CN110928237B - Vibration signal-based numerical control machining center flutter online identification method - Google Patents
Vibration signal-based numerical control machining center flutter online identification method Download PDFInfo
- Publication number
- CN110928237B CN110928237B CN201911327809.4A CN201911327809A CN110928237B CN 110928237 B CN110928237 B CN 110928237B CN 201911327809 A CN201911327809 A CN 201911327809A CN 110928237 B CN110928237 B CN 110928237B
- Authority
- CN
- China
- Prior art keywords
- scale
- numerical control
- tree model
- machining center
- signal
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B19/00—Programme-control systems
- G05B19/02—Programme-control systems electric
- G05B19/18—Numerical control [NC], i.e. automatically operating machines, in particular machine tools, e.g. in a manufacturing environment, so as to execute positioning, movement or co-ordinated operations by means of programme data in numerical form
- G05B19/406—Numerical control [NC], i.e. automatically operating machines, in particular machine tools, e.g. in a manufacturing environment, so as to execute positioning, movement or co-ordinated operations by means of programme data in numerical form characterised by monitoring or safety
- G05B19/4065—Monitoring tool breakage, life or condition
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B2219/00—Program-control systems
- G05B2219/30—Nc systems
- G05B2219/37—Measurements
- G05B2219/37616—Use same monitoring tools to monitor tool and workpiece
Landscapes
- Engineering & Computer Science (AREA)
- Human Computer Interaction (AREA)
- Manufacturing & Machinery (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Automatic Control Of Machine Tools (AREA)
Abstract
The invention belongs to the field of online identification of chatter in the cutting process of a numerical control machining center, and particularly discloses an online identification method of chatter of the numerical control machining center based on vibration signals. The method comprises the following steps: the method comprises the steps of collecting a main shaft vibration signal generated when a numerical control machining center performs cutting machining, preprocessing the vibration signal, constructing a multi-scale arrangement entropy and a multi-scale power spectrum entropy of the preprocessed signal, splicing the multi-scale arrangement entropy and the multi-scale power spectrum entropy to be used as a feature vector to be input into a constructed gradient lifting tree model, performing iterative training on the gradient lifting tree model to obtain an optimal gradient lifting tree model, preprocessing the main shaft vibration signal generated in the machining process, and then inputting the preprocessed main shaft vibration signal into the optimal gradient lifting tree model, so that online identification of flutter of the numerical control machining center is realized. The invention can realize the extraction of the flutter sensitivity characteristics of the machine tool and the identification of whether the numerical control machining center flutters or not and the flutter severity, and has the advantages of high monitoring real-time performance, high identification accuracy, good generalization capability and the like.
Description
Technical Field
The invention belongs to the technical field of monitoring of a machining process of a numerical control machining center, and particularly relates to a vibration signal-based online identification method for chatter of the numerical control machining center.
Background
Flutter is a self-excited vibration that has instability, confusion, and anomaly. The on-line flutter identification of the numerical control machining center refers to that a computer or other computing equipment judges and identifies whether the numerical control machining center generates flutter or not by detecting signal changes of various sensors in the process of machining workpieces by the numerical control machining center. The flutter identification process of the numerical control machining center is essentially a pattern identification process, and a complete flutter online identification system consists of a research object (the numerical control machining center), machining conditions, a sensor network, time sequence signal processing, sensitive feature extraction, pattern identification and the like.
If flutter occurs in the machining process, poor part machining surface quality, low material removal rate, cutter abrasion and machine tool utilization rate reduction are directly caused if the flutter occurs, and a workpiece is scrapped and a cutter is damaged if the flutter occurs, and related machine parts of the machine tool can be damaged even in extreme cases or even the machine tool is stopped for maintenance. Therefore, it is necessary to quickly identify the state of the machine tool in real time and monitor whether the machine tool vibrates.
Through years of development, the flutter monitoring technology of the numerical control machining center makes certain progress in the aspects of breadth and depth, but the monitoring time and the identification precision do not meet the actual monitoring requirement at present, the application range is limited, the requirements of automation and intellectualization are not met, and the flutter monitoring technology still has certain limitations in the aspect of actual application, such as the problems of insufficient identification precision, poor monitoring real-time performance, incapability of coping with complex machining scenes and machining conditions and the like.
Therefore, there is a need in the art to provide an on-line vibration identification method for a nc machining center based on vibration signals, which can accurately, quickly and in real time identify whether the nc machining center vibrates and the severity of the vibration.
Disclosure of Invention
Aiming at the defects or improvement requirements of the prior art, the invention provides the numerical control machining center flutter online identification method based on the combination of the multi-scale entropy sensitivity and the gradient lifting tree algorithm, the method can avoid manually adjusting relevant flutter identification algorithm parameters under the condition of variable machining scenes and variable machining parameters, can realize extraction of machine tool flutter sensitivity characteristics and identification of whether the numerical control machining center flutters and the flutter severity, has the advantages of high monitoring real-time performance, high identification accuracy, good generalization capability and the like, and can accurately, quickly and real-timely identify whether the numerical control machining center flutters and the flutter severity.
In order to achieve the purpose, the invention provides a vibration signal-based numerical control machining center flutter online identification method, which comprises the following steps of:
s1, collecting a main shaft vibration signal when the numerical control machining center performs cutting machining, and performing equal-length segmentation and normalization on the main shaft vibration signal according to the main shaft rotation period;
s2, eliminating harmonic frequency signals in the spindle vibration signals after equal length segmentation and regularization processing to obtain residual component signals;
s3, calculating the multi-scale permutation entropy of the residual component signals by adopting a multi-scale permutation entropy characteristic algorithm, calculating the multi-scale power spectrum entropy of the residual component signals by adopting a multi-scale power spectrum entropy characteristic algorithm, and splicing the multi-scale permutation entropy and the multi-scale power spectrum entropy to serve as feature vectors;
s4, constructing a gradient lifting tree model, inputting the characteristic vector serving as a sample into the gradient lifting tree model, performing iterative training on the gradient lifting tree model to enable the loss value of the gradient lifting tree model to reach a preset value, and ending the iterative training on the gradient lifting tree model to obtain an optimal gradient lifting tree model;
s5, the spindle vibration signal in the machining process is processed in steps S1 to S3 and then input into the optimal gradient lifting tree model, and therefore online identification of chatter of the numerical control machining center is achieved.
More preferably, step S1 specifically includes the following steps: installing a vibration sensor on a main shaft of a numerical control machining center, acquiring a three-way vibration signal of the main shaft, intercepting the main shaft vibration signal in the machining process from the vibration signal according to the rotation period of the main shaft, performing sliding framing on the main shaft vibration signal, and dividing the main shaft vibration signal into a plurality of equilong signal sections, thereby completing equilong division and normalization of the vibration signal; each signal segment has a class value and a spindle rotation period corresponding thereto.
More preferably, in step S2, the harmonic signals in the spindle vibration signals after the equal length segmentation and regularization are removed by using an angle synchronous averaging algorithm.
More preferably, step S3 specifically includes the following steps:
s31, aiming at the selected M scale factors, carrying out coarse graining on the residual component signals according to the scale factors to obtain M coarse graining signals;
s32, reconstructing the coarse grained signals under a certain scale factor to obtain a reconstruction matrix under the scale factor, and then using each row vector in the reconstruction matrix as an original sequence to perform ascending arrangement on the original sequence;
s33, comparing the sequence after the ascending sequence with the original sequence to obtain an index sequence after the ascending sequence of the elements of the vector corresponding to the original sequence;
s34, counting the times of the same index sequence in the reconstruction matrix, and calculating the permutation entropy of the residual component signals under the scale factor by adopting a multi-scale permutation entropy characteristic algorithm;
s35, calculating the power spectrum entropy of the residual component signals under a certain scale factor by adopting a multi-scale power spectrum entropy characteristic algorithm for the coarse grained signals under the scale factor;
s36 repeating the steps S32 to S35 to obtain multi-scale arrangement entropy and multi-scale power spectrum entropy of the residual component signals under M scale factors;
s37, splicing the multi-scale arrangement entropy and the multi-scale power spectrum entropy to be used as the feature vector.
Preferably, the model of the multi-scale permutation entropy feature algorithm is as follows:
the model of the multi-scale power spectrum entropy characteristic algorithm is as follows:
wherein s is a scale factor when calculating the multi-scale permutation entropy, m is an embedding dimension when reconstructing the residual component signal after coarse graining, and N islAs a source of the coarse grained signalThe number of times the start sequence l occurs,
n is the length of the residual component signal, py(wi) Is a coarse grained signal x at a certain scale factoriThe corresponding normalized power spectrum.
More preferably, in step S4, the gradient lifting tree model is:
rm,i,l＝yi,l-pm,l(xi)
wherein R ism,l,jThe jth region is divided for the mth round and corresponding to the leaf node of the gradient lifting tree model with the class l, and I (x belongs to Rm,l,j) To indicate a function, when x belongs to Rm,l,jWhen it is 1, otherwise it is 0, om,i,lIs the ith sample x in the mth iteration processiLeaf node output, y, for a gradient lifting tree model of class lcIs the class value of sample x, if ycSample x belongs to class c, p, 1c(x) The probability that x belongs to the class C is judged for a gradient lifting tree model f (x), C is the number of the class of the flutter of the numerical control machining center, C is more than or equal to 0 and less than or equal to C-1, rm,i,lFor the m-th, i-th sample xiFor negative gradient errors of class l, each sample xiIs composed of feature vectors, yi,lFor the ith sample xiTrue probability, p, for class lm,l(xi) For the ith sample xiPredict probability of class L during the mth iteration, L (y, f (x))As a function of the logarithmic loss.
More preferably, step S4 specifically includes the following steps:
s41 inputting feature vectors into the gradient lifting tree model;
s42 calculating loss value according to log loss function L (y, f (x)) in the gradient lifting tree model, and then adopting negative gradient error function r according to the loss valuem,i,lCalculating a negative gradient error;
s43 inputs the negative gradient error to the leaf node output function om,i,lUpdating decision tree leaf node output, and updating the gradient lifting tree model according to leaf node output;
and S44, the steps are circulated to carry out iterative updating until the loss value calculated by the loss function L (y, f (x)) reaches a preset value, and the training of the gradient lifting tree model is finished to obtain the optimal gradient lifting tree model.
Further preferably, the harmonic component of each signal segment is represented by Sp(θ) and is constructed using the formula:
wherein u ═ { u ═1,u2,…,uNThe spindle vibration signal is a spindle vibration signal fragment subjected to equal-length segmentation and regularization; k is the number of signal sections divided by the main shaft vibration signal, K is more than or equal to 0 and less than or equal to K-1, and K is a subscript of a kth section signal; theta is the rotation period of the main shaft.
Generally, compared with the prior art, the above technical solution conceived by the present invention mainly has the following technical advantages:
1. the method establishes the flutter online identification system with high identification rate and high real-time property by establishing a gradient lifting tree model, extracting the multi-scale arrangement entropy and the multi-scale power spectrum entropy of the vibration signal after removing harmonic frequency as characteristic vector input and performing iterative update on the gradient lifting tree model by utilizing a logarithmic loss function and a related update mode.
2. The method can avoid manually adjusting related flutter identification algorithm parameters under the conditions of variable processing scenes and variable processing parameters, can realize extraction of the flutter sensitivity characteristics of the machine tool and identification of whether the numerical control processing center flutters and the flutter severity, has the advantages of high monitoring real-time performance, high identification accuracy, good generalization capability and the like, and can accurately, quickly and real-timely identify whether the numerical control processing center flutters and the flutter severity.
3. The method can eliminate harmonic frequency components in the vibration signal firstly, and residual components of the harmonic frequency components are noise signals and non-periodic component signals, so that the method is favorable for improving the ratio of signal components related to flutter, and then extracts multi-scale arrangement entropy and multi-scale power spectrum entropy as flutter sensitivity characteristics, and the gradient promotes the good generalization capability and classification capability of the tree model, so that the flutter state can be identified with high accuracy.
4. The method collects the vibration signals of the small and medium-sized vertical numerical control machining center in the actual machining process during variable cutting depth machining, and after the vibration signals are processed, the result shows that the method can effectively identify the chatter vibration of the numerical control machining center. In addition, the real-time performance of the actual processing and identification process is high, and online flutter identification can be realized, so that the machining state of the machine tool can be effectively monitored, and the selection and adjustment of machining parameters can be guided.
Drawings
FIG. 1 is a flow chart of a method for identifying chatter vibration of a CNC machining center on line based on vibration signals according to an embodiment of the present invention;
FIG. 2 is a flow chart of main shaft vibration signal acquisition, i.e., preprocessing, in a vibration signal-based numerical control machining center chatter online identification method according to an embodiment of the present invention;
fig. 3 (a) is a time-domain waveform diagram of the vibration signal when the nc machining center is stable, slightly flutter, and severely flutter, and fig. 3 (b) is a fast fourier transform time-frequency diagram of the vibration signal when the nc machining center is stable, slightly flutter, and severely flutter;
fig. 4 (a), (b), and (c) are a time domain amplitude map, an FFT amplitude map, and an FFT amplitude map of the signal segment S1 of fig. 3 (a), respectively, and an FFT amplitude map of the signal from which the harmonic component is removed;
fig. 5 (a), (b), and (c) are a time domain amplitude map, an FFT amplitude map, and an FFT amplitude map of the signal segment S3 of fig. 3 (a), respectively;
fig. 6 (a), (b), and (c) are a time domain amplitude map, an FFT amplitude map, and an FFT amplitude map of the signal segment S4 of fig. 3 (a), respectively, and an FFT amplitude map of the signal from which the harmonic component is removed;
the surface topography of the machined workpiece when the numerically controlled machining center corresponding to the signal segments S1, S3 and S4 of (a) in fig. 3 respectively is in stable, slight flutter and severe flutter in (a), (b) and (c) in fig. 7 respectively;
FIG. 8 (a) is a graph showing the result of the flutter identification test using the gradient lifting tree according to the present invention, FIG. 8 (b) is a graph showing the result of the identification test using the decision tree, and FIG. 8 (c) is a graph showing the result of the identification test using the support vector machine;
fig. 9 (a) and (b) are a time domain waveform diagram and a chattering recognition result diagram of the spindle vibration signal, respectively.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
As shown in figures 1 and 2, the invention discloses a chatter online identification method for a numerical control machining center based on vibration signals, which comprises the steps of collecting three-way vibration signals of a main shaft position in a machining process, then removing harmonic frequency components in the vibration signals by using an angle synchronous averaging method, then extracting chatter sensitive features by using a multi-scale arrangement entropy and a multi-scale power spectrum entropy extraction algorithm respectively, and finally inputting feature vectors into a gradient lifting tree model for chatter identification. The method can eliminate harmonic frequency components in the vibration signal, and residual components of the harmonic frequency components are noise signals and non-periodic component signals, so that the method is favorable for improving the ratio of signal components related to flutter, and then extracts multi-scale arrangement entropy and multi-scale power spectrum entropy as flutter sensitivity characteristics, and the gradient promotes the good generalization capability and classification capability of the tree model, so that the flutter state can be identified with high accuracy. In order to verify the method, the vibration signal generated when the small and medium-sized vertical numerical control machining center performs variable cutting depth machining in the actual machining process is collected, and after the vibration signal is processed, the result shows that the method can effectively perform flutter identification on the numerical control machining center. In addition, the method has high real-time performance in the actual processing and identification process, and can realize on-line flutter identification, thereby effectively monitoring the machine tool machining state and guiding the selection and adjustment of machining parameters.
As shown in fig. 3, 4, 5 and 6, the method establishes a gradient lifting tree model, extracts the multi-scale arrangement entropy and the multi-scale power spectrum entropy of the vibration signal without harmonic frequency as feature input, and iteratively updates the gradient lifting tree model by using a logarithmic loss function and a related updating mode, thereby establishing the flutter online identification system with high identification rate and high real-time property.
The method specifically comprises the following steps:
(1) acquiring a signal sample set:
a vibration sensor is arranged on a main shaft of a numerical control machining center, an acceleration sensor is used for collecting three-way vibration signals at the position of the main shaft in the machining process of the numerical control machining center, then the signals in the machining process are intercepted according to the machining stage, then the three-way vibration signals are subjected to sliding framing into multiple sections of signals according to the analysis requirement to be segmented and normalized, and if the signals with the window length of 25.6ms are used as one section of signals. And then, corresponding each section of signal to a processing process according to the actual processing condition, wherein each section of processing process has a corresponding flutter degree as the flutter category of the section of signal, so as to obtain the label of the section of signal, such as stable cutting, slight flutter and severe flutter.
(2) Signal preprocessing and feature extraction:
in order to extract the characteristics which are more sensitive to the vibration of the machine tool and enhance the generalization and the accuracy of the identification method, the related preprocessing and characteristic extraction work needs to be carried out on the main shaft vibration signals after equal-length segmentation and regularization. In order to facilitate inputting of each data into the gradient lifting tree model for training, harmonic frequency components in the spindle vibration signal after equal-length segmentation and regularization are removed to obtain a residual component signal, and then a flutter sensitivity characteristic of the residual component signal is extracted. Firstly, carrying out harmonic frequency component elimination on a main shaft vibration signal subjected to equal length segmentation and regularization processing on a first section to obtain a residual component signal of the signal; and then extracting multi-scale arrangement entropy and multi-scale power spectrum entropy from the residual component signals to be used as flutter sensitivity characteristics to be output, performing harmonic frequency elimination and flutter sensitivity characteristic extraction on the second section of the main shaft vibration signals subjected to equal length segmentation and normalization, and performing the operations on all the main shaft vibration signals subjected to equal length segmentation and normalization by analogy to obtain the characteristics of all the sample sets.
Specifically, harmonic frequency component elimination and flutter sensitivity characteristic extraction are carried out in the following modes:
and (2.1) calculating by using an angle synchronous averaging method to obtain harmonic frequency components in each section of the main shaft vibration signal after equal length segmentation and regularization, then removing the harmonic frequency components from the original signal section to obtain a residual component signal, and taking the residual component signal as the input of a multi-scale arrangement entropy and multi-scale power spectrum entropy extraction algorithm.
(2.2) after obtaining the residual component signal of each section of signal, firstly carrying out coarse graining treatment on the residual component signal according to a plurality of selected scale factors to obtain a series of coarse grained signals corresponding to the residual component signal of the section; and then respectively calculating permutation entropies under different scales according to a permutation entropy calculation method, finally splicing the permutation entropies under different scales, and outputting a multi-scale permutation entropy feature vector.
(2.3) calculating a power spectrum for each coarsely granulated signal, calculating power spectral entropies under different scales by using a power spectral entropy calculation method, finally splicing the power spectral entropies under different scales, and outputting a multi-scale power spectral entropy feature vector; and finally, splicing the multi-scale arrangement entropy and the multi-scale power spectrum entropy to obtain a feature vector under the scale, and taking the feature vector as a sample as an input feature of a subsequent identification algorithm.
(2.4) repeating steps (2.1), (2.2) and (2.3) for all signal segments, thereby extracting the feature vectors of all signal segments as a set, which is used as an input for obtaining a subsequent recognition algorithm.
Wherein the harmonic frequency component of each signal segment is represented by Sp(θ) and is constructed using the formula:
wherein u is { u ═ u { (R) }1,u2,…,uNThe preprocessed signal fragment is obtained; k is the number of the divided signal segments, K is more than or equal to 0 and less than or equal to K-1, and K is a subscript of the kth segment signal; theta is the rotation period of the main shaft.
The multi-scale permutation entropy calculation formula is represented by MPE, the multi-scale power spectrum entropy is represented by MPSE, and the calculation is carried out by adopting the following formula:
where s is a scale factor, m is an embedding dimension when reconstructing the coarsely-grained signal,
is the probability that the sequence l occurs, and NlThe reconstructed signals are sorted in ascending order, the number of times sequence l appears,
n is the length of the residual component signal, py(wi) Is a certain rulerCoarse grained signal x at the scale factoriThe corresponding normalized power spectrum.
(3) Pre-training a flutter identification algorithm:
as shown in fig. 7, after obtaining the label information of the flutter degree of the signal sample and the feature vector of the signal, the flutter identification algorithm (i.e. gradient lifting tree algorithm) needs to be pre-trained to achieve high-accuracy identification of flutter. Before training begins, the category of the flutter degree is defined as C, such as stable cutting, slight flutter and severe flutter, and the loss function type of the gradient lifting tree algorithm is designated as a logarithmic loss function. And then, carrying out iterative updating on each sample aiming at each decision tree, stopping algorithm training when the preset requirements are finally met, and outputting a pre-training model.
The specific training process is as follows:
(3.1) input dataset D { (x)1,y1),(x2,y2),…,(xN,yN) Determining a loss function, and initializing a gradient lifting tree model algorithm according to the loss function;
(3.2) for each sample, iteratively calculating a residual error by using a negative gradient error function, and forming a new data set according to the data set input of the corresponding sample and the calculated residual error:
{(x1,rm,1),(x2,rm,2),…,(xN,rm,N)}
training an m-th regression tree, wherein leaf nodes of the regression tree can be divided into a block of region in a sample set;
(3.3) calculating a leaf node output value according to a leaf node output function aiming at each leaf node of the regression tree and carrying out one-round updating on the regression tree;
(3.4) circulating the steps (3.2) and (3.3) to finish the 1 st round of training and iteration of the C-total regression tree;
and (3.5) carrying out iterative training in the steps (3.2), (3.3) and (3.4) circularly until the preset iteration times (such as 1000 rounds) or the preset error requirement is met, stopping the model training and finally outputting the pre-trained gradient lifting tree model.
Wherein, it is toThe number loss function is denoted by L (y, f (x)), and the negative gradient error function by rm,i,lRepresenting the leaf node output function by om,i,lExpressed, calculated using the following equations, respectively:
rm,i,l＝yi,l-pm,l(xi)
wherein R ism,l,jThe jth region is divided for the mth round and corresponding to the leaf node of the gradient lifting tree model with the class l, and I (x belongs to Rm,l,j) To indicate a function, when x belongs to Rm,l,jWhen it is 1, otherwise it is 0, om,i,lIs the ith sample x in the mth iteration processiLeaf node output, y, for a gradient lifting tree model of class lcIs the class value of sample x, if ycSample x belongs to class c, p, 1c(x) The probability that x belongs to the class C is judged for a gradient lifting tree model f (x), C is the number of the class of the flutter of the numerical control machining center, C is more than or equal to 0 and less than or equal to C-1, rm,i,lFor the m-th, i-th sample xiFor negative gradient errors of class l, each sample xiIs composed of feature vectors, yi,lFor the ith sample xiTrue probability, p, for class lm,l(xi) For the ith sample xiPredicting the probability of class L in the mth iteration process, wherein L (y, f (x)) is a logarithmic loss function, and fm(x) Is the m-th iteration, updatedGradient lifting tree model, fm-1(x) And (5) iterating the m-1 th round, and updating the updated gradient lifting tree model.
The method of the present invention will be described below by taking as an example a milling process performed in a small vertical nc machining center with a variable cutting depth.
As shown in FIG. 1, the embodiment of the present invention has the following steps:
(1) the method comprises the steps of arranging a PCB three-way acceleration sensor on a main shaft seat of a numerical control machining center, collecting three-way vibration signal data of a main shaft part, segmenting and organizing the signals according to preset segmentation parameters, taking each segmented signal as a signal segment sample, marking the sample flutter category according to whether flutter occurs or the severity of the flutter occurs in the actual machining process, taking the category as the basis of a follow-up flutter identification algorithm during training, and taking each signal segment sample as a sample for follow-up data preprocessing and feature extraction. The amplitude of the vibration signal during stable cutting, slight vibration and severe vibration is shown in fig. 3 and 7, and it can be seen from the amplitude of the vibration signal that the amplitude of the vibration signal of the spindle does not change much when vibration occurs, and the vibration is difficult to identify from the amplitude of the vibration signal of the spindle. Fig. 4, 5 and 6 show fast fourier transform time-frequency diagrams of the spindle vibration signal, from which it can be seen that the frequency domain characteristics of the signal at different times do not change much.
(2) And (3) removing harmonic frequency components in each signal segment sample by using an angle synchronous averaging method to obtain a residual component signal, and taking the residual component signal as the input of a subsequent multi-scale entropy feature extraction algorithm. Wherein the harmonic frequency component is represented by Sp(θ) and is constructed using the formula:
wherein u is { u ═ u { (R) }1,u2,…,uNThe preprocessed signal fragment is obtained; k is the number of the divided signal segments, and K is more than or equal to 0 and less than or equal to K-1; theta is the rotation period of the main shaft.
Fig. 4, 5, and 6 show time domain diagrams, frequency domain diagrams of signals without removing harmonic components, and frequency domain diagrams of residual component signals with removing harmonic components of the signal segments S1, S3, and S4, respectively. Comparing the frequency domain diagram of the residual component signal after removing the harmonic frequency components with the frequency domain diagram of the signal without removing the harmonic frequency components, the angular synchronous averaging method effectively removes the harmonic frequency components in the signal.
(3) After the residual component signals are obtained, the residual component signals are subjected to coarse graining according to the selected scale factors, and then multi-scale arrangement entropy and multi-scale power spectrum entropy are extracted to be used as feature vectors to be output. The calculation formula of the multi-scale permutation entropy MPE and the multi-scale power spectrum entropy MPSE is as follows:
wherein s is a scale factor; m is an embedding dimension when reconstructing the coarsely granulated signal;
is the probability that the sequence l occurs, and NlThe reconstructed signals are sorted in ascending order, the number of times sequence l appears,
n is the length of the residual component signal; p is a radical ofy(wi) Is the coarse grained signal x at a certain scale factoriThe corresponding normalized power spectrum.
Notably, the scale factors selected to calculate the multi-scale permutation entropy and the multi-scale power spectral entropy may be different.
(4) After the feature vectors of all samples are obtained, the feature vectors are used as the input of the gradient lifting tree model with the determined parameters such as the loss function type, the category and the like, and the gradient lifting tree model can be subjected to iterative training by using the flutter degree label corresponding to the samples. In each iteration, one sample is sequentially input, and each regression tree forming the gradient lifting tree model is subjected to cyclic iterative updating until all samples are traversed. And then finishing the specified iteration times, or stopping updating when the error reaches a preset error range, and finishing the pre-training of the gradient lifting tree model.
According to the trained optimal gradient lifting tree model, framing, detuning and flutter sensitivity characteristic extraction are carried out on a section of main shaft vibration signal in a new machining process, and then the main shaft vibration signal is input into the trained gradient lifting tree model, so that whether flutter occurs in the machining process corresponding to the section of signal or the severity of flutter after the flutter occurs can be identified. In the identification process, due to the generalization and stability of the model, the identification accuracy of the model is high, and the comparison result with other methods is shown in fig. 5, wherein the number of test samples is 30, by using the method provided by the present invention, the identification results of 30 samples are all correct, and other methods are not completely correct. The vibration time domain signal in (a) of fig. 9 is identified, and the identification result is shown in (b) of fig. 9, where only a part of the signal segment identification result is in error, and not a plurality of consecutive samples are in error, so as to meet the actual identification requirement. Therefore, the method provided by the invention can identify whether the machining center generates the flutter or the severity of the flutter, so as to guide the adjustment of the machining parameters, avoid the problems of surface quality reduction of products, accelerated tool abrasion and the like.
Specifically, in the variable cutting depth milling scene, the adopted end mill can obtain the corresponding flutter degree by giving a vibration signal through model training and identification according to the processing technology requirement of the part and the performance of the milling cutter.
Fig. 3, 4, 5 and 6 are surface topography graphs of the workpiece under cutting conditions S1, S3 and S4, respectively, of the cutting depth and the spindle rotation speed as shown in table 1, when milling is performed under the variable cutting depth parameter studied by the present invention, from which it can be found that there are great differences between the surface topography of the workpiece under different parameters, which correspond to different degrees of chattering of the machine tool. As can be seen from the figure, when the surface topography of the workpiece is as shown in fig. 4 (a), the surface topography is regular and flat, the machine tool machining is in a stable cutting state, when the cutting parameters are changed, the surface topography of the workpiece is as shown in fig. 7 (b), irregular streaks begin to appear and high ridges appear in a partial region, resulting in surface unevenness, and the machine tool is in a slight chattering state at this time, and when the cutting parameters are changed to S4 cutting conditions in table 1, the surface topography of the workpiece is as shown in fig. 7 (c), the surface of which has a large number of irregular streaks and high ridges and deep furrows are spread over the entire surface of the workpiece, and the machine tool is in a severe chattering state at this time. At the moment, the machine tool must give an alarm for prompting the situation, intervene and adjust the machining process of the machine tool, and avoid the situations of serious abrasion and damage of the cutter, even fault of machine tool parts and the like.
TABLE 1
In conclusion, the method provided by the invention has the advantages that harmonic frequency components of the main shaft vibration signal in the machining process of the numerical control machining center are removed, the flutter sensitivity characteristics are extracted, the main shaft vibration signal is input into the gradient lifting tree model for training, and after the flutter state is identified, the method has great significance for the research of the targeted numerical control machining center flutter online identification and machine tool state monitoring in terms of the accuracy of flutter identification and the stability and the high efficiency in the application aspect.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (8)
1. A vibration signal-based numerical control machining center flutter online identification method is characterized by comprising the following steps:
s1, collecting a main shaft vibration signal when the numerical control machining center performs cutting machining, and performing equal-length segmentation and normalization on the main shaft vibration signal according to the main shaft rotation period;
s2, eliminating harmonic frequency signals in the spindle vibration signals after equal length segmentation and regularization processing to obtain residual component signals;
s3, calculating the multi-scale permutation entropy of the residual component signals by adopting a multi-scale permutation entropy characteristic algorithm, calculating the multi-scale power spectrum entropy of the residual component signals by adopting a multi-scale power spectrum entropy characteristic algorithm, and splicing the multi-scale permutation entropy and the multi-scale power spectrum entropy to serve as feature vectors;
s4, constructing a gradient lifting tree model, inputting the feature vector into the gradient lifting tree model, performing iterative training on the gradient lifting tree model to enable the loss value of the gradient lifting tree model to reach a preset value, and ending the iterative training on the gradient lifting tree model to obtain an optimal gradient lifting tree model;
s5, the spindle vibration signal in the machining process is processed in steps S1 to S3 and then input into the optimal gradient lifting tree model, and therefore online identification of chatter of the numerical control machining center is achieved.
2. The method for on-line identification of chatter of a numerical control machining center based on vibration signals as claimed in claim 1, wherein the step S1 specifically comprises the steps of: installing a vibration sensor on a main shaft of a numerical control machining center, acquiring a three-way vibration signal of the main shaft, intercepting the main shaft vibration signal in the machining process from the vibration signal according to the rotation period of the main shaft, performing sliding framing on the main shaft vibration signal, and dividing the main shaft vibration signal into a plurality of equilong signal sections, thereby completing equilong division and normalization of the vibration signal; each signal segment has a class value and a spindle rotation period corresponding thereto.
3. The on-line identification method for the chatter of the numerical control machining center based on the vibration signals as claimed in claim 1, wherein in step S2, an angle synchronous averaging algorithm is used to remove harmonic signals in the spindle vibration signals after equal length segmentation and regularization.
4. The method for on-line identification of chatter of a numerical control machining center based on vibration signals as claimed in claim 1, wherein the step S3 specifically comprises the steps of:
s31, aiming at the selected M scale factors, carrying out coarse graining on the residual component signals according to the scale factors to obtain M coarse graining signals;
s32, reconstructing the coarse grained signals under a certain scale factor to obtain a reconstruction matrix under the scale factor, and then using each row vector in the reconstruction matrix as an original sequence to perform ascending arrangement on the original sequence;
s33, comparing the sequence after the ascending sequence with the original sequence to obtain an index sequence after the ascending sequence of the elements of the vector corresponding to the original sequence;
s34, counting the times of the same index sequence in the reconstruction matrix, and calculating the permutation entropy of the residual component signals under the scale factor by adopting a multi-scale permutation entropy characteristic algorithm;
s35, calculating the power spectrum entropy of the residual component signals under a certain scale factor by adopting a multi-scale power spectrum entropy characteristic algorithm for the coarse grained signals under the scale factor;
s36 repeating the steps S32 to S35 to obtain multi-scale arrangement entropy and multi-scale power spectrum entropy of the residual component signals under M scale factors;
s37, splicing the multi-scale arrangement entropy and the multi-scale power spectrum entropy to be used as the feature vector.
5. The vibration signal-based chattering online identification method for the numerical control machining center according to claim 4, wherein the model of the multi-scale permutation entropy characteristic algorithm is as follows:
the model of the multi-scale power spectrum entropy characteristic algorithm is as follows:
wherein s is a scale factor when calculating the multi-scale permutation entropy, m is an embedding dimension when reconstructing the residual component signal after coarse graining, and N islThe number of times the original sequence l of the coarsely grained signal appears,
n is the length of the residual component signal, py(wi) Is a coarse grained signal x at a certain scale factoriThe corresponding normalized power spectrum.
6. The method for on-line identification of chatter of a numerical control machining center based on vibration signals as claimed in claim 1, wherein in step S4, the gradient lifting tree model is:
rm，i，l＝yi，l-pm，l(xi)
wherein R ism，l，jThe jth region is divided for the mth round and corresponding to the leaf node of the gradient lifting tree model with the class l, and I (x belongs to Rm，l，j) To indicate a function, when x belongs to Rm，l，jWhen it is 1, otherwise it is 0, Om，i，lIs the ith sample x in the mth iteration processiLeaf node output, y, for a gradient lifting tree model of class lcIs the class value of sample x, if ycSample x belongs to class c, p, 1c(x) The probability that x belongs to the class C is judged for a gradient lifting tree model f (x), C is the number of the class of the flutter of the numerical control machining center, C is more than or equal to 0 and less than or equal to C-1, rm，i，lFor the m-th, i-th sample xiFor negative gradient errors of class l, each sample xiIs composed of feature vectors, yi，lFor the ith sample xiTrue probability, p, for class lm，l(xi) For the ith sample xiThe probability of class i is predicted during the mth iteration, and L (y, f (x)) is a logarithmic loss function.
7. The method for on-line identification of chatter of a numerical control machining center based on vibration signals as claimed in claim 6, wherein the step S4 specifically comprises the steps of:
s41 inputting feature vectors into the gradient lifting tree model;
s42 calculating loss value according to log loss function L (y, f (x)) in the gradient lifting tree model, and then adopting negative gradient error function r according to the loss valuem，i，lCalculating a negative gradient error;
s43 inputs the negative gradient error to the leaf node output function om，i，lUpdating decision tree leaf node output, and updating the gradient lifting tree model according to leaf node output;
and S44, the steps are circulated to carry out iterative updating until the loss value calculated by the loss function L (y, f (x)) reaches a preset value, and the training of the gradient lifting tree model is finished to obtain the optimal gradient lifting tree model.
8. The method for on-line identification of chatter of a numerical control machining center based on vibration signals as claimed in any one of claims 1 to 7,
harmonic frequency component of each signal segment is Sp(θ) and is constructed using the formula:
wherein u ═ { u ═1，u2，…，uNThe spindle vibration signal is a spindle vibration signal fragment subjected to equal-length segmentation and regularization; k is the number of signal sections divided by the main shaft vibration signal, K is more than or equal to 0 and less than or equal to K-1, and K is a subscript of a kth section signal; theta is the rotation period of the main shaft.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911327809.4A CN110928237B (en) | 2019-12-20 | 2019-12-20 | Vibration signal-based numerical control machining center flutter online identification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911327809.4A CN110928237B (en) | 2019-12-20 | 2019-12-20 | Vibration signal-based numerical control machining center flutter online identification method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110928237A CN110928237A (en) | 2020-03-27 |
| CN110928237B true CN110928237B (en) | 2020-12-29 |
Family
ID=69863496
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911327809.4A Active CN110928237B (en) | 2019-12-20 | 2019-12-20 | Vibration signal-based numerical control machining center flutter online identification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110928237B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112394642B (en) * | 2020-11-27 | 2022-09-20 | 华中科技大学 | Robot milling chatter identification method based on super-parameter optimization |
| CN112529099A (en) * | 2020-12-24 | 2021-03-19 | 华中科技大学 | Robot milling chatter identification method |
| CN113313198B (en) * | 2021-06-21 | 2022-08-19 | 西北工业大学 | Cutter wear prediction method based on multi-scale convolution neural network |
| CN114800042B (en) * | 2022-04-28 | 2023-05-05 | 华中科技大学 | Robot milling vibration type identification method based on power spectrum entropy difference |
| CN114800041B (en) * | 2022-05-06 | 2023-05-09 | 嘉兴南湖学院 | Cutter state monitoring method and monitoring device thereof |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104849050A (en) * | 2015-06-02 | 2015-08-19 | 安徽工业大学 | Rolling bearing fault diagnosis method based on composite multi-scale permutation entropies |
| CN110110680A (en) * | 2019-05-14 | 2019-08-09 | 中国人民解放军陆军装甲兵学院 | Planetary transmission fault-signal classification method based on DWT-LPP |
| CN110333078A (en) * | 2019-08-21 | 2019-10-15 | 佛山科学技术学院 | A kind of rolling bearing degenerate state stage determines method |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9329836B2 (en) * | 2014-04-17 | 2016-05-03 | International Business Machines Corporation | Extracting entropy from the vibration of multiple machines |
| TW201633025A (en) * | 2015-03-10 | 2016-09-16 | Anderson Ind Corp | Diagnostic method for malfunction mode of machine tool main shaft and system thereof |
| CN105436981B (en) * | 2015-09-28 | 2018-02-16 | 上海诺倬力机电科技有限公司 | Flutter Control method and numerical control processing apparatus based on vibration monitoring |
| CN109297713B (en) * | 2018-08-07 | 2019-12-31 | 浙江大学 | Steam turbine fault diagnosis method based on stable and non-stable vibration signal characteristic selection |
| CN109489800A (en) * | 2018-12-14 | 2019-03-19 | 广东世港信息科技有限公司 | A kind of disturbance event recognition methods in distribution optic cable vibration safety pre-warning system |
| CN109858352B (en) * | 2018-12-26 | 2020-09-18 | 华中科技大学 | Fault diagnosis method based on compressed sensing and improved multi-scale network |
-
2019
- 2019-12-20 CN CN201911327809.4A patent/CN110928237B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104849050A (en) * | 2015-06-02 | 2015-08-19 | 安徽工业大学 | Rolling bearing fault diagnosis method based on composite multi-scale permutation entropies |
| CN110110680A (en) * | 2019-05-14 | 2019-08-09 | 中国人民解放军陆军装甲兵学院 | Planetary transmission fault-signal classification method based on DWT-LPP |
| CN110333078A (en) * | 2019-08-21 | 2019-10-15 | 佛山科学技术学院 | A kind of rolling bearing degenerate state stage determines method |
Non-Patent Citations (2)
| Title |
|---|
| Mao Ge et al..An Effective Bearing Fault Diagnosis Technique via Local Robust Principal Component Analysis and Multi-Scale Permutation Entropy.《Entropy》.2019, * |
| 基于多尺度排列熵的滚动轴承故障诊断;陈长征 等;《机械工程师》;20190910;第17-22页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110928237A (en) | 2020-03-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110928237B (en) | Vibration signal-based numerical control machining center flutter online identification method | |
| US9508042B2 (en) | Method for predicting machining quality of machine tool | |
| CN110647943B (en) | Cutting tool wear monitoring method based on evolution data cluster analysis | |
| CN112508105B (en) | Fault detection and retrieval method for oil extraction machine | |
| WO2016155241A1 (en) | Method, system and computer device for capacity prediction based on kalman filter | |
| CN111898443B (en) | Flow monitoring method for wire feeding mechanism of FDM type 3D printer | |
| CN114619292B (en) | Milling cutter wear monitoring method based on fusion of wavelet denoising and attention mechanism with GRU network | |
| CN110561191A (en) | Numerical control machine tool cutter abrasion data processing method based on PCA and self-encoder | |
| CN112207631B (en) | Method for generating tool detection model, method, system, device and medium for detecting tool detection model | |
| CN110682159A (en) | Cutter wear state identification method and device | |
| CN112528955B (en) | High-frequency element machining size precision prediction method and system | |
| CN110716500A (en) | Method and system for determining segmented modeling points of temperature sensitive interval | |
| CN114429152A (en) | Rolling bearing fault diagnosis method based on dynamic index antagonism self-adaption | |
| CN110348540B (en) | Clustering-based method and device for screening transient power angle stability faults of power system | |
| WO2022132898A1 (en) | Monitoring system for estimating useful life of a machine component | |
| CN115659665A (en) | Voltage sag source signal noise reduction and identification method based on Trojan horse optimization algorithm | |
| CN111144440A (en) | Method and device for analyzing daily power load characteristics of special transformer user | |
| CN114925728A (en) | Rolling bearing fault diagnosis method, rolling bearing fault diagnosis device, electronic device and storage medium | |
| CN115922553A (en) | Method for online monitoring polishing processing state of silicon carbide wafer | |
| CN110930012B (en) | Energy consumption abnormity positioning method based on sensitivity analysis and improved negative selection method | |
| CN113449254A (en) | Method for analyzing monitoring stability of arbitrary net type deformation and method for determining position of monitoring point | |
| CN115424635B (en) | Cement plant equipment fault diagnosis method based on sound characteristics | |
| CN114021275B (en) | Rolling bearing fault diagnosis method based on deep convolution fuzzy system | |
| CN116226693A (en) | Gaussian mixture model nuclear power operation condition division method based on density peak clustering | |
| Ding et al. | Machine tool fault classification diagnosis based on audio parameters |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |