A connector fretting wear fault classification method based on electrical signal fusion

Abstract

This invention discloses a connector fretting wear fault classification method based on electrical signal fusion, relating to the field of data processing technology. The method includes: acquiring multi-signal data and environmental disturbance data of the connector, and performing frequency domain expansion on the multi-signal data to generate a spectrum matrix; constructing a coupling feature matrix based on the spectrum matrix and environmental disturbance data, and performing bidirectional orthogonal decomposition on the coupling feature matrix to generate a wear-dominant representation matrix and an environmental coupling representation matrix; constructing an initial classification boundary matrix based on the wear-dominant representation matrix, and generating a drift mapping matrix based on the environmental coupling representation matrix. This invention dynamically corrects the initial classification boundary matrix by constructing a decoupling representation relationship between the wear-dominant representation matrix and the environmental coupling representation matrix, combined with the drift mapping matrix, so that the stable classification boundary matrix can maintain a consistent discrimination scale as the environmental coupling state changes, thereby achieving stable discrimination of fretting wear fault classification.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and specifically to a method for classifying connector micro-motion wear faults based on electrical signal fusion. Background Technology

[0002] In long-term vibration service scenarios such as rail transit equipment, power connectors, automotive wiring harness connection structures, and avionics interfaces, connectors are prone to fretting wear under alternating loads and micro-displacement. Contact voltage signal data, contact current signal data, and contact resistance signal data can directly reflect the electrical change characteristics of the contact interface. At the same time, vibration acceleration data, ambient temperature data, and normal load data can characterize the external operating conditions. Frequency domain analysis based on multiple electrical signal data and feature fusion combined with environmental disturbance data have become an important technical path for identifying connector fretting wear conditions and provide a data foundation for fault early warning and life assessment.

[0003] Currently, during the operation of connectors in a complex environment with continuous changes in disturbances, vibration acceleration data, ambient temperature data, and normal load data exhibit a superposition of periodic and random fluctuations. The frequency domain characteristics of the multi-signal data are synchronously shifted, making the feature representation based on the spectrum matrix prone to coupling with environmental disturbances. When the environmental coupling state accumulates and changes over a long period of time, the distance relationship between the wear-dominant representation vector and the category center vector may systematically shift, making it difficult to maintain a stable discrimination scale at the classification boundary, thereby affecting the consistency of the fretting wear fault category determination.

[0004] Secondly, under long-term operating conditions, the distribution structure between different fault categories will gradually migrate as the environmental coupling state changes. If the classification boundary lacks a dynamic correction mechanism for the drift trend, the matching relationship between the initial classification boundary matrix and the actual feature space will gradually weaken, making it difficult for the new wear-dominant representation vector to maintain a unified discrimination standard in the stable classification boundary matrix. Consequently, during the multi-condition cyclic switching process, the discrimination results fluctuate and the classification center of gravity drifts. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a method for classifying connector fretting wear faults based on electrical signal fusion.

[0006] A connector fretting wear fault classification method based on electrical signal fusion includes:

[0007] Collect multi-signal data and environmental disturbance data from the connector, and perform frequency domain expansion on the multi-signal data to generate a spectrum matrix;

[0008] A coupled feature matrix is ​​constructed based on the spectrum matrix and environmental disturbance data, and a bidirectional orthogonal decomposition is performed on the coupled feature matrix to generate a wear-dominant representation matrix and an environmental coupling representation matrix;

[0009] An initial classification boundary matrix is ​​constructed based on the wear-dominant representation matrix, and a drift mapping matrix is ​​generated based on the environment coupling representation matrix;

[0010] A stable classification boundary matrix is ​​generated by performing boundary projection correction on the initial classification boundary matrix using the drift mapping matrix, and the fretting wear fault classification result is output based on the stable classification boundary matrix.

[0011] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0012] This invention constructs a coupled feature matrix based on the spectrum matrix and environmental disturbance data in an operating scenario with continuously changing environmental disturbances, and forms an independent representation relationship between the wear-dominant representation matrix and the environmental coupled representation matrix. This makes the criteria for judging the fretting wear state no longer directly affected by the synchronous interference of vibration acceleration data, ambient temperature data and normal load data fluctuations, thereby maintaining the stability of wear feature expression under the condition of long-term drift at the classification boundary.

[0013] Furthermore, this invention also performs boundary offset constraints and consistency correction on the initial classification boundary matrix based on the drift mapping matrix, so that the stable classification boundary matrix can be adaptively adjusted according to the changing direction of the environmental coupling state. When new multi-electric signal data enters the discrimination process, the boundary discrimination calculation is completed according to the correspondence between the stable classification boundary matrix and the category center matrix, maintaining the consistency of the fretting wear fault classification results under different working conditions and suppressing the cumulative impact of classification center offset on discrimination accuracy during long-term operation.

[0014] In summary, under operating conditions where environmental disturbances are constantly changing and classification boundaries drift over a long period of time, this invention constructs a decoupled representation relationship between the wear-dominant representation matrix and the environmental coupling representation matrix, and combines this with a drift mapping matrix to dynamically correct the initial classification boundary matrix. This allows the stable classification boundary matrix to maintain a consistent discrimination scale as the environmental coupling state changes, thereby achieving stable discrimination of fretting wear fault classification. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0016] Figure 1The flowchart shows a connector fretting wear fault classification method based on electrical signal fusion provided by the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see Figure 1 As shown in the figure, this embodiment discloses a connector fretting wear fault classification method based on electrical signal fusion, the method including:

[0019] S11: Collect multi-signal data and environmental disturbance data of the connector, and perform frequency domain expansion on the multi-signal data to generate a spectrum matrix;

[0020] In one specific embodiment, the connector's multiple electrical signal data includes contact voltage signal data, contact current signal data, and contact resistance signal data, and the environmental disturbance data includes vibration acceleration data, ambient temperature data, and normal load data;

[0021] Among them, the contact voltage signal data is acquired through a voltage acquisition probe, the contact current signal data is acquired through a current sensor, and the contact resistance signal data is obtained by converting the contact voltage signal data and the contact current signal data.

[0022] The formula for calculating contact resistance signal data is as follows:

[0023]

[0024] In the formula, For the first Contact resistance signal data at each sampling time. For the first Contact voltage signal data at each sampling time. For the first Contact current signal data at each sampling time. Number the sampling time;

[0025] It should be noted that when the contact current signal data is less than the preset lower current limit at the corresponding sampling time, the contact resistance signal data at that sampling time is set to 0.

[0026] Considering that the original acquisition frequencies of connector contact voltage signal data, contact current signal data, ambient temperature data, and normal load data may be inconsistent, before performing time alignment, the data is first collected based on the total number of sampling points. Perform linear interpolation or resampling on the environmental disturbance data to ensure its time base is perfectly aligned with the multi-signal data, guaranteeing that each set of multi-signal data is in sync with the original data. There are corresponding environmental disturbance characteristics at all times.

[0027] The representation of multi-signal data is as follows:

[0028]

[0029] in, For multiple electrical signal data, This represents the total number of sampling times.

[0030] Vibration acceleration data is collected using an accelerometer, ambient temperature data is collected using a temperature sensor, and normal load data is collected using a load sensor.

[0031] Contact voltage signal data, contact current signal data, and contact resistance signal data are arranged in channel order to obtain multi-electrical signal data. Vibration acceleration data, ambient temperature data, and normal load data are time-aligned according to the same sampling time base as the multi-electrical signal data to obtain environmental disturbance data.

[0032] Specifically, the steps for generating the spectrum matrix are as follows:

[0033] S111: Perform sliding window segmentation using multiple electrical signal data to generate a set of signal segments;

[0034] In a specific embodiment, the multi-electric signal data is divided into sliding window segments according to a preset window length and a preset sliding step size to obtain signal segments corresponding to multiple consecutive time intervals.

[0035] Among them, the The formula for calculating the starting position of a signal segment is:

[0036]

[0037] In the formula, For the first The starting position of each signal segment Preset sliding step size;

[0038] It should be noted that the preset sliding step size... These are parameters preset based on the sampling frequency and time-varying characteristics of the multi-signal data.

[0039] No. The formula for calculating the end position of a signal segment is:

[0040]

[0041] For the first The end position of each signal segment This is the preset window length.

[0042] It should be noted that the preset window length These are parameters preset based on the sampling frequency and required frequency resolution of the multi-signal data.

[0043] The formula for calculating the total number of signal segments is:

[0044]

[0045] No. The formula for extracting a signal segment is:

[0046]

[0047] In the formula, For the first a signal segment

[0048] The representation of a set of signal segments based on multiple signal segments is as follows:

[0049]

[0050] in, It is a set of signal segments.

[0051] S112: Perform amplitude normalization based on the multi-signal data and signal segment set to generate a normalized signal segment set;

[0052] In a specific embodiment, since the contact voltage signal data, contact current signal data and contact resistance signal data in the multi-electrical signal data have different dimensions, amplitude normalization is performed on each signal segment in the signal segment set according to the channel to obtain a normalized signal segment set.

[0053] The maximum and minimum values ​​required for amplitude normalization are taken from the corresponding channel data in the corresponding signal segment.

[0054] The first In the signal segment, the first The first channel in the The original values ​​of each position are denoted as ;

[0055] Then the first In the signal segment, the first The formula for calculating the minimum value of each channel is:

[0056]

[0057] No. In the signal segment, the first The formula for calculating the maximum value of each channel is:

[0058]

[0059] Normalized In the signal segment, the first The first channel in the The formula for calculating the value at each position is:

[0060]

[0061] The normalized signal segment set is represented as follows:

[0062]

[0063] in, For the normalized first In the signal segment, the first The first channel in the The value at each position, For the first In the signal segment, the first The minimum value of each channel, For the first In the signal segment, the first The maximum value of each channel. To prevent real constants with a denominator of zero, For the first A normalized signal segment, It is a set of normalized signal segments.

[0064] It should be noted that the normalized values ​​range from 0 to 1. At that time, the first In the signal segment, the first The normalization result of each channel is uniformly assigned a value of 0.

[0065] S113: Perform windowing calculations on the normalized set of signal segments to generate a windowed signal matrix.

[0066] In a specific embodiment, windowing calculation is performed on each normalized signal segment in the set of normalized signal segments to obtain the corresponding windowed signal matrix;

[0067] It should be noted that windowing calculation is used to reduce the impact of signal segment boundary truncation on the frequency domain expansion results.

[0068] Specifically, the steps for generating the windowed signal matrix are as follows:

[0069] S113.1: Construct a window function sequence based on the set of normalized signal segments, and generate a window function matrix.

[0070] In one specific embodiment, a window function sequence is constructed based on the length of each normalized signal segment in the set of normalized signal segments, and then the window function sequence is copied along the channel direction to generate a window function matrix with the same dimension as the normalized signal segments.

[0071] The formula for calculating the window function sequence is:

[0072]

[0073] The window function matrix is ​​represented as follows:

[0074]

[0075] in, For the window function sequence at the th The value at each position, It is the window function matrix.

[0076] It should be noted that this embodiment uses a Hamming window to construct the window function sequence. When the number of channels for the multi-signal data is... When the preset window length is denoted as L, the number of rows in the window function matrix is ​​set to... The number of columns in the window function matrix is ​​set to .

[0077] S113.2: Generate the initial windowed signal matrix by performing element-wise multiplication between the normalized signal segment set and the window function matrix.

[0078] In a specific embodiment, the first segment in the set of normalized signal segments... The normalized signal segment is element-wise multiplied with the window function matrix to obtain the... An initial windowed signal matrix;

[0079] No. The values ​​at each position in the initial windowed signal matrix are determined by the first... The value at the corresponding position in a normalized signal segment is obtained by multiplying the value at the corresponding position in the window function matrix.

[0080] No. The formula for calculating the initial windowed signal matrix is:

[0081]

[0082] No. In the initial windowed signal matrix, the first... The first channel in the The formula for calculating the value at each position is:

[0083]

[0084] in, For the first An initial windowed signal matrix, This is an element-wise multiplication operation. For the first In the nth windowed signal matrix The first channel in the The value at each position.

[0085] S113.3: Perform amplitude range correction based on the initial windowed signal matrix to generate a windowed signal matrix.

[0086] In a specific embodiment, amplitude range correction is performed on the initial windowed signal matrix obtained in step S113.2 to obtain the corrected windowed signal matrix;

[0087] Amplitude range correction is used to compensate for the overall amplitude reduction introduced by the window function, so that the windowed signal matrix corresponding to different signal segments maintains a consistent amplitude scale.

[0088] The formula for calculating the average weight of the window function sequence is:

[0089]

[0090] The corrected first The formula for calculating a windowed signal matrix is:

[0091]

[0092] The corrected first In the nth windowed signal matrix The first channel in the The formula for calculating the value at each position is:

[0093]

[0094] in, For average weight, For the corrected first A windowed signal matrix, For the corrected first In the nth windowed signal matrix The first channel in the The value at each position.

[0095] S114: Perform a fast Fourier transform based on the windowed signal matrix to generate a spectrum matrix.

[0096] In a specific embodiment, fast Fourier transform is performed on each windowed signal matrix output in step S113.3 by channel to obtain the spectral moments of the corresponding signal segments;

[0097] Then, arrange all the spectral moments in the order of the signal segment numbers to generate a spectral matrix.

[0098] No. The spectral moment of the th The first channel in the The formula for calculating the complex spectrum value at each frequency position is:

[0099]

[0100] No. The spectral moment of the th The first channel in the The formula for calculating the spectral amplitude at each frequency position is:

[0101]

[0102] The formula for calculating the number of effective frequency points is:

[0103]

[0104] No. The spectral moments are represented as follows:

[0105]

[0106] The spectrum matrix is ​​represented as follows:

[0107]

[0108] in, For the first The spectral moment of the th The first channel in the Complex spectrum values ​​at each frequency position, For the first The spectral moment of the th The first channel in the Spectral amplitude at each frequency position, The number of effective frequency points, For the first One spectral moment, For the spectrum matrix, It is the imaginary unit.

[0109] It should be noted that the effective frequency points correspond to a frequency position range from the first frequency position to the second. Each frequency position;

[0110] The frequency location data within this range is retained to remove repetitive spectral results caused by conjugate symmetry.

[0111] S12: Construct a coupled feature matrix based on the spectrum matrix and environmental disturbance data, and perform bidirectional orthogonal decomposition on the coupled feature matrix to generate a wear-dominant representation matrix and an environmental coupled representation matrix.

[0112] In one specific embodiment, the spectrum matrix output in step S11 includes multiple spectrum moments arranged in the order of signal segments, and the environmental disturbance data includes vibration acceleration data, ambient temperature data, and normal load data.

[0113] After aligning the frequency domain features corresponding to each spectral moment in the spectrum matrix with the environmental disturbance data within the same signal segment time interval, feature splicing is performed to generate a coupled feature matrix. Then, based on the coupled feature matrix, mean elimination, covariance calculation, and bidirectional orthogonal decomposition are performed sequentially to generate the wear-dominant representation matrix and the environmental coupled representation matrix.

[0114] Specifically, the logic for generating the wear-dominant representation matrix and the environment-coupled representation matrix is ​​as follows:

[0115] S121: Perform feature concatenation based on the spectrum matrix and environmental disturbance data to generate a coupled feature matrix.

[0116] In a specific embodiment, vector expansion is first performed on each spectral moment in the spectral matrix to obtain the spectral features corresponding to each signal segment;

[0117] Then, the environmental disturbance data is divided into windows according to the start and end positions in step S111, and the window mean values ​​of vibration acceleration data, ambient temperature data and normal load data are calculated in each window to obtain the environmental disturbance characteristics corresponding to each signal segment.

[0118] Finally, the spectral features and environmental disturbance features corresponding to the same signal segment are concatenated in the column direction to generate a coupled feature matrix.

[0119] No. The formula for calculating the spectral eigenvector corresponding to each spectral moment is:

[0120]

[0121] in, For the first The spectral eigenvectors corresponding to each spectral moment This is a column-wise expansion operation.

[0122] No. The formula for calculating the mean value of the vibration acceleration data window corresponding to each signal segment is:

[0123]

[0124] No. The formula for calculating the mean value of the ambient temperature data window corresponding to each signal segment is:

[0125]

[0126] No. The formula for calculating the mean value of the normal load data window corresponding to each signal segment is:

[0127]

[0128] No. The environmental disturbance feature vector corresponding to each signal segment is represented as follows:

[0129]

[0130] No. The coupled feature vector corresponding to each signal segment is represented as follows:

[0131]

[0132] The coupling characteristic matrix is ​​represented as follows:

[0133]

[0134] in, For the first The mean of the vibration acceleration data window corresponding to each signal segment For the first The mean of the ambient temperature data window corresponding to each signal segment. For the first Mean value of normal load data window corresponding to each signal segment For the first The environmental disturbance feature vector corresponding to each signal segment For the first The coupling feature vector corresponding to each signal segment This is the coupling characteristic matrix.

[0135] S122: Mean elimination is performed by coupling the feature matrix to generate a centered feature matrix.

[0136] In one specific embodiment, mean elimination is performed on the coupled feature matrix column by column. First, the feature mean of each column is calculated. Then, the feature mean of the corresponding column is subtracted from the value of each column in the coupled feature matrix to obtain an initial centered matrix. Finally, numerical range constraints are applied to the initial centered matrix to generate a centered feature matrix.

[0137] Specifically, the steps for generating the centered feature matrix are as follows:

[0138] S122.1: Calculate the mean values ​​of each column of features based on the coupled feature matrix, and generate a feature mean vector.

[0139] In one specific embodiment, the column mean is calculated for each column in the coupling feature matrix to generate a feature mean vector;

[0140] Each position in the feature mean vector corresponds one-to-one with the corresponding column in the coupled feature matrix.

[0141] No. The formula for calculating the mean of column features is:

[0142]

[0143] The characteristic mean vector is represented as follows:

[0144]

[0145] in, For the first Column feature mean, The feature mean vector, The column number of the coupling characteristic matrix. The first element in the coupling characteristic matrix Line number The values ​​in the column.

[0146] S122.2 generates an initial centered matrix by subtracting the feature mean vector from the coupled feature matrix.

[0147] In one specific embodiment, the feature mean vector is copied row by row into a matrix form with the same number of rows as the coupled feature matrix, and then the copied result corresponding to the feature mean vector is subtracted from the coupled feature matrix to generate the initial centered matrix.

[0148] The replication matrix corresponding to the feature mean vector is represented as:

[0149]

[0150] The formula for calculating the initial centered matrix is:

[0151]

[0152] in, This is the copy matrix corresponding to the eigenvalue mean vector. This is the initial centered matrix.

[0153] S122.3: Perform numerical range constraints based on the initial centralized matrix to generate a centralized feature matrix.

[0154] In one specific embodiment, numerical range constraints are applied to the values ​​at each position in the initial centered matrix to generate a centered feature matrix;

[0155] Numerical range constraints are used to limit the impact of abnormal offset values ​​on the subsequent covariance matrix calculation results.

[0156] The centered feature matrix Line number The formula for calculating column values ​​is:

[0157]

[0158] in, The centered feature matrix Line number The values ​​in the column, This is an upper limit constraint on the numerical range;

[0159] It should be noted that the upper limit of the numerical range constraint Statistical settings are made based on the historical overload data fluctuation range of the connector in a laboratory environment.

[0160] The centered feature matrix is ​​represented as follows:

[0161]

[0162] It should be noted that the range of values ​​for each position in the centered feature matrix is ​​as follows: .

[0163] S123: Calculate the covariance matrix based on the centered feature matrix to generate the covariance matrix.

[0164] In one specific embodiment, a transpose matrix multiplication is performed based on the centered feature matrix to generate a covariance matrix;

[0165] The covariance matrix is ​​used to characterize the joint fluctuation relationship between the columns of the centered feature matrix.

[0166] The formula for calculating the covariance matrix is:

[0167]

[0168] in, Let covariance matrix be the variance matrix. It is the transpose of the centered feature matrix.

[0169] S124: Perform bidirectional orthogonal decomposition using the covariance matrix to generate the wear-dominant subspace matrix and the environment coupling subspace matrix.

[0170] In a specific embodiment, the covariance matrix is ​​first divided into blocks according to the concatenation order of the spectral feature vector and the environmental disturbance feature vector in step S121.

[0171] Then, perform orthogonal eigenvalue decomposition on the covariance submatrix corresponding to the spectral features and the covariance submatrix corresponding to the environmental disturbance features, respectively;

[0172] Finally, the feature vectors corresponding to the principal feature values ​​are extracted to generate the wear-dominant subspace matrix and the environment coupling subspace matrix.

[0173] Let the dimension of the spectral feature vector be... The dimension of the environmental disturbance feature vector is Then we have:

[0174]

[0175] The block form of the covariance matrix is:

[0176]

[0177] in, The covariance submatrix corresponding to the spectral features. This is the covariance submatrix corresponding to the environmental disturbance characteristics. This is the cross-covariance submatrix between spectral features and environmental disturbance features. It is the cross-covariance submatrix between environmental disturbance characteristics and spectral characteristics.

[0178] The orthogonal eigenvalue decomposition formula for the covariance submatrix corresponding to the spectral features is:

[0179]

[0180] The orthogonal eigenvalue decomposition formula for the covariance submatrix corresponding to environmental disturbance characteristics is:

[0181]

[0182] in, This is the eigenvector matrix corresponding to the spectral features. This is the eigenvalue matrix corresponding to the spectral features. This represents the eigenvector matrix corresponding to the environmental disturbance characteristics. This is the eigenvalue matrix corresponding to the environmental disturbance characteristics.

[0183] The wear-dominant subspace matrix is ​​represented as follows:

[0184]

[0185] The representation of the environment coupling subspace matrix is ​​as follows:

[0186]

[0187] in, For wear-dominant subspace matrix, The environment coupling subspace matrix;

[0188] for Sort by eigenvalue in descending order. 1 eigenvector for Sort by eigenvalue in descending order. 1 eigenvector The number of columns in the wear-dominant subspace matrix. denoted as the number of columns in the environment coupling subspace matrix.

[0189] It should be noted that: and The preset positive integer is used to control the number of eigenvectors retained in the wear-dominant subspace matrix and the environment coupling subspace matrix;

[0190] Number of columns in the wear-dominant subspace matrix The selection principle is to ensure that the cumulative contribution rate of the corresponding eigenvalues ​​reaches 85% to 95%; the number of columns of the environmental coupling subspace matrix. The value is usually set to the number of channels for environmental disturbance data, and is controlled by... and The value of can effectively reduce the dimensionality of high-dimensional features while retaining the core physical information.

[0191] S125: Perform feature projection on the centered feature matrix based on the wear-dominant subspace matrix and the environment coupling subspace matrix to generate the wear-dominant representation matrix and the environment coupling representation matrix;

[0192] In one specific embodiment, the wear-dominant subspace matrix is ​​used as a projection basis to project the column data corresponding to the spectral features in the centered feature matrix to a low-dimensional space to obtain the wear-dominant representation vector corresponding to each signal segment.

[0193] The formula for calculating the wear-dominant representation matrix is ​​as follows:

[0194]

[0195] In the formula, The wear-dominant representation matrix, For wear-dominant subspace matrix, For the center of the feature matrix, the first Spectral feature sub-blocks composed of columns;

[0196] Similarly, by projecting the column data corresponding to the environmental disturbance features using the environmental coupling subspace matrix, we can obtain the environmental coupling representation vector corresponding to each signal segment.

[0197] The formula for calculating the environment coupling representation matrix is ​​as follows:

[0198]

[0199] In the formula, The environment coupling representation matrix, In the centered feature matrix The environmental disturbance feature sub-blocks are composed of columns. The environment coupling subspace matrix;

[0200] It should be noted that the projection operation maps the original high-dimensional features to the physical attribute space defined by the orthogonal basis, realizing a deep decoupling between wear features and environmental disturbance features, and providing a data foundation for the subsequent construction of a stable classification boundary.

[0201] S13: Construct an initial classification boundary matrix based on the wear-dominant representation matrix, and generate a drift mapping matrix based on the environment coupling representation matrix;

[0202] In a specific embodiment, the wear-dominant representation matrix output in step S12 is used to characterize the distribution of each signal segment in the wear-dominant direction, and the environmental coupling subspace matrix is ​​used to characterize the distribution of each signal segment in the environmental coupling direction.

[0203] First, calculate the set of category center vectors based on the wear-dominant representation matrix and the fault category labels corresponding to each signal segment. Then, calculate the boundary spacing matrix based on the set of category center vectors and solve the initial classification boundary matrix based on the boundary spacing matrix. Finally, perform time series difference calculation based on the environmental coupling representation matrix to generate the drift mapping matrix.

[0204] It should be noted that the fault category label for each signal segment is given by the labeled historical samples of the connector fretting wear test platform. Each signal segment corresponds to a fault category label, which is used to distinguish different fretting wear fault categories.

[0205] Specifically, the steps for constructing the initial classification boundary matrix and generating the drift mapping matrix are as follows:

[0206] S131: Calculate the set of category center vectors based on the wear-dominant representation matrix, and generate the category center matrix.

[0207] In one specific embodiment, the wear-dominant representation matrix is... Each row serves as a wear-dominant representation vector corresponding to a signal segment, i.e., let the wear-dominant representation matrix be... The Behavior , as the first The wear-dominant representation vector corresponding to each signal segment;

[0208] Then, the wear-dominant representation vectors are classified and summarized according to the fault category labels, and the mean vectors corresponding to each fault category are calculated to obtain the set of category center vectors;

[0209] Then, the set of category center vectors is arranged in order of fault category number to generate a category center matrix.

[0210] Based on the wear-dominant representation matrix Total number of signal segments Wear-dominant representation of the number of columns in the matrix , No. The wear-dominant representation vector corresponding to each signal segment is denoted as ;

[0211] Let the total number of fault categories be , No. The set of sample numbers corresponding to each fault category is denoted as . Then the first The formula for calculating the category center vector corresponding to each fault category is:

[0212]

[0213] The category center matrix is ​​represented as follows:

[0214]

[0215] in, For the first The wear-dominant representation vector corresponding to each signal segment For the first The set of sample numbers corresponding to each fault category For the first The number of samples corresponding to each fault category For the first The category center vector corresponding to each fault category This is the category center matrix.

[0216] S132: Perform vector spacing calculation using the class center matrix to generate the boundary spacing matrix.

[0217] In one specific embodiment, firstly, the class center vector pairs are extracted based on the class center matrix, then Euclidean distance is calculated for each class center vector pair to generate an initial distance matrix, and finally, symmetry correction is performed based on the initial distance matrix to generate a boundary spacing matrix.

[0218] Specifically, the steps for generating the boundary spacing matrix are as follows:

[0219] S132.1: Extract the center vector pairs of each category based on the category center matrix, and generate a set of center vector pairs.

[0220] In a specific embodiment, category center vectors are extracted from the category center matrix by combining them pairwise according to category numbers to generate a set of center vector pairs;

[0221] Each element in the set of center vector pairs corresponds to a set of category center vector pairs between different fault categories.

[0222] The central vector pair set is represented as follows:

[0223]

[0224] in, For the set of central vector pairs, For the first The category center vector corresponding to each fault category For the first The category center vector corresponding to each fault category.

[0225] S132.2: Perform Euclidean distance calculation on the set using the center vector to generate the initial distance matrix.

[0226] In a specific embodiment, vector difference calculation is first performed on each category of center vector pair in the center vector pair set to obtain a difference vector set. Then, the sum of squares is calculated on the difference vector set to obtain a distance square sequence. Finally, the square root calculation is performed on the distance square sequence to generate an initial distance matrix.

[0227] Specifically, the steps for generating the initial distance matrix are as follows:

[0228] S132.2.1: Perform vector difference calculation on the set based on the center vector to generate a set of difference vectors.

[0229] In one specific embodiment, a subtraction operation is performed on each group of category center vector pairs in the center vector pair set to generate a set of difference vectors.

[0230] The set of difference vectors is represented as follows:

[0231]

[0232] in, It is a set of difference vectors. For the first The category center vector corresponding to the fault category is the first one. The difference vector of the category center vectors corresponding to each fault category.

[0233] S132.2.2: Perform sum of squares calculation using the set of difference vectors to generate a sequence of squared distances.

[0234] In one specific embodiment, the summation operation is performed on each difference vector in the difference vector set, with each position squared, to generate a distance squared sequence.

[0235] No. The fault category and the first The formula for calculating the squared distance between each fault category is:

[0236]

[0237] The squared distance sequence is represented as follows:

[0238]

[0239] in, For the first The fault category and the first The squared distance between each fault category Difference vector In the The value at each position, It is a sequence of squared distances.

[0240] S132.2.3: Perform square root calculation based on the squared distance sequence to generate the initial distance matrix.

[0241] In one specific embodiment, the square root operation is performed on each squared distance value in the distance square sequence, and the matrix is ​​written according to the positional relationship corresponding to the fault category number to generate an initial distance matrix.

[0242] No. The fault category and the first The formula for calculating the initial distance between each fault category is:

[0243]

[0244] The initial distance matrix is ​​represented as follows:

[0245]

[0246] in, For the first The fault category and the first Initial distance values ​​between each fault category This is the initial distance matrix.

[0247] S132.3: Perform symmetry correction based on the initial distance matrix to generate the boundary spacing matrix.

[0248] In one specific embodiment, transpose completion is performed on the initial distance matrix, copying the initial distance values ​​in the upper triangular region to the corresponding positions in the lower triangular region, and uniformly assigning values ​​to the diagonal positions. Generate the boundary spacing matrix.

[0249] The first boundary spacing matrix Line number The formula for calculating column values ​​is:

[0250]

[0251] The boundary spacing matrix is ​​represented as follows:

[0252]

[0253] in, The first in the boundary spacing matrix Line number The values ​​in the column, This is the boundary spacing matrix.

[0254] S133: Perform boundary calculation based on the boundary spacing matrix to generate the initial classification boundary matrix.

[0255] In a specific embodiment, based on the spacing results between each fault category in the boundary spacing matrix, a corresponding boundary position value is constructed for each fault category, and then all boundary position values ​​are arranged in order of fault category to generate an initial classification boundary matrix.

[0256] No. The fault category and the first The boundary location value between each fault category is taken as half the distance between them.

[0257] No. The fault category and the first The formula for calculating the boundary location values ​​between fault categories is:

[0258]

[0259] The first classification boundary matrix Line number The formula for calculating column values ​​is:

[0260]

[0261] The initial classification boundary matrix is ​​represented as follows:

[0262]

[0263] in, For the first The fault category and the first Boundary position values ​​between fault categories The first in the initial classification boundary matrix Line number The values ​​in the column, This is the initial classification boundary matrix.

[0264] S134: Perform time series difference calculation based on the environment coupling representation matrix to generate a drift mapping matrix.

[0265] In one specific embodiment, the environment is coupled to the representation matrix. Each row serves as an environment coupling representation vector corresponding to a signal segment, where the first row... The environmental coupling representation vector corresponding to each signal segment is denoted as . ;

[0266] Perform differential calculations between adjacent time segments in sequence according to their signal segment numbers to generate the environmental drift vector. The formula for calculating the environmental drift vector corresponding to each signal segment is:

[0267]

[0268] in, For the first The environmental drift vector corresponding to each signal segment ;

[0269] An environmental drift sequence is constructed based on multiple sets of environmental drift vectors and in the order of signal segment numbers.

[0270] A drift mapping matrix is ​​generated by performing column-by-column averaging based on the environmental drift sequence.

[0271] Among them, the first in the drift mapping matrix The formula for calculating the value at each position is:

[0272]

[0273] The drift mapping matrix is ​​represented as follows:

[0274]

[0275] in, For the drift mapping matrix in the th The value at each position, For the drift mapping matrix, For the first The environmental drift vector corresponding to the signal segment is at the _th The value at each position.

[0276] It should be noted that step S134 uses the time series difference results between adjacent signal segments to characterize the direction and magnitude of change in the environmental coupling state;

[0277] The drift mapping matrix is ​​used to characterize the nonlinear shift trend of environmental disturbances on the signal feature space, reflecting the evolution law of fretting wear state under different environmental couplings. By using this matrix to perform boundary projection correction on the initial classification boundary matrix in step S14, the classification center shift caused by environmental fluctuations can be dynamically offset, improving the stability of classification results under varying working conditions.

[0278] S14: Perform boundary projection correction on the initial classification boundary matrix using the drift mapping matrix to generate a stable classification boundary matrix, and output the fretting wear fault classification result based on the stable classification boundary matrix;

[0279] In a specific embodiment, matrix mapping calculation is first performed based on the drift mapping matrix and the initial classification boundary matrix to generate a boundary offset matrix. Then, norm normalization is performed on the boundary offset matrix to generate a normalized boundary matrix. Subsequently, boundary consistency correction is performed based on the normalized boundary matrix to generate a stable classification boundary matrix. Finally, boundary discrimination calculation is performed on the new multi-electrical signal data based on the stable classification boundary matrix to generate the fretting wear fault classification result.

[0280] Specifically, the steps for outputting the fretting wear fault classification results are as follows:

[0281] S141: Perform matrix mapping calculation based on the drift mapping matrix and the initial classification boundary matrix to generate the boundary offset matrix.

[0282] In one specific embodiment, the mapping coefficient matrix is ​​first extracted based on the drift mapping matrix, then matrix multiplication is performed between the mapping coefficient matrix and the initial classification boundary matrix to generate the initial offset matrix, and finally magnitude constraints are performed based on the initial offset matrix to generate the boundary offset matrix.

[0283] Specifically, the steps for generating the boundary offset matrix are as follows:

[0284] S141.1: Extract the mapping coefficient matrix from the drift mapping matrix and generate the mapping coefficient matrix.

[0285] In one specific embodiment, absolute value transformation and diagonal writing are performed on the values ​​at each position in the drift mapping matrix to generate a mapping coefficient matrix;

[0286] The values ​​at the diagonal positions in the mapping coefficient matrix correspond to the magnitude results of the values ​​at each position in the drift mapping matrix, while the values ​​at the off-diagonal positions are uniformly assigned a value. .

[0287] Let the drift mapping matrix be The drift mapping matrix of the first The value at each position is The total number of positions in the drift mapping matrix is Then the first in the mapping coefficient matrix Line number The formula for calculating column values ​​is:

[0288]

[0289] The mapping coefficient matrix is ​​represented as follows:

[0290]

[0291] in, The first in the mapping coefficient matrix Line number The values ​​in the column, This is the mapping coefficient matrix.

[0292] It should be noted that: It is a diagonal matrix. When the dimension of the initial classification boundary matrix is ​​inconsistent with the dimension of the mapping coefficient matrix, the dimension of the mapping coefficient matrix is ​​expanded.

[0293] The dimension expansion method is to sequentially write the diagonal values ​​in the mapping coefficient matrix into the extended diagonal matrix according to the number of rows and columns of the initial classification boundary matrix. The expanded mapping coefficient matrix maintains the same dimension as the initial classification boundary matrix.

[0294] For example, if the dimension of the environmental disturbance characteristics is That is, the drift mapping matrix The total number of positions is 3, and the corresponding drift characteristic value is ;

[0295] Meanwhile, assuming the total number of fault categories is Then the initial classification boundary matrix The dimension is ,because For the mapping coefficient matrix Perform dimensional expansion to make it from Expand to ;

[0296] During the expansion process, diagonal elements are arranged according to... The sequence is filled cyclically, specifically as follows:

[0297]

[0298] This yields the extended mapping coefficient matrix:

[0299]

[0300] By using the above-mentioned cyclic filling method, the classification boundary corresponding to each fault category can receive the projection feedback of environmental drift characteristics.

[0301] Even the classification space dimension Higher than the perturbation feature dimension Matrix operations can still be performed. Achieve dynamic correction of classification boundaries across dimensional spaces.

[0302] S141.2: Generate the initial offset matrix by performing matrix multiplication between the mapping coefficient matrix and the initial classification boundary matrix;

[0303] In a specific embodiment, the mapping coefficient matrix output in step S141.1 and the initial classification boundary matrix output in step S133 are multiplied by matrix to generate an initial offset matrix.

[0304] The values ​​at each position in the initial offset matrix are used to characterize the boundary offset result of the initial classification boundary matrix under the action of drift mapping.

[0305] The formula for calculating the initial offset matrix is:

[0306]

[0307] in, This is the initial offset matrix. The mapping coefficient matrix, This is the initial classification boundary matrix.

[0308] S141.3: Perform magnitude constraints based on the initial offset matrix to generate the boundary offset matrix.

[0309] In one specific embodiment, magnitude constraints are applied to the values ​​at each position in the initial offset matrix to generate a boundary offset matrix;

[0310] Amplitude constraints are used to limit the impact of excessively large values ​​at a single position in the initial offset matrix on the subsequent boundary normalization results.

[0311] Let the upper limit of the amplitude constraint be Then the first element in the boundary offset matrix Line number The formula for calculating column values ​​is:

[0312]

[0313] The boundary offset matrix is ​​represented as follows:

[0314]

[0315] in, The first in the boundary offset matrix Line number The values ​​in the column, This is the boundary offset matrix. This represents the upper limit of the amplitude constraint.

[0316] It should be noted that: the upper limit of the amplitude constraint The threshold parameter is pre-set based on the statistical distribution of boundary offsets in historical training samples. The value range of each position in the boundary offset matrix is ​​as follows: .

[0317] S142: Perform norm normalization using the boundary offset matrix to generate a normalized boundary matrix.

[0318] In one specific embodiment, the boundary offset matrix is ​​subjected to row-wise L2 norm calculation, and then the values ​​of each row are divided by the corresponding row's L2 norm to generate a normalized boundary matrix. Norm normalization is used to eliminate the overall scale difference between the boundary offset magnitudes of different fault categories.

[0319] The first boundary offset matrix The formula for calculating the L2 norm of a row is:

[0320]

[0321] The first normalized boundary matrix Line number The formula for calculating column values ​​is:

[0322]

[0323] The normalized boundary matrix is ​​represented as follows:

[0324]

[0325] in, The first in the boundary offset matrix The 2-norm of the row, The first in the normalized boundary matrix Line number The values ​​in the column, For the normalized boundary matrix, To prevent real constants with a denominator of zero.

[0326] It should be noted that the L2 norm result for each row in the normalized boundary matrix is ​​normalized to [value missing]. nearby;

[0327] When the L2 norm of a certain row in the boundary offset matrix is At that time, the normalization result corresponding to that row is uniformly assigned a value. .

[0328] S143, perform boundary consistency correction based on the normalized boundary matrix to generate a stable classification boundary matrix.

[0329] In one specific embodiment, symmetric consistency correction and diagonal zeroing are performed on the normalized boundary matrix to generate a stable classification boundary matrix;

[0330] Symmetric consistency correction is used to ensure that the boundary results of different fault categories satisfy the correspondence in the matrix position.

[0331] The stable classification boundary matrix of the first Line number The formula for calculating column values ​​is:

[0332]

[0333] The stable classification boundary matrix is ​​represented as follows:

[0334]

[0335] in, For the stable classification boundary matrix, the first... Line number The values ​​in the column, To stabilize the classification boundary matrix.

[0336] S144: Perform boundary discrimination calculation on the new multi-electrical signal data based on the stable classification boundary matrix to generate fretting wear fault classification results.

[0337] In a specific embodiment, the new multi-electric signal data is first processed by sliding window segmentation, amplitude normalization, windowing calculation and fast Fourier transform according to the processing steps S111 to S114 to obtain a new spectrum matrix.

[0338] The new spectrum matrix is ​​then vector-expanded according to the processing method in step S121, and feature-concatenated with the environmental disturbance data within the same time interval to obtain a new coupled feature matrix.

[0339] The new coupling feature matrix is ​​then subjected to mean elimination according to the processing method in step S122, and projected according to the feature vector matrix corresponding to the spectral features in step S124 to obtain a new wear-dominant representation vector.

[0340] Finally, based on the distance between the new wear-dominant representation vector and the category center vector corresponding to each fault category, and combined with the stable classification boundary matrix, boundary discrimination calculation is performed to generate the fretting wear fault classification result.

[0341] Let the new wear dominant representation vector be , No. The category center vectors corresponding to each fault category are: Then the new wear-dominant representation vector is the same as the first... The formula for calculating the distance between the category center vectors corresponding to each fault category is:

[0342]

[0343] Let the fault category number corresponding to the minimum distance value be . Then we have:

[0344]

[0345] In a specific embodiment, the distance values ​​between the new wear-dominant representation vector and the category center vectors corresponding to the remaining fault categories are respectively compared with the values ​​in the stable classification boundary matrix. Compare the boundary values ​​at corresponding positions in the row;

[0346] When the distance value and the corresponding boundary value meet the preset boundary discrimination condition, output the first... Each fault category is used as the classification result for fretting wear faults.

[0347] No. The boundary discrimination conditions corresponding to each fault category are expressed as follows:

[0348]

[0349] in, To remove Other than the fault category number.

[0350] The classification results of fretting wear faults are represented as follows:

[0351]

[0352] in, For the new wear-dominant representation vector and the first The distance between the category center vectors corresponding to each fault category. The fault category number corresponding to the smallest distance value is assigned. The results show the classification of fretting wear faults.

[0353] It should be noted that when the new wear-dominant representation vector meets the boundary discrimination conditions corresponding to multiple fault categories simultaneously, the fault category corresponding to the smallest distance value is selected as the fretting wear fault classification result.

[0354] When the new wear-dominant representation vector does not meet the boundary discrimination conditions corresponding to all fault categories, the fault category corresponding to the smallest distance value is taken as the fretting wear fault classification result.

[0355] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A connector fretting wear fault classification method based on electrical signal fusion, characterized in that, The method includes: Collect multi-signal data and environmental disturbance data from the connector, and perform frequency domain expansion on the multi-signal data to generate a spectrum matrix; A coupled feature matrix is ​​constructed based on the spectrum matrix and environmental disturbance data, and a bidirectional orthogonal decomposition is performed on the coupled feature matrix to generate a wear-dominant representation matrix and an environmental coupling representation matrix; An initial classification boundary matrix is ​​constructed based on the wear-dominant representation matrix, and a drift mapping matrix is ​​generated based on the environment coupling representation matrix; A stable classification boundary matrix is ​​generated by performing boundary projection correction on the initial classification boundary matrix using the drift mapping matrix, and the fretting wear fault classification result is output based on the stable classification boundary matrix.

2. The connector fretting wear fault classification method based on electrical signal fusion according to claim 1, characterized in that, The steps to generate the spectrum matrix are as follows: A set of signal segments is generated by performing sliding window segmentation on multiple electrical signal data; Amplitude normalization is performed based on the multi-signal data and the set of signal segments to generate a normalized set of signal segments; Windowing calculations are performed on a normalized set of signal segments to generate a windowed signal matrix; Perform a Fast Fourier Transform on the windowed signal matrix to generate a spectrum matrix.

3. The connector fretting wear fault classification method based on electrical signal fusion according to claim 2, characterized in that, The steps for generating the windowed signal matrix are as follows: Construct a sequence of window functions from the set of normalized signal segments, and generate a window function matrix; An initial windowed signal matrix is ​​generated by performing element-wise multiplication between the set of normalized signal segments and the window function matrix. Amplitude range correction is performed based on the initial windowed signal matrix to generate a windowed signal matrix.

4. The connector fretting wear fault classification method based on electrical signal fusion according to claim 1, characterized in that, The logic for generating the wear-dominant representation matrix and the environment-coupled representation matrix is ​​as follows: A coupled feature matrix is ​​generated by performing feature concatenation based on the spectrum matrix and environmental disturbance data. Mean elimination is performed by coupling the feature matrix to generate a centered feature matrix; Calculate the covariance matrix based on the centered feature matrix, and generate the covariance matrix; Bidirectional orthogonal decomposition is performed using the covariance matrix to generate the wear-dominant subspace matrix and the environment coupling subspace matrix; Based on the wear-dominant subspace matrix and the environment coupling subspace matrix, feature projection is performed on the centered feature matrix to generate the wear-dominant representation matrix and the environment coupling representation matrix.

5. The connector fretting wear fault classification method based on electrical signal fusion according to claim 4, characterized in that, The steps to generate the centered feature matrix are as follows: Calculate the mean values ​​of each column of features based on the coupling feature matrix, and generate a feature mean vector; An initial centered matrix is ​​generated by subtracting the feature mean vector from the coupled feature matrix; Numerical range constraints are applied based on the initial centralized matrix to generate a centralized feature matrix.

6. The connector fretting wear fault classification method based on electrical signal fusion according to claim 1, characterized in that, The steps for constructing the initial classification boundary matrix and generating the drift mapping matrix are as follows: Calculate the set of category center vectors based on the wear-dominant representation matrix, and generate the category center matrix; The vector spacing is calculated using the category center matrix to generate the boundary spacing matrix; Perform boundary calculations based on the boundary spacing matrix to generate an initial classification boundary matrix; Perform time series difference calculations based on the environment coupling representation matrix to generate a drift mapping matrix.

7. The connector fretting wear fault classification method based on electrical signal fusion according to claim 6, characterized in that, The steps to generate the boundary spacing matrix are as follows: Extract the center vector pairs of each category from the category center matrix to generate a set of center vector pairs; The initial distance matrix is ​​generated by performing Euclidean distance calculation on the set using the center vector; Symmetry correction is performed based on the initial distance matrix to generate the boundary spacing matrix.

8. The connector fretting wear fault classification method based on electrical signal fusion according to claim 7, characterized in that, The steps to generate the initial distance matrix are as follows: Perform vector difference calculation on the set based on the central vector to generate a set of difference vectors; A sequence of squared distances is generated by performing a sum of squares calculation on a set of difference vectors. Perform square root calculation based on the squared distance sequence to generate the initial distance matrix.

9. The connector fretting wear fault classification method based on electrical signal fusion according to claim 1, characterized in that, The steps for outputting the fretting wear fault classification results are as follows: Perform matrix mapping calculations based on the drift mapping matrix and the initial classification boundary matrix to generate the boundary offset matrix; Norm normalization is performed using the boundary offset matrix to generate a normalized boundary matrix; Perform boundary consistency correction based on the normalized boundary matrix to generate a stable classification boundary matrix; Based on the stable classification boundary matrix, boundary discrimination calculations are performed on the new multi-electrical signal data to generate fretting wear fault classification results.

10. The connector fretting wear fault classification method based on electrical signal fusion according to claim 9, characterized in that, The steps to generate the boundary offset matrix are as follows: Extract the mapping coefficient matrix from the drift mapping matrix, and generate the mapping coefficient matrix; The initial offset matrix is ​​generated by performing matrix multiplication between the mapping coefficient matrix and the initial classification boundary matrix. Amplitude constraints are applied based on the initial offset matrix to generate the boundary offset matrix.

Images