Variable step size compression sampling method, device and equipment for gear box oil temperature signal

Abstract

The application belongs to the technical field of signal processing, and relates to a variable step length compression sampling method of a gearbox oil temperature signal, comprising the following steps: collecting a plurality of historical gearbox oil temperature signals, performing segmentation and adaptive clustering, and obtaining a plurality of signal classes similar in shape; constructing a training matrix of each signal class; extracting an approximate part and a detail part of all signal segments, and constructing a plurality of mixed dictionaries according to the approximate part of the same training matrix and the detail part of each training matrix; calculating the information entropy and the required sampling point number of each signal segment in the training matrix, constructing a measurement matrix with equal time sampling intervals according to the sampling point number, and obtaining a compression sampling model; collecting an online gearbox oil temperature signal and inputting the compression sampling model to obtain a plurality of compression signals with variable step lengths; and reconstructing the compression signals to obtain reconstructed signals. The method can be used for variable step length compression sampling of aperiodic variable condition gearbox oil temperature signals.

Description

Technical Field

[0001] This application relates to the field of signal processing technology, and in particular to a variable step size compression sampling method, apparatus and device for gearbox oil temperature signals. Background Technology

[0002] Gearboxes are widely used in the mechanical transmission systems of rotating machinery due to their compact structure and torque transmission capabilities. In some situations, gearboxes operate under harsh conditions and are subjected to alternating loads during variable operation. Under high-speed, heavy-load conditions, various faults can easily occur, leading to the failure of the entire transmission system. This can not only cause equipment downtime but also result in equipment and personnel accidents. Therefore, health monitoring and fault diagnosis of moving parts in equipment, such as gearboxes, are crucial.

[0003] Oil temperature can reflect the health status of a gearbox. When only the overall health status of the gearbox needs to be calculated, fault detection technology based on oil temperature signals is a reliable method. However, oil temperature signals are slowly varying signals with very high redundancy. Therefore, with the continuous development of productivity and the increasing number of mechanical equipment, the large volume of gearbox oil temperature monitoring signals will pose a significant challenge to the data processing, transmission, and storage capabilities of electronic equipment. Therefore, compressed sampling of the oil temperature signal is essential.

[0004] Currently, the main challenge in applying compressed sensing technology to gearbox oil temperature compression sampling is that most rotating machinery frequently changes its operating conditions, resulting in non-periodic variations in the oil temperature signal. Therefore, the sparse dictionary trained using historical oil temperature signals cannot adequately represent new oil temperature signals that differ significantly from historical signals. Furthermore, because the sparsity of each sample differs within the same transform domain, using a fixed number of sampling points will cause oversampling of signals with low sparsity and undersampling of signals with high sparsity. Summary of the Invention

[0005] Therefore, it is necessary to provide a variable step size compression sampling method, apparatus, and device for gearbox oil temperature signals to address the above-mentioned technical problems. This method can perform variable step size compression sampling on gearbox oil temperature signals under non-periodic variable operating conditions, with high accuracy and strong generalization.

[0006] The variable step size compression sampling method for gearbox oil temperature signals includes:

[0007] Collect multiple historical gearbox oil temperature signals;

[0008] Multiple historical gearbox oil temperature signals are segmented to obtain multiple signal segments; adaptive clustering is performed on multiple signal segments to obtain multiple groups of signal classes with similar shapes; and a training matrix is ​​constructed for each group of signal classes.

[0009] Extract the approximate and detailed parts of all signal segments in each training matrix, and construct multiple mixed dictionaries based on the approximate parts of the same training matrix and the detailed parts of each training matrix;

[0010] Calculate the information entropy and the number of sampling points required for each signal segment in the training matrix, construct a measurement matrix with equal time sampling intervals based on the number of sampling points, and obtain a compressed sampling model based on multiple measurement matrices and the hybrid dictionary;

[0011] The online gearbox oil temperature signal is collected and input into the compression sampling model to obtain multiple compression signals with varying step sizes; the compression signals are then reconstructed to obtain the reconstructed signal.

[0012] In one embodiment, adaptive clustering is performed on multiple signal segments to obtain multiple groups of signal classes with similar morphologies, including:

[0013] Calculate the Euclidean distance between each signal segment and the cluster center; calculate the correlation distance based on the signal segment, cluster center, average value of the signal segment, and average value of the cluster center; calculate the curve shape trend characteristics based on the sorting results of the distance between the signal segment, cluster center, and maximum value.

[0014] Calculate the dual-scale similarity metric matrix based on Euclidean distance, correlation distance, and curve shape trend characteristics;

[0015] Calculate the variable value coefficients, cardinality, and weighted average of cardinality for each signal segment, and calculate the membership matrix based on the dual-scale similarity metric matrix;

[0016] The objective function is calculated based on the membership matrix. The process stops when the objective function is less than a preset threshold or when the preset maximum number of iterations is reached, thus obtaining the signal class.

[0017] In one embodiment, the Euclidean distance is calculated based on each signal segment and cluster center; the correlation distance is calculated based on the signal segment, cluster center, average value of the signal segment, and average value of the cluster center; and the curve shape trend characteristics are calculated based on the sorting results of the signal segment, cluster center, and maximum distance, including:

[0018]

[0019]

[0020]

[0021] In the formula, For Euclidean distance, y i Represents the i-th cluster center, y j Represents the j-th signal segment. For the relevant distance, and y i and y j The average value, To represent the trend characteristics of the curve shape, sort(·) sorts the values ​​from largest to smallest and takes the first Z values.

[0022] Based on Euclidean distance, correlation distance, and curve shape trend characteristics, a dual-scale similarity metric matrix is ​​calculated, including:

[0023]

[0024] In the formula, d ij This is a two-scale similarity metric matrix;

[0025] Calculate the variable value coefficients, cardinality, and weighted average of cardinality for each signal segment, and calculate the membership matrix based on the dual-scale similarity metric matrix, including:

[0026] η(k)=η0exp(-τ / k)

[0027]

[0028]

[0029]

[0030]

[0031] In the formula, k is the current algorithm iteration number, η0 and τ are fixed value coefficients, η(k) is the first variable value coefficient, α is the second variable value coefficient, and N i Let be the cardinality of the i-th type of signal. U is a weighted average of the base values. ij This is the membership matrix;

[0032] Based on the membership matrix and signal segments, update the cluster centers, including:

[0033]

[0034] In the formula, y i As cluster center;

[0035] The objective function is calculated based on the membership matrix. The process stops when the objective function is less than a preset threshold or when the preset number of iterations is reached, resulting in signal classes, including:

[0036]

[0037] In the formula, J is the objective function.

[0038] In one embodiment, approximate and detailed portions of all signal segments in each training matrix are extracted, and multiple hybrid dictionaries are constructed based on the approximate portions of the same training matrix and the detailed portions of each training matrix, including:

[0039] Decompose each signal segment in all training matrices to obtain the corresponding multiple intrinsic mode functions;

[0040] Calculate the correlation coefficient between each signal segment in the training matrix and its corresponding intrinsic mode function, and obtain the category dictionary and shared dictionary of the training matrix;

[0041] Based on the category dictionary of the training matrix, update the corresponding residual matrix, and update the atoms of the category dictionary and the corresponding sparse matrix. Stop when the current iteration count reaches the preset second iteration count, and obtain the updated category dictionary and the current sparse matrix.

[0042] Based on the current sparse matrix, reconstruct the approximate and detailed parts of the training matrix, and merge the detailed parts of all training matrices into a merged matrix;

[0043] Based on the shared dictionary of the training matrix and the merged matrix, update the corresponding residual matrix, and update the atoms and corresponding sparse matrix of the shared dictionary. Stop when the current iteration number reaches the preset third iteration number to obtain the updated shared dictionary.

[0044] The updated category dictionary and the updated shared dictionary are merged into a mixed dictionary.

[0045] In one embodiment, based on the category dictionary of the training matrix, the corresponding residual matrix is ​​updated, and the atoms of the category dictionary and the corresponding sparse matrix are updated, including:

[0046] The process of learning the category dictionary of the training matrix is ​​transformed into an optimization function:

[0047]

[0048] In the formula, Y c For the training matrix, D c For a category dictionary, X c This is the corresponding sparse matrix;

[0049] Extract the non-zero positions from the corresponding sparse matrix in the residual matrix and update the corresponding residual matrix:

[0050]

[0051] In the formula, L is the category dictionary D c Number of atoms;

[0052] The corresponding residual matrix is ​​decomposed using singular value decomposition.

[0053]

[0054] In the formula, Let U be the residual matrix. c V is the left singular matrix of the residual matrix. c Let Δ be the right singular matrix of the residual matrix. c It is a diagonal matrix;

[0055] Update the atoms in the category dictionary and the corresponding sparse matrix based on the decomposition results:

[0056]

[0057]

[0058] In the formula, l represents the number of atomic columns and the number of rows in the sparse matrix.

[0059] In one embodiment, based on the current sparse matrix, the approximate and detailed portions of the training matrix are reconstructed, and the detailed portions of all training matrices are merged into a merged matrix, including:

[0060] Based on the current sparse matrix, reconstruct an approximate portion of the training matrix:

[0061]

[0062] Details of reconstructing the training matrix:

[0063]

[0064] Merge all the details of the training matrices:

[0065] R = [R1, R2, ..., R] C ]

[0066] In the formula, R is the merging matrix.

[0067] In one embodiment, the information entropy of each signal segment in the training matrix is ​​calculated:

[0068]

[0069] In the formula, Q represents the signal segment y. i The number of possible values ​​for ρ q Represents y i The probability of taking the value q;

[0070] Calculate the number of sampling points required for each signal segment in the training matrix:

[0071]

[0072] In the formula, m is the number of sampling points required.

[0073] In one embodiment, constructing a measurement matrix with equal-time sampling intervals based on the number of sampling points includes:

[0074]

[0075] In the formula, λ m The sampling weight for each data point;

[0076] A compressed sampling model is obtained based on multiple measurement matrices and the hybrid dictionary, including:

[0077]

[0078] In the formula, This is the compression signal of the s-th segment in the oil temperature signal. This is the sparse matrix corresponding to the mixed dictionary.

[0079] A variable-step compression sampling device for gearbox oil temperature signals includes:

[0080] The acquisition module is used to acquire multiple historical gearbox oil temperature signals;

[0081] The segmentation module is used to segment multiple historical gearbox oil temperature signals to obtain multiple signal segments; adaptive clustering is performed on multiple signal segments to obtain multiple groups of signal classes with similar shapes; and a training matrix is ​​constructed for each group of signal classes.

[0082] The extraction module is used to extract the approximate and detailed parts of all signal segments in each training matrix, and construct multiple mixed dictionaries based on the approximate parts of the same training matrix and the detailed parts of each training matrix.

[0083] The calculation module is used to calculate the information entropy of each signal segment in the training matrix and the number of sampling points required, construct a measurement matrix with equal time sampling intervals based on the number of sampling points, and obtain a compressed sampling model based on multiple measurement matrices and the hybrid dictionary.

[0084] The sampling module is used to acquire online gearbox oil temperature signals and input them into the compression sampling model to obtain multiple compression signals with varying step sizes; the compression signals are then reconstructed to obtain the reconstructed signals.

[0085] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program performing the following steps:

[0086] Collect multiple historical gearbox oil temperature signals;

[0087] Multiple historical gearbox oil temperature signals are segmented to obtain multiple signal segments; adaptive clustering is performed on multiple signal segments to obtain multiple groups of signal classes with similar shapes; and a training matrix is ​​constructed for each group of signal classes.

[0088] Extract the approximate and detailed parts of all signal segments in each training matrix, and construct multiple mixed dictionaries based on the approximate parts of the same training matrix and the detailed parts of each training matrix;

[0089] Calculate the information entropy and the number of sampling points required for each signal segment in the training matrix, construct a measurement matrix with equal time sampling intervals based on the number of sampling points, and obtain a compressed sampling model based on multiple measurement matrices and the hybrid dictionary;

[0090] The online gearbox oil temperature signal is collected and input into the compression sampling model to obtain multiple compression signals with varying step sizes; the compression signals are then reconstructed to obtain the reconstructed signal.

[0091] The aforementioned variable step-size compression sampling method, apparatus, and equipment for gearbox oil temperature signals propose a hybrid dictionary that extracts the commonalities and differences of different signals, demonstrating strong signal representation and generalization capabilities. It adaptively adjusts the number of sampling points and the sampling step size based on the information entropy of each signal segment, achieving a balance between minimizing the number of sampling points and ensuring signal reconstruction quality. By acquiring a small number of signal points using compressed sensing, compressed sampling and accurate reconstruction of the signal can be achieved, making the process simple and fast. Attached Figure Description

[0092] Figure 1 This is a flowchart illustrating a variable step size compression sampling method for gearbox oil temperature signals in one embodiment.

[0093] Figure 2 This is a graph of the original temperature signal collected in this invention;

[0094] Figure 3 The results of segmenting and adaptive clustering of the collected raw temperature signals are the signal classes with similar shapes. (a) is the first signal class, (b) is the second signal class, (c) is the third signal class, (d) is the fourth signal class, and (e) is the fifth signal class.

[0095] Figure 4 To use the EMD method to decompose signal segments of different shapes to obtain corresponding approximate and detailed parts;

[0096] Figure 5 This is a schematic diagram of a hybrid dictionary learning method based on intrinsic mode basis functions;

[0097] Figure 6The images show the reconstruction results of the training signal after sparse decomposition using a hybrid dictionary. (a) shows a comparison between the original and reconstructed signals; (b) shows the reconstruction error.

[0098] Figure 7 The images show the reconstruction results of the test signal after sparse decomposition using a hybrid dictionary. (a) shows a comparison between the original and reconstructed signals; (b) shows the reconstruction error.

[0099] Figure 8 Here are the histograms of the relationships between the original signals: (a) is the histogram of the relationship between the information entropy of the original signals; (b) is the histogram of the relationship between the number of sampling points of the compressed samples of the original signals.

[0100] Figure 9 This is a variable-step-size compressed sampling diagram of the original signal;

[0101] Figure 10 The images show the reconstructed results of the compressed training and testing signals. (a) is a comparison of the original and reconstructed signals; (b) is a reconstruction error diagram.

[0102] Figure 11 This is a structural block diagram of a variable step-size compression sampling device for gearbox oil temperature signals in one embodiment;

[0103] Figure 12 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0104] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application.

[0105] It should be noted that all directional indicators (such as up, down, left, right, front, back, etc.) in the embodiments of this application are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicator will also change accordingly.

[0106] Furthermore, the use of terms such as "first" and "second" in this application is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of those features. In the description of this application, "multiple sets" means at least two sets, such as two sets, three sets, etc., unless otherwise explicitly specified.

[0107] In this application, unless otherwise expressly specified and limited, the terms "connection," "fixed," etc., should be interpreted broadly. For example, "fixed" can mean a fixed connection, a detachable connection, or an integral part; it can mean a mechanical connection, an electrical connection, a physical connection, or a wireless communication connection; it can mean a direct connection or an indirect connection through an intermediate medium; it can mean the internal communication of two elements or the interaction between two elements, unless otherwise expressly limited. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0108] Furthermore, the technical solutions of the various embodiments of this application can be combined with each other, but only if they are based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by this application.

[0109] This application provides a variable step size compression sampling method for gearbox oil temperature signals, such as... Figure 1 As shown, in one embodiment, it includes:

[0110] Step 102: Collect multiple historical gearbox oil temperature signals.

[0111] In this step, the data acquisition method is existing technology, and the acquired historical gearbox oil temperature signals are specifically as follows: Figure 2 As shown.

[0112] Step 104: Segment the multiple historical gearbox oil temperature signals to obtain multiple signal segments; perform adaptive clustering on the multiple signal segments to obtain multiple groups of signal classes with similar shapes; construct the training matrix for each group of signal classes.

[0113] Specifically:

[0114] Calculate the Euclidean distance between each signal segment and its cluster center; calculate the correlation distance based on the signal segment, cluster center, average value of the signal segment, and average value of the cluster center; calculate the curve shape trend characteristics based on the sorting results of the signal segment, cluster center, and maximum distance; calculate the dual-scale similarity metric matrix based on the Euclidean distance, correlation distance, and curve shape trend characteristics; calculate the variable value coefficient, cardinality, and weighted average of cardinality for each signal segment, and calculate the membership matrix based on the dual-scale similarity metric matrix; calculate the objective function based on the membership matrix, and stop when the objective function is less than a preset threshold or the preset maximum number of iterations is reached, thus obtaining the signal class.

[0115] More specifically:

[0116] The acquired signal is segmented into n signal segments y1, y2, ..., y n Each signal segment has a length of N;

[0117] For the n signal segments y1, y2, ..., y n Adaptive clustering is performed to obtain C groups of signal classes with similar morphologies. Each signal class is then used to construct a separate training matrix Yc = [y c1 ,y c2 ,…,y cni ], where c = 1, ..., C,

[0118] The specific adaptive clustering process can employ a competitive clustering algorithm with dual-scale similarity measurement, and the results are as follows: Figure 3 As shown.

[0119] Set the maximum number of iterations K, the stopping threshold ε, and initialize the current iteration number k = 1;

[0120] Calculate the Euclidean distance between each signal segment and the cluster center:

[0121]

[0122] In the formula, For Euclidean distance, y i Represents the i-th cluster center, where i is the index of the current cluster center, i = 1, 2, ..., C, y j Represents the j-th signal segment, where j is the number of the current signal segment, j = 1, 2, ..., S, t represents the position of the sampling point in the signal segment, and N represents the number of sampling points contained in each signal segment;

[0123] Calculate the correlation distance based on the signal segment, cluster center, average value of the signal segment, and average value of the cluster center:

[0124]

[0125] In the formula, For the relevant distance, and y i and y j The average value;

[0126] Based on the sorting results of signal segments, cluster centers, and maximum distances, the trend characteristics of the curve shape are calculated:

[0127]

[0128] In the formula, The approximate curve shape is represented by sort(·), which sorts the values ​​from largest to smallest and takes the top Z values, which is the sorted result based on the distance of the maximum value, Z = 0.06N;

[0129] Based on Euclidean distance, correlation distance, and curve shape trend characteristics, a dual-scale similarity metric matrix is ​​calculated, including:

[0130]

[0131] In the formula, d ij This is a two-scale similarity metric matrix;

[0132] Calculate the variable value coefficients, cardinality, and weighted average of cardinality for each signal segment, and calculate the membership matrix based on the dual-scale similarity metric matrix, including:

[0133] η(k)=η0exp(-τ / k)

[0134]

[0135]

[0136]

[0137]

[0138] In the formula, k is the current algorithm iteration number, η0 and τ are fixed value coefficients, η(k) is the first variable value coefficient, α is the second variable value coefficient, C is the number of cluster centers, S is the number of signal segments, and N i Let be the cardinality of the i-th type of signal. U is a weighted average of the base values. ij This is the membership matrix;

[0139] Update the cluster centers based on the membership matrix and signal segments:

[0140]

[0141] In the formula, y i As cluster center;

[0142] The objective function is calculated based on the membership matrix. The clustering process stops when the objective function is less than a preset threshold or when the preset first iteration number is reached. At this point, the clustering accuracy meets the requirements. Each signal segment is assigned a signal class based on the maximum value of each column element in the membership matrix, including:

[0143]

[0144] In the formula, J is the objective function.

[0145] Determine whether the objective function J is less than the threshold ε:

[0146] If J>ε, let k=k+1, continue iterating, calculate the variable value coefficient, cardinality, and weighted average of cardinality for each signal segment, calculate the membership matrix, update the cluster centers, calculate the objective function and re-evaluate, until J<ε or k=K;

[0147] If J < ε or k = K, then stop iterating and exit the iteration.

[0148] Step 106: Extract the approximate and detailed parts of all signal segments in each training matrix, and construct multiple mixed dictionaries based on the approximate parts of the same training matrix and the detailed parts of each training matrix.

[0149] Specifically:

[0150] Decompose each signal segment in all training matrices to obtain the corresponding multiple intrinsic mode functions (IMFs); calculate the correlation coefficient between each signal segment in the training matrix and the corresponding IMF, and obtain the category dictionary and shared dictionary of the training matrix; update the corresponding residual matrix based on the category dictionary of the training matrix, and update the atoms and corresponding sparse matrix of the category dictionary, stopping when the current iteration number reaches the preset second iteration number, obtaining the updated category dictionary and the current sparse matrix; reconstruct the approximate and detailed parts of the training matrix based on the current sparse matrix, and merge the detailed parts of all training matrices into a merged matrix; update the corresponding residual matrix based on the shared dictionary and merged matrix of the training matrix, and update the atoms and corresponding sparse matrix of the shared dictionary, stopping when the current iteration number reaches the preset third iteration number, obtaining the updated shared dictionary; merge the updated category dictionary and the updated shared dictionary into a mixed dictionary.

[0151] More specifically:

[0152] The EMD method was used to decompose all training matrices Y1,…,Y1 respectively. c ,…,Y C Each sample y in c1 ,y c2 ,…,y cni Several intrinsic mode functions (IMFs) are obtained.

[0153] like Figure 4 As shown, two signal segments within the same training matrix have similar curve shapes, and each signal segment can be considered as consisting of an approximate part and details. After decomposition using the EMD method, the approximate parts of the two signals are very similar, with the difference lying in the details. Therefore, it can be assumed that by adding the approximate part of signal segment 1 to the details of signal segment 2 and adjusting the combined weights of the approximate and details, signal segment 2 can be reconstructed.

[0154] The correlation coefficient of each sample's IMF is calculated using the Pearson correlation coefficient (PCC). IMFs with a correlation coefficient greater than 0.8 are selected from the training matrix Yc to construct a category dictionary D. c Select all training matrices Y1,…,Y c ,…,Y C The IMFs with correlation coefficients of 0.2 to 0.8 are used to construct a shared dictionary D. s Finally, the category dictionary D... c The shared dictionary Ds forms a mixed dictionary, with the corresponding dictionary Dc mixed for each training matrix, where Dc mixed = [D c D s ].

[0155] Specifically, such as Figure 5 As shown.

[0156] Set the maximum number of iterations K1, the maximum sparsity T1, the sparse matrix element threshold μ1, and initialize the current iteration number k1 = 1;

[0157] Each training matrix Y c Category dictionary D c The learning process is transformed into an optimization function:

[0158]

[0159] In the formula, Y c For the training matrix, D c For a category dictionary, X c This is the corresponding sparse matrix;

[0160] When it is necessary to update the atom in the l-th column, the l-th column of the dictionary is excluded from the signal construction, and the corresponding non-zero positions in the sparse matrix of the residual matrix are extracted and the corresponding residual matrix is ​​updated:

[0161]

[0162] In the formula, L is the category dictionary D c Number of atoms;

[0163] The corresponding residual matrix is ​​decomposed using singular value decomposition.

[0164]

[0165] In the formula, Let U be the residual matrix. c V is the left singular matrix of the residual matrix. c Let Δ be the right singular matrix of the residual matrix. c It is a diagonal matrix;

[0166] Based on the decomposition results, update the l-th column atoms of the category dictionary and the corresponding l-th row sparse matrix using an alternating iterative method:

[0167]

[0168]

[0169] In the formula, l represents the number of atomic columns and the number of rows in the sparse matrix.

[0170] Determine if the current iteration count k1 exceeds the maximum iteration count K1. If so, exit the iteration and obtain the updated category dictionary; then convert the sparse matrix X... c Elements smaller than the threshold μ1 are set to 0 to obtain the current sparse matrix. If not, the learning process of the category dictionary of the current sparse matrix is ​​transformed into an optimization function, and the iteration is restarted until the current iteration number exceeds the maximum iteration number K1.

[0171] Based on the current sparse matrix, reconstruct an approximate portion of the training matrix:

[0172]

[0173] Details of reconstructing the training matrix:

[0174]

[0175] Merge all the details of the training matrices:

[0176] R = [R1, R2, ..., R] C ]

[0177] In the formula, R is the merging matrix.

[0178] Set the maximum number of iterations K2, the maximum sparsity T2, the sparse matrix element threshold μ2, and initialize the current iteration number k2 = 1;

[0179] Based on the merge matrix, each training matrix Y is... c Shared dictionary D s The learning process is transformed into an optimization function:

[0180]

[0181] In the formula, Y c For the training matrix, D s For shared dictionaries, X s This is the corresponding sparse matrix;

[0182] When it is necessary to update the atom in the l-th column, the l-th column of the dictionary is excluded from the signal composition, and the non-zero positions in the corresponding sparse matrix of the residual matrix are extracted to obtain the newly updated residual matrix:

[0183]

[0184] In the formula, L is the shared dictionary D s Number of atoms;

[0185] Singular value decomposition is used to analyze the corresponding residual matrix. Decompose:

[0186]

[0187] In the formula, Let U be the residual matrix. s V is the left singular matrix of the residual matrix. s Let Δ be the right singular matrix of the residual matrix. s It is a diagonal matrix;

[0188] Based on the decomposition results, update the l-th column atoms and the corresponding l-th row sparse matrix of the shared dictionary using an alternating iterative method:

[0189]

[0190]

[0191] In the formula, l represents the number of atomic columns and the number of rows in the sparse matrix.

[0192] Determine if the current iteration count k2 exceeds the maximum iteration count K2. If so, exit the iteration and obtain the updated shared dictionary; then convert the sparse matrix X... s Elements smaller than the threshold μ2 are set to 0 to obtain the current sparse matrix. If not, the learning process of the shared dictionary of the current sparse matrix is ​​transformed into an optimization function, and the iteration is restarted until the current iteration number exceeds the maximum iteration number K2.

[0193] The updated category dictionary and the updated shared dictionary are merged, that is, the C category dictionaries corresponding to each training matrix and a shared dictionary shared by all training matrices are merged into a mixed dictionary Dc mixed for each training matrix.

[0194] like Figure 6 and Figure 7 The figures shown are the reconstruction results of the training and test signals after sparse decomposition using the hybrid dictionary. (a) is a comparison of the original and reconstructed signals; (b) is a reconstruction error diagram. Figure 6 The reconstruction error graph shows that the reconstruction error increases in signal segments with complex curve shapes, but the maximum reconstruction error does not exceed 0.4°C. Figure 7The reconstruction error graph shows that the hybrid dictionary has a strong expressive power and generalization ability for the test signal, and the maximum reconstruction error does not exceed 0.2℃.

[0195] Step 108: Calculate the information entropy of each signal segment in the training matrix and the number of sampling points required. Construct a measurement matrix with equal time sampling intervals based on the number of sampling points, and obtain a compressed sampling model based on multiple measurement matrices and a mixed dictionary.

[0196] Specifically:

[0197] Calculate the training matrix Y c The information entropy of each signal segment is calculated, and the maximum value is taken:

[0198]

[0199] In the formula, Q represents the signal segment y. i The number of possible values ​​for ρ q Represents y i The probability of taking the value q;

[0200] Calculate the number of sampling points required for each signal segment in the training matrix:

[0201]

[0202] In the formula, m is the number of sampling points required.

[0203] Construct a measurement matrix with equal time sampling intervals based on the number of sampling points, including:

[0204]

[0205] In the formula, λ m The sampling weight for each data point can be determined using RIP constraints and the singular value method.

[0206] A compressed sampling model is obtained based on multiple measurement matrices and the hybrid dictionary, including:

[0207]

[0208] In the formula, This is the compression signal of the s-th segment in the oil temperature signal. This is the sparse matrix corresponding to the mixed dictionary.

[0209] like Figure 8 This is a histogram showing the relationship between the information entropy of the original signal and the number of compressed samples, i.e., the value of m.

[0210] like Figure 9As shown, each signal segment undergoes adaptive sampling with equal time intervals, resulting in a variable step-size compression sampling diagram of the original signal. In signal segments with low information entropy, reducing the number of sampling points and increasing the sampling step size reduces computer memory usage. Conversely, in signal segments with high information entropy, increasing the number of sampling points and decreasing the sampling step size ensures signal reconstruction accuracy.

[0211] Step 110: Input the online gearbox oil temperature signal into the compression sampling model to obtain multiple compression signals with varying step sizes; reconstruct the compression signals to obtain the reconstructed signals.

[0212] In this step, the Orthogonal Matching Pursuit (OMP) method is used to reconstruct the compressed signal.

[0213] Figure 10 The image shows the reconstruction effect of the compressed training and testing signals. As can be seen from the image, the maximum reconstruction error does not exceed 0.4℃, which verifies the effectiveness of the technical solution in the invention.

[0214] In this embodiment, the method is applicable to various aperiodic signals. The compressed sensing method is efficient, simple and universal, and the decompression method avoids signal sparsity estimation, thus having practical value.

[0215] The variable step-size compression sampling method for gearbox oil temperature signals described above proposes a hybrid dictionary that extracts the commonalities and differences of different signals, demonstrating strong signal representation and generalization capabilities. It adaptively adjusts the number of sampling points and the sampling step size based on the information entropy of each signal segment, achieving a balance between minimizing the number of sampling points and ensuring signal reconstruction quality. By acquiring a small number of signal points using compressed sensing, compressed sampling and accurate reconstruction of the signal can be achieved, making the process simple and fast.

[0216] It should be understood that, although Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but may be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but may be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0217] This application also provides a variable step compression sampling device for gearbox oil temperature signals, such as... Figure 11As shown, in one embodiment, it includes: a data acquisition module 1102, a segmentation module 1104, an extraction module 1106, a calculation module 1108, and a sampling module 1110, wherein:

[0218] Acquisition module 1102 is used to acquire multiple historical gearbox oil temperature signals;

[0219] The segmentation module 1104 is used to segment multiple historical gearbox oil temperature signals to obtain multiple signal segments; perform adaptive clustering on multiple signal segments to obtain multiple groups of signal classes with similar shapes; and construct a training matrix for each group of signal classes.

[0220] Extraction module 1106 is used to extract the approximate and detailed parts of all signal segments in each training matrix, and construct multiple mixed dictionaries based on the approximate parts of the same training matrix and the detailed parts of each training matrix.

[0221] The calculation module 1108 is used to calculate the information entropy of each signal segment in the training matrix and the number of sampling points required, construct a measurement matrix with equal time sampling intervals based on the number of sampling points, and obtain a compressed sampling model based on multiple measurement matrices and a mixed dictionary.

[0222] The sampling module 1110 is used to acquire online gearbox oil temperature signals into the compression sampling model to obtain multiple compression signals with varying step sizes; the compression signals are then reconstructed to obtain reconstructed signals.

[0223] Specific limitations regarding the variable-step compression sampling device for gearbox oil temperature signals can be found in the above description of the variable-step compression sampling method for gearbox oil temperature signals, and will not be repeated here. Each module in the aforementioned device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in hardware or independently of the processor in a computer device, or stored in software in the memory of a computer device, so that the processor can call and execute the corresponding operations of each module.

[0224] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 12As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements a variable-step compression sampling method for gearbox oil temperature signals. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0225] Those skilled in the art will understand that Figure 12 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0226] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the method described above.

[0227] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0228] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A variable step size compressed sampling method for gear case oil temperature signals, characterized by, The method comprises the following steps: collecting a plurality of historical gearbox oil temperature signals; segmenting the plurality of historical gearbox oil temperature signals to obtain a plurality of signal segments; performing adaptive clustering on the plurality of signal segments to obtain a plurality of groups of signal classes similar in shape; constructing a training matrix for each group of signal classes; extracting the approximate part and the detail part of all signal segments in each training matrix, and constructing a plurality of mixed dictionaries according to the approximate part of the same training matrix and the detail part of each training matrix; calculating the information entropy and the required sampling points of each signal segment in the training matrix, constructing a measurement matrix with equal time sampling intervals according to the sampling points, and obtaining a compressed sampling model according to the plurality of measurement matrices and the mixed dictionaries; inputting the online gearbox oil temperature signal into the compressed sampling model to obtain a plurality of compressed signals with variable step lengths; reconstructing the compressed signals to obtain reconstructed signals; extracting the approximate part and the detail part of all signal segments in each training matrix, and constructing a plurality of mixed dictionaries according to the approximate part of the same training matrix and the detail part of each training matrix, comprising: decomposing each signal segment in all training matrices to obtain a plurality of corresponding intrinsic modal functions; calculating the correlation coefficient of each signal segment in the training matrix and the corresponding intrinsic modal function, and obtaining a category dictionary and a shared dictionary of the training matrix; updating the corresponding residual matrix according to the category dictionary of the training matrix, and updating the atoms of the category dictionary and the corresponding sparse matrix, stopping when the current iteration number reaches a preset second iteration number, and obtaining the updated category dictionary and the current sparse matrix; reconstructing the approximate part and the detail part of the training matrix according to the current sparse matrix, and merging the detail parts of all training matrices into a merged matrix; updating the corresponding residual matrix according to the shared dictionary of the training matrix and the merged matrix, and updating the atoms of the shared dictionary and the corresponding sparse matrix, stopping when the current iteration number reaches a preset third iteration number, and obtaining the updated shared dictionary; merging the updated category dictionary and the updated shared dictionary into a mixed dictionary.

2. The variable step size compression sampling method of a gear case oil temperature signal according to claim 1, wherein, performing adaptive clustering on the plurality of signal segments to obtain a plurality of groups of signal classes similar in shape, comprising: calculating the Euclidean distance between each signal segment and the clustering center; calculating the correlation distance between the signal segment, the clustering center, the average value of the signal segment, and the average value of the clustering center; and calculating the curve shape trend feature according to the signal segment, the clustering center, and the maximum distance sorting result; calculating the double-scale similarity measurement matrix according to the Euclidean distance, the correlation distance, and the curve shape trend feature; calculating the variable value coefficient, the base, and the weighted average of the base of each signal segment, and calculating the membership matrix according to the double-scale similarity measurement matrix; calculating the objective function according to the membership matrix, stopping when the objective function is less than a preset threshold or reaches a preset maximum iteration number, and obtaining the signal class.

3. The variable step size compression sampling method of a gear case oil temperature signal according to claim 2, wherein, calculating the Euclidean distance between each signal segment and the clustering center comprises: wherein is the Euclidean distance, y i represent the first i cluster center, y j represent the first j signal segment; calculating the correlation distance between the signal segment, the clustering center, the average value of the signal segment, and the average value of the clustering center comprises: wherein is the correlation distance, and are the average values of y i and y j respectively. calculating the curve shape trend feature according to the signal segment, the clustering center, and the maximum distance sorting result comprises: In the formula, is a curve shape trend feature, sort(·) is sorted from large to small according to the numerical value, and the first Z values are taken. According to the Euclidean distance, the correlation distance and the curve shape trend characteristics, a two-scale similarity measurement matrix is calculated, including: In the formula, is a two-scale similarity measure matrix; The variable value coefficient, the base and the weighted average of the base of each signal segment are calculated, and a membership matrix is calculated according to the two-scale similarity measurement matrix, including: In the formula, k This represents the current algorithm iteration number. η 0 and τ For fixed value coefficients, η ( k ) represents the first variable coefficient. α The second variable value coefficient, For the first i The cardinality of the signal type The weighted average of the base values. U ij This is the membership matrix; The cluster center is updated according to the membership matrix and the signal segment, including: In the formula, y i is the cluster center; The objective function is calculated according to the membership matrix, and the signal class is obtained when the objective function is less than a preset threshold or reaches a preset first iteration number, including: In the formula, J is the objective function.

4. The variable step size compressive sampling of a gearbox oil temperature signal method of claim 1, wherein, The corresponding residual matrix is updated according to the category dictionary of the training matrix, and the atom of the category dictionary and the corresponding sparse matrix are updated, including: The learning process of the category dictionary of the training matrix is converted into an optimization function: wherein Y c is a training matrix, D c is a class dictionary, X c is a corresponding sparse matrix; The corresponding non-zero position in the corresponding sparse matrix in the residual matrix is extracted, and the corresponding residual matrix is updated: In the formula, L is a category dictionary D c atomic number; The corresponding residual matrix is decomposed using the singular value decomposition method: wherein is a residual matrix, U c is a left singular matrix of the residual matrix, V c is a right singular matrix of the residual matrix, Δ c is a diagonal matrix; The atom of the category dictionary and the corresponding sparse matrix are updated according to the decomposition result: In the formula, l is the number of atomic columns and the number of sparse matrix rows.

5. A variable step size compressive sampling method of a gearbox oil temperature signal according to any one of claims 1 to 3, characterized in that, According to the current sparse matrix, the approximate part and the detail part of the training matrix are reconstructed, and the detail parts of all training matrices are combined into a combined matrix, including: According to the current sparse matrix, the approximate part of the training matrix is reconstructed: The detail part of the training matrix is reconstructed: The detail parts of all training matrices are combined: R =[ R 1, R 2,…, R C ] In the formula, R is a merge matrix.

6. A variable step size compressive sampling method of a gearbox oil temperature signal according to any one of claims 1 to 3, characterized in that, The information entropy of each signal segment in the training matrix is calculated: In the formula, Q is a signal segment y i The number of values, ρ q represents y i The probability that the value is q ; The number of sampling points required by each signal segment in the training matrix is calculated: In the formula, m is the number of sampling points desired.

7. A variable step size compressive sampling method of a gearbox oil temperature signal according to any one of claims 1 to 3, characterized in that, A measurement matrix with equal time sampling intervals is constructed according to the number of sampling points, including: In the formula, λ m is the sampling weight for each data point; A compressed sampling model is obtained according to the multiple measurement matrices and the mixed dictionary, including: In the formula, is the compressed signal of the s-th signal segment in the oil temperature signal, is the sparse matrix corresponding to the mixed dictionary.

8. A variable step size compressive sampling apparatus for gear case oil temperature signals, characterized by, The variable step length compression sampling method of the gearbox oil temperature signal of any one of claims 1-7 is adopted, including: The acquisition module is used to acquire a plurality of historical gearbox oil temperature signals; The segmentation module is used to segment the plurality of historical gearbox oil temperature signals to obtain a plurality of signal segments; to perform adaptive clustering on the plurality of signal segments to obtain a plurality of groups of signal classes; and to construct a training matrix for each group of signal classes; The extraction module is used to extract the approximate part and the detail part of all signal segments in each training matrix, and to construct a plurality of mixed dictionaries according to the approximate part of the same training matrix and the detail part of each training matrix; The calculation module is used to calculate the information entropy and the number of sampling points required by each signal segment in the training matrix, to construct a measurement matrix with equal time sampling intervals according to the number of sampling points, and to obtain a compressed sampling model according to the multiple measurement matrices and the mixed dictionary; The sampling module is used to input the online gearbox oil temperature signal into the compressed sampling model to obtain a plurality of compressed signals with variable step lengths; and to reconstruct the compressed signals to obtain reconstructed signals. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. The processor executes the computer program to implement the steps of the method of any one of claims 1-7. The processor executes the computer program to implement the steps of the method of any one of claims 1-7.

Images