A vibration signal recognition method and system based on improved multi-scale permutation entropy

Abstract

The present invention proposes a vibration signal recognition method and system based on improved multi-scale permutation entropy. The method includes: collecting vibration signals under different working states, performing multi-scale analysis based on sliding windows and sample quantiles on the collected vibration signals under different working states, and obtaining a sequence after multi-scale analysis; obtaining the permutation entropy of the sequence after multi-scale analysis, obtaining a multi-scale permutation entropy value, and using the value of the multi-scale permutation entropy as a feature vector of the vibration signal under different fault states; inputting the feature vector into a classifier based on machine learning, thereby realizing the recognition of vibration signals under different fault states and obtaining a fault state recognition result; and verifying the rationality of the recognition method according to the fault state recognition result. The present invention solves the defect that the conventional multi-scale permutation entropy fault diagnosis algorithm has insufficient ability in fault feature extraction, and has a short calculation time, thereby improving the fault diagnosis efficiency.

Description

Technical Field

[0001] The invention belongs to the field of signal recognition, and in particular relates to a vibration signal recognition method and system based on improved multi-scale permutation entropy. Background Art

[0002] At present, with the development of science and technology, mechanical systems are showing a trend of intelligence and integration. The good operating state of mechanical systems plays an important role in ensuring the quality and efficiency of industrial development. As a key component in the mechanical system, the good operating state of rolling bearings determines the state performance of the entire mechanical system. Therefore, it is necessary to identify the working state of rolling bearings. Among the existing rolling bearing fault diagnosis methods based on signal analysis, the diagnosis method based on vibration signals has the characteristics of high frequency of use, simplicity and effectiveness. However, due to the working environment of rolling bearings, the bearings will be restricted by environmental interference, their own processes and other factors during operation, resulting in strong nonlinearity and non-stationarity of their vibration signals. Therefore, traditional signal analysis methods based on time domain and frequency domain cannot accurately and effectively extract the characteristic information of vibration signals. With the deepening of nonlinear theory, nonlinear dynamics technology has gradually been well applied in time series analysis. Among them, entropy theory, such as permutation entropy, has the characteristics of simple calculation, strong anti-noise ability, and can effectively detect the randomness and dynamic mutation of time series. It is widely used in mechanical signal analysis, EEG signal analysis, myoelectric signal analysis, and geographical hydrological signal analysis. However, due to the complexity of mechanical systems, the characteristic information of vibration signals is not only manifested at a single scale. Therefore, in order to improve the ability of permutation entropy to extract fault characteristics of vibration signals, it is necessary to analyze vibration signals at multiple scales. Multiscale permutation entropy theory coarsens the vibration signal time series at multiple scales to obtain coarse-grained sequences at multiple scales, thereby realizing multi-scale analysis of vibration signals.

[0003] The core idea of ​​existing multi-scale analysis methods, such as multi-scale permutation entropy and multi-scale fuzzy entropy, is to multi-scale the time series, that is, to obtain the characteristic information of the vibration signal at different scales by coarse-graining the time series of the vibration signal at different scales to form the characteristic vector. The essence of the coarse-graining process is averaging.

[0004] Implementation of existing technology

[0005] First, the permutation entropy algorithm is defined. The principle of the permutation entropy algorithm is: for a time series {x(i), i=1,2,...,N} of length N, the phase space of x(i) is reconstructed:

[0006]

[0007] Where m is the embedding dimension, τ is the delay time, and k=N-(m-1)τ is the number of reconstructed vectors.

[0008] X j The elements in are arranged in ascending order:

[0009] x[i+(j1-1)τ]≤…≤x[i+(j m -1)τ]

[0010] In the formula, j1, j2, …j m is the index of the column in the phase space where each element is located before sorting.

[0011] If X j If two elements in are equal, they are arranged in the original order. By analyzing the time series after phase space reconstruction, for any X j , we can get the corresponding symbol sequence S l ={j1,j2,…,j m}, where l = 1, 2, ..., k, and k ≤ m! . The permutation entropy can be defined as:

[0012]

[0013] Where P j is the probability of the symbol sequence, and

[0014] like Then H p (m) = ln(m!), the value is the largest. p (m) Standardization can be obtained:

[0015]

[0016] At this time, H p The value of is [0,1], which can reflect the complexity of the time series. p The larger it is, the more complex the sequence is.

[0017] The permutation entropy algorithm is simple to calculate and summarizes the characteristic information of the vibration signal by measuring the complexity of the time series. In terms of parameter selection, according to the suggestions of Bandt et al., the proposer of the algorithm, the embedding dimension m should not be too small or too large; if the embedding dimension m is too small, the reconstructed vector contains too few states and cannot effectively reflect the dynamic mutation of the time series, and the algorithm loses its meaning; if the embedding dimension m is too large, the reconstruction process of the phase space will homogenize the time series, so it is generally taken as 3 to 7; and the time delay τ has little effect on the calculation, and τ=1 is generally taken. The length of the time series refers to the general selection method in the process of rolling bearing vibration signal extraction, and the length is taken as N=2048.

[0018] In order to improve the effectiveness of conventional permutation entropy algorithm for vibration signal fault feature extraction, it is necessary to perform multi-scale analysis on it. The conventional multi-scale method based on coarse-grained process is:

[0019] Define a time series x(N)=x(1),x(2),...,x(N) of length N, first perform coarse-grained analysis on it and construct a coarse-grained sequence

[0020]

[0021] In the formula, τ is a positive integer. Generally, τ is defined as a scale factor. For a time series x(N) with a length of N, when the scale factor τ = 1, the coarse-grained result is p j (1) = x(N), the resulting coarse-grained sequence is the original time series, and the coarse-graining process is not performed at this time. When the scale factor τ≠1, the time series x(N) is coarse-grained into a coarse-grained sequence of length [N / τ]. ([N / τ] is defined as when N / τ is not an integer, it is rounded to zero). Taking the scale factor τ = 2 and the scale factor τ = 3 as examples, the coarse-graining process analysis is as follows Figure 2 shown.

[0022] Disadvantages of existing technology

[0023] (1) The core idea of ​​conventional multi-scaling methods is the coarse-graining process, that is, finding the average value at different scales. The average value of data in a time window is not enough to accurately summarize the data information in the window.

[0024] (2) According to the characteristics of conventional multi-scaling methods, when the scale factor is τ, the time series x(N) is coarse-grained into a coarse-grained sequence of length [N / τ]. For example, when the scale factor τ = 2, the original time series of length 2048 becomes a coarse-grained sequence of length 1024. Therefore, in the conventional multi-scale analysis process, the length of the time series shortens as the scale factor increases. Too short time series have the following problems: first, a large amount of information is lost; second, when performing permutation entropy calculation, as the time series shortens, the parameters need to be adjusted accordingly, which affects the adaptability of the fault diagnosis process.

[0025] (3) The fault diagnosis algorithm based on conventional multi-scale permutation entropy has the defect of insufficient ability in fault feature extraction. It is necessary to extract permutation entropy values ​​at multiple scales as feature vectors to effectively summarize the feature information. Too much feature information will cause redundancy of feature vectors and increase the calculation time, thus affecting the efficiency of fault diagnosis. Summary of the invention

[0026] In order to solve the above technical problems, the present invention proposes a technical solution of a vibration signal recognition method based on improved multi-scale permutation entropy to solve the above technical problems.

[0027] The first aspect of the present invention discloses a vibration signal recognition method based on improved multi-scale permutation entropy; the method comprises:

[0028] Step S1, collecting vibration signals under different working states, performing multi-scale analysis based on sliding windows and sample quantiles on the collected vibration signals under different working states, and obtaining a sequence after multi-scale analysis;

[0029] Step S2, obtaining the permutation entropy of the sequence after the multi-scale analysis to obtain a multi-scale permutation entropy value, and using the value of the multi-scale permutation entropy as a feature vector of the vibration signal under different fault states;

[0030] Step S3, inputting the feature vector into a classifier based on machine learning, thereby realizing recognition of vibration signals of different fault states and obtaining a fault state recognition result;

[0031] Step S4: verify the rationality of the identification method according to the fault state identification result.

[0032] According to the method of the first aspect of the present invention, in step S1, the specific method of multi-scale analysis based on sliding window and sample quantile includes:

[0033] Step S11: Perform sliding window analysis on the sequence x(N)=x(1),x(2),...,x(N) of vibration signals under different working conditions, taking the window length, i.e. the scale factor, as τ, and obtaining the first sequence {x1,x2,...,x τ}, the first sequence is {x1,x2,...,x τ}, and the sample quantile of the first sequence is obtained as p1(τ);

[0034] Step S12: Move the sliding window forward by one data point to obtain the second sequence {x2, x3, ..., x τ+1}, the sample quantile of the second sequence is obtained as p2(τ);

[0035] Step S13, and so on, apply the method of step S12 to traverse the entire sequence of vibration signals until the last sequence is {x N-τ+1 ,x N-τ+2 ,...,x N}, find the sample quantile of the last sequence and get p N-τ+1 (τ);

[0036] Step S14, obtain the sequence {p1(τ), p2(τ), ..., p N-τ+1 (τ)}, the sequence {p1(τ),p2(τ),...,p N-τ+1 (τ)} is the sequence after the multi-scale analysis.

[0037] According to the method of the first aspect of the present invention, in the step S1, the value range of the window length, that is, the scale factor τ is specifically:.

[0038] According to the method of the first aspect of the present invention, in step S1, the method for obtaining the sample quantile is:

[0039] p j (τ) = quantile(x j , x j+1 ...x j+τ-1 )1≤J≤N-τ+1

[0040] in,

[0041] p j (τ): the sample quantile of the jth sequence;

[0042] x j : The data of the jth sequence of the original sequence x(N);

[0043] J: The number of data points by which the sliding window moves forward;

[0044] N: the length of the original sequence x(N);

[0045] quantile(·): Calculates the sample quantile process of the sliding window.

[0046] According to the method of the first aspect of the present invention, in step S3, the feature vector is input into a classifier based on machine learning to achieve recognition of vibration signals of different fault states, and the method for obtaining a fault state recognition result includes:

[0047] The feature vector is input into a classifier based on a support vector machine, thereby realizing the recognition of vibration signals of different fault states and obtaining a fault state recognition result.

[0048] According to the method of the first aspect of the present invention, in step S3, before the recognition of vibration signals of different fault states is realized, the method further includes:

[0049] Construct an SVM recognition model. When constructing the SVM recognition model, in order to adaptively select the kernel function parameters and penalty factors of the SVM, an optimization algorithm is selected for optimization analysis.

[0050] According to the method of the first aspect of the present invention, in step S3, the optimization algorithm applies a PSO algorithm, ie, a particle swarm optimization algorithm.

[0051] The second aspect of the present invention discloses a vibration signal recognition system based on improved multi-scale permutation entropy; the system comprises:

[0052] The first processing module is configured to collect vibration signals under different working states, perform multi-scale analysis based on sliding windows and sample quantiles on the collected vibration signals under different working states, and obtain a sequence after multi-scale analysis;

[0053] The second processing module is configured to obtain the permutation entropy of the sequence after the multi-scale analysis, obtain the multi-scale permutation entropy value, and use the multi-scale permutation entropy value as the feature vector of the vibration signal under different fault states;

[0054] A third processing module is configured to input the feature vector into a classifier based on machine learning, so as to realize recognition of vibration signals of different fault states and obtain a fault state recognition result;

[0055] The fourth processing module is configured to verify the rationality of the identification method according to the fault state identification result.

[0056] According to the system of the second aspect of the present invention, the first processing module is specifically configured as follows: a specific method of multi-scale analysis based on sliding windows and sample quantiles includes:

[0057] The vibration signal sequence x(N)=x(1),x(2),...,x(N) under different working conditions is subjected to sliding window analysis, and the window length, i.e., the scale factor, is taken as τ. The first sequence is obtained as {x1,x2,...,x τ}, and the sample quantile of the first sequence is obtained as p1(τ);

[0058] Step S12: Move the sliding window forward by one data point to obtain the second sequence {x2, x3, ..., x τ+1}, the sample quantile of the second sequence is obtained as p2(τ);

[0059] Similarly, the method of step S12 is applied to traverse the entire sequence of vibration signals until the last sequence {x N-τ+1 ,x N-τ+2 ,...,x N}, find the sample quantile of the last sequence and get p N-τ+1 (τ);

[0060] Get the sequence {p1(τ),p2(τ),...,p N-τ+1(τ)}, the sequence {p1(τ),p2(τ),...,p N-τ+1 (τ)} is the sequence after the multi-scale analysis;

[0061] The window length, i.e., the scale factor τ, has a specific value range of: 1≤τ≤30;

[0062] The method to obtain the sample quantile is:

[0063] p j (τ) = quantile(x j , x j+1 ...x j+τ-1 )1≤J≤N-τ+1

[0064] in,

[0065] p j (τ): the sample quantile of the jth sequence;

[0066] x j : The data of the jth sequence of the original sequence x(N);

[0067] J: The number of data points by which the sliding window moves forward;

[0068] N: the length of the original sequence x(N);

[0069] quantile(·): Calculates the sample quantile process of the sliding window.

[0070] According to the system of the second aspect of the present invention, the third processing module is specifically configured to input the feature vector into a classifier based on machine learning, so as to realize the recognition of vibration signals of different fault states, and the method for obtaining the fault state recognition result includes:

[0071] The feature vector is input into a classifier based on a support vector machine, thereby realizing the recognition of vibration signals of different fault states and obtaining a fault state recognition result;

[0072] In step S3, before the recognition of vibration signals of different fault states is realized, the method further includes:

[0073] When constructing the SVM recognition model, in order to adaptively select the kernel function parameters and penalty factors of the SVM, an optimization algorithm is selected for optimization analysis;

[0074] The optimization algorithm uses PSO algorithm, namely particle swarm optimization algorithm.

[0075] The third aspect of the present invention discloses an electronic device. The electronic device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps in any one of the vibration signal recognition methods based on improved multi-scale permutation entropy in the first aspect of the present disclosure are implemented.

[0076] The fourth aspect of the present invention discloses a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in any one of the vibration signal recognition methods based on improved multi-scale permutation entropy in the first aspect of the present disclosure are implemented.

[0077] It can be seen that the solution proposed by the present invention,

[0078] (1) The fault identification efficiency is improved while ensuring the fault identification accuracy.

[0079] (2) The sequence length changes less, which avoids the problem of adjusting the permutation entropy calculation parameters due to changes in sequence length.

[0080] In summary, the solution proposed in the present invention can solve the defect of insufficient ability of conventional multi-scale permutation entropy fault diagnosis algorithm in fault feature extraction, and has a short calculation time, thereby improving the fault diagnosis efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0082] Figure 1 A flowchart of a vibration signal recognition method based on improved multi-scale permutation entropy according to an embodiment of the present invention;

[0083] Figure 2 It is a multi-scaling process diagram of the prior art when τ=2 and τ=3;

[0084] Figure 3 is a diagram of an improved multi-scaling process when τ=2 and τ=3 according to an embodiment of the present invention;

[0085] Figure 4 A diagram showing vibration signals of a normal bearing and a bearing with a fault according to an embodiment of the present invention;

[0086] Figure 5 A single-scale permutation entropy fault identification result diagram according to an embodiment of the present invention;

[0087] Figure 6 Graphs showing fault identification results at the first two scales according to an embodiment of the present invention;

[0088] Figure 7 Graphs showing fault identification results at the first three scales according to an embodiment of the present invention;

[0089] Figure 8 is a fault identification result diagram with a scale factor of 10 according to an embodiment of the present invention;

[0090] Fig. 9 Graphs showing fault identification results with scale factors of 10 and 30 according to an embodiment of the present invention;

[0091] Fig.10 A structural diagram of a vibration signal recognition system based on improved multi-scale permutation entropy according to an embodiment of the present invention;

[0092] Fig.11 The figure is a structural diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0093] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0094] The first aspect of the present invention discloses a vibration signal recognition method based on improved multi-scale permutation entropy. Figure 1 FIG. 4 is a flow chart of a vibration signal recognition method based on improved multi-scale permutation entropy according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0095] Step S1, collecting vibration signals under different working states, performing multi-scale analysis based on sliding windows and sample quantiles on the collected vibration signals under different working states, and obtaining a sequence after multi-scale analysis;

[0096] Step S2, obtaining the permutation entropy of the sequence after the multi-scale analysis to obtain a multi-scale permutation entropy value, and using the value of the multi-scale permutation entropy as a feature vector of the vibration signal under different fault states;

[0097] Step S3, inputting the feature vector into a classifier based on machine learning, thereby realizing recognition of vibration signals of different fault states and obtaining a fault state recognition result;

[0098] Step S4: verify the rationality of the identification method according to the fault state identification result.

[0099] In step S1, vibration signals under different working states are collected, and multi-scale analysis based on sliding windows and sample quantiles is performed on the collected vibration signals under different working states to obtain a sequence after multi-scale analysis.

[0100] In some embodiments, in step S1, the specific method of multi-scale analysis based on sliding window and sample quantiles includes:

[0101] Step S11: Perform sliding window analysis on the sequence x(N)=x(1),x(2),...,x(N) of vibration signals under different working conditions, taking the window length, i.e. the scale factor, as τ, and obtaining the first sequence {x1,x2,...,x τ}, and the sample quantile of the first sequence is obtained as p1(τ);

[0102] Step S12: Move the sliding window forward by one data point to obtain the second sequence {x2, x3, ..., x τ+1}, the sample quantile of the second sequence is obtained as p2(τ);

[0103] Step S13, and so on, apply the method of step S12 to traverse the entire sequence of vibration signals until the last sequence is {x N-τ+1 ,x N-τ+2 ,...,x N}, find the sample quantile of the last sequence and get p N-τ+1 (τ);

[0104] Step S14, obtain the sequence {p1(τ), p2(τ), ..., p N-τ+1 (τ)}, the sequence {p1(τ),p2(τ),...,p N-τ+1 (τ)} is the sequence after the multi-scale analysis;

[0105] In some embodiments, in step S1, the window length, i.e., the scale factor τ, has a value range of: 1≤τ≤30;

[0106] In some embodiments, in step S1, the method for obtaining the sample quantile is:

[0107] p j (τ) = quantile(x j , x j+1 ...x j+τ-1 )1≤J≤N-τ+1

[0108] in,

[0109] p j (τ): the sample quantile of the jth sequence;

[0110] x j : The data of the jth sequence of the original sequence x(N);

[0111] J: The number of data points by which the sliding window moves forward;

[0112] N: the length of the original sequence x(N);

[0113] quantile(·): Calculates the sample quantile process of the sliding window.

[0114] In step S2, the permutation entropy of the sequence after the multi-scale analysis is obtained to obtain a multi-scale permutation entropy value, and the value of the scale permutation entropy is used as a feature vector of the vibration signal under different fault states.

[0115] In step S3, the feature vector is input into a classifier based on machine learning, so as to realize the recognition of vibration signals of different fault states and obtain a fault state recognition result.

[0116] In some embodiments, in step S3, the feature vector is input into a classifier based on machine learning to achieve recognition of vibration signals of different fault states. The method for obtaining the fault state recognition result includes:

[0117] The feature vector is input into a classifier based on a support vector machine, thereby realizing the recognition of vibration signals of different fault states and obtaining a fault state recognition result;

[0118] In some embodiments, in step S3, before the recognition of vibration signals of different fault states is achieved, the method further includes:

[0119] When constructing the SVM recognition model, in order to adaptively select the kernel function parameters and penalty factors of the SVM, an optimization algorithm is selected for optimization analysis;

[0120] In some embodiments, the optimization algorithm applies a PSO algorithm, ie, a particle swarm optimization algorithm.

[0121] In step S4, the rationality of the identification method is verified based on the fault state identification result.

[0122] Specifically:

[0123] (1) Test data

[0124] The test data used in this paper are selected from the public rolling bearing data set of Case Western Reserve University in the United States. The bearing model is SKF 6205-2RS JEM deep groove ball bearing. The bearings with different fault types are arranged by electrospark technology, and the vibration signal sampling frequency is 12KHz. In addition to the vibration signal under the normal state (denoted as NORM), there are three fault state samples including inner ring fault (IRF), outer ring fault (ORF), and rolling element fault (REF), and the data length is 2048. The vibration acceleration signals of the bearings under the four states are as follows: Figure 4 shown.

[0125] (2) Analysis of test results

[0126] Take 80 groups of vibration signals in normal state and three fault states, totaling 320 groups of data. In the fault identification process, select 40 groups of 80 groups of data in each of the four states as training sets (50%), and the remaining 40 groups as test sets (50%) to verify the effectiveness of the algorithm. In the process of fault identification using SVM, set label 1, label 2, label 3, label 4 to correspond to normal state, inner ring fault, rolling element fault and outer ring fault respectively.

[0127] Verification 1: Verify the performance of the conventional single-scale permutation entropy algorithm.

[0128] The permutation entropy is selected as the fault feature, and the feature vector is solved for the training set and the test set data, and the feature vector is input into the SVM recognition model optimized by PSO. The fault recognition accuracy of the vibration signal is 87.5% (140 / 160). The fault recognition results are as follows: Figure 5 As shown in the figure, 3 samples in label 2 are misclassified as label 4; 2 samples in label 3 are misclassified as label 2; 15 samples in label 4 are misclassified as label 2; a total of 20 samples cannot be accurately identified.

[0129] Verification 2: Verify the performance of the conventional multi-scale permutation entropy algorithm.

[0130] Multi-scale permutation entropy is selected as the fault feature, the feature vector is solved for the training set and test set data, and the feature vector is input into the SVM recognition model optimized by PSO.

[0131] Conventional multiscale decomposition can obtain corresponding multiscale sequences by selecting different scale factors, and then the permutation entropy values ​​of the coarse-grained sequences under different scale factors are used as multidimensional feature vectors to realize fault identification. Generally, the first few multiscale sequences are selected as multiscale sequences containing fault information and their permutation entropy values ​​are extracted. In order to verify the effectiveness of the proposed improved multiscale permutation entropy algorithm, the permutation entropy values ​​of 1-4 multiscale sequences are selected as feature vectors. The fault identification results are shown in Table 1. The values ​​corresponding to each state in the table are the number of correct identifications.

[0132] Table 1 Fault identification results analysis

[0133]

[0134] From Table 1, we can draw the following conclusions: (1) When the scale factor is 1, no coarse-graining is performed, that is, the original time series. At this time, the fault identification result is equivalent to the conventional single-scale permutation entropy; (2) The increase in the dimension of the feature vector has no direct relationship with the improvement of the fault identification accuracy. If the extracted feature vector is not sufficient to summarize the characteristic information of the vibration signal, the fault identification accuracy cannot be improved. For example, when the sequence permutation entropy with scale factors of 1 and 2 is selected as the feature vector, its fault identification accuracy is slightly lower than the single-scale permutation entropy. (3) When the sequence under the first four scale factors is selected and its permutation entropy value is obtained as the feature vector, the fault characteristic information of the vibration signal can be effectively summarized and different types of fault characteristic signals can be accurately identified.

[0135] Taking the permutation entropy values ​​of the coarse-grained sequences of the first two scales and the first three scales as the feature vector as an example, the vibration signal fault recognition results are as follows: Figure 6 , Figure 7 shown.

[0136] Verification 3: Verify the performance of the improved multi-scale permutation entropy algorithm proposed in this paper.

[0137] The improved multi-scale permutation entropy algorithm proposed in this paper takes the following two factors into consideration when selecting the scale factor: (1) The sample quantile is used to characterize the clustering of data at a certain point. Therefore, the scale factor should not be too small during the calculation process. This paper selects the initial scale factor as 10. (2) The sliding window can characterize the internal details between continuous data. Therefore, in the process of increasing the scale factor, this paper believes that the scale factor increase step size based on the conventional multi-scale permutation entropy is 1, and the increase process is too slow (τ = 1, 2, ..., n). Therefore, the multi-scale increase step size of the improved multi-scale permutation entropy is set to 20 (τ = 10, 30, ..., n).

[0138] The permutation entropy values ​​of the multi-scaled sequences with a scale factor of 10, a scale factor of 10, and a scale factor of 30 are selected as feature information to form the feature vector of the vibration signal and input into the classifier for fault identification. The results are as follows Figure 8 , Fig. 9 shown.

[0139] When the sequence permutation entropy value with a scale factor of 10 is selected as the fault feature vector, the fault recognition accuracy of the vibration signal is 92.5% (148 / 160). Among them, 1 sample in label 2 is misclassified as label 4; 5 samples in label 3 are misclassified as label 4; 5 samples in label 4 are misclassified as label 3, and 1 sample is misclassified as label 2; a total of 12 samples cannot be accurately identified.

[0140] By selecting the sequence permutation entropy values ​​with a scale factor of 10 and a scale factor of 30 as the fault feature vector, accurate classification of all state samples can be achieved.

[0141] Conventional multi-scale permutation entropy algorithms need to extract sequence permutation entropy values ​​at four scales as feature vectors to achieve state recognition of all vibration signals. In contrast, the improved multi-scale permutation entropy proposed in this paper only needs to extract sequence permutation entropy values ​​at the first two scales (scale factors are 10 and 30) to complete the fault state recognition process. Under the premise of ensuring fault recognition accuracy, the fault recognition efficiency is improved.

[0142] In the conventional multi-scale permutation entropy algorithm, as the scale factor increases, the length of the coarse-grained time series is [N / τ]; when the scale factor is 2, the length of the sequence is 1024; when the scale factor is 4, the length of the sequence is 512; and according to the suggestion of Bandt, the proposer of the permutation entropy algorithm, as the time series changes, the corresponding parameters should be adjusted in the permutation entropy calculation process. In the conventional multi-scale permutation entropy calculation process, the parameters are assumed to be unchanged for permutation entropy calculation, which theoretically reduces the rigor. In the improved multi-scale permutation entropy calculation process proposed in this paper, as the scale factor increases, the length of the coarse-grained time series is N-τ+1; when the scale factor is 10, the length of the sequence is 2039; when the scale factor is 30, the length of the sequence is 2019; the sequence length changes less, avoiding the problem of adjusting the permutation entropy calculation parameters due to changes in sequence length.

[0143] In summary, the solution proposed in the present invention can solve the defect of insufficient ability of conventional multi-scale permutation entropy fault diagnosis algorithm in fault feature extraction, and has a short calculation time, thereby improving the fault diagnosis efficiency.

[0144] The second aspect of the present invention discloses a vibration signal recognition system based on improved multi-scale permutation entropy. Fig.10is a structural diagram of a vibration signal recognition system based on improved multi-scale permutation entropy according to an embodiment of the present invention; Fig.10 As shown, the system 100 includes:

[0145] The first processing module 101 is configured to collect vibration signals under different working states, perform multi-scale analysis based on sliding windows and sample quantiles on the collected vibration signals under different working states, and obtain a sequence after multi-scale analysis;

[0146] The second processing module 102 is configured to obtain the permutation entropy of the sequence after the multi-scale analysis, obtain the multi-scale permutation entropy value, and use the multi-scale permutation entropy value as the feature vector of the vibration signal under different fault states;

[0147] The third processing module 103 is configured to input the feature vector into a classifier based on machine learning, so as to realize the recognition of vibration signals of different fault states and obtain a fault state recognition result;

[0148] The fourth processing module 104 is configured to verify the rationality of the identification method according to the fault state identification result.

[0149] According to the system of the second aspect of the present invention, the first processing module 101 is specifically configured as follows: a specific method of multi-scale analysis based on sliding windows and sample quantiles includes:

[0150] The vibration signal sequence x(N)=x(1),x(2),...,x(N) under different working conditions is subjected to sliding window analysis, and the window length, i.e., the scale factor, is taken as τ. The first sequence is obtained as {x1,x2,...,x τ}, and the sample quantile of the first sequence is obtained as p1(τ);

[0151] Step S12: Move the sliding window forward by one data point to obtain the second sequence {x2, x3, ..., x τ+1}, the sample quantile of the second sequence is obtained as p2(τ);

[0152] Similarly, the method of step S12 is applied to traverse the entire sequence of vibration signals until the last sequence {x N-τ+1 ,x N-τ+2 ,...,x N}, find the sample quantile of the last sequence and get p N-τ+1 (τ);

[0153] Get the sequence {p1(τ),p2(τ),...,p N-τ+1 (τ)}, the sequence {p1(τ),p2(τ),...,pN-τ+1 (τ)} is the sequence after the multi-scale analysis;

[0154] The window length, i.e., the scale factor τ, has a specific value range of: 1≤τ≤30;

[0155] like Figure 3 As shown, the method for taking sample quantiles is:

[0156] p j (τ) = quantile(x j , x j+1 ...x j+τ-1 )1≤J≤N-τ+1

[0157] in,

[0158] p j (τ): the sample quantile of the jth sequence;

[0159] x j : The data of the jth sequence of the original sequence x(N);

[0160] J: The number of data points by which the sliding window moves forward;

[0161] N: the length of the original sequence x(N);

[0162] quantile(·): Calculates the sample quantile process of the sliding window.

[0163] According to the system of the second aspect of the present invention, the third processing module 103 is specifically configured to input the feature vector into a classifier based on machine learning, so as to realize the recognition of vibration signals of different fault states, and the method for obtaining the fault state recognition result includes:

[0164] The feature vector is input into a classifier based on a support vector machine, thereby realizing the recognition of vibration signals of different fault states and obtaining a fault state recognition result;

[0165] In step S3, before the recognition of vibration signals of different fault states is realized, the method further includes:

[0166] When constructing the SVM recognition model, in order to adaptively select the kernel function parameters and penalty factors of the SVM, an optimization algorithm is selected for optimization analysis;

[0167] The optimization algorithm uses PSO algorithm, namely particle swarm optimization algorithm.

[0168] The third aspect of the present invention discloses an electronic device. The electronic device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of any one of the vibration signal recognition methods based on improved multi-scale permutation entropy disclosed in the first aspect of the present invention are implemented.

[0169] Fig.11 is a structural diagram of an electronic device according to an embodiment of the present invention, such as Fig.11 As shown, the electronic device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. Among them, the processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the electronic device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, near field communication (NFC) or other technologies. The display screen of the electronic device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the electronic device can be a touch layer covered on the display screen, or a button, a trackball or a touch pad set on the housing of the electronic device, or an external keyboard, touch pad or mouse, etc.

[0170] Those skilled in the art will understand that Fig.11 The structure shown in the figure is only a structural diagram of the part related to the technical solution of the present disclosure, and does not constitute a limitation on the electronic device to which the technical solution of the present application is applied. The specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0171] The fourth aspect of the present invention discloses a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the vibration signal recognition method based on improved multi-scale permutation entropy disclosed in any one of the first aspects of the present invention are implemented.

[0172] Please note that the technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification. The above embodiments only express several implementation methods of the present application, and their descriptions are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that for ordinary technicians in this field, without departing from the concept of the present application, several variations and improvements can be made, which all belong to the scope of protection of the present application. Therefore, the scope of protection of the patent in this application shall be based on the attached claims.

Claims

1. A vibration signal recognition method based on improved multi-scale permutation entropy, characterized in that: The method comprises: Step S1, collecting vibration signals under different working states, performing multi-scale analysis based on sliding windows and sample quantiles on the collected vibration signals under different working states, and obtaining a sequence after multi-scale analysis; Step S2, obtaining the permutation entropy of the sequence after the multi-scale analysis to obtain a multi-scale permutation entropy value, and using the value of the multi-scale permutation entropy as a feature vector of the vibration signal under different fault states; Step S3, inputting the feature vector into a classifier based on machine learning, thereby realizing recognition of vibration signals of different fault states and obtaining a fault state recognition result; Step S4: verifying the rationality of the identification method according to the fault state identification result; In step S1, the specific method of multi-scale analysis based on sliding window and sample quantile includes: Step S11: Perform sliding window analysis on the sequence x(N)=x(1),x(2),...,x(N) of vibration signals under different working conditions, taking the window length, i.e. the scale factor, as τ, and obtaining the first sequence {x1,x2,...,x τ }, and the sample quantile of the first sequence is obtained as p1(τ); Step S12: Move the sliding window forward by one data point to obtain the second sequence {x2, x3, ..., x τ+1 }, the sample quantile of the second sequence is obtained as p2(τ); Step S13, and so on, apply the method of step S12 to traverse the entire sequence of vibration signals until the last sequence is {x N-τ+1 ,x N-τ+2 ,...,x N }, the sample quantile of the last sequence is obtained as p N-τ+1 (τ); Step S14, obtain the sequence {p1(τ), p2(τ), ..., p N-τ+1 (τ)}, the sequence {p1(τ),p2(τ),...,p N-τ+1 (τ)} is the sequence after the multi-scale analysis.

2. A vibration signal recognition method based on improved multi-scale permutation entropy according to claim 1, characterized in that: In the step S1, the window length, that is, the scale factor τ, has a value range of 1≤τ≤30.

3. The vibration signal recognition method based on improved multi-scale permutation entropy according to claim 2 is characterized in that: In step S1, the method for obtaining the sample quantile is: p j (τ)=quantile(x j ,x j+1 ...x j+τ-1 ) 1≤J≤N-τ+1 in, p j (τ): the sample quantile of the jth sequence; x j : The data of the jth sequence of the original sequence x(N); J: The number of data points by which the sliding window moves forward; N: the length of the original sequence x(N); quantile(·): Calculates the sample quantile process function of a sliding window.

4. The vibration signal recognition method based on improved multi-scale permutation entropy according to claim 1, characterized in that: In step S3, the feature vector is input into a classifier based on machine learning to achieve recognition of vibration signals of different fault states. The method for obtaining the fault state recognition result includes: The feature vector is input into a classifier based on a support vector machine, thereby realizing the recognition of vibration signals of different fault states and obtaining a fault state recognition result.

5. The vibration signal recognition method based on improved multi-scale permutation entropy according to claim 4 is characterized in that: In step S3, before the recognition of vibration signals of different fault states is realized, the method further includes: Construct an SVM recognition model. When constructing the SVM recognition model, in order to adaptively select the kernel function parameters and penalty factors of the SVM, an optimization algorithm is selected for optimization analysis.

6. The vibration signal recognition method based on improved multi-scale permutation entropy according to claim 5, characterized in that: The optimization algorithm uses PSO algorithm, namely particle swarm optimization algorithm.

7. A vibration signal recognition system based on improved multi-scale permutation entropy, characterized in that: The system comprises: The first processing module is configured to collect vibration signals under different working states, perform multi-scale analysis based on sliding windows and sample quantiles on the collected vibration signals under different working states, and obtain a sequence after multi-scale analysis; The second processing module is configured to obtain the permutation entropy of the sequence after the multi-scale analysis, obtain the multi-scale permutation entropy value, and use the multi-scale permutation entropy value as the feature vector of the vibration signal under different fault states; A third processing module is configured to input the feature vector into a classifier based on machine learning, so as to realize recognition of vibration signals of different fault states and obtain a fault state recognition result; A fourth processing module is configured to verify the rationality of the identification system according to the fault state identification result; Among them, the multi-scale analysis based on sliding windows and sample quantiles specifically includes: The vibration signal sequence x(N)=x(1),x(2),...,x(N) under different working conditions is subjected to sliding window analysis, and the window length, i.e., the scale factor, is taken as τ. The first sequence is {x1,x2,...,x τ }, and the sample quantile of the first sequence is obtained as p1(τ); Move the sliding window forward by one data point to obtain the second sequence {x2,x3,...,x τ+1 }, the sample quantile of the second sequence is obtained as p2(τ); Similarly, the method of step S12 is applied to traverse the entire sequence of vibration signals until the last sequence {x N-τ+1 ,x N-τ+2 ,...,x N }, the sample quantile of the last sequence is obtained as p N-τ+1 (τ); Get the sequence {p1(τ),p2(τ),...,p N-τ+1 (τ)}, the sequence {p1(τ),p2(τ),...,p N-τ+1 (τ)} is the sequence after the multi-scale analysis.

8. An electronic device, characterized in that: The electronic device includes a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, the steps in the vibration signal recognition method based on improved multi-scale permutation entropy described in any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the steps in the vibration signal recognition method based on improved multi-scale permutation entropy according to any one of claims 1 to 6 are implemented.

Images