A bearing fault diagnosis method and system with improved initial clustering centers and weights

Abstract

The present invention provides a bearing fault diagnosis method and system with improved initial cluster centers and weights, wherein the method comprises: step 1: calculating the data dissimilarity between the first data points in the bearing data set; step 2: determining the second data point according to the data dissimilarity; step 3: determining the attribute weight of the first data point; step 4: calculating the weighted Euclidean distance according to the attribute weight, and dividing the first data point into a target cluster; step 5: determining the median of the first data point in the target cluster, and determining the subsequent cluster center; step 6: outputting the clustering result according to the initial cluster center and the subsequent cluster center; step 7: determining the bearing fault diagnosis result according to the clustering result. The bearing fault diagnosis method and system with improved initial cluster centers and weights of the present invention introduces the standard deviation coefficient method to assign weights, retains more information, and has a higher classification accuracy; uses the median instead of the mean iteration to eliminate the influence of outliers, and makes the diagnosis more accurate.

Description

Technical Field

[0001] The present invention relates to the technical field of bearing detection, and in particular to a bearing fault diagnosis method and system for improving initial cluster centers and weights. Background Art

[0002] At present, with the development of big data technology, mining valuable information from complex data is becoming more and more important. Unsupervised cluster analysis is one of the important technologies of data mining. The goal of clustering is to maximize the similarity of data objects in the same cluster and minimize the similarity of data objects in different clusters. It is widely used in mechanical fault diagnosis.

[0003] For a given data set, the traditional K-means clustering method determines the K value in advance, that is, the final number of classification clusters. At the same time, the initial class center is randomly determined, and the points are classified into the closest cluster by calculating the Euclidean distance. Then the center of each cluster is recalculated, and the classification operation is repeated until the cluster center no longer changes, and the classification is finally completed. It can be seen that the selection of a reasonable initial cluster center is the prerequisite for correct clustering. The more similar the selected initial cluster center is to the actual center, the better the clustering effect is, and the fewer iterations are required to achieve the specified accuracy; conversely, the worse the effect is and the lower the efficiency is. At the same time, based on the actual situation and the degree of difference between indicators, using the traditional K-means clustering method to analyze the data of all indicators may result in unsatisfactory clustering results. If some unimportant indicators are removed, information loss may occur.

[0004] The invention patent with application number: CN202211264745.X discloses a bearing fault diagnosis method based on a fuzzy width learning model, wherein the method includes: constructing an initial fuzzy width learning model based on a width learning system and a fuzzy system; training the output fuzzy width learning model through training set data to obtain a target fuzzy width learning model, wherein the training set data includes multiple bearing vibration signal data with fault type labels; using the target fuzzy width learning model to calculate the membership value of the vibration signal data of the bearing to be tested, and judging the fault type of the bearing to be tested according to the membership value. The above invention not only greatly shortens the learning time, but also uses this model to judge the fault type of the bearing to be tested, with strong robustness, fast diagnosis speed, and high fault diagnosis accuracy.

[0005] However, when the above-mentioned prior art judges the fault type of the bearing to be tested based on the membership value, it is necessary to determine the membership value interval corresponding to the fault type in advance. Historical data is needed to determine the membership value interval, but historical data is not always reliable. There may be outliers when determining the membership value interval based on historical data. In this case, the accuracy of diagnosis is low.

[0006] In view of this, there is an urgent need for a bearing fault diagnosis method and system with improved initial cluster centers and weights to at least solve the above-mentioned deficiencies. Summary of the invention

[0007] One of the purposes of the present invention is to provide a bearing fault diagnosis method with improved initial clustering centers and weights, calculate the data dissimilarity between the first data points of the acquired bearing data set, determine the attribute weight of the first data point according to the standard deviation coefficient method, calculate the weighted Euclidean distance according to the attribute weight, determine the initial clustering center according to the weighted Euclidean distance and the second data point, and introduce the standard deviation coefficient method to assign weights. While fully retaining information, the classification can also maintain a high accuracy rate; in addition, the median in the first data point is determined instead of the mean for iteration, which eliminates the influence of outliers and further improves the accuracy of diagnosis.

[0008] An embodiment of the present invention provides a bearing fault diagnosis method for improving initial cluster centers and weights, comprising:

[0009] Step 1: Obtain a preset bearing data set, and calculate the data dissimilarity between different first data points in the bearing data set;

[0010] Step 2: Determine the second data point as the initial cluster center based on data dissimilarity;

[0011] Step 3: Determine the attribute weight of the first data point according to the standard deviation coefficient method;

[0012] Step 4: Calculate the weighted Euclidean distance according to the attribute weights, and divide the first data point into the corresponding target cluster according to the weighted Euclidean distance and the second data point;

[0013] Step 5: Determine the median of the first data point in the target cluster, and determine the subsequent cluster center based on the median;

[0014] Step 6: Output the clustering results based on the initial cluster center and the subsequent cluster center;

[0015] Step 7: Determine the bearing fault diagnosis result based on the clustering results.

[0016] Preferably, step 1: obtaining a preset bearing data set, and calculating data dissimilarity between different first data points in the bearing data set, comprises:

[0017] Get the dissimilarity matrix;

[0018] Calculate the mean dissimilarity of the first data point and the overall dissimilarity of the bearing data set based on the bearing data set and the dissimilarity matrix;

[0019] The mean dissimilarity and the overall dissimilarity are taken together as the data dissimilarity.

[0020] Preferably, step 2: determining a second data point as an initial cluster center according to data dissimilarity, comprises:

[0021] Step 2.1: Analyze data dissimilarity and obtain mean dissimilarity;

[0022] Step 2.2: Obtain the third data point with the largest mean dissimilarity, use the third data point as the first initial cluster subcenter, and replace the mean dissimilarity of the third data point with 0 for updating, and obtain the fourth data point with the largest updated mean dissimilarity;

[0023] Step 2.3: Determine whether the target dissimilarity between the third data point and the fourth data point is greater than a preset dissimilarity threshold. If so, use the fourth data point as the second initial cluster subcenter and replace the mean dissimilarity of the fourth data point with 0 for updating;

[0024] Step 2.4: Determine whether the center sum value of the first initial cluster subcenter and the second initial cluster subcenter is equal to the preset number of cluster clusters. If so, output the first initial cluster subcenter and the second initial cluster subcenter and use them as the initial cluster center. Otherwise, continue to execute steps 2.3 to 2.4 until the center sum value is equal to the number of cluster clusters.

[0025] Preferably, step 3: determining the attribute weight of the first data point according to the standard deviation coefficient method includes:

[0026] Obtain the total number of first data points, and at the same time, obtain target data of the first data point of each attribute;

[0027] Calculate the attribute mean of the target data of the same attribute of the first data point;

[0028] The standard deviation of each attribute is determined according to the total number of first data points, the total number of attributes of the first data points, and the attribute mean, wherein the calculation formula of the standard deviation is as follows:

[0029]

[0030] Among them, x ik is the target data of the kth attribute of the ith first data point, is the attribute mean of the kth attribute, n is the total number of first data points, σ k is the standard deviation of the kth attribute;

[0031] The attribute weight is calculated based on the standard deviation and the total number of attributes of the first data point, where the attribute weight is:

[0032]

[0033] Where m is the total number of attributes of the first data point, ω k is the attribute weight of the kth attribute.

[0034] Preferably, step 4: calculating the weighted Euclidean distance according to the attribute weight, and dividing the first data point into a corresponding target cluster according to the weighted Euclidean distance and the second data point, comprises:

[0035] The weighted Euclidean distance between each first data point and the initial cluster center is calculated according to the attribute weight. The weighted Euclidean distance is expressed as follows:

[0036]

[0037] Among them, d ij For x ik With x jk The weighted Euclidean distance between ;

[0038] Based on the nearest distance principle, the target cluster to which the first data point belongs is determined according to the weighted Euclidean distance and the second data point.

[0039] Preferably, step 7: determining the bearing fault diagnosis result according to the clustering result includes:

[0040] According to the clustering results, construct fault detection templates corresponding to different clustering results;

[0041] Obtaining the operating parameters of the target bearing that needs to be diagnosed for a specific bearing fault;

[0042] Conduct spectrum analysis on operating parameters to obtain observation spectrum diagram;

[0043] Determine the target main frequency in the observed spectrum graph and the frequency position corresponding to the target main frequency;

[0044] Get the frequency multiple position of the target main frequency;

[0045] Determine the inspection point on the target bearing according to the frequency position and the frequency multiplication position;

[0046] Obtain the operating conditions of the target bearing;

[0047] Based on the fault detection template and working conditions, the maintenance type of the maintenance point is determined according to the target main frequency corresponding to the frequency position and the multiplier position;

[0048] Based on the preset maintenance strategy generation model, the maintenance strategy is generated according to the maintenance type, and corresponding maintenance is performed according to the maintenance strategy.

[0049] Preferably, obtaining the frequency multiple position of the target main frequency includes:

[0050] Obtain multiple bearing fault spectrum annotation records;

[0051] According to the bearing fault spectrum annotation records, train the frequency doubling position annotation model;

[0052] Get the bearing type of the target bearing;

[0053] Determine the labeled image according to the bearing type, target main frequency, observed spectrum diagram and frequency doubling position labeling model;

[0054] Parse the annotated image and obtain the octave positions.

[0055] Preferably, determining the labeled image according to the bearing type, the target main frequency, the observed spectrum diagram and the frequency doubling position labeling model includes:

[0056] Marking the target main frequency on the observed spectrum and characterizing it based on a preset first characterization rule to obtain a marked feature;

[0057] Input the annotation features into the frequency octave position annotation model to obtain a preprocessed image;

[0058] Characterizing the bearing type based on a preset second characterization rule to obtain a cleaning feature;

[0059] Generating rules and cleaning features according to the cleaning template to obtain the cleaning template;

[0060] The preprocessed image is cleaned using the cleaning template to obtain the labeled image.

[0061] A bearing fault diagnosis method for improving initial cluster centers and weights provided by an embodiment of the present invention also includes:

[0062] Step 8: Perform a supplementary crawling necessity analysis on the first data point for data dissimilarity calculation. If the analysis result of the supplementary crawling necessity analysis indicates that supplementary crawling is required, perform supplementary crawling;

[0063] Among them, step 8: performing a supplementary crawling necessity analysis on the first data point for data dissimilarity calculation, and if the analysis result of the supplementary crawling necessity analysis is that supplementary crawling is needed, performing supplementary crawling includes:

[0064] Get the data type of the first data point;

[0065] Determine, according to the data type, a first classification cluster capable of performing data dissimilarity matching;

[0066] Obtaining the data dimension of the first data point in each first classification cluster;

[0067] Determine, according to the data dimension, a second classification cluster capable of performing data dissimilarity matching, wherein the second classification cluster is a sub-classification cluster of the first classification cluster;

[0068] Determine the first crawled data of the full dimension of the third classification cluster according to the first classification cluster;

[0069] Determine, according to the second classification cluster corresponding to the first classification cluster, second crawled data of the missing dimension of the first classification cluster;

[0070] Integrate the first crawled data and the second crawled data to obtain target crawled data;

[0071] Obtain the crawling platform according to the classification information and dimension information of the target crawling data, and send a crawling request to the crawling platform for supplementary crawling.

[0072] An embodiment of the present invention provides a bearing fault diagnosis system with improved initial cluster centers and weights, comprising:

[0073] A data dissimilarity calculation subsystem, used for acquiring a preset bearing data set, and calculating data dissimilarity between different first data points in the bearing data set;

[0074] An initial cluster center determination subsystem, used for determining a second data point as an initial cluster center according to data dissimilarity;

[0075] An attribute weight determination subsystem, used to determine the attribute weight of the first data point according to a standard deviation coefficient method;

[0076] A clustering subsystem, for calculating a weighted Euclidean distance according to attribute weights, and for classifying the first data point into a corresponding target cluster according to the weighted Euclidean distance and the second data point;

[0077] A subsequent cluster center determination subsystem is used to determine the median of the first data point in the target cluster, and determine the subsequent cluster center according to the median;

[0078] The clustering result output subsystem is used to output the clustering results according to the initial cluster center and the subsequent cluster center;

[0079] The diagnosis result determination subsystem is used to determine the bearing fault diagnosis result based on the clustering result.

[0080] The beneficial effects of the present invention are:

[0081] The present invention calculates the data dissimilarity between the first data points of the acquired bearing data set, determines the attribute weight of the first data point according to the standard deviation coefficient method, calculates the weighted Euclidean distance according to the attribute weight, determines the initial cluster center according to the weighted Euclidean distance and the second data point, and introduces the standard deviation coefficient method to assign weights. While fully retaining information, the classification can also maintain a high accuracy rate. In addition, the median in the first data point is determined to replace the mean for iteration, which eliminates the influence of outliers and further improves the accuracy of diagnosis.

[0082] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings.

[0083] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0085] Figure 1 A schematic diagram of a bearing fault diagnosis method for improving initial cluster centers and weights in an embodiment of the present invention;

[0086] Figure 2 It is a schematic diagram of a bearing fault diagnosis system with improved initial cluster centers and weights in an embodiment of the present invention. DETAILED DESCRIPTION

[0087] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0088] The embodiment of the present invention provides a bearing fault diagnosis method with improved initial cluster centers and weights, such as Figure 1 As shown, including:

[0089] Step 1: Obtain a preset bearing data set, and calculate the data dissimilarity between different first data points in the bearing data set; wherein the bearing data set is: a bearing data set of Case Western Reserve University in the United States; wherein the first data point is: a single data in the bearing data set; and the data dissimilarity is: the difference or diversity between different first data points;

[0090] Step 2: According to the data dissimilarity, determine the second data point as the initial cluster center; wherein the second data point is: the first data point with the largest data dissimilarity;

[0091] Step 3: Determine the attribute weight of the first data point according to the standard deviation coefficient method; wherein the attribute weight is: the weight of the indicator determined according to the standard deviation coefficient method;

[0092] Step 4: Calculate the weighted Euclidean distance according to the attribute weights, and divide the first data point into the corresponding target cluster according to the weighted Euclidean distance and the second data point; wherein the weighted Euclidean distance is: the distance between the first data point and the initial cluster center; the target cluster is: the set of the first data points divided together;

[0093] Step 5: Determine the median of the first data point in the target cluster, and determine the subsequent cluster center based on the median;

[0094] Step 6: Output the clustering results based on the initial cluster center and the subsequent cluster center. When outputting the clustering results based on the initial cluster center and the subsequent cluster center, determine whether the initial cluster center and the subsequent cluster center are consistent; if so, output the clustering results; otherwise, replace the initial cluster center in step 4 with the subsequent cluster center, and output the clustering results;

[0095] Step 7: According to the clustering results, determine the bearing fault diagnosis results. The bearing fault diagnosis results are the integrated results of the fault types corresponding to the clustering data, and the fault types include: inner ring damage, outer ring damage, ball damage and normal.

[0096] The working principle and beneficial effects of the above technical solution are:

[0097] The present application calculates the data dissimilarity between the first data points of the bearing data set, determines the attribute weight of the first data point according to the standard deviation coefficient method, calculates the weighted Euclidean distance according to the attribute weight, determines the initial cluster center according to the weighted Euclidean distance and the second data point, and introduces the standard deviation coefficient method to assign weights. While fully retaining information, the classification can also maintain a high accuracy rate; in addition, the median of the first data point is determined instead of the mean for iteration, which eliminates the influence of outliers and further improves the accuracy of diagnosis.

[0098] In specific applications, the system obtains the monitoring signal of the bearing that needs to be detected for bearing fault, and then outputs the fault diagnosis result.

[0099] In one embodiment, step 1: obtaining a preset bearing data set, and calculating data dissimilarity between different first data points in the bearing data set, includes:

[0100] Obtaining a dissimilarity matrix; wherein the dissimilarity matrix stores the proximity between multiple data;

[0101] Calculate the mean dissimilarity of the first data point and the overall dissimilarity of the bearing data set based on the bearing data set and the dissimilarity matrix;

[0102] The specific proximity is:

[0103]

[0104] The mean difference is specifically:

[0105]

[0106] The overall differences are as follows:

[0107]

[0108] Among them, dis(x i ,x j ) is x i With x j The difference between i is the first data point of i, x j is the jth first data point, x ik is the target data of the kth attribute of the ith first data point, x jk is the target data of the kth attribute of the jth first data point, and m is the total number of attributes of the first data point; Adis(x i ) is x i The mean dissimilarity of , Tdis is the total dissimilarity, dis(x i -x j ) is x i With x j The distance between them, n is the total number of first data points in the bearing data set;

[0109] The mean dissimilarity and the overall dissimilarity are taken together as the data dissimilarity.

[0110] The working principle and beneficial effects of the above technical solution are:

[0111] This application introduces a dissimilarity matrix, determines the mean dissimilarity and the overall dissimilarity according to the bearing data set and the dissimilarity matrix, and uses the mean dissimilarity and the overall dissimilarity together as data dissimilarity, thereby improving the comprehensiveness of dissimilarity judgment.

[0112] In one embodiment, step 2: determining a second data point as an initial cluster center according to data dissimilarity, comprises:

[0113] Step 2.1: Analyze data dissimilarity and obtain mean dissimilarity;

[0114] Step 2.2: Obtain the third data point with the largest mean dissimilarity, use the third data point as the first initial cluster subcenter, and replace the mean dissimilarity of the third data point with 0 for updating, and obtain the fourth data point with the largest updated mean dissimilarity;

[0115] Step 2.3: Determine whether the target dissimilarity between the third data point and the fourth data point is greater than a preset dissimilarity threshold. If so, use the fourth data point as the second initial cluster subcenter and replace the mean dissimilarity of the fourth data point with 0 for updating; wherein the preset dissimilarity threshold is manually preset;

[0116] Step 2.4: Determine whether the center sum value of the first initial cluster subcenter and the second initial cluster subcenter is equal to the preset number of cluster clusters. If so, output the first initial cluster subcenter and the second initial cluster subcenter and use them as the initial cluster center. Otherwise, continue to execute steps 2.3 to 2.4 until the center sum value is equal to the number of cluster clusters. The center sum value is the sum of the number of the first initial cluster subcenter and the number of the second initial cluster subcenter; the preset number of cluster clusters is manually input in advance, for example: 4.

[0117] The working principle and beneficial effects of the above technical solution are:

[0118] This application introduces mean dissimilarity, determines and updates the third data point with the largest mean dissimilarity, obtains the updated initial cluster center with the largest mean dissimilarity, and the process of obtaining the initial cluster center is more reasonable.

[0119] In one embodiment, step 3: determining the attribute weight of the first data point according to the standard deviation coefficient method includes:

[0120] Obtain the total number of first data points, and at the same time, obtain target data of the first data point of each attribute;

[0121] Calculate the attribute mean of the target data of the same attribute of the first data point;

[0122] The standard deviation of each attribute is determined according to the total number of first data points, the total number of attributes of the first data points, and the attribute mean, wherein the calculation formula of the standard deviation is as follows:

[0123]

[0124] Among them, x ik is the target data of the kth attribute of the ith first data point, is the attribute mean of the kth attribute, n is the total number of first data points, σ k is the standard deviation of the kth attribute;

[0125] The attribute weight is calculated based on the standard deviation and the total number of attributes of the first data point, where the attribute weight is:

[0126]

[0127] Where m is the total number of attributes of the first data point, ωk is the attribute weight of the first data point of the kth attribute.

[0128] The working principle and beneficial effects of the above technical solution are:

[0129] The attributes of this application are weighted using the standard deviation coefficient method, and the classification can maintain a higher accuracy while fully retaining the information.

[0130] In one embodiment, step 4: calculating the weighted Euclidean distance according to the attribute weight, and dividing the first data point into a corresponding target cluster according to the weighted Euclidean distance and the second data point, includes:

[0131] The weighted Euclidean distance between each first data point and the initial cluster center is calculated according to the attribute weight. The weighted Euclidean distance is expressed as follows:

[0132]

[0133] Among them, d ij For x ik With x jk The weighted Euclidean distance between ;

[0134] Based on the nearest distance principle, the target cluster to which the first data point belongs is determined according to the weighted Euclidean distance and the second data point.

[0135] The working principle and beneficial effects of the above technical solution are:

[0136] The present application introduces the weighted Euclidean distance and the nearest distance principle to determine the target cluster to which the first data point belongs, thereby improving the accuracy and rationality of classification.

[0137] In one embodiment, step 7: determining the bearing fault diagnosis result according to the clustering result includes:

[0138] According to the clustering results, fault detection templates corresponding to different clustering results are constructed; wherein the fault detection templates corresponding to different clustering results are: fault detection templates corresponding to different fault types, such as: a fault detection template that constrains only bearing inner ring damage detection, and another example: a fault detection template that constrains only ball bearing damage detection;

[0139] Obtaining the operating parameters of the target bearing for which a bearing fault diagnosis is required; wherein the operating parameters are: a vibration signal of the bearing operation;

[0140] Performing spectrum analysis on the operating parameters to obtain an observed spectrum diagram; wherein the spectrum analysis is: fast Fourier transform; and the observed spectrum diagram is: a spectrum diagram of the operating parameters;

[0141] Determine the target main frequency in the observed spectrum graph and the frequency position corresponding to the target main frequency;

[0142] Obtain the frequency multiple position of the target main frequency; wherein the frequency multiple position is: the position where the peak value in the spectrum graph is an integer multiple of the target main frequency;

[0143] According to the frequency position and the frequency multiplication position, the inspection point on the target bearing is determined; wherein the inspection point is: the position where the fault may exist on the target bearing;

[0144] Obtaining the operating conditions of the target bearing; wherein the operating conditions include: bearing type and working condition, etc.;

[0145] Based on the fault detection template and the working conditions, the maintenance type of the maintenance point is determined according to the target main frequency corresponding to the frequency position and the frequency multiplication position; when determining the maintenance type of the maintenance point, the target main frequency corresponding to the frequency position and the frequency multiplication position can be compared with the main frequency corresponding to the working conditions listed in the fault detection template. If the frequency match is met, it is determined that the fault type corresponding to the fault detection template may exist;

[0146] Based on the preset maintenance strategy generation model, the maintenance strategy is generated according to the maintenance type, and the corresponding maintenance is carried out according to the maintenance strategy. Among them, the maintenance strategy is: what kind of maintenance tools need to be brought and how to perform maintenance;

[0147] Among them, according to the frequency position and the frequency multiplication position, the inspection point on the target bearing is determined, including:

[0148] Obtaining observed spectrum features according to the frequency position and the frequency octave position; wherein the observed spectrum features are: spectrum coefficients of the observed spectrum graph at the frequency position and the frequency octave position, such as: Fourier transform coefficients;

[0149] Based on the preset critical sorting algorithm, the processing results of the observed spectrum features processed by the BN algorithm are critically sorted, and the processing results are selected based on the sorting and screening rules to obtain the selected features; wherein the critical sorting algorithm is: random forest algorithm; the BN algorithm is a normalization algorithm; the sorting and screening rules are: screen out the first number of processing results;

[0150] According to the ranking of the selected features in the processing results, the feature dispersion in the feature distribution is determined; wherein, the higher the ranking, the lower the feature dispersion, and the lower the feature dispersion, the denser the selection of the selected features;

[0151] Determine the feature extraction frequency according to the feature discreteness and the selected feature distribution; wherein the feature extraction frequency is the extraction position of the supplementary extraction in the observed spectrum graph;

[0152] Determine the supplementary feature according to the feature extraction frequency; wherein the supplementary feature is: the spectrum coefficient corresponding to the extraction position in the observed spectrum graph;

[0153] Determine the fusion feature distribution according to the observed spectrum features and the supplementary features; wherein the fusion feature distribution is: the spectrum feature distribution of the fault position including the frequency position information and the spectrum coefficient;

[0154] Determine the maintenance points based on the fused feature distribution.

[0155] The working principle and beneficial effects of the above technical solution are:

[0156] In general, the main frequency of the vibration signal can be obtained based on the operating parameters of the target bearing. The main frequency vibration signal corresponds to the fault type and cause. The above main frequency can provide some clues, but it is impossible to directly determine the fault type and cause. Therefore, this application draws an observation spectrum diagram based on the operating parameters of the target bearing, and determines the frequency position of the target main frequency and the frequency multiple position of the target main frequency in the observation spectrum diagram. The maintenance point is determined according to the frequency position and the frequency multiple position. When determining, the observation spectrum characteristics are determined according to the frequency position and the frequency multiple position, and then the criticality sorting algorithm and the BN algorithm are introduced to determine the processing result sequence of the normalized processing. According to the criticality of the processing result, the feature discreteness in the feature distribution is determined, and then the feature extraction frequency of the supplementary feature is determined. After supplementary sampling, the maintenance point is determined according to the fused feature distribution, which further improves the accuracy of the maintenance point determination process. The operating conditions and fault detection templates are introduced to determine the maintenance type of the maintenance point, and the fault detection efficiency is higher and more accurate. The maintenance strategy generation model is introduced to generate the maintenance strategy corresponding to the maintenance type, which improves the suitability of the maintenance.

[0157] In one embodiment, obtaining the frequency multiple position of the target main frequency includes:

[0158] Acquire multiple bearing fault spectrum annotation records; wherein the bearing fault spectrum annotation record is: a record of the frequency multiple position corresponding to the main frequency of the spectrum annotation of the bearing fault signal manually analyzed;

[0159] According to the bearing fault spectrum annotation records, the frequency multiple position annotation model is trained; the frequency multiple position annotation model is an intelligent model that quickly analyzes the frequency multiple position of the main frequency based on the known main frequency;

[0160] Obtain the bearing type of the target bearing; wherein the bearing type is: the type of bearing;

[0161] Determine the labeled image according to the bearing type, the target main frequency, the observed spectrum diagram and the frequency multiple position labeling model; wherein the labeled image is: the observed spectrum diagram of the frequency multiple position labeled by the frequency multiple position labeling model;

[0162] Parse the annotated image and obtain the octave positions.

[0163] The working principle and beneficial effects of the above technical solution are:

[0164] This application introduces a bearing fault spectrum annotation record to train a frequency dot position annotation model. In addition, the bearing type is introduced as a conditional variable. The annotation image is determined according to the bearing type, target main frequency, observed spectrum diagram and frequency dot position annotation model. The annotation image is analyzed to obtain the annotation image according to the bearing type, target main frequency, observed spectrum diagram and frequency dot position annotation model. The process of determining the frequency dot position is more intelligent and more efficient.

[0165] In one embodiment, determining the labeled image according to the bearing type, the target main frequency, the observed spectrum diagram and the frequency doubling position labeling model includes:

[0166] The target main frequency is marked on the observed spectrum graph and characterized based on a preset first characterization rule to obtain a marked feature; wherein the preset first characterization rule is: a rule for characterizing an image; the marked feature is: where the target main frequency is marked on the observed spectrum graph and how much the target main frequency is, etc.;

[0167] Input the annotation features into the frequency octave position annotation model to obtain a preprocessed image; wherein the preprocessed image is: an image of the frequency octave position annotated based on the annotation features and output by the frequency octave position annotation model;

[0168] Characterize the bearing type based on the preset second characterization rule to obtain the cleaning feature; wherein the second characterization rule is: a rule for semantically characterizing the semantics of the bearing type; the cleaning feature is: what semantics the bearing type corresponds to;

[0169] According to the cleaning template generation rule and the cleaning feature, a cleaning template is obtained; wherein the cleaning template generation rule is: what kind of cleaning template is generated based on what kind of cleaning feature, for example: constraining the semantics of what kind of bearing type to generate a template for cleaning what kind of frequency multiplication data;

[0170] The preprocessed image is cleaned using the cleaning template to obtain the labeled image.

[0171] The working principle and beneficial effects of the above technical solution are:

[0172] This application obtains annotation features and inputs a frequency dot position annotation model to determine an annotated image. At the same time, cleaning features and cleaning template generation rules are introduced to determine a cleaning template, and the cleaning template is used to clean the preprocessed image to obtain an annotated image, thereby further improving the accuracy of the annotation of the annotated image.

[0173] The embodiment of the present invention provides a bearing fault diagnosis method for improving initial cluster centers and weights, further comprising:

[0174] Step 8: Perform a supplementary crawling necessity analysis on the first data point for data dissimilarity calculation. If the analysis result of the supplementary crawling necessity analysis indicates that supplementary crawling is required, perform supplementary crawling;

[0175] Among them, step 8: performing a supplementary crawling necessity analysis on the first data point for data dissimilarity calculation, and if the analysis result of the supplementary crawling necessity analysis is that supplementary crawling is needed, performing supplementary crawling includes:

[0176] Get the data type of the first data point;

[0177] According to the data type, a first classification cluster capable of data dissimilarity matching is determined; wherein the first classification cluster capable of data dissimilarity matching is: a set of first data point combinations with the same data type;

[0178] Obtaining the data dimension of the first data point in each first classification cluster;

[0179] According to the data dimension, a second classification cluster capable of data dissimilarity matching is determined, wherein the second classification cluster is a sub-classification cluster of the first classification cluster; wherein the second classification cluster capable of data dissimilarity matching is: a set of first data points with the same data dimension in the first classification cluster;

[0180] Determine the first crawled data of all dimensions of the third classification cluster according to the first classification cluster; wherein the first crawled data is: crawled data of all dimensions of a certain data type;

[0181] According to the second classification cluster corresponding to the first classification cluster, determine the second crawled data of the missing dimension of the first classification cluster; wherein the second crawled data is: crawled data of a certain dimension of a certain data type;

[0182] Integrate the first crawled data and the second crawled data to obtain target crawled data;

[0183] According to the classification information and dimension information of the target crawled data, the crawling platform is obtained, and a crawling request is sent to the crawling platform for additional crawling. Among them, the classification information is: the relevant information of what type of data; the dimension information is: the relevant information of what data dimension; the crawling platform is: the big data platform for sharing bearing data.

[0184] The working principle and beneficial effects of the above technical solution are:

[0185] When performing data dissimilarity analysis based on the bearing data set, the analyzed data may be incomplete, which in turn affects the subsequent clustering. Therefore, the present application introduces data types and data dimensions to perform a necessity analysis for supplementary crawling of the first data point. During the analysis, all dimensional information of the missing data type and the missing dimensional information in the first classification cluster are determined, and the crawling platform is determined to perform corresponding crawling based on the classification information and dimensional information, which improves the targeted nature of data crawling and avoids the problem of inaccurate subsequent diagnosis caused by incomplete comparison data of data dissimilarity.

[0186] The embodiment of the present invention provides a bearing fault diagnosis system with improved initial cluster centers and weights, such as Figure 2 As shown, including:

[0187] The data dissimilarity calculation subsystem 1 is used to obtain a preset bearing data set and calculate the data dissimilarity between different first data points in the bearing data set;

[0188] An initial cluster center determination subsystem 2, used for determining a second data point as an initial cluster center according to data dissimilarity;

[0189] The attribute weight determination subsystem 3 is used to determine the attribute weight of the first data point according to the standard deviation coefficient method;

[0190] A clustering subsystem 4, for calculating a weighted Euclidean distance according to an attribute weight, and classifying the first data point into a corresponding target cluster according to the weighted Euclidean distance and the second data point;

[0191] A subsequent cluster center determination subsystem 5 is used to determine the median of the first data point in the target cluster, and determine the subsequent cluster center according to the median;

[0192] The clustering result output subsystem 6 is used to output the clustering result according to the initial cluster center and the subsequent cluster center;

[0193] The diagnosis result determination subsystem 7 is used to determine the bearing fault diagnosis result according to the clustering result.

[0194] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A bearing fault diagnosis method with improved initial cluster centers and weights, characterized in that: include: Step 1: Obtain a preset bearing data set, and calculate the data dissimilarity between different first data points in the bearing data set; Step 2: Determine the second data point as the initial cluster center based on data dissimilarity; Step 3: Determine the attribute weight of the first data point according to the standard deviation coefficient method; Step 4: Calculate the weighted Euclidean distance according to the attribute weights, and divide the first data point into the corresponding target cluster according to the weighted Euclidean distance and the second data point; Step 5: Determine the median of the first data point in the target cluster, and determine the subsequent cluster center based on the median; Step 6: Output the clustering results based on the initial cluster center and the subsequent cluster center; Step 7: Determine the bearing fault diagnosis result based on the clustering result; The step 3: determining the attribute weight of the first data point according to the standard deviation coefficient method includes: Obtain the total number of first data points, and at the same time, obtain target data of the first data point of each attribute; Calculate the attribute mean of the target data of the same attribute of the first data point; The standard deviation of each attribute is determined according to the total number of first data points, the total number of attributes of the first data points, and the attribute mean, wherein the calculation formula of the standard deviation is as follows: Among them, x ik is the target data of the kth attribute of the ith first data point, is the attribute mean of the kth attribute, n is the total number of first data points, σ k is the standard deviation of the kth attribute; The attribute weight is calculated based on the standard deviation and the total number of attributes of the first data point, where the attribute weight is: Where m is the total number of attributes of the first data point, ω k is the attribute weight of the kth attribute; The step 4: calculating the weighted Euclidean distance according to the attribute weight, and dividing the first data point into a corresponding target cluster according to the weighted Euclidean distance and the second data point, includes: The weighted Euclidean distance between each first data point and the initial cluster center is calculated according to the attribute weight. The weighted Euclidean distance is expressed as follows: Among them, d ij For x ik With x jk The weighted Euclidean distance between ; Based on the nearest distance principle, the target cluster to which the first data point belongs is determined according to the weighted Euclidean distance and the second data point.

2. A bearing fault diagnosis method with improved initial cluster centers and weights as claimed in claim 1, characterized in that: Step 1: Obtain a preset bearing data set, and calculate the data dissimilarity between different first data points in the bearing data set, including: Get the dissimilarity matrix; Calculate the mean dissimilarity of the first data point and the overall dissimilarity of the bearing data set based on the bearing data set and the dissimilarity matrix; The mean dissimilarity and the overall dissimilarity are taken together as the data dissimilarity.

3. A bearing fault diagnosis method with improved initial cluster centers and weights as claimed in claim 1, characterized in that: Step 2: Determine the second data point as the initial cluster center based on data dissimilarity, including: Step 2.1: Analyze data dissimilarity and obtain mean dissimilarity; Step 2.2: Obtain the third data point with the largest mean dissimilarity, use the third data point as the first initial cluster subcenter, and replace the mean dissimilarity of the third data point with 0 for updating, and obtain the fourth data point with the largest updated mean dissimilarity; Step 2.3: Determine whether the target dissimilarity between the third data point and the fourth data point is greater than a preset dissimilarity threshold. If so, use the fourth data point as the second initial cluster subcenter and replace the mean dissimilarity of the fourth data point with 0 for updating; Step 2.4: Determine whether the center sum value of the first initial cluster subcenter and the second initial cluster subcenter is equal to the preset number of cluster clusters. If so, output the first initial cluster subcenter and the second initial cluster subcenter and use them as the initial cluster center. Otherwise, continue to execute steps 2.3 to 2.4 until the center sum value is equal to the number of cluster clusters.

4. A bearing fault diagnosis method with improved initial cluster centers and weights as claimed in claim 1, characterized in that: Step 7: According to the clustering results, determine the bearing fault diagnosis results, including: According to the clustering results, construct fault detection templates corresponding to different clustering results; Obtaining the operating parameters of the target bearing that needs to be diagnosed for a specific bearing fault; Conduct spectrum analysis on operating parameters to obtain observation spectrum diagram; Determine the target main frequency in the observed spectrum graph and the frequency position corresponding to the target main frequency; Get the frequency multiple position of the target main frequency; Determine the inspection point on the target bearing according to the frequency position and the frequency multiplication position; Obtain the operating conditions of the target bearing; Based on the fault detection template and working conditions, the maintenance type of the maintenance point is determined according to the target main frequency corresponding to the frequency position and the multiplier position; Based on the preset maintenance strategy generation model, the maintenance strategy is generated according to the maintenance type, and corresponding maintenance is performed according to the maintenance strategy.

5. A bearing fault diagnosis method with improved initial cluster centers and weights as claimed in claim 4, characterized in that: Get the frequency multiple position of the target main frequency, including: Obtain multiple bearing fault spectrum annotation records; According to the bearing fault spectrum annotation records, train the frequency doubling position annotation model; Get the bearing type of the target bearing; Determine the labeled image according to the bearing type, target main frequency, observed spectrum diagram and frequency doubling position labeling model; Parse the annotated image and obtain the octave positions.

6. A bearing fault diagnosis method with improved initial cluster centers and weights as claimed in claim 5, characterized in that: According to the bearing type, target main frequency, observed spectrum and frequency doubling position annotation model, the annotation image is determined, including: Marking the target main frequency on the observed spectrum and characterizing it based on a preset first characterization rule to obtain a marked feature; Input the annotation features into the frequency octave position annotation model to obtain a preprocessed image; Characterizing the bearing type based on a preset second characterization rule to obtain a cleaning feature; Generating rules and cleaning features according to the cleaning template to obtain the cleaning template; The preprocessed image is cleaned using the cleaning template to obtain the labeled image.

7. A bearing fault diagnosis method with improved initial cluster centers and weights as claimed in claim 1, characterized in that: Also includes: Step 8: Perform a supplementary crawling necessity analysis on the first data point for data dissimilarity calculation. If the analysis result of the supplementary crawling necessity analysis indicates that supplementary crawling is required, perform supplementary crawling; Among them, step 8: performing a supplementary crawling necessity analysis on the first data point for data dissimilarity calculation, and if the analysis result of the supplementary crawling necessity analysis is that supplementary crawling is needed, performing supplementary crawling includes: Get the data type of the first data point; Determine, according to the data type, a first classification cluster capable of performing data dissimilarity matching; Obtaining the data dimension of the first data point in each first classification cluster; Determine, according to the data dimension, a second classification cluster capable of performing data dissimilarity matching, wherein the second classification cluster is a sub-classification cluster of the first classification cluster; Determine the first crawled data of the full dimension of the third classification cluster according to the first classification cluster; Determine, according to the second classification cluster corresponding to the first classification cluster, second crawled data of the missing dimension of the first classification cluster; Integrate the first crawled data and the second crawled data to obtain target crawled data; Obtain the crawling platform according to the classification information and dimension information of the target crawling data, and send a crawling request to the crawling platform for supplementary crawling.

8. A bearing fault diagnosis system with improved initial cluster centers and weights, characterized in that: include: A data dissimilarity calculation subsystem, used for acquiring a preset bearing data set, and calculating data dissimilarity between different first data points in the bearing data set; An initial cluster center determination subsystem, used for determining a second data point as an initial cluster center according to data dissimilarity; An attribute weight determination subsystem, used to determine the attribute weight of the first data point according to a standard deviation coefficient method; A clustering subsystem, for calculating a weighted Euclidean distance according to attribute weights, and for classifying the first data point into a corresponding target cluster according to the weighted Euclidean distance and the second data point; A subsequent cluster center determination subsystem is used to determine the median of the first data point in the target cluster, and determine the subsequent cluster center according to the median; The clustering result output subsystem is used to output the clustering results according to the initial cluster center and the subsequent cluster center; A diagnosis result determination subsystem is used to determine the bearing fault diagnosis result according to the clustering result; The attribute weight determination subsystem determines the attribute weight of the first data point according to the standard deviation coefficient method, including: Obtain the total number of first data points, and at the same time, obtain target data of the first data point of each attribute; Calculate the attribute mean of the target data of the same attribute of the first data point; The standard deviation of each attribute is determined according to the total number of first data points, the total number of attributes of the first data points, and the attribute mean, wherein the calculation formula of the standard deviation is as follows: Among them, x ik is the target data of the kth attribute of the ith first data point, is the attribute mean of the kth attribute, n is the total number of first data points, σ k is the standard deviation of the kth attribute; The attribute weight is calculated based on the standard deviation and the total number of attributes of the first data point, where the attribute weight is: Where m is the total number of attributes of the first data point, ω k is the attribute weight of the kth attribute; The clustering subsystem calculates the weighted Euclidean distance according to the attribute weights, and divides the first data point into a corresponding target cluster according to the weighted Euclidean distance and the second data point, including: The weighted Euclidean distance between each first data point and the initial cluster center is calculated according to the attribute weight. The weighted Euclidean distance is expressed as follows: Among them, d ij For x ik With x jk The weighted Euclidean distance between ; Based on the nearest distance principle, the target cluster to which the first data point belongs is determined according to the weighted Euclidean distance and the second data point.

Images