Tool abnormal state monitoring method for machining process based on permutation spectrum matrix

Abstract

The application discloses a machining process tool abnormal state monitoring method based on a permutation spectrum matrix, and comprises the following steps: collecting state monitoring signals of parts in a cutting process, and dividing the state monitoring signals into a template selection data set and a test data set according to a certain proportion; obtaining all permutation patterns according to an embedding dimension parameter in a permutation pattern parameter combination; reconstructing monitoring signals of a given length into a phase space form through different permutation pattern parameter combinations; calculating corresponding permutation spectrum features according to different time delay parameters, and composing a permutation spectrum matrix; calculating the permutation spectrum matrix of all samples in the template selection data set, as a permutation spectrum matrix template set; selecting an un-identified sample from the test data set, calculating the permutation spectrum matrix, obtaining the matching degree of the sample with each sample in the permutation spectrum matrix template set, obtaining a template matching degree and a maximum template matching degree vector; and selecting the sample corresponding to the maximum template matching degree, and the processing state of the sample is the processing state of the un-identified sample.

Description

TECHNICAL FIELD

[0001] The present application relates to a machine tool abnormal state monitoring method, more particularly to a method for monitoring abnormal state of a tool in a cutting process by using multi-source sensing data in a machine tool processing process, combining permutation spectrum matrix and template matching strategy. BACKGROUND

[0002] The rapid development of data acquisition and artificial intelligence technology accelerates the manufacturing industry from automatic production to intelligent manufacturing mode. In the intelligent manufacturing mode, processing state monitoring is a key factor to improve the quality of numerical control machining. Abnormal processing state will lead to poor product surface quality, low machining precision and waste of raw materials. The tool is an important component of the numerical control machine tool, and the tool state monitoring directly affects the product machining quality, production efficiency and production cost. The tool monitoring system in intelligent manufacturing can effectively identify and distinguish the abnormal state of the tool by using the state signal of the processing process, and take measures such as tool replacement, parameter modification or shutdown according to the identification result to ensure the workpiece quality.

[0003] In the metal cutting process, due to the contact between the tool and the workpiece, tool wear and even damage phenomenon will inevitably occur, affecting the machining precision and workpiece quality. In addition, thin-walled parts are prone to deformation during machining, which in turn affects the size of cutting force, leading to tool wear and tilt and other problems. In recent years, tool wear, machining deformation, cutting chatter and other metal cutting problems have been widely concerned. However, most of the researches only focus on a single abnormal state, without considering the coupling phenomenon of multiple machining abnormalities, which does not conform to the actual machining process, especially the machining process of thin-walled parts.

[0004] The machining abnormality of the tool is manifested as a certain fixed data fluctuation pattern in the signal, and the fluctuation patterns of different abnormal states have great differences. Traditional tool abnormal monitoring methods mostly only focus on the global information of the signal, ignoring the detailed fluctuation pattern, while these detailed fluctuation patterns are the key to distinguish different abnormal machining states of the tool. Therefore, there is a need in the art for a tool monitoring system that can accurately distinguish various single and mixed abnormalities of parts in the cutting process. SUMMARY

[0005] To address the aforementioned problems, an embodiment of the present invention provides a method for monitoring abnormal tool states during machining processes based on permutation spectrum matrices, comprising the following steps: Step 1: Signal acquisition and dataset partitioning: Acquire state monitoring signals of parts during the cutting process and partition them into a template selection dataset and a test dataset according to a certain ratio; Step 2: Constructing permutation patterns and reconstructing phase space: Obtain all permutation patterns according to the embedding dimension parameters in the permutation pattern parameter combinations. For monitoring signals of a given length, reconstruct them into phase space form through different permutation pattern parameter combinations; Step 3: Generating permutation spectra and permutation spectrum matrices: Calculate the corresponding permutation spectrum features according to different time delay parameters and form a permutation spectrum matrix; Step 4: Generating template sets: Obtain the permutation spectrum matrices of all samples in the template selection dataset and use these matrices as the permutation spectrum matrix template set; Step 5: Obtaining template matching degree: Select unidentified samples from the test dataset, obtain the permutation spectrum matrix of the sample, and obtain its matching degree with each sample in the permutation spectrum matrix template set, obtaining the template matching degree and the maximum template matching degree vector; Step 6: Machining state identification: Based on the obtained template matching degree, select the sample corresponding to the maximum template matching degree. The machining state of this sample is the machining state of the unidentified sample.

[0006] In an optional implementation, the status monitoring signal mentioned in step one is a data signal of the part during the cutting process collected by the internal or external sensors of the CNC machine tool, including current, voltage, temperature, and vibration signals.

[0007] In an optional implementation, the division into template selection dataset and test dataset according to a certain ratio in step one includes dividing the collected monitoring signals under different processing states into two sets, template selection dataset and test dataset, according to the set sample allocation ratio z1:z2. The template selection dataset M is used to construct feature templates and calculate the matching degree, while the test dataset T consists of unidentified samples used for model verification.

[0008] First, represent all samples as the total dataset F = (F1, F2, ..., F...). u ), where u represents the number of states in the dataset, F u Let M represent the subset of data representing the u-th state. Assume each subset contains z samples, and z1 + z2 = z. Based on a set ratio z1:z2, each subclass is divided into two parts, forming the template selection dataset M = (M1, M2, ..., M...). u ) and the test dataset T = (T1, T2, ..., T u ),in:

[0009]

[0010]

[0011] In the formula, M u and T u These represent the template selection data subset and the test data subset corresponding to the u-th state data, respectively. This indicates that the template selects a subset of data M. u The z1th sample in the dataset, Represents a subset T of test data u The z2th sample in the data is z1+z2=z, where z is the total number of samples for the uth state.

[0012] In an optional implementation, the arrangement mode parameter combination mentioned in step two includes the parameters used when constructing the arrangement mode and reconstructing the phase space of the monitoring signal, which includes three parameters [m, τ, N], where m is the embedding dimension, τ is the time delay, and N is the data length. The value of the data length N is affected by the embedding dimension m and should satisfy N≥5m!. The range of parameter τ is determined according to parameter m and satisfies τ∈[1,m!].

[0013] In an optional implementation, the arrangement pattern described in step two includes determining the data fluctuation type based on the parameter embedding dimension m. The arrangement pattern consists of an m-dimensional vector containing positive integers [1, 2, ..., m], where each number appears only once. There are m! arrangements of m positive integers, and each arrangement represents an arrangement pattern, denoted as Π. m (k), where k represents the index of the permutation pattern, k = 1, 2, ..., m!, and the set of all permutations is represented by Π. m ;

[0014] The reconstruction into phase space form described in step two includes reconstructing a state monitoring signal of length N into multiple sub-sequences based on the parameter embedding dimension m and the parameter time delay τ, and converting them into corresponding permutation patterns. This further includes the following steps:

[0015] For a given state monitoring signal X = {x1, x2, ..., xn} of length N, N}, embed X into the m-dimensional original phase space S m S m It contains n subsequences

[0016]

[0017] Let m represent the i-th subsequence, where each subsequence has a length of m.

[0018] Calculate the number of subsequences n based on the parameter combination [m,τ,N):

[0019] n = N - (m - 1)τ

[0020] The elements in the sub-sequence are arranged in ascending order, and converted into a vector containing [1, 2, …, m] and each positive integer appears only once according to the order relationship of the elements The arrangement mode is called the sub-sequence r j represents a positive integer from 1 to m, j = 1, 2, …, m;

[0021] When there are elements with assigned equal values in the sub-sequence , that is To ensure the uniqueness of the result, let r l r l+1 , l represents the position corresponding to a certain value point in the sub-sequence, l = 1, 2, …, m-1;

[0022] After all the sub-sequences in the original phase space S m are converted into corresponding arrangement modes, the corresponding symbolic phase space π m is obtained, so as to convert the state monitoring signal into the form of symbolic phase space based on the arrangement mode, so as to count the number of arrangement modes.

[0023] In an optional embodiment, the arrangement spectrum feature in step three is a feature vector based on the number of arrangement modes in the symbolic phase space Π m There are m! arrangement modes in the arrangement mode set Π m , each arrangement mode is denoted as Π m (k), k = 1, 2, …, m!, the number of each arrangement mode is counted and denoted as The probability of each arrangement mode appearing is Where n is the number of sub-sequences, and the probability distribution of the arrangement mode of the state monitoring signal X with length N is represented as:

[0024] The probability distribution of the arrangement mode of the signal X is weighted using the amplitude characteristic, and the arrangement spectrum feature

[0025]

[0026] In the formula, represents the arrangement spectrum feature of the signal X, represents the arrangement spectrum feature value corresponding to the kth arrangement mode Π m (k), k = 1, 2, …, m!, and the w represents the amplitude feature of the signal, which is represented by the variance of the signal:

[0027]

[0028] wherein, represents the mean value of all data points in the signal, x q represents the data points in the signal, q = 1, 2, …, N, which represents the total number of data points in the signal. Arrangement spectrum characteristics At the same time, the amplitude information and the sequential structure unit information of the signal are considered, so that the overall and detailed characteristics of the signal can be more fully reflected.

[0029] The arrangement spectrum matrix described in step three is a matrix composed of multiple arrangement spectrum characteristics obtained by using the same parameter embedding dimension m and different parameter time delay τ. When the embedding dimension parameter m is determined, τ ∈ [1, m!], so that m! groups of arrangement spectrum characteristics are obtained, which are denoted as arrangement spectrum matrix P:

[0030]

[0031] wherein, represents the arrangement spectrum characteristics of the signal X at the time delay parameter τ = h, h = 1, 2, …, m!.

[0032] In an optional embodiment, the arrangement spectrum matrix template set described in step four includes a set PM composed of arrangement spectrum matrices calculated by using all samples in the template selection data set M = (M1, M2, …, M u According to the data division method of step one, the template selection data set M contains u kinds of state data, each kind of data contains z1 samples, and a total of u × z1 samples are contained. The arrangement spectrum matrix P of each sample is calculated as a template, and a arrangement spectrum matrix template set composed of u × z1 arrangement spectrum matrices is obtained represents the arrangement spectrum matrix corresponding to the i-th sample of the j-th state data in the template selection data set M.

[0033] In an optional embodiment, the unidentified sample described in step five is a sample in the test data set T = (T1, T2, …, T u When the state is identified, an arbitrary signal sample is selected from the test data set T as an unidentified sample, denoted as t, and it is considered that the processing state of the sample is unknown. The arrangement spectrum matrix of the sample is denoted as P t According to the data division method of step one, the test data set T contains u kinds of state data, each kind of data contains z2 samples, and the test data set T contains a total of u × z2 samples unidentified samples;

[0034] The template matching degree vector described in step five is a vector calculated by using the Euclidean distance between the row vectors of the matrix to calculate the arrangement spectrum matrix P tThe matching degree of the permutation spectrum matrix of each sample in the permutation spectrum matrix template set PM The matching degree of the permutation spectrum matrix of each sample in the permutation spectrum matrix template set PM The permutation spectrum matrix P of the un-identified sample t t The matching degree of the permutation spectrum matrix of the i-th sample corresponding to the j-th state data in the template selection data set M The higher the value, the higher the similarity between the two matrices, and the formula for calculating the matching degree is as follows:

[0035]

[0036]

[0037] In the formula, A and B represent two matrices with the same dimension, in the present application, B represents the permutation spectrum matrix P corresponding to the un-identified sample t t A represents the permutation spectrum matrix corresponding to the i-th sample of the j-th state data in the template selection data set M a v,w and b v,w respectively represent the element values of the v-th row and the w-th column in the matrices A and B, v=1, 2, …, m!, w=1, 2, …, m!, and m represents the embedding dimension.

[0038] By the method provided by the embodiment of the present application, the state monitoring signal collected by the partial cutting process can be used to establish a numerical control machine tool cutting process abnormal state monitoring model. The proposed permutation spectrum matrix operation is simple and can reflect the underlying data fluctuations of the signal under different sampling intervals and contain as much state information as possible. The template matching strategy method can timely and accurately determine the processing state to which the current un-identified sample belongs, realize the abnormal state recognition of the numerical control machine tool processing process, and provide technical guidance for the on-site staff to implement tool replacement, shutdown maintenance and other decisions, so as to ensure the processing quality of the product.

[0039] According to the embodiment of the present application, at least the following advantages are included. The present application adopts a machining process tool abnormal state monitoring method based on a permutation spectrum matrix. The permutation spectrum matrix contains permutation pattern distribution of signals under multiple sampling intervals, can express fluctuation patterns of signals under different states in detail, and can accurately identify the machining state of an unidentified sample under a small sample condition. The permutation spectrum matrix is a result of permutation spectrum characteristics corresponding to multiple parameter combinations, so that parameter optimization is not needed to select the best combination, a large amount of operation cost is saved, the calculation method is simple and the calculation time is short. Since the permutation pattern pays more attention to the order relationship between adjacent data values rather than the numerical values themselves, the dependence on the amplitude threshold value can be avoided, and the noise resistance is strong. At the same time, since the parameter optimization link is omitted, the model has stronger migration ability. The method is a guiding numerical control machine tool cutting process abnormal state monitoring method, has the characteristics of simple calculation, strong migration, good noise resistance, etc. The method has certain openness in practical application, and is suitable for real-time state monitoring and post-fault detection of product machining process or mechanical equipment in operation. BRIEF DESCRIPTION OF DRAWINGS

[0040] The foregoing features of the present application can be more readily understood from the following detailed description taken in connection with the accompanying drawings, in which:

[0041] Figure 1 A flow chart of a machining process tool abnormal state monitoring method based on a permutation spectrum matrix according to an embodiment of the present application is shown;

[0042] Figure 2 A flow chart of a permutation spectrum matrix construction process in a machining process tool abnormal state monitoring method based on a permutation spectrum matrix according to an embodiment of the present application is shown;

[0043] Figure 3 Time domain graphs of vibration signals of four machining states of a certain thin-walled cylindrical part cutting process in an example applying an embodiment of the present application and corresponding permutation spectrum graphs (wherein m=4, τ=1, 10) are shown;

[0044] Figure 4 Gray scale graphs of permutation spectrum matrices of vibration signals of four machining states of a certain thin-walled cylindrical part cutting process in an example applying an embodiment of the present application are shown (wherein m=4, τ=1-24);

[0045] Figure 5 Matching results of an unidentified sample and a permutation spectrum matrix template set of vibration signals of four machining states in a certain thin-walled cylindrical part cutting process in an example applying an embodiment of the present application are shown;

[0046] Figure 6Fig. 4 shows a confusion matrix of the recognition results of all unidentified samples of the vibration signals of four machining states of a thin-walled cylindrical part cutting process in one example of an embodiment of the present application.

[0047] The sequence numbers, symbols and codes in the figures are explained as follows:

[0048] X: represents a state monitoring signal of a machining process;

[0049] m: a parameter of arranging the spectrum feature, representing the embedding dimension

[0050] τ: a parameter of arranging the spectrum feature, representing the time delay Probability distribution of the permutation pattern of the monitoring signal X corresponding to the parameter m and the parameter τ Permutation spectrum feature of the monitoring signal X corresponding to the parameter m and the parameter τ = 1 Permutation spectrum feature of the monitoring signal X corresponding to the parameter m and the parameter τ = 2 Permutation spectrum feature of the monitoring signal X corresponding to the parameter m and the parameter τ = m!

[0051] P: permutation spectrum matrix

[0052] m!-dim: represents that the dimension of the data is m! dimension DETAILED DESCRIPTION

[0053] Exemplary embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the drawings, it is understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the present disclosure to those skilled in the art. It should be noted that, unless otherwise specified, the technical terms or scientific terms used in the present disclosure should be understood as their general meanings understood by those skilled in the art to which the present disclosure belongs. In one embodiment of the present disclosure, the specific implementation method is described in detail with reference to the accompanying drawings, but the present disclosure is not limited by the specific implementation method. In one embodiment of the present disclosure, the specific implementation method is described in detail with reference to the accompanying drawings, but the present disclosure is not limited by the specific implementation method.

[0054] The permutation spectrum matrix is composed of permutation spectrum (PS) features of different parameter combinations, and the matrix directly acts on the original signal without data conversion. Moreover, since the optimal feature parameter combination does not need to be found, the matrix can retain most of the machining state information. The template matching strategy is a method of selecting a template based on the matrix matching degree of the PSM training set, which is used to identify the machining state category of the unidentified signals in the test set.

[0055] Therefore, multi-source sensing signals generated in the cutting process of the part are collected, the arrangement spectrum features of the signals are obtained by phase space reconstruction, and multiple PSMs of the same signal are obtained according to different parameter combinations. The PSM template set of the monitoring signals of different states is constructed by using the template selection data set, and then the method based on the template matching strategy is used to identify the best matching template of the PSM of the test signal and the PSM template set of the training set, and the processing state of the best matching template is the processing state of the test signal. Based on this, the embodiment of the application provides a machining process tool abnormal state monitoring method based on an arrangement spectrum matrix. According to the machining process tool abnormal state monitoring method based on the arrangement spectrum matrix provided by the embodiment of the application, the original signal is divided into different phase spaces through different parameter combinations. On this basis, the arrangement spectrum feature vector corresponding to each phase space is calculated, and the arrangement spectrum features obtained by different parameter combinations are combined to form an arrangement spectrum matrix (PSM). The state recognition method based on the best matching template is proposed. A certain proportion of samples are selected as the training set, and the PSM of each sample in the training set is calculated as a template. The matching degree of the test sample and each sample in the template set is calculated by using the Euclidean distance between the row vectors of the matrix, and the best template matching is performed. The processing state corresponding to the best matching template is the processing state of the test sample. According to the recognition result, the current processing state of the tool can be judged, and technical basis is provided for subsequent tool replacement, cutting parameter adjustment and shutdown health management measures.

[0056] The machining process tool abnormal state monitoring method based on the arrangement spectrum matrix provided by the embodiment of the application is described in detail below with reference to the accompanying drawings. Figure 1 The machining process tool abnormal state monitoring method based on the arrangement spectrum matrix provided by the embodiment of the application is described in detail below with reference to the accompanying drawings. Figure 2 The machining process tool abnormal state monitoring method based on the arrangement spectrum matrix provided by the embodiment of the application is described in detail below with reference to the accompanying drawings.

[0057] Reference Figures 1-2According to an embodiment of the present invention, a method for monitoring abnormal tool states during machining processes based on permutation spectrum matrices is provided, comprising the following steps: Step 1: Signal acquisition and dataset division: Acquire state monitoring signals of parts during the cutting process and divide them into template selection datasets and test datasets according to a certain ratio; Step 2: Constructing permutation patterns and reconstructing phase space: Obtain all permutation patterns according to the embedding dimension parameters in the permutation pattern parameter combinations. For monitoring signals of a given length, reconstruct them into phase space form through different permutation pattern parameter combinations; Step 3: Generating permutation spectra and permutation spectrum matrices: Calculate the corresponding permutation spectrum features according to different time delay parameters and form a permutation spectrum matrix; Step 4: Generating template sets: Obtain the permutation spectrum matrices of all samples in the template selection dataset and use these matrices as permutation spectrum matrix template sets; Step 5: Obtaining template matching degree: Select unidentified samples from the test dataset, obtain the permutation spectrum matrix of the sample, and calculate its matching degree with each sample in the permutation spectrum matrix template set to obtain a template matching degree vector; Step 6: Machining state identification: Based on the obtained template matching degree vector, select the sample corresponding to the template matching strategy. The machining state of the sample is the machining state of the unidentified sample.

[0058] Optionally, the status monitoring signal mentioned in step one refers to the data signal of the part during the cutting process collected by internal or external sensors of the CNC machine tool, including but not limited to current, voltage, temperature, vibration signals, etc. In the exemplary embodiments of the present invention applied below, vibration signal is used as an example for explanation.

[0059] Optionally, the division into template selection dataset and test dataset according to a certain ratio mentioned in step one refers to dividing the collected monitoring signals under different processing states into two sets according to the set sample allocation ratio z1:z2. The template selection dataset M is used to construct feature templates and calculate subsequent matching degrees, while the test dataset T consists of unidentified samples used for model validation.

[0060] First, represent all samples as the total dataset F = (F1, F2, ..., F...). u ), where u represents the number of states in the dataset, F u Let M represent the subset of data representing the u-th state. Assume each subset contains z samples (where z1 + z2 = z). Based on a predetermined ratio z1:z2, each subclass is divided into two parts, forming the template selection dataset M = (M1, M2, ..., M...). u ) and the test dataset T = (T1, T2, ..., T u ).in:

[0061]

[0062]

[0063] wherein M u and T u denote the template selection data subset and the test data subset corresponding to the u-th state data, respectively. denote the z1-th sample in the template selection data subset M u . denote the z2-th sample in the test data subset T u , and z1+z2=z, z being the total number of samples of the u-th state data.

[0064] Optionally, the permutation pattern parameter combination in step two mainly refers to the parameters used for constructing the permutation pattern and monitoring the phase space reconstruction, including three parameters [m, τ, N]. Among them, m is the embedding dimension, τ is the time delay, and N is the data length. Among them, for the embedding dimension m, when the parameter m is too small, the permutation pattern has fewer types, which is not conducive to representing the state information in the data; when the parameter m is too large, it leads to too large permutation pattern type, which greatly increases the calculation cost, which is not conducive to the real-time diagnosis requirement. In an exemplary embodiment, m=4 can be selected for discussion. In addition, the value of the data length N is affected by the parameter m, and should satisfy N≥5m!. The range of parameter τ is determined according to parameter m, which satisfies τ∈[1,m!] and is a positive integer.

[0065] Optionally, the permutation pattern in step two mainly refers to the data fluctuation type determined according to the parameter embedding dimension m. The permutation pattern is composed of an m-dimensional vector containing positive integers [1, 2, …, m] and each number appearing only once. There are m! permutation ways for m positive integers, each permutation way represents a permutation pattern, denoted as Π m (k), k represents the serial number of the permutation pattern, k=1, 2, …, m!. The set of all permutation patterns is denoted as Π m .

[0066] Optionally, the reconstruction in step two is in the form of phase space, which mainly refers to reconstructing a length N state monitoring signal into multiple sub-sequences according to the parameter embedding dimension m and the parameter time delay τ, and converting it into the corresponding permutation pattern, which can further include the following steps:

[0067] For a given length N state monitoring signal X={x1, x2, …, x N}, embed X into the m-dimensional original phase space S m , S m contains n sub-sequences i=1, 2, …, n. denotes the ith sub-sequence, each of which has length m. The number of sub-sequences n is calculated according to the parameter combination [m, τ, N]:

[0068] n = N - (m - 1)τ

[0069] The elements in the sub-sequence are arranged in ascending order, and are converted into a vector containing [1, 2, …, m] and each positive integer appearing only once according to the order relationship of the elements is called the permutation pattern of the sub-sequence , r j denotes a positive integer from 1 to m, j = 1, 2, …, m. When there are elements with equal values in the sub-sequence (i.e.: ), in order to ensure the uniqueness of the result, let r l < r l+1 , l denotes the position corresponding to a certain value point in the sub-sequence, l = 1, 2, …, m - 1. For example, for the sub-sequence there are x2≤x3≤x1, when m = 3, τ = 1, the corresponding permutation pattern is

[0070] After all the sub-sequences in the original phase space S m are converted into the corresponding permutation patterns, the corresponding symbolic phase space π m can be obtained. Through the above step two, the state monitoring signal is converted into the form of the symbolic phase space based on the permutation pattern, which is convenient for the statistics of the permutation pattern times.

[0071] Optionally, the permutation spectrum feature described in step three mainly refers to a feature vector based on the number of permutation patterns in the symbolic phase space Π m and the amplitude characteristics of the signal. There are m! permutation patterns in the permutation pattern set Π m , each of which is denoted as Π m (k), k = 1, 2, …, m!. The number of each permutation pattern appearing is counted, denoted as , and the probability of each permutation pattern appearing is . Among them, n is the number of sub-sequences. The probability distribution of the permutation pattern of the state monitoring signal X with length N is represented as:

[0072] Considering that there can be differences in amplitude between signals with the same permutation spectrum, the permutation spectrum alone cannot effectively distinguish such signals. For example, the ordered patterns corresponding to signals {1, 4, 6} and {1, 8, 9} are both [1, 2, 3], but the mean and variance of the two signals are significantly different. Therefore, in an embodiment, the probability distribution of the permutation pattern of the signal X can be weighted using the amplitude characteristics to obtain the permutation spectrum feature :

[0073]

[0074] In the formula, w represents the amplitude characteristics of the signal, and the variance of the signal is used to represent the amplitude characteristics of the signal:

[0075]

[0076] In the formula, represents the mean of all data points in the signal, and x q represents the data points in the signal, q = 1, 2, …, N, and N represents the total number of data points in the signal. The permutation spectrum feature The amplitude information and the order structure unit information of the signal are considered at the same time, which can more fully reflect the overall and detailed characteristics of the signal.

[0077] The permutation spectrum matrix described in step three is a matrix composed of multiple permutation spectrum features obtained by using the same parameter embedding dimension m and different parameter time delay τ. When the embedding dimension parameter m is determined, τ ∈ [1, m!], so m! groups of permutation spectrum features are obtained, which are represented as the permutation spectrum matrix P:

[0078]

[0079] In the formula, represents the permutation spectrum feature of the signal X at the time delay parameter τ = h, h = 1, 2, …, m!.

[0080] The above step three is the construction process of the permutation spectrum matrix. From the perspective of permutation order, the fluctuation characteristics of the signal are extracted, and the fluctuation patterns under different parameters can be considered at the same time. Not only is the best parameter selection link optimized, but the amount of information contained in the obtained permutation spectrum matrix is also more abundant, which is beneficial to obtain more accurate recognition results.

[0081] Optionally, the permutation spectrum matrix template set described in step four refers to the permutation spectrum matrix template set obtained by using the template selection data set M = (M1, M2, …, M uFor all samples in the dataset M, calculate the set PM consisting of the permutation spectrum matrix of each sample. Based on the data partitioning method in step one, the template selection dataset M contains u types of state data, each type containing z1 samples, for a total of u×z1 samples. Calculate the permutation spectrum matrix P of each sample as the template to obtain the permutation spectrum matrix template set consisting of u×z1 permutation spectrum matrices. This represents the permutation spectrum matrix corresponding to the i-th sample of the j-th state data in the template selection dataset M.

[0082] Optionally, the unidentified samples mentioned in step five refer to the test dataset T = (T1, T2, ..., T...). u The samples in the dataset. During state recognition, an arbitrary signal sample is selected from the test dataset T as an unrecognized sample, denoted as t. The processing state of this sample is considered unknown, and its permutation spectrum matrix is ​​denoted as P. t According to the data partitioning method in step one, the test dataset T contains u types of state data, each type of data contains z2 samples, and the test dataset T contains u×z2 unidentified samples.

[0083] Optionally, the template matching degree vector mentioned in step five refers to the permutation spectrum matrix P of unidentified samples calculated using the Euclidean distance between matrix row vectors. t Permutation spectrum matrix of each sample in the permutation spectrum matrix template set PM The matching degree is calculated, and all the matching degrees obtained are called the template matching degree vector. P represents the permutation spectrum matrix of the unidentified sample t. t The permutation spectrum matrix corresponding to the i-th sample of the j-th state data in the template selection dataset M The matching degree is the degree of similarity between the two matrices; a higher value indicates a higher degree of similarity. The formula for calculating the matching degree is as follows:

[0084]

[0085]

[0086] In the formula, A and B represent two matrices of the same dimension. In this invention, matrix B represents the permutation spectrum matrix P corresponding to the unidentified sample t. t Let matrix A represent the permutation spectrum matrix corresponding to the i-th sample of the j-th state data in the template selection dataset M. a v,w and b v,w Let v and w represent the values ​​of the elements in the v-th row and w-th column of matrices A and B, respectively, where v = 1, 2, ..., m! and w = 1, 2, ..., m!, and m represents the embedding dimension.

[0087] Step five is a matching degree calculation process of the unknown sample t and the samples in the permutation spectrum matrix template set P. The matching degree represents the similarity of the samples, and the higher the value, the more similar the two samples are, and the more likely they belong to the same processing state.

[0088] Optionally, the sample corresponding to the template matching strategy in step six is a method for selecting the processing state corresponding to the best matching template as the processing state of the unknown sample according to the matching strategy. The c matching degrees with the largest values in the template matching degree vector D are calculated c≥1. The c matching degrees correspond to the processing state with the highest frequency, i.e., the state of the unknown sample t. When the frequencies of two states are the same, the processing state corresponding to the maximum matching degree is taken as the state of the unknown sample t. In actual application examples, for example, the value c can be set according to specific circumstances and needs.

[0089] Through the method provided by the above embodiments of the application, the state monitoring signal collected during the partial cutting process can be used to establish an abnormal state monitoring model of the cutting process of the numerical control machine tool. The permutation spectrum matrix operation is simple, can reflect the underlying data fluctuations of the signal under different sampling intervals, and contains as much state information as possible. The template matching strategy method can timely and accurately determine the processing state to which the current unknown sample belongs, realize the abnormal state recognition of the machining process of the numerical control machine tool, and provide technical guidance for the on-site staff to make decisions such as tool replacement and shutdown maintenance, so as to ensure the machining quality of the product.

[0090] The following refers to the accompanying drawings, and a detailed description of an exemplary embodiment of a machining process tool abnormal state monitoring method based on a permutation spectrum matrix according to an embodiment of the application is provided.

[0091] Figure 3 The time domain graphs of the vibration signals of four processing states of the cutting process of a certain thin-walled cylindrical part and the corresponding permutation spectrum graphs (where m=4 and τ=1, 10) in an example according to an embodiment of the application are shown. Figure 4 The gray scale graphs of the permutation spectrum matrices of the vibration signals of four processing states of the cutting process of a certain thin-walled cylindrical part in an example according to an embodiment of the application are shown (where m=4 and τ=1-24). Figure 5 The matching results of an unknown sample and a permutation spectrum matrix template set of four processing state vibration signals in the cutting process of a certain thin-walled cylindrical part in an example according to an embodiment of the application are shown. Figure 6 The recognition result confusion matrix graph of all unknown samples of the vibration signals of four processing states of the cutting process of a certain thin-walled cylindrical part in an example according to an embodiment of the application is shown.

[0092] Referring toFigures 3-6 In an exemplary embodiment of the present invention, vibration signals from four machining states during the cutting of a thin-walled cylindrical part were used as experimental data for analysis. A PCB 356A32 vibration sensor was used to acquire the signals at a sampling frequency of 10.24 kHz. The thin-walled cylindrical material was 7109 aluminum alloy, and the cutting tool was made of cemented carbide. The cutting depth was 1 mm, the spindle speed was 180 r / min, and the feed rate was 0.1 mm / r.

[0093] Two cylinder thicknesses were designed: one 6mm thick, less prone to deformation, and the other 2mm thick, more prone to deformation. Furthermore, two types of cutting tools were used: a new tool and a worn tool. Based on the above design, four machining states were defined: normal state, machining deformation state, tool wear state, and a mixed state where machining deformation and tool wear occur simultaneously. Therefore, including the normal state data, four data sets were obtained, as shown in Table 1, with each state containing 100 samples.

[0094] Table 1 Description of Vibration Signal Set

[0095]

[0096] Figure 1 A flowchart of a method for monitoring abnormal tool conditions in machining processes based on a permutation spectrum matrix, according to an embodiment of the present invention, is shown. Figure 2 A flowchart illustrating the permutation spectrum matrix construction process in a machining process tool abnormality monitoring method based on a permutation spectrum matrix according to an embodiment of the present invention is shown. The following references... Figure 1 and Figure 2 This paper describes a method for monitoring abnormal tool conditions in machining processes based on a permutation spectrum matrix, according to one embodiment of the present invention. The method for monitoring abnormal tool conditions in machining processes based on a permutation spectrum matrix, according to one embodiment of the present invention, includes the following steps:

[0097] Step 1: Signal Acquisition and Phase Space Reconstruction: The vibration signal of a thin-walled cylindrical part during the cutting process is acquired as a state monitoring signal, and all samples are divided into template selection dataset M and test dataset T according to the ratio of z1:z2=80:20. Figure 3 The time-domain waveforms of the raw data of four state monitoring signals during the cutting process of the thin-walled cylindrical part are shown. The data length of each sample is N=1024.

[0098] Step two: Constructing permutation pattern and phase space reconstruction: First, determine all permutation pattern categories, for example, with embedding dimension m = 4, 24 permutation patterns can be obtained. According to m = 4, the sample is divided into subsequences with a length of 4, and reconstructed into a phase space. Another parameter τ ∈ [1, m!] = [1, 24], different parameter combinations [m, τ] can obtain different phase space forms.

[0099] Step three: Calculate permutation spectrum and permutation spectrum matrix: According to step two, the permutation spectrum characteristics of a sample under different parameter combinations [m, τ] can be obtained. Figure 3 The permutation spectrum of a certain sample corresponding to the parameter combination [m, τ] = [4, 1] and the parameter combination [m, τ] = [4, 10] of the four machining states of the thin-walled cylindrical part cutting process is shown. From the figure, it can be seen that when τ = 1, the permutation spectrum of the normal state and the wear state, the machining deformation state and the mixed state all have high similarity. When τ = 10, the difference of the permutation spectrum of the four state monitoring signals is obviously increased. It can be seen that different time delay parameters τ have different sensitivities to different state signals, and using multiple τ corresponding permutation spectrum for identification is beneficial to improve the identification result of the model.

[0100] The parameter τ ∈ [1, m!] = [1, 24], so 24 columns of permutation spectrum characteristics can be obtained for each sample, which together form a permutation spectrum matrix with a size of 24 x 24. Figure 4 The gray scale diagram of the permutation spectrum matrix corresponding to m = 4 of a certain sample of the four machining states of the thin-walled cylindrical part cutting process is shown.

[0101] Step four: Generate template set: Calculate the permutation spectrum matrix of all samples in the template selection data set M, there are 4 state monitoring signals, each state monitoring signal has 80 samples in the data set M, a total of 4 x 80 = 320 permutation spectrum matrix templates. The 320 matrix templates are taken as the permutation spectrum matrix template set PM;

[0102] Step five: Calculate template matching degree: Select an unidentified sample t from the test data set T, calculate the permutation spectrum matrix B corresponding to the unidentified sample t, and calculate the matching degree d between the permutation spectrum matrix B and each sample in the permutation spectrum matrix template set PM according to the Euclidean distance between the matrix row vectors. The matching degree d of a certain unidentified sample t is shown in Table 2. t t The matching degree value of 320 samples is shown in Table 2. Figure 5

[0103] Table 2 Template matching degree of a certain unidentified sample t

[0104] Sample 1 2 3 ... 78 79 80 Normal state 0.3376 0.3366 0.3403 … 0.3350 0.3317 0.3346 Processing deformation 0.8276 0.8334 0.8384 … 0.6980 0.7340 0.8348 Worn state 0.3392 0.3375 0.3367 … 0.3354 0.3370 0.3375 Mixed state 0.4100 0.5709 0.4855 … 0.4947 0.5631 0.4270

[0105] ​​In the present example, let c=3. According to the template matching strategy, the top c templates with the largest matching degrees are all in the machining deformation state, so the machining state corresponding to the unclassified sample t is the machining deformation state.

[0106] Step six: machining state recognition: according to the calculated template matching degree and template matching strategy, it is determined that the machining state of the unclassified sample is the machining deformation state. In the present example, there are 4x20=80 unclassified samples, Figure 6 The recognition result confusion matrix diagram of all unclassified samples is shown. It can be seen that the recognition rate reaches 100%

[0107] According to the above recognition result, the on-site staff can monitor the machine tool cutting process, and in the early stage of abnormal state, timely identify the occurrence of abnormal state and accurately judge the type of abnormal state, avoid the problems in the product machining process caused by tool wear, chatter or workpiece deformation, etc., to ensure the product quality. According to the diagnosis result, the relevant staff can be guided to perform tool changing, processing parameter adjustment or shutdown maintenance operations on the numerical control machine tool, to ensure the safety and reliability of the entire production process.

[0108] The machining process tool abnormal state monitoring method based on the permutation spectrum matrix proposed in the present application can utilize the state monitoring signals (including but not limited to vibration signals) generated by the numerical control machine tool in the product machining process, timely discover different data fluctuation modes caused by tool wear, chatter, workpiece deformation and their coupled states, and obtain the corresponding permutation spectrum and permutation spectrum matrix. The permutation spectrum feature has high sensitivity to data fluctuation changes, and is simple to calculate and has good noise resistance. The permutation spectrum matrix optimizes the best parameter selection process of the permutation spectrum feature, and contains more state information, so the recognition effect is better. The model proposed in the present application is not only suitable for the abnormal state monitoring of the thin-walled cylindrical cutting process in the case, but also suitable for the production and machining processes of other products and parts and the operation fault diagnosis process of equipment. The present application has good expandability, and provides certain reference for other technical personnel in the technical field.

[0109] It should be noted that the flowchart and block diagrams in the drawings show architectural, functional and chronological arrangement of systems, methods and computer program products according to the embodiments of the present application. In this regard, each block in the flowchart or block diagrams can represent a module, a segment, or a portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that in some alternative implementations, the functions noted in the blocks can occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks can sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and / or flowchart illustrations, and combinations thereof, can be implemented by a dedicated hardware-based system that performs the specified functions or acts, or combinations of hardware and software.

[0110] In addition, the foregoing merely illustrates some embodiments and that a person of ordinary skill in the art can make changes, modifications, additions and / or alterations without departing from the scope and spirit of the disclosed embodiments, which are illustrative rather than restrictive. In addition, the described embodiments relate to currently considered most practical and most preferred embodiments, which should be understood as the embodiments should not be limited to the disclosed embodiments, but rather are intended to cover various modifications and equivalent arrangements included in the spirit and scope of the embodiments. In addition, the above-described various embodiments can be applied together with other embodiments, for example, aspects of one embodiment can be combined with aspects of another embodiment to achieve another embodiment. In addition, each individual feature or component of any given component can constitute another embodiment.

[0111] The above embodiments are only used to illustrate the technical solutions of the present application, and not to limit it; although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions recorded in the above embodiments can still be modified, or some or all of the technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application, and they should be covered in the scope of the claims and the specification of the present application.​​

Claims

1. A method for monitoring abnormal state of a cutting tool in a machining process based on a permutation spectrum matrix, characterized by The method comprises the following steps: Step 1: signal acquisition and data set division: acquire the state monitoring signals of the parts in the cutting process, and divide them into a template selection data set and a test data set according to a certain proportion; Step 2: constructing permutation patterns and phase space reconstruction: obtain all permutation patterns according to the embedding dimension parameter in the permutation pattern parameter combination, and reconstruct the monitoring signals of a given length into a phase space form through different permutation pattern parameter combinations; Step 3: generating permutation spectrum and permutation spectrum matrix: calculate the corresponding permutation spectrum features according to different time delay parameters, and compose a permutation spectrum matrix; Step 4: generating a template set: obtain the permutation spectrum matrix of all samples in the template selection data set, and take these matrices as the permutation spectrum matrix template set; Step 5: obtaining template matching degree: select an unidentified sample from the test data set, obtain the permutation spectrum matrix of the sample, and obtain the matching degree of the sample with each sample in the permutation spectrum matrix template set, to obtain the template matching degree and the maximum template matching degree vector; Step 6: processing state recognition: according to the obtained template matching degree, select the sample corresponding to the maximum template matching degree, and the processing state of the sample is the processing state of the unidentified sample; In step 1, the division into a template selection data set and a test data set according to a certain proportion comprises dividing the acquired monitoring signals under different processing states into two sets, a template selection data set M and a test data set T, according to a set sample allocation proportion z1:z2, wherein the template selection data set M is used to construct a feature template and subsequent matching degree calculation, and the test data set T is an unidentified sample and is used for model verification; Firstly, all samples are expressed as total dataset F = (F1, F2, …, F u ), u represents data with u kinds of states in the dataset, F u represents a sub-dataset composed of the u-th state data, it is assumed that each sub-dataset contains z samples, and z1+z2=z, according to the set proportion z1:z2, each sub-class is divided into two parts to constitute a template selection dataset M = (M1, M2, …, M u ) and a test dataset T = (T1, T2, …, T u ), wherein: M u = [M u,1 , M u,2 ,..., M u,z1 ] T u = [T u,1 , T u,2 ,..., T u,z2 ] wherein M u and T u respectively represent a template selection data subset and a test data subset corresponding to the u-th state data, represents the z1-th sample in the template selection data subset M u represents the z2-th sample in the test data subset T u , z1+z2=z, z being the total number of samples of the u-th state data.​ In step four, the permutation spectrum matrix template set is arranged, including calculating the set PM of permutation spectrum matrix of each sample by using all samples in the template selection data set M=(M1, M2, …, M u ) according to the data division manner in step one, the template selection data set M contains u kinds of state data, each kind of data contains z1 samples, and a total of u×z1 samples, and the permutation spectrum matrix P of each sample is calculated as a template to obtain the permutation spectrum matrix template set composed of u×z1 permutation spectrum matrices indicates the permutation spectrum matrix corresponding to the i-th sample of the j-th state data in the template selection data set M.

2. The machining process tool abnormal state monitoring method based on the permutation spectrum matrix according to claim 1, characterized in that: The state monitoring signals in step 1 are data signals of the parts in the cutting process acquired by using internal sensors or external sensors of a numerical control machine tool, including current, voltage, temperature, and vibration signals.

3. The machining process tool abnormal state monitoring method based on the permutation spectrum matrix according to claim 2, characterized in that: The permutation pattern parameter combination in step 2 comprises parameters used when constructing permutation patterns and reconstructing monitoring signals into a phase space form, including three parameters [m, τ, N], wherein m is an embedding dimension, τ is a time delay, and N is a data length, the value of the data length N is affected by the embedding dimension m, and satisfies N≥5m!, and the range of the parameter τ is determined according to the parameter m, and satisfies τ∈[1,m!].

4. The machining process tool abnormal state monitoring method based on the permutation spectrum matrix according to claim 3, characterized in that: The arrangement mode described in step two, including determining the data fluctuation type according to the parameter embedding dimension m, the arrangement mode is composed of an m-dimensional vector containing positive integers [1, 2, …, m] and each number appears only once, there are m! Arrangement ways of m positive integers, each arrangement way represents an arrangement mode, denoted as Π m (k), k represents the serial number of the arrangement mode, k = 1, 2, …, m!, the set of all arrangement modes is denoted as Π m ; The reconstruction into a phase space form in step 2 comprises reconstructing a state monitoring signal with a length of N into multiple sub-sequences according to the parameter embedding dimension m and the parameter time delay τ, and converting the sub-sequences into corresponding permutation patterns, and further comprises the following steps: For a given length N state monitoring signal X = {x1, x2,..., xN}, X is embedded into an m-dimensional original phase space S N m S m contains n sub-sequences​ represents the ith subsequence, each of length m, Calculate the number n of sub-sequences according to the parameter combination [m, τ, N]: n=N-(m-1)τ The elements in the subsequence are arranged in ascending order, and converted into a vector containing [1, 2, …, m] and each positive integer appearing only once according to the order relationship of the elements , called the arrangement mode of the subsequence , r j represents the positive integer from 1 to m, j = 1, 2, …, m; When there are elements with assigned values in the subsequence, i.e. To ensure the uniqueness of the result, let r l l+1 l represents the position corresponding to a certain value point in the subsequence, l = 1, 2, …, m-1;​​ Original phase space S m After all the sub-sequences are converted into corresponding permutation patterns, the corresponding symbolic phase space π is obtained m Thus, the state monitoring signal is converted into the form of symbolic phase space based on permutation patterns, so as to perform the statistics of permutation pattern frequencies.

5. The machining process tool abnormal state monitoring method based on the permutation spectrum matrix according to claim 4, characterized in that: The arrangement spectrum features described in step three are based on symbol phase space Π m The feature vector is the number of arrangement patterns in Π m There are m! arrangement patterns in the arrangement pattern set Π m (k), k = 1, 2, …, m!, and the number of occurrences of each arrangement pattern is counted and denoted as The probability of occurrence of each arrangement pattern is Where n is the number of subsequences, and the probability distribution of the arrangement patterns of the state monitoring signal X with length N is represented as: using the amplitude characteristic to weight the probability distribution of the permutation pattern of the signal X, to obtain a permutation spectrum characteristic where, represents the permutation spectrum feature of signal X, represents the kth permutation pattern m (k) the corresponding permutation spectrum feature value, k = 1, 2, …, m!, and w represents the amplitude feature of the signal, which is represented by the variance of the signal: wherein represents the mean of all data points in the signal, x q represents the data points in the signal, q = 1, 2,..., N, which N represents the total number of data points in the signal; and The permutation spectrum matrix described in step three is a matrix composed of multiple permutation spectrum feature groups obtained by using the same parameter embedding dimension m and different parameter time delay τ. When the embedding dimension parameter m is determined, the parameter τ ∈ [1, m!], a total of m! permutation spectrum features are obtained, which is denoted as a permutation spectrum matrix P: wherein denotes the alignment spectrum feature of signal X at time delay parameter τ = h, h = 1, 2,..., m!.

6. The permutation spectrum matrix-based machine process tool abnormal state monitoring method according to claim 5, characterized in that: The unidentified sample described in step five is a sample in the test data set T = (T1, T2, …, T u ), and when the state is identified, a signal sample is randomly selected from the test data set T as an unidentified sample, denoted as t, and it is considered that the processing state of the sample is unknown. The arrangement spectrum matrix of the sample is denoted as P t , according to the data division mode described in step one, the test data set T contains u kinds of state data, each kind of data contains z2 samples, and the test data set T contains u×z2 unidentified samples. The template matching degree vector described in step five is the arrangement spectrum matrix P of the un-identified sample calculated by the Euclidean distance between the matrix row vectors t The matching degree of the arrangement spectrum matrix of each sample in the arrangement spectrum matrix template set PM The matching degree of the arrangement spectrum matrix of each sample in the arrangement spectrum matrix template set PM The matching degree of the arrangement spectrum matrix of each sample in the arrangement spectrum matrix template set PM t The matching degree of the arrangement spectrum matrix of each sample in the arrangement spectrum matrix template set PM The matching degree of the arrangement spectrum matrix of each sample in the arrangement spectrum matrix template set PM where A and B represent two matrices with the same dimension, B represents the permutation spectrum matrix corresponding to the unclassified sample t t A represents the permutation spectrum matrix corresponding to the i-th sample of the j-th state data in the template selection data set M a v,w and b v,w represent the element value of the v-th row and w-th column in the matrix A and B, v = 1, 2, …, m!, w = 1, 2, …, m!, and m represents the embedding dimension.

Images