Clustering method based on time sequence and life cycle analysis method of bearing

Abstract

The invention discloses a clustering method based on a time sequence and a life cycle analysis method of a bearing. The clustering method in one embodiment comprises the following steps: respectively carrying out phase-space reconstruction on received operation data sequences of a plurality of devices and generating phase-space matrixes in one-to-one correspondence with the plurality of devices; decomposing each phase space matrix and generating an orthogonal matrix and a dynamic matrix; the orthogonal matrixes of all the operation data sequences are utilized to form a Stepepol manifold, the multiple devices are clustered according to the Stepepol manifold, and a first clustering result is generated; and based on the first clustering result, performing clustering in combination with the dynamic matrix of each operation data sequence, and generating a second clustering result of the plurality of devices. According to the embodiment provided by the invention, the phase-space matrixes are generated by performing phase-space reconstruction on the multiple pieces of equipment operation data based on the time sequence, and comprehensive clustering is performed by using the orthogonal matrixes and the dynamic matrixes obtained by decomposing the phase-space matrixes, so that the clustering accuracy is effectively improved.

Description

Technical Field

[0001] This invention relates to the field of signal processing technology, and in particular to a time series-based clustering method and a bearing lifecycle analysis method. Background Technology

[0002] In related technologies, the paper "Improved time series clustering based on new geometric frameworks" proposes a method for clustering time series. It reconstructs the phase space of multiple time series, performs QR decomposition on the trajectory matrix of the reconstructed phase space, selects orthogonal matrices Q to form a Steifel manifold, calculates the point distance and mean in the Steifel manifold, and uses the umap and HDBSCAN methods for clustering to obtain the clustering results.

[0003] However, in practical applications, there are issues that need to be improved in clustering results. How to improve the clustering accuracy of time-series device data has become a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] To address at least one of the aforementioned problems, a first embodiment of the present invention provides a time-series-based clustering method, comprising:

[0005] The phase space of the received operating data sequences from multiple devices is reconstructed to generate a phase space matrix corresponding to each of the multiple devices.

[0006] Each phase space matrix is ​​decomposed and an orthogonal matrix and a dynamic matrix are generated;

[0007] A Steiffel manifold is constructed using the orthogonal matrices of each running data sequence, and the multiple devices are clustered according to the Steiffel manifold to generate a first clustering result;

[0008] Based on the first clustering result, the dynamic matrix of each running data sequence is combined to perform clustering and generate the second clustering result of the multiple devices.

[0009] For example, in some embodiments of the clustering method provided in this application, the step of using the orthogonal matrices of each running data sequence to form a Stevorl manifold, and clustering the multiple devices according to the Stevorl manifold to generate a first clustering result further includes:

[0010] A Steifel manifold is constructed using the orthogonal matrices of each running data sequence, wherein the Steifel manifold includes the first point representing each orthogonal matrix;

[0011] Calculate and obtain the point spacing between each first point, and the average point spacing between any two points among the first points;

[0012] Based on the point spacing and average spacing of each first point, the multiple devices are clustered using a pre-set clustering operation to generate the first clustering result.

[0013] For example, in some embodiments of the clustering method provided in this application, the step of clustering based on the first clustering result and combining the dynamic matrix of each running data sequence to generate a second clustering result for the multiple devices further includes:

[0014] The first clustering result includes multiple clustering sub-results, each of which stretches the dynamic matrix of the running data sequence corresponding to the clustering sub-result into a first vector;

[0015] Each running data sequence's first vector is projected onto a unit sphere, the unit sphere including a second point representing each dynamic matrix;

[0016] Calculate and obtain the point spacing between each second point, and the average point spacing between any two points among the second points;

[0017] Based on the point spacing and average spacing of each second point, the multiple devices are clustered using a preset clustering operation to generate the second clustering result.

[0018] For example, in some embodiments of the clustering method provided in this application, the step of clustering based on the first clustering result and combining the dynamic matrix of each running data sequence to generate a second clustering result for the multiple devices further includes:

[0019] Each dynamic matrix of the running data sequence is stretched into a second vector;

[0020] The second vector of each running data sequence is projected onto a unit sphere, which includes a third point representing each dynamic matrix;

[0021] Calculate and obtain the point spacing between each third point, and the average point spacing between any two points among the third points;

[0022] Based on the point spacing and average spacing of each third point, the multiple devices are clustered using a preset clustering operation to generate the third clustering result;

[0023] The second clustering result is generated by combining the first clustering result and the third clustering result according to the preset weight ratio.

[0024] For example, in the clustering method provided in some embodiments of this application, the step of decomposing each phase space matrix and generating an orthogonal matrix and a dynamic matrix further includes:

[0025] Each phase space matrix is ​​decomposed into a first orthogonal matrix and a first dynamic matrix.

[0026] For example, in the clustering method provided in some embodiments of this application, the step of decomposing each phase space matrix and generating an orthogonal matrix and a dynamic matrix further includes:

[0027] The phase space matrices are decomposed into second orthogonal matrices and second dynamic matrices respectively.

[0028] For example, in the clustering method provided in some embodiments of this application, when the device includes multiple operating data sequences, the step of reconstructing the phase space of the received operating data sequences of the multiple devices and generating a phase space matrix corresponding one-to-one with the multiple devices further includes:

[0029] The phase space is reconstructed for multiple operating data sequences of the device, and an intermediate phase space matrix corresponding to each operating data sequence is generated.

[0030] The phase space matrix of the device is generated by splicing together multiple intermediate phase space matrices of the device.

[0031] For example, in some embodiments of the clustering method provided in this application, the same phase space reconstruction parameters are used to reconstruct the phase space of multiple operating data sequences of the device.

[0032] For example, in some embodiments of the clustering method provided in this application, the step of reconstructing the phase space of the received operating data sequences of multiple devices and generating a phase space matrix corresponding one-to-one with the multiple devices further includes:

[0033] Receive operational data sequences from multiple devices, and process each operational data sequence to generate a standard operational data sequence;

[0034] The phase space is reconstructed for each standard operating data sequence and the phase space matrix is ​​generated.

[0035] A second embodiment of the present invention provides a method for analyzing the life cycle of a bearing, comprising:

[0036] The phase space of the received operating data sequences of multiple bearings is reconstructed to generate a bearing phase space matrix that corresponds one-to-one with the multiple bearings.

[0037] The phase space matrix of each bearing is decomposed to generate the bearing orthogonal matrix and the bearing dynamic matrix;

[0038] The orthogonal matrices of the bearings in each running data sequence are used to form a Steiffer manifold and the corresponding first bearing evolution path is formed. The dynamic matrices of the bearings in each running data sequence are then projected to form the corresponding second bearing evolution path.

[0039] Each first bearing evolution path is compared with the preset first bearing full-cycle path for similarity, and the corresponding first similarity result is generated.

[0040] Based on the first similarity result, a second similarity result is generated by comparing the similarity of each second bearing evolution path with the preset second bearing full-cycle path, and the life cycle stages of the multiple bearings are generated according to the second similarity result.

[0041] For example, in the life cycle analysis method provided in some embodiments of this application, the step of comparing the similarity of each first bearing evolution path with a preset first bearing full life cycle path and generating a corresponding first similarity result further includes: sorting the first bearing evolution paths according to the first similarity results of each first bearing evolution path in descending order of similarity.

[0042] The step of generating a second similarity result based on the first similarity result, by comparing the similarity of each second bearing evolution path with a preset second bearing full-cycle path, and generating the life cycle stages of the multiple bearings based on the second similarity result, further includes: selecting a first preset number of first bearing evolution paths with the highest similarity from the first similarity ranking; comparing the second bearing evolution paths of the selected first bearing evolution paths with the preset second bearing full-cycle paths to generate a second similarity result; arranging the second similarity results in descending order of similarity to generate a second similarity ranking; and generating the life cycle stages of the multiple bearings based on the second similarity ranking.

[0043] For example, in the life cycle analysis method provided in some embodiments of this application, the step of comparing the similarity of each first bearing evolution path with a preset first bearing full life cycle path and generating a corresponding first similarity result further includes: sorting the first bearing evolution paths according to the first similarity results of each first bearing evolution path in descending order of similarity.

[0044] The step of generating a second similarity result based on the first similarity result, by comparing the similarity of each second bearing evolution path with a preset second bearing full-cycle path, and generating the life cycle stages of the multiple bearings based on the second similarity result, further includes:

[0045] Each second bearing evolution path is compared with a preset second bearing full-cycle path for similarity and a third similarity result is generated. Based on a preset weight ratio, the first similarity result and the third similarity result are combined to generate a second similarity result. The second similarity results are sorted in descending order of similarity to generate a second similarity ranking. The life cycle stages of the multiple bearings are generated based on the second similarity ranking.

[0046] For example, in some embodiments of the lifecycle analysis method provided in this application, the lifecycle analysis method further includes:

[0047] The operating and development speed of each bearing is determined based on the geodesic distance between the evolution path of the first bearing and the geodesic distance between the full cycle path of the first bearing.

[0048] Clustering is performed on the multiple bearings based on their life cycle stages and operational development speed, and a fourth clustering result is generated.

[0049] A third embodiment of the present invention provides a time-series-based clustering device, comprising a phase space reconstruction unit, a decomposition unit, a first clustering unit, and a second clustering unit, wherein...

[0050] The phase space reconstruction unit is used to reconstruct the phase space of the received operating data sequences of multiple devices and generate a phase space matrix corresponding to each of the multiple devices.

[0051] The decomposition unit is used to decompose each phase space matrix and generate an orthogonal matrix and a dynamic matrix respectively;

[0052] The first clustering unit is used to form a Steiffer manifold using the orthogonal matrices of each running data sequence, and to cluster the multiple devices according to the Steiffer manifold and generate a first clustering result;

[0053] The second clustering unit is used to perform clustering based on the first clustering result, combined with the dynamic matrix of each running data sequence, and generate the second clustering result of the multiple devices.

[0054] A fourth embodiment of the present invention provides a bearing lifecycle analysis device, including a phase space reconstruction module, a decomposition module, a path formation module, a first similarity module, and a second similarity module, wherein...

[0055] The phase space reconstruction module is used to reconstruct the phase space of the received operating data sequences of multiple bearings and generate a bearing phase space matrix that corresponds one-to-one with the multiple bearings.

[0056] The decomposition module is used to decompose the phase space matrix of each bearing and generate the bearing orthogonal matrix and the bearing dynamic matrix respectively.

[0057] The path forming module is used to form a Steiffel manifold using the bearing orthogonal matrices of each running data sequence and to form a corresponding first bearing evolution path, and to project the bearing dynamic matrix of each running data sequence to form a corresponding second bearing evolution path.

[0058] The first similarity module is used to compare the similarity of each first bearing evolution path with the preset first bearing full cycle path and generate the corresponding first similarity result.

[0059] The first similarity module is used to generate a second similarity result based on the first similarity result, by comparing the similarity between each second bearing evolution path and a preset second bearing full-cycle path, and then generate the life cycle stages of the multiple bearings based on the second similarity result.

[0060] The fifth embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first embodiment;

[0061] or

[0062] When the program is executed by the processor, it implements the method described in the second embodiment.

[0063] The sixth embodiment of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor.

[0064] When the processor executes the program, it implements the method as described in the first embodiment;

[0065] or

[0066] When the processor executes the program, it implements the method described in the second embodiment.

[0067] The beneficial effects of this invention are as follows:

[0068] This invention addresses existing problems by developing a time-series-based clustering method and a bearing lifecycle analysis method. It generates phase space matrices by reconstructing the phase space of multiple devices' operational data based on time series data, and then uses the orthogonal and dynamic matrices obtained from the decomposition of each phase space matrix for comprehensive clustering. This overcomes the problems in existing technologies, effectively improves clustering accuracy, and has broad application prospects. Attached Figure Description

[0069] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0070] Figure 1 A flowchart illustrating a clustering method according to an embodiment of the present invention is shown;

[0071] Figure 2 A structural block diagram of the clustering device according to an embodiment of the present invention is shown;

[0072] Figure 3 A flowchart illustrating a lifecycle analysis method according to an embodiment of the present invention is shown;

[0073] Figure 4 This diagram shows a structural block diagram of a life cycle analysis device according to an embodiment of the present invention;

[0074] Figure 5 A schematic diagram of the bearing path according to an embodiment of the present invention is shown;

[0075] Figure 6 A schematic diagram of the structure of a computer device according to another embodiment of the present invention is shown. Detailed Implementation

[0076] To more clearly illustrate the present invention, the following description, in conjunction with preferred embodiments and accompanying drawings, further explains the invention. Similar components in the drawings are indicated by the same reference numerals. Those skilled in the art should understand that the specific description below is illustrative rather than restrictive and should not be construed as limiting the scope of protection of the present invention.

[0077] In response to the above situation, such as Figure 1 As shown, one embodiment of the present invention provides a time-series-based clustering method, including:

[0078] The phase space of the received operating data sequences from multiple devices is reconstructed to generate a phase space matrix corresponding to each of the multiple devices.

[0079] Each phase space matrix is ​​decomposed and an orthogonal matrix and a dynamic matrix are generated;

[0080] A Steiffel manifold is constructed using the orthogonal matrices of each running data sequence, and the multiple devices are clustered according to the Steiffel manifold to generate a first clustering result;

[0081] Based on the first clustering result, the dynamic matrix of each running data sequence is combined to perform clustering and generate the second clustering result of the multiple devices.

[0082] In this embodiment, for the time-series-based operating data sequences of multiple devices, phase space reconstruction is performed on each operating data sequence to obtain a phase space matrix. The phase space matrix is ​​decomposed into an orthogonal matrix representing orthogonal vectors of the operating data sequence and a dynamic matrix representing the dynamic information of the operating data sequence. A Steifel manifold is formed using the orthogonal matrices to generate a clustering result. Then, based on the clustering result generated by the orthogonal matrix and combined with the dynamic matrix, a clustering result integrating the dynamic information of the phase space matrix and the orthogonal vectors is generated, thereby effectively improving the accuracy of clustering the operating data sequences of each device.

[0083] To further illustrate the specific implementation of this application, a specific example is given, using the clustering of operational data sequences for multiple bearing devices as an example, including the following steps:

[0084] The first step is to reconstruct the phase space of the received operating data sequences from multiple devices and generate phase space matrices that correspond one-to-one with the multiple devices.

[0085] In this embodiment, the time series data of 1000 bearing devices during operation are received as operating data sequences. Each operating data sequence is processed to generate a standard operating data sequence. Then, each standard operating data sequence undergoes phase space reconstruction to generate the phase space matrix. The operating data sequences are data collected by sensors installed on the bearing devices based on time series data. For example, if a vibration sensor is installed on the bearing device, the operating data sequence is the vibration signal collected by the vibration sensor. A series of signals collected based on the time series data form a one-dimensional operating data sequence.

[0086] First, data processing is performed on each running data sequence, including:

[0087] First, the running data sequence is cleaned to remove noise and outliers, reducing the impact of noise and outliers on the running data sequence to improve clustering accuracy.

[0088] Secondly, in order to facilitate phase space reconstruction, it is necessary to ensure that all running data sequences have the same length and sampling rate.

[0089] Next, further denoising and normalization are performed on each running data sequence. For example, the min-max normalization scaling method is used to normalize the running data sequence, with the formula: (x i -x min ) / (x max -x min ), where xi x is a variable, that is, a value in the running data sequence of this embodiment. min Let x be the minimum value of the variable. max This represents the maximum value of the variable.

[0090] Finally, noise removal is performed on the normalized running data sequence using at least one of the following methods, not limited to mean filtering, median filtering, weighted average filtering, 3 standard deviation denoising, binning denoising, and DBSCAN denoising, to obtain a standard running data sequence.

[0091] Second, phase space reconstruction is performed based on the standard operating data sequence after data processing.

[0092] The standard operating data sequence is a one-dimensional time series. It is transformed into points in a high-dimensional space to capture the dynamic characteristics of the time series. For example, a phase space matrix, i.e., delayed coordinate embedding, is constructed by selecting an appropriate delay time (τ) and embedding dimension (m). Phase space reconstruction is performed on 1000 bearing data points, where m and τ can be obtained using the Gram-Pianka method and mutual information method. The same m and τ are used consistently during phase space reconstruction; however, m and τ can also be calculated separately for different bearing operating time data, and then the largest m and smallest τ can be selected, or the m and τ of a single data point can be used. By performing phase space reconstruction on each of the 1000 bearing data points, reconstructed phase space matrices PSR1, PSR2, PSR3…PSRN corresponding one-to-one with each bearing data point are obtained.

[0093] The second step is to decompose each phase space matrix and generate orthogonal matrices.

[0094] In this embodiment, the phase space matrix of 1000 bearing data is decomposed into an orthogonal matrix for representing the orthogonal vectors of the running data sequence and a dynamic matrix for representing the dynamic information of the running data sequence.

[0095] In an optional embodiment, QR decomposition is performed on the phase space matrix of 1000 bearing data points. QR decomposition is a method of decomposing a matrix into the product of two matrices. For an m×n matrix A, it is decomposed into an m×m orthogonal matrix Q and an m×n upper triangular matrix R, such that A = QR. QR decomposition can be viewed as a transformation that preserves the vector lengths and angles of the data. Each column vector of Q is a unit vector, and the main operation is a rotation of the original vectors. The diagonal elements of the upper triangular matrix R are the projection lengths of the column vectors of the original matrix A onto the column vectors of Q. These diagonal elements reflect the degree of "stretching" of the column vectors in the corresponding orthogonal directions, i.e., the energy or intensity of the original signal in those directions.

[0096] In this embodiment, the phase space matrix of 1000 bearing data points is decomposed using QR decomposition to obtain an orthogonal matrix Q and an upper triangular matrix R. Specifically, the phase space is decomposed using QR decomposition to obtain Q1, R1, Q2, R2, Q3, R3, ..., QN, RN.

[0097] In another optional embodiment, the phase space matrix of 1000 bearing data points is decomposed using SVD. SVD decomposition decomposes the phase space matrix into an orthogonal matrix U, a diagonal matrix Σ, and an orthogonal matrix V. Similar to QR decomposition, in SVD decomposition A = UΣV T In this model, V and U can both be viewed as rotation matrices, while Σ represents stretching. Specifically, V is the orthonormal basis of the original domain, representing a rotation of the original data space; Σ is a diagonal matrix, representing stretching or compressing the rotated data; that is, the U matrix represents the orthogonal vectors of the time series data, while the Σ matrix contains the dynamic information of the time series. SVD decomposition can be seen as a process of first rotating the original data through V, then stretching or compressing it through Σ, and finally rotating it again through U. Performing SVD decomposition on the phase space matrix of the bearing data yields U1, U2, U3…UN, and Σ1, Σ2, Σ3…ΣN.

[0098] The third step involves using the orthogonal matrices of each running data sequence to form a Steiffer manifold, and then clustering the multiple devices based on the Steiffer manifold to generate the first clustering result.

[0099] In this embodiment, a first clustering result is formed by constructing a Steifel manifold using orthogonal matrices and studying the point spacing and mean spacing on the Steifel manifold. Specifically, the following steps are included:

[0100] First, a Steifel manifold is constructed using the orthogonal matrices of each running data sequence, the Steifel manifold including a first point representing each orthogonal matrix.

[0101] In this embodiment, if the phase space matrix decomposition uses QR decomposition, then the orthogonal matrix Q is used to form a Stevorl manifold, that is, the time series is embedded into the Stevorl manifold. The Q matrices of 1000 bearing data points are used to form the Stevorl manifold, thus forming 1000 points on the Stevorl manifold. If the phase space matrix decomposition uses SVD decomposition, then the orthogonal matrix U is used to form the Stevorl manifold, that is, the time series is embedded into the Stevorl manifold. The U matrices of 1000 bearing data points are used to form the Stevorl manifold, thus forming 1000 points on the Stevorl manifold.

[0102] Second, calculate and obtain the point spacing between each first point, and the average distance between any two points among the first points.

[0103] Third, based on the point spacing and average spacing of each first point, the multiple devices are clustered using a pre-set clustering operation to generate the first clustering result.

[0104] In this embodiment, the point spacing between each point and the average spacing between any two points in each of the second points are calculated. Specifically, the point spacing is the distance between a point and each of the other points, and the average spacing is the average distance between any two points in the multiple points. The point spacing is the distance formed by combining any two points in the multiple points, and the average spacing is the average of all the point spacings formed. In this embodiment, based on the point spacing and the average spacing, the dimensionality reduction algorithm UMAP combined with the clustering algorithm HDBSCAN is used to cluster 1000 bearing data points, for example, forming clustering result 1, clustering result 2, and clustering result 3. For example, the 1000 bearing data points are divided into 3 categories. The first category corresponding to clustering result 1 includes 200 bearing data points, corresponding to 200 bearing devices; the second category corresponding to clustering result 2 includes 300 bearing data points, corresponding to 300 bearing devices; and the third category corresponding to clustering result 3 includes 500 bearing data points, corresponding to 500 bearing devices.

[0105] The fourth step involves clustering the multiple devices based on the first clustering result and combining the dynamic matrix of each running data sequence to generate the second clustering result.

[0106] In this embodiment, based on the first clustering result, a corresponding dynamic matrix is ​​obtained according to the different classifications of the first clustering result, and the dynamic matrix is ​​projected onto the unit space for secondary clustering. Specifically, the following steps are included.

[0107] First, the first clustering result includes multiple clustering sub-results, and the dynamic matrix of the running data sequence corresponding to each clustering sub-result is stretched into a first vector.

[0108] In this embodiment, the dynamic matrix is ​​the R matrix obtained from QR decomposition. For each cluster of data, the R matrix information corresponding to the original time series is obtained, stretched into a vector, and projected onto a unit space, such as onto a Sn unit sphere. For example, the corresponding R matrix is ​​obtained for clustering result 1:

[0109]

[0110] Stretch R into a vector v. In this embodiment, the stretching operation can be horizontal stretching, vertical stretching, or a combination of both. For example:

[0111] Lateral stretching: v = [r11, r12, r13, r22, r23, r33];

[0112] or

[0113] Longitudinal stretching: v = [r11, r12, r22, r13, r23, r33];

[0114] or

[0115] Combined transverse and longitudinal stretching: v = [r11,r12,r13,r22,r23,r33,r11,r12,r22,r13,r23,r33].

[0116] Second, the first vector of each running data sequence is projected onto a unit sphere, which includes a second point representing each dynamic matrix.

[0117] In this embodiment, the stretched vector v is projected onto the unit sphere, that is, the vector v is transformed into a point with a norm of 1. The Euclidean norm of the vector, i.e., its length, is calculated, and then the vector is divided by the Euclidean norm to normalize the vector to a unit length, that is, projected onto the unit sphere and formed as a point on the unit sphere.

[0118] Third, calculate and obtain the point spacing of each second point, and the average distance between any two points in each second point;

[0119] Fourth, based on the point spacing and average spacing of each second point, the multiple devices are clustered using a preset clustering operation to generate the second clustering result.

[0120] In this embodiment, the point spacing and mean spacing between each point are calculated. Based on the point spacing and mean spacing, the dimensionality reduction algorithm UMAP combined with the clustering algorithm HDBSCAN is used to perform secondary clustering on the 200 bearing data points of clustering result 1, for example, to form clustering result 4 and clustering result 5. For example, the 200 bearing data are divided into 2 categories, with clustering result 4 corresponding to 50 bearing data points and clustering result 5 corresponding to 150 bearing devices.

[0121] Similarly, the 300 bearing data points using clustering result 2 and the 500 bearing data points using clustering result 3 are projected using the corresponding dynamic matrix to perform a second clustering, resulting in clustering results respectively.

[0122] It should be noted that when the dynamic matrix is ​​the Σ matrix obtained by SVD decomposition, for each type of cluster data, the Σ matrix information corresponding to the original time series is obtained, stretched into a vector, and projected onto the unit space for secondary clustering operation. The specific implementation method is similar to that of QR decomposition, and will not be described in detail here.

[0123] Thus, the clustering operation of the operational data sequence of 1000 bearings is completed, resulting in a clustering result with higher accuracy than that achieved using only orthogonal matrices obtained from phase space matrix decomposition. This embodiment generates phase space matrices by reconstructing the phase space of multiple equipment operational data based on time series, and then performs a first-order clustering using the orthogonal matrices obtained from each phase space matrix decomposition. Based on the first-order clustering result, a second-order clustering is performed using a dynamic matrix, resulting in a more accurate and detailed clustering result.

[0124] To further illustrate the specific implementation of this application, another specific example is given, using the clustering of operational data sequences for multiple bearing devices as an example, including the following steps:

[0125] The first step is to reconstruct the phase space of the received operating data sequences from multiple devices and generate phase space matrices that correspond one-to-one with the multiple devices.

[0126] In this embodiment, the time series data of 1000 bearing devices during operation are received as operating data sequences. Each operating data sequence is processed to generate a standard operating data sequence. Then, each standard operating data sequence undergoes phase space reconstruction to generate the phase space matrix. The specific implementation of this embodiment is similar to the previous embodiments and will not be repeated here.

[0127] The second step is to decompose each phase space matrix and generate orthogonal matrices.

[0128] In this embodiment, the phase space matrix of 1000 bearing data points is decomposed into an orthogonal matrix representing orthogonal vectors of the running data sequence and a dynamic matrix representing the dynamic information of the running data sequence. The specific implementation of this embodiment is similar to the previous embodiments and will not be repeated here.

[0129] The third step involves using the orthogonal matrices of each running data sequence to form a Steiffer manifold, and then clustering the multiple devices based on the Steiffer manifold to generate the first clustering result.

[0130] In this embodiment, a Steifel manifold is constructed using orthogonal matrices, and the point spacing and mean are studied on the Steifel manifold to form the first clustering result. The specific implementation of this embodiment is similar to the previous embodiments and will not be repeated here.

[0131] The fourth step involves clustering the multiple devices based on the first clustering result and combining the dynamic matrix of each running data sequence to generate the second clustering result.

[0132] In this embodiment, the dynamic matrix obtained by decomposing the phase space matrix of each running data sequence is projected onto a unit space to perform clustering to form a third clustering result. The second clustering result is then generated by combining the first and third clustering results. Specifically, the following steps are included.

[0133] First, stretch the dynamic matrix of each running data sequence into a second vector.

[0134] In this embodiment, the dynamic matrix is ​​the R matrix obtained from QR decomposition. The R matrix is ​​stretched into a vector. During the stretching process, horizontal stretching, vertical stretching, or a combination of both can be used. The specific implementation method of this embodiment is similar to the previous embodiments and will not be repeated here.

[0135] Second, the second vector of each running data sequence is projected onto a unit sphere, which includes a third point representing each dynamic matrix.

[0136] In this embodiment, the stretched second vector is projected onto a unit space, for example, onto a unit sphere Sn, forming a point on the unit sphere. The specific implementation of this embodiment is similar to the previous embodiments and will not be repeated here.

[0137] Third, calculate and obtain the point spacing between each third point, as well as the average distance between any two points among the third points.

[0138] Fourth, based on the point spacing and average spacing of each third point, the multiple devices are clustered using a pre-set clustering operation to generate the third clustering result.

[0139] In this embodiment, the point spacing and mean spacing between each point are calculated. Based on the point spacing and mean spacing, the dimensionality reduction algorithm UMAP combined with the clustering algorithm HDBSCAN is used to cluster the 1000 bearing data points to generate a third clustering result.

[0140] Fifth, based on the preset weight ratio, the second clustering result is generated by combining the first clustering result and the third clustering result.

[0141] In this embodiment, the phase space matrix generated by reconstructing the phase space of the operating data sequence of each bearing device, the orthogonal matrix and the dynamic matrix formed by decomposing the phase space matrix are used to perform clustering based on the orthogonal matrix and the dynamic matrix respectively, and two relatively independent clustering results are generated. The final clustering result is generated by combining the two according to the weight ratio of the different clustering results.

[0142] Specifically, the weight of the first clustering result formed by the orthogonal matrix is ​​set to w1, and the weight of the second clustering result formed by the dynamic matrix is ​​set to w2. The final clustering result is obtained by calculating the weighted average of the two. Second clustering result = First clustering result * w1 + Third clustering result * w2.

[0143] It is worth noting that this embodiment is only used to illustrate the specific implementation of this application. This application does not specifically limit the weight ratio and fusion method of the orthogonal matrix and the dynamic matrix. Those skilled in the art should set the weight ratio according to actual application needs, such as according to at least one of the clustering quality, data characteristics or domain knowledge, and then form the final clustering result through weighted average or other fusion methods.

[0144] Thus, the clustering operation of the operational data sequence of 1000 bearings is completed, resulting in a clustering result with higher accuracy than that obtained by using orthogonal matrices derived solely from phase space matrix decomposition. This embodiment generates phase space matrices by reconstructing the phase space of multiple equipment operational data based on time series. The first clustering result is generated using the orthogonal matrices obtained from each phase space matrix decomposition, and the third clustering result is generated using the dynamic matrices obtained from each phase space matrix decomposition. Finally, a weighted average fusion method using set weight ratios generates the final second clustering result, which exhibits higher accuracy and greater detail.

[0145] Considering the case where multiple sensors are installed on the bearing device, in an optional embodiment, when the device includes multiple operating data sequences, the step of reconstructing the phase space of the received operating data sequences of the multiple devices and generating a phase space matrix corresponding to each of the multiple devices further includes: reconstructing the phase space of the multiple operating data sequences of the device and generating an intermediate phase space matrix corresponding to each operating data sequence; and concatenating the multiple intermediate phase space matrices of the device to generate the phase space matrix of the device.

[0146] In this embodiment, to ensure stable and efficient operation of the bearing and to promptly detect potential faults, multiple sensors are installed on the bearing equipment to monitor its fault status. For example, vibration sensors, temperature sensors, acoustic sensors, and displacement sensors are installed on the bearing equipment to monitor its operating status. The signals collected by multiple sensors form multiple one-dimensional operating data sequences. Specifically, the following steps are included:

[0147] First, the phase space of multiple operating data sequences of the device is reconstructed to generate an intermediate phase space matrix that corresponds one-to-one with each operating data sequence.

[0148] In this embodiment, a bearing device is used as an example: signals collected by multiple sensors of a bearing are used to form multiple one-dimensional operating data sequences. Each one-dimensional operating data sequence is processed, such as noise removal and normalization, to improve the accuracy of the data. Then, phase space reconstruction is performed on the different one-dimensional operating data sequences to obtain the corresponding phase space matrix.

[0149] In an optional embodiment, the same phase space reconstruction parameters are used to reconstruct the phase space of multiple operating data sequences of the device.

[0150] Specifically, in this embodiment, the most suitable embedding dimension *m* and time delay *τ* are calculated for different sensor data of each bearing device. For example, the Grassberger-Procaccia (GP) method or mutual information method is used to determine the embedding dimension *m* and time delay *τ*. After calculating *m* and *τ* for all sensors, the largest *m* and the smallest *τ* are selected as unified parameters for phase space reconstruction, thereby ensuring that the reconstructed phase space can contain as much important information as possible from all sensor data. Finally, the selected *m* and *τ* are used to reconstruct the phase space of multiple one-dimensional sensor data for each bearing device, generating a phase space matrix that reveals the dynamic characteristics of the data.

[0151] Second, the multiple intermediate phase space matrices of the device are spliced ​​together to generate the phase space matrix of the device.

[0152] In this embodiment, the phase space matrices of the sensors from multiple bearing devices are spliced ​​together to form a larger matrix as the phase space matrix. Subsequent steps are then performed, as described in the previous embodiment, and will not be repeated here.

[0153] The clustering method based on the above embodiments, such as Figure 2 As shown, one embodiment of the present invention also provides a time-series-based clustering device, including a phase space reconstruction unit, a decomposition unit, a first clustering unit, and a second clustering unit, wherein...

[0154] The phase space reconstruction unit is used to reconstruct the phase space of the received operating data sequences of multiple devices and generate a phase space matrix corresponding to each of the multiple devices.

[0155] The decomposition unit is used to decompose each phase space matrix and generate an orthogonal matrix and a dynamic matrix respectively;

[0156] The first clustering unit is used to form a Steiffer manifold using the orthogonal matrices of each running data sequence, and to cluster the multiple devices according to the Steiffer manifold and generate a first clustering result;

[0157] The second clustering unit is used to perform clustering based on the first clustering result, combined with the dynamic matrix of each running data sequence, and generate the second clustering result of the multiple devices.

[0158] In this embodiment, for the time-series-based operating data sequences of multiple devices, a phase space reconstruction unit reconstructs the phase space of each operating data sequence to obtain a phase space matrix. A decomposition unit decomposes each phase space matrix into an orthogonal matrix representing orthogonal vectors of the operating data sequence and a dynamic matrix representing the dynamic information of the operating data sequence. A first clustering unit uses the orthogonal matrices to form a Steifel manifold to generate a clustering result. Then, a second clustering unit uses the clustering result generated by the orthogonal matrix and combines it with the dynamic matrix to generate a clustering result that integrates the dynamic information of the phase space matrix and the orthogonal vectors, thereby effectively improving the accuracy of clustering the operating data sequences of each device.

[0159] The clustering approach based on the clustering method in the above embodiments is as follows: Figure 3 As shown, one embodiment of the present invention also provides a bearing lifecycle analysis method, including:

[0160] The phase space of the received operating data sequences of multiple bearings is reconstructed to generate a bearing phase space matrix that corresponds one-to-one with the multiple bearings.

[0161] The phase space matrix of each bearing is decomposed to generate the bearing orthogonal matrix and the bearing dynamic matrix;

[0162] The orthogonal matrices of the bearings in each running data sequence are used to form a Steiffer manifold and the corresponding first bearing evolution path is formed. The dynamic matrices of the bearings in each running data sequence are then projected to form the corresponding second bearing evolution path.

[0163] Each first bearing evolution path is compared with the preset first bearing full-cycle path for similarity, and the corresponding first similarity result is generated.

[0164] Based on the first similarity result, a second similarity result is generated by comparing the similarity of each second bearing evolution path with the preset second bearing full-cycle path, and the life cycle stages of the multiple bearings are generated according to the second similarity result.

[0165] In this embodiment, the life cycle of bearing equipment is divided into the following stages: normal operation, initial wear, initial failure, failure development, severe failure, and final failure. The life cycle stage of each bearing equipment is determined using the aforementioned life cycle analysis method. Specifically, this embodiment generates corresponding bearing phase space matrices by reconstructing the phase space of multiple bearing equipment operation data sequences based on time series. The bearing orthogonal matrix and bearing dynamic matrix obtained by decomposing each bearing phase space matrix are then used to form a Steiffel manifold and a first bearing evolution path. The bearing dynamic matrix is ​​projected into space to form a second bearing evolution path. A first similarity result is generated by comparing the first bearing evolution path with a preset full-cycle bearing path. A second similarity result is generated by comparing the second bearing evolution path with the preset full-cycle bearing path. The life cycle stage of each bearing is determined based on the second similarity result.

[0166] In this embodiment, by using the operating data sequences of each bearing device, orthogonal matrices representing orthogonal vectors of the operating data sequences and dynamic matrices representing dynamic information of the operating data sequences are obtained through phase space reconstruction and phase space matrix decomposition. The characteristics of the orthogonal matrices and dynamic matrices are used to form corresponding bearing evolution paths. By comparing with the preset bearing full-cycle paths, the similarity results are determined, and the life cycle of each bearing is determined by the similarity results. This helps users to understand the status of bearing devices and provide fault warnings, which has practical application value.

[0167] In an optional embodiment, the step of comparing the similarity of each first bearing evolution path with a preset first bearing full-cycle path and generating corresponding first similarity results further includes: sorting the first bearing evolution paths according to the first similarity results of each first bearing evolution path in descending order of similarity.

[0168] The step of generating a second similarity result based on the first similarity result, by comparing the similarity of each second bearing evolution path with a preset second bearing full-cycle path, and generating the life cycle stages of the multiple bearings based on the second similarity result, further includes: selecting a first preset number of first bearing evolution paths with the highest similarity from the first similarity ranking; comparing the second bearing evolution paths of the selected first bearing evolution paths with the preset second bearing full-cycle paths to generate a second similarity result; arranging the second similarity results in descending order of similarity to generate a second similarity ranking; and generating the life cycle stages of the multiple bearings based on the second similarity ranking.

[0169] In this embodiment, as Figure 5As shown, path 0 is the first bearing full-cycle path, and paths 1, 2, and 3 are the first bearing evolution paths of different bearing devices in this embodiment. Based on orthogonal matrices, a Steifel manifold is formed to create the corresponding first bearing evolution path. The first similarity result is formed by comparing the similarity with the first bearing full-cycle path. The first similarity result is sorted to form a first similarity ranking. The bearing devices with higher similarity are obtained from the first similarity ranking. The corresponding second bearing evolution path formed by dynamic data is determined according to the bearing device. The second similarity result is generated by comparing the similarity between the second bearing evolution path and the second bearing full-cycle path. The life cycle stage of each bearing is determined according to the second similarity ranking formed by the second similarity result.

[0170] Specifically, the following steps are included:

[0171] First, data processing and phase space reconstruction.

[0172] In this embodiment, the received operating data of multiple bearing devices are first subjected to noise reduction and normalization processing to improve data accuracy. Then, the phase space is reconstructed based on the processed data to generate the corresponding phase space matrix.

[0173] Second, phase space decomposition

[0174] Third, the path is formed.

[0175] In this implementation, each phase space matrix is ​​decomposed, for example, by QR decomposition or SVD decomposition. Considering the entire life cycle of the bearing, the orthogonal matrix obtained by phase space matrix decomposition is used to form a time-varying path on the Steifel manifold, namely the first bearing evolution path. Then, the dynamic matrix obtained by phase space matrix decomposition is used to perform Sn projection to form a reference path, namely the second bearing evolution path.

[0176] Fourth, establish the first similarity.

[0177] Fifth, forming a second similarity.

[0178] Sixth, life cycle assessment

[0179] In this embodiment, based on a pre-set first bearing full-cycle path, the similarity of the first bearing evolution path on the Steifel manifold is compared with that of the first bearing full-cycle path. Specifically, the similarity is calculated using the Dynamic Time Warp (DTW) algorithm. Similarly, based on the second bearing full-cycle path, the similarity of the second bearing evolution path on the projection space is compared with that of the second bearing full-cycle path. Specifically, in this embodiment, the similarity scores are sorted during the similarity calculation. For example, the paths composed of orthogonal matrices of the operating data of the bearing to be analyzed are sorted after similarity calculation with the first bearing full-cycle path, and the top J data with higher similarity are selected; similarity scores SQ1, SQ2, ..., SQJ are obtained. For these J data, the paths composed of dynamic matrices of the operating data of the corresponding bearing to be analyzed are calculated, and the similarity is calculated with the second bearing full-cycle path. The data with the highest similarity is selected, and the life cycle stage of the corresponding bearing equipment is determined based on the data with the highest similarity.

[0180] To further improve the accuracy of life cycle analysis, in an optional embodiment, the operating development speed of each bearing is determined based on the geodesic distance between the first bearing evolution path and the geodesic distance between the first bearing full life cycle path; clustering is performed based on the life cycle stage and operating development speed of the multiple bearings to generate a fourth clustering result.

[0181] In this embodiment, since orthogonal matrices represent changes in feature directions, and the corresponding time series with the highest similarity has been selected based on the first similarity result formed by the orthogonal matrices, the geodesic distance of the first bearing evolution path and the geodesic distance of the first bearing full-cycle path can be used to determine the operating development speed of each bearing device. That is, by introducing the geodesic distance of the orthogonal matrix of the bearing device on the manifold to determine the operating development speed of the bearing device, the indicators characterizing the operating state of the bearing device are further increased, effectively improving the detection dimension of the bearing device's operating state. This embodiment performs clustering again based on the life cycle stage and operating development speed of the bearing device. That is, based on the already formed life cycle stages of the bearing device, the bearing devices are further classified by indicators characterizing the speed of their operating state. For example, if a bearing device is in the normal operating stage, it can be further refined into bearing devices in the normal operating stage with a relatively fast operating development speed based on its operating development speed. Based on the clustering results, the frequency of checking the bearing device can be increased to avoid the bearing entering the failure stage due to a rapid development speed, thus affecting production operations, improving the accuracy of life cycle analysis, effectively stabilizing production, and improving the yield of manufactured products.

[0182] It is worth noting that this application does not impose specific limitations on the calculation of geodesic distances. Those skilled in the art should select an appropriate calculation method based on actual application needs, which will not be elaborated here.

[0183] In another optional embodiment, the step of comparing the similarity of each first bearing evolution path with a preset first bearing full-cycle path and generating a corresponding first similarity result further includes: sorting the first bearing evolution paths according to the first similarity results of each first bearing evolution path in descending order of similarity.

[0184] The step of generating a second similarity result based on the first similarity result, by comparing the similarity of each second bearing evolution path with a preset second bearing full-cycle path, and generating the life cycle stages of the multiple bearings based on the second similarity result, further includes:

[0185] Each second bearing evolution path is compared with a preset second bearing full-cycle path for similarity and a third similarity result is generated. Based on a preset weight ratio, the first similarity result and the third similarity result are combined to generate a second similarity result. The second similarity results are sorted in descending order of similarity to generate a second similarity ranking. The life cycle stages of the multiple bearings are generated based on the second similarity ranking.

[0186] In this embodiment, a first bearing evolution path is formed by constructing a Steiffel manifold based on orthogonal matrices, and a second bearing evolution path is formed based on dynamic data. Similarity comparisons are performed to obtain corresponding first and third similarity results. The final second similarity result is generated by using a weighted average fusion method with set weight ratios. The second similarity results are used to form a second similarity ranking from largest to smallest, and the life cycle stage of each bearing is determined based on the second similarity ranking.

[0187] Specifically, the following steps are included:

[0188] First, data processing and phase space reconstruction.

[0189] Second, phase space decomposition

[0190] Third, the path is formed.

[0191] In this embodiment, the above steps are similar to those in the previous embodiments. For specific implementation details, please refer to the previous embodiments, which will not be repeated here.

[0192] Fourth, establish the first similarity.

[0193] Fifth, forming a third similarity.

[0194] Sixth, life cycle assessment

[0195] In this embodiment, a first similarity result is generated by comparing the similarity of the first bearing evolution path formed by orthogonal matrices and the first bearing full-cycle path. This first similarity result is then ranked, for example, selecting the top J data points with high similarity. Similarly, a third similarity result is generated by comparing the similarity of the second bearing evolution path formed by dynamic matrices and the second bearing full-cycle path. This third similarity result is then ranked, for example, selecting the top K data points with high similarity. An appropriate weighting ratio is selected to calculate the path with the highest similarity. For example, J similarity scores SQ1, SQ2, ..., SQJ from orthogonal matrices and K similarity scores SR1, SR2, ..., SRK from dynamic matrices are selected. A weighted average similarity, i.e., the second similarity result, is calculated for each path. The sequence with the highest weighted average similarity is selected as the corresponding sequence. Specifically, the weighted average similarity is:

[0196]

[0197] Where: Sweighted is the weighted average similarity, S Q It is the similarity score of orthogonal matrices, S R It is the similarity score of the dynamic matrix, w Q It is the weight of the similarity of orthogonal matrices, w R It is the weight of the dynamic matrix similarity, weight w Q and w R The following conditions should be met: w Q ≥0 and w R ≥0, w Q +w R >0.

[0198] In this embodiment, a phase space matrix is ​​generated by reconstructing the phase space of the operating data sequence of each bearing device, and an orthogonal matrix and a dynamic matrix are generated by decomposing the phase space matrix. A first similarity result and a third similarity result are generated based on the similarity comparison of the orthogonal matrix and the dynamic matrix, respectively. A second similarity is formed according to the different similarity results and the corresponding weight ratio. A second similarity sort is generated based on the second similarity and arranged in descending order of similarity. The life cycle stages of the multiple bearings are generated based on the second similarity sort.

[0199] It is worth noting that this embodiment can also determine the operating development speed of each bearing by comparing the geodesic distance of the first bearing evolution path with the geodesic distance of the first bearing's full life cycle path. The obtained bearing equipment life cycle stages and operating development speeds can then be clustered again to improve the accuracy of the bearing equipment life cycle analysis. For specific implementation details of this embodiment, please refer to the foregoing embodiments, which will not be repeated here.

[0200] The lifecycle analysis method based on the above embodiments, such as Figure 4 As shown, one embodiment of the present invention also provides a bearing lifecycle analysis device, including a phase space reconstruction module, a decomposition module, a path formation module, a first similarity module, and a second similarity module, wherein...

[0201] The phase space reconstruction module is used to reconstruct the phase space of the received operating data sequences of multiple bearings and generate a bearing phase space matrix that corresponds one-to-one with the multiple bearings.

[0202] The decomposition module is used to decompose the phase space matrix of each bearing and generate the bearing orthogonal matrix and the bearing dynamic matrix respectively.

[0203] The path forming module is used to form a Steiffel manifold using the bearing orthogonal matrices of each running data sequence and to form a corresponding first bearing evolution path, and to project the bearing dynamic matrix of each running data sequence to form a corresponding second bearing evolution path.

[0204] The first similarity module is used to compare the similarity of each first bearing evolution path with the preset first bearing full cycle path and generate the corresponding first similarity result.

[0205] The first similarity module is used to generate a second similarity result based on the first similarity result, by comparing the similarity between each second bearing evolution path and a preset second bearing full-cycle path, and then generate the life cycle stages of the multiple bearings based on the second similarity result.

[0206] This embodiment uses a phase space reconstruction module to reconstruct the phase space of multiple bearing equipment operation data sequences based on time series, generating corresponding bearing phase space matrices. A decomposition module then decomposes each bearing phase space matrix to obtain bearing orthogonal matrices and bearing dynamic matrices. A path formation module uses the bearing orthogonal matrices to form a Steiffel manifold and create a first bearing evolution path. The bearing dynamic matrix is ​​then projected into space to create a second bearing evolution path. A first similarity module compares the first bearing evolution path with a preset bearing full-cycle path to generate a first similarity result. A second similarity module combines the second bearing evolution path with the preset bearing full-cycle path to generate a second similarity result. Based on the second similarity result, the lifecycle stage of each bearing is determined.

[0207] Another embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a time-series-based clustering method or a bearing lifecycle analysis method of the present application.

[0208] In practical applications, the computer-readable storage medium can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. For example, a computer-readable storage medium can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this embodiment, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0209] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0210] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0211] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0212] like Figure 6 As shown, another embodiment of the present invention provides a structural schematic diagram of a computer device. Figure 6 The computer device T12 shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0213] like Figure 6 As shown, the computer device T12 is represented in the form of a general-purpose computing device. The components of the computer device T12 may include, but are not limited to: one or more processors or processing units T16, system memory T28, and bus T18 connecting different system components (including system memory T28 and processing unit T16).

[0214] Bus T18 represents one or more of several bus architectures, including memory buses or memory controllers, peripheral buses, graphics acceleration ports, processors, or local buses using any of the various bus architectures. Examples of these architectures include, but are not limited to, the Industry Standard Architecture (ISA) bus, the Micro Channel Architecture (MAC) bus, the Enhanced ISA bus, the Video Electronics Standards Association (VESA) local bus, and the Peripheral Component Interconnect (PCI) bus.

[0215] Computer device T12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by computer device T12, including volatile and non-volatile media, removable and non-removable media.

[0216] System memory T28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) T30 and / or cache memory T32. Computer device T12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system T34 may be used to read and write non-removable, non-volatile magnetic media (…). Figure 6 Not shown; usually referred to as a "hard drive"). Although Figure 6 Not shown, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk") and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to the bus T18 via one or more data media interfaces. The memory T28 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the embodiments of the present invention.

[0217] A program / utility T40 having a set (at least one) of program modules T42 can be stored, for example, in a memory T28. Such program modules T42 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. Program modules T42 typically perform the functions and / or methods described in the embodiments of the present invention.

[0218] Computer device T12 can also communicate with one or more external devices T14 (e.g., keyboard, pointing device, display T24, etc.), and with one or more devices that enable a user to interact with computer device T12, and / or with any device that enables computer device T12 to communicate with one or more other computing devices (e.g., network card, modem, etc.). This communication can be performed through input / output (I / O) interface T22. Furthermore, computer device T12 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) through network adapter T20. Figure 6 As shown, network adapter T20 communicates with other modules of computer device T12 via bus T18. It should be understood that, although... Figure 6 As not shown, it can be used in conjunction with computer device T12 with other hardware and / or software modules, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0219] The processor unit T16 executes various functional applications and data processing by running programs stored in the system memory T28, such as implementing a time series-based clustering method or a bearing life cycle analysis method provided in the embodiments of the present invention.

[0220] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. For those skilled in the art, other variations or modifications can be made based on the above description. It is impossible to exhaustively list all the implementation methods here. All obvious variations or modifications derived from the technical solutions of the present invention are still within the protection scope of the present invention.

Claims

1. A time-series-based clustering method, characterized in that, include: The phase space of the received operating data sequences from multiple devices is reconstructed to generate a phase space matrix corresponding to each of the multiple devices. Each phase space matrix is ​​decomposed and an orthogonal matrix and a dynamic matrix are generated; A Steiffel manifold is constructed using the orthogonal matrices of each running data sequence, and the multiple devices are clustered according to the Steiffel manifold to generate a first clustering result; Based on the first clustering result, the dynamic matrix of each running data sequence is combined to perform clustering and generate the second clustering result of the multiple devices.

2. The clustering method according to claim 1, characterized in that, The step of constructing a Steiffer manifold using orthogonal matrices of each running data sequence, and clustering the multiple devices based on the Steiffer manifold to generate a first clustering result further includes: A Steifel manifold is constructed using the orthogonal matrices of each running data sequence, wherein the Steifel manifold includes the first point representing each orthogonal matrix; Calculate and obtain the point spacing between each first point, and the average point spacing between any two points among the first points; Based on the point spacing and average spacing of each first point, the multiple devices are clustered using a pre-set clustering operation to generate the first clustering result.

3. The clustering method according to claim 2, characterized in that, The step of clustering based on the first clustering result and combining the dynamic matrix of each running data sequence to generate a second clustering result for the multiple devices further includes: The first clustering result includes multiple clustering sub-results, each of which stretches the dynamic matrix of the running data sequence corresponding to the clustering sub-result into a first vector; Each running data sequence's first vector is projected onto a unit sphere, the unit sphere including a second point representing each dynamic matrix; Calculate and obtain the point spacing between each second point, and the average point spacing between any two points among the second points; Based on the point spacing and average spacing of each second point, the multiple devices are clustered using a preset clustering operation to generate the second clustering result.

4. The clustering method according to claim 2, characterized in that, The step of clustering based on the first clustering result and combining the dynamic matrix of each running data sequence to generate a second clustering result for the multiple devices further includes: Each dynamic matrix of the running data sequence is stretched into a second vector; The second vector of each running data sequence is projected onto a unit sphere, which includes a third point representing each dynamic matrix; Calculate and obtain the point spacing between each third point, and the average point spacing between any two points among the third points; Based on the point spacing and average spacing of each third point, the multiple devices are clustered using a preset clustering operation to generate the third clustering result; The second clustering result is generated by combining the first clustering result and the third clustering result according to the preset weight ratio.

5. The clustering method according to claim 1, characterized in that, The step of decomposing each phase space matrix and generating orthogonal and dynamic matrices further includes: Perform a first decomposition on each phase space matrix to generate a first orthogonal matrix and a first dynamic matrix; or The phase space matrices are decomposed into second orthogonal matrices and second dynamic matrices respectively.

6. The clustering method according to claim 1, characterized in that, When the device includes multiple operating data sequences, the step of reconstructing the phase space of the received operating data sequences of the multiple devices and generating a phase space matrix corresponding one-to-one with the multiple devices further includes: The phase space is reconstructed for multiple operating data sequences of the device, and an intermediate phase space matrix corresponding to each operating data sequence is generated. The phase space matrix of the device is generated by splicing together multiple intermediate phase space matrices of the device.

7. The clustering method according to claim 6, characterized in that, The same phase space reconstruction parameters are used to reconstruct the phase space of multiple operating data sequences of the device.

8. The clustering method according to any one of claims 1-7, characterized in that, The step of reconstructing the phase space of the received operating data sequences from multiple devices and generating a phase space matrix corresponding to each of the multiple devices further includes: Receive operational data sequences from multiple devices, and process each operational data sequence to generate a standard operational data sequence; The phase space is reconstructed for each standard operating data sequence and the phase space matrix is ​​generated.

9. A method for analyzing the life cycle of a bearing, characterized in that, include: The phase space of the received operating data sequences of multiple bearings is reconstructed to generate a bearing phase space matrix that corresponds one-to-one with the multiple bearings. The phase space matrix of each bearing is decomposed to generate the bearing orthogonal matrix and the bearing dynamic matrix; The orthogonal matrices of the bearings in each running data sequence are used to form a Steiffer manifold and the corresponding first bearing evolution path is formed. The dynamic matrices of the bearings in each running data sequence are then projected to form the corresponding second bearing evolution path. Each first bearing evolution path is compared with the preset first bearing full-cycle path for similarity, and the corresponding first similarity result is generated. Based on the first similarity result, a second similarity result is generated by comparing the similarity of each second bearing evolution path with the preset second bearing full-cycle path, and the life cycle stages of the multiple bearings are generated according to the second similarity result.

10. The life cycle analysis method according to claim 9, characterized in that, The step of comparing the similarity of each first bearing evolution path with the preset first bearing full-cycle path and generating the corresponding first similarity result further includes: sorting the first bearing evolution paths according to the first similarity results of each first bearing evolution path in descending order of similarity. The step of generating a second similarity result based on the first similarity result, by comparing the similarity of each second bearing evolution path with a preset second bearing full-cycle path, and generating the life cycle stages of the multiple bearings based on the second similarity result, further includes: selecting a first preset number of first bearing evolution paths with the highest similarity from the first similarity ranking; comparing the second bearing evolution paths of the selected first bearing evolution paths with the preset second bearing full-cycle paths to generate a second similarity result; arranging the second similarity results in descending order of similarity to generate a second similarity ranking; and generating the life cycle stages of the multiple bearings based on the second similarity ranking.

11. The life cycle analysis method according to claim 9, characterized in that, The step of comparing the similarity of each first bearing evolution path with the preset first bearing full-cycle path and generating the corresponding first similarity result further includes: sorting the first bearing evolution paths according to the first similarity results of each first bearing evolution path in descending order of similarity. The step of generating a second similarity result based on the first similarity result, by comparing the similarity of each second bearing evolution path with a preset second bearing full-cycle path, and generating the life cycle stages of the multiple bearings based on the second similarity result, further includes: Each second bearing evolution path is compared with a preset second bearing full-cycle path for similarity and a third similarity result is generated. Based on a preset weight ratio, the first similarity result and the third similarity result are combined to generate a second similarity result. The second similarity results are sorted in descending order of similarity to generate a second similarity ranking. The life cycle stages of the multiple bearings are generated based on the second similarity ranking.

12. The life cycle analysis method according to claim 9, characterized in that, The life cycle analysis method also includes: The operating and development speed of each bearing is determined based on the geodesic distance between the evolution path of the first bearing and the geodesic distance between the full cycle path of the first bearing. Clustering is performed on the multiple bearings based on their life cycle stages and operational development speed, and a fourth clustering result is generated.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-8; or When the program is executed by the processor, it implements the method as described in any one of claims 9-12.

14. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1-8; or When the processor executes the program, it implements the method as described in any one of claims 9-12.

Images