sample set is taken as its relative distance. For each remaining state feature vector, the Euclidean distances between it and all other state feature vectors with higher local densities are obtained, and the minimum value of these Euclidean distances is taken as the relative distance of that state feature vector. Based on the local density, relative distance, and total sample size of each state feature vector, a centrality index is calculated. The state feature vector with the highest centrality index is taken as the cluster center for the stable cutting states.

[0045] Specifically, the central index satisfies the following relation:

[0046] ;

[0047] In the formula, For the first The central index of each state feature vector For the first Local density of a state feature vector, For the first The relative distance between the state feature vectors The total number of samples in the historical feature sample set. Set a preset confidence threshold for the number of samples. For adjustment coefficients, The empirical value range is [0.01, 0.1]. In this embodiment, it is an exponential function with the natural constant as its base. It is 200. The value is 0.05, which can be determined by the implementers based on the actual situation. and In other embodiments, implementers can set adjustment coefficients according to the actual implementation situation. For example, when the processing conditions are relatively simple and the data features converge quickly, the adjustment coefficient can be appropriately increased to accelerate the model into a mature monitoring state; when the processing conditions are complex or the data fluctuates greatly, the adjustment coefficient can be appropriately reduced so that the model can accumulate more samples before giving a high confidence level, thereby enhancing the robustness of the system.

[0048] in: This reflects whether a state feature vector is suitable as a center. The larger the value, the denser the surrounding area of ​​this state feature vector, and the farther away it is from other dense areas, indicating that it is more likely to be the center of an independent stable state pattern. This reflects the impact of data volume on the reliability of the results. When the total number of samples... Much larger When the sample size is small, the value is close to 1, indicating sufficient data and a valid score; when the total sample size is small, the value is closer to 0, leading to a decrease in the centrality index. The smaller the value, the better, thus preventing the incorrect selection of a center when data is insufficient.

[0049] S4. Obtain the early warning index of chatter based on the distance between the real-time state feature vector and the cluster center of the stable cutting state.

[0050] It should be noted that when chatter occurs, the feature vector shifts outward from the central region representing the stable state. Whether an offset has occurred can be determined by calculating the distance between the real-time feature vector and the stable center. However, normal parameter adjustments can also cause slight shifts; only when the offset exceeds a certain range does it signify a real risk. Therefore, this invention obtains an early warning index for chatter based on the distance between the real-time state feature vector and the cluster center of the stable cutting state.

[0051] Specifically, the Euclidean distance between the current state feature vector and the cluster centers of the stable cutting states is calculated; the cluster centers of the stable cutting states are designated as the stable classes; for all state feature vectors except those of the stable cutting states, each state feature vector is traversed in descending order of local density; for the currently traversed state feature vector, among all state feature vectors with a local density higher than that vector, the vector with the closest Euclidean distance to that vector is found and used as the guiding vector. The class of the currently traversed state feature vector is then labeled as the same as that of the guiding vector.

[0052] It should be added that for the current state feature vector, the local density of its guiding vector is higher, so the guiding vector has been processed and obtained the class label.

[0053] Further, all state feature vectors ultimately labeled as stable are selected, and the average Euclidean distance between these vectors and the cluster centers of stable cutting states is calculated. This average distance is used as the cluster radius of the cluster centers of stable cutting states. The Euclidean distance between the state feature vectors of the current spindle vibration signal sequence and the cluster centers of stable cutting states is calculated. Based on this Euclidean distance and the cluster radius of the cluster centers of stable cutting states, the chatter early warning index of the spindle vibration signal sequence is obtained. In response to the chatter early warning index exceeding a preset alarm value, an early warning control command is generated, for example, an alarm value of 0.8. The implementer can determine the alarm value according to the actual situation.

[0054] Specifically, the flutter early warning index satisfies the following relationship:

[0055] ;

[0056] In the formula, The flutter early warning index is the feature vector of the current state. Let Euclidean distance be the feature vector of the current state and the cluster center of the stable cutting state. The cluster radius, To activate the gain coefficient, For the boundary tolerance factor, It is the hyperbolic tangent function; The empirical value range is [1.0, 10.0]. The empirical value range is [1.1, 1.5]. In this embodiment, It is 5.0. The value is 1.2; in other embodiments, the implementer may set it according to the actual implementation situation. For example, when the system is required to instantly trigger a high alarm value after a security boundary is breached, the value can be appropriately increased. When it is desired to maintain a certain transition buffer zone near the boundary, the size can be appropriately reduced. When the machine tool has been in service for a long time and the mechanical clearance is large, resulting in a wide range of foundation vibration fluctuations, the foundation vibration level can be appropriately increased. To reduce the false alarm rate; when performing high-precision finishing and requiring extremely high surface quality, the frequency can be appropriately reduced. To improve the detection rate of subtle abnormalities.

[0057] in, It reflects the degree to which the current state deviates from the stable center. The larger the value, the further the current state feature vector deviates from the cluster center of the stable cutting state, indicating a more unstable system and a higher risk of flutter early warning index. The value increases rapidly. It functions as a switch for adjustment. When the distance... It did not exceed the expanded boundary. hour, Approaching 0, thus suppressing the early warning index of flutter and ignoring normal fluctuations; when Exceeding the boundary hour, The rapid increase in flutter index relieves the inhibition, allowing the early warning index to be primarily determined by the degree of deviation, thus enabling sensitive detection of abnormal shifts.

[0058] For example, Figure 2 The diagram shows a comparison of the early warning effects in this invention. It can be seen from the diagram that there is a significant lag in flutter warning based solely on vibration signals. However, the flutter early warning index used in this invention can keenly capture subtle changes in the frequency domain energy distribution of the signal. It shows a sharp upward trend before the amplitude increases dramatically, thus enabling early and accurate warning of flutter in floor boring processes.

Claims

1. An early warning method for chatter in floor-mounted boring machining based on vibration signal feature clustering, characterized in that, include: Obtain the spindle vibration signal sequence of the floor-type boring machine; perform variational mode decomposition on the spindle vibration signal sequence to obtain the intrinsic mode function components; The frequency band energy entropy of the principal shaft vibration signal sequence is obtained based on the energy proportion of the intrinsic mode function components, the standard deviation of the high-frequency noise components, and the mean effective amplitude, satisfying the following relationship: , The frequency band energy entropy of the main shaft vibration signal sequence, For the first The energy percentage of each eigenmode function component The standard deviation of the high-frequency noise component in the spindle vibration signal sequence. The effective amplitude mean of the spindle vibration signal sequence The total number of intrinsic mode function components. It is an exponential function with the natural constant as its base; The weighted spectral kurtosis of the principal shaft vibration signal sequence is obtained based on the energy proportion and spectral kurtosis of the intrinsic mode function components; a state feature vector is constructed based on the frequency band energy entropy and the weighted spectral kurtosis. The local density is defined as the number of other vectors contained within a preset distance range of the state feature vector; the relative distance of the state feature vector is obtained based on the local density; the central index of the state feature vector is obtained based on the local density, relative distance, and total number of samples, satisfying the following relationship: , For the first The central index of a state feature vector For the first Local density of a state feature vector, For the first The relative distance between each state feature vector. The total number of samples in the historical feature sample set. Set a preset confidence threshold for the number of samples. This is the adjustment coefficient; The cluster centers of the stable cutting state are determined based on the centrality index; the flutter early warning index is obtained based on the Euclidean distance between the current state feature vector and the cluster centers, and the cluster radius of the cluster centers of the stable cutting state, satisfying the following relationship: , The flutter early warning index is the feature vector of the current state. Let Euclidean distance be the feature vector of the current state and the cluster center of the stable cutting state. The cluster radius, To activate the gain coefficient, For the boundary tolerance factor, The value is a hyperbolic tangent function; the cluster radius is obtained based on the distance between the historical state feature vector and the cluster center. Early warning control instructions are generated based on the flutter early warning index.

2. The early warning method for chatter in floor-mounted boring machining based on vibration signal feature clustering according to claim 1, characterized in that, The process of obtaining the spindle vibration signal sequence of the floor-type boring machine includes: acquiring the analog voltage signal output by the piezoelectric accelerometer installed on the end face of the spindle box of the floor-type boring machine; converting the analog voltage signal into a standard voltage signal through a charge amplifier; filtering out interference signals with excessively high frequencies in the standard voltage signal using a filter; and acquiring the processed signal at fixed time intervals through an analog-to-digital conversion module to obtain the spindle vibration signal sequence.

3. The early warning method for chatter in floor boring based on vibration signal feature clustering according to claim 1, characterized in that, The method for obtaining the standard deviation of the high-frequency noise component and the mean effective amplitude includes: selecting the intrinsic mode function component with the highest center frequency, using the standard deviation of this component as the standard deviation of the high-frequency noise component, and using the average of the absolute values ​​of all data points in the main shaft vibration signal sequence as the mean effective amplitude.

4. The early warning method for chatter in floor-mounted boring machining based on vibration signal feature clustering according to claim 1, characterized in that, The weighted spectral kurtosis satisfies the following relation: ； In the formula, The weighted spectral kurtosis characteristics of the main shaft vibration signal sequence, The total number of intrinsic mode function components. For the first The energy percentage of each eigenmode function component For the first Spectral kurtosis of each intrinsic mode function component.

5. The early warning method for chatter in floor-mounted boring machining based on vibration signal feature clustering according to claim 1, characterized in that, The step of obtaining the relative distance of state feature vectors based on local density includes: for the state feature vector with the highest local density, taking the maximum value of all distance values ​​in the historical feature sample set as its relative distance; for each remaining state feature vector, obtaining the Euclidean distance between the state feature vector and all state feature vectors with local densities higher than that state feature vector, and taking the minimum value of the Euclidean distance as the relative distance of that state feature vector.

6. The early warning method for chatter in floor boring based on vibration signal feature clustering according to claim 1, characterized in that, Before obtaining the cluster radius, the method further includes: marking the cluster centers of stable cutting states as stable classes; for all state feature vectors other than the cluster centers of stable cutting states, traversing each state feature vector in descending order of local density; for the currently traversed state feature vector, finding the vector with the closest Euclidean distance to the vector among all state feature vectors with a local density higher than the vector, using it as a guide vector, and marking the category of the currently traversed state feature vector as the same category as the guide vector.

7. The early warning method for chatter in floor boring based on vibration signal feature clustering according to claim 6, characterized in that, The method for obtaining the cluster radius further includes: filtering out all state feature vectors that are ultimately marked as stable classes; calculating the average Euclidean distance between all the filtered state feature vectors and the cluster centers of the stable cutting states, and using this average value as the cluster radius of the cluster centers of the stable cutting states.

Images