A method and device for estimating time-varying wear of a plunger pump slipper pair

Abstract

The present application provides a method and device for estimating the time-varying wear of a plunger pump slipper pair, which belongs to the field of estimating the time-varying wear of mechanical equipment. The method provided in the present application includes: forming a wear table based on the wear of the plunger pump slipper pair at different times, and forming a wear sample based on the wear table and an operating parameter signal; calculating the wear characterization features and fusion weights of the operating parameter signal, and generating fusion data based on the fusion weights and the time-frequency graph; analyzing the correlation of the fusion data based on an unsupervised network, and constructing an adjacency matrix based on the correlation; forming an alternating integrated network with the adjacency matrix, and training the alternating integrated network with the wear sample, wherein the trained alternating integrated network is used to estimate the wear of the plunger pump slipper pair in real time. The method and device for estimating the time-varying wear of the plunger pump slipper pair provided in the present application can realize accurate estimation of the time-varying wear of the plunger pump slipper pair.

Description

Technical Field

[0001] The present application relates to the technical field of estimating the time-varying wear of mechanical equipment, and in particular to a method and device for estimating the time-varying wear of a plunger pump slipper pair. Background Art

[0002] The plunger hydraulic pump (plunger pump for short) is a key component of the aircraft hydraulic system. Its reliability and life directly affect the safety and stability of the aircraft hydraulic system. Wear is a typical progressive failure of aircraft plunger pumps. According to statistics, more than 80% of aircraft plunger pump failures are caused by wear. Accurately identifying the wear status of aircraft plunger pumps and doing maintenance and replacement work in advance can ensure the safe flight of the aircraft and avoid catastrophic accidents.

[0003] The friction pair of the slipper and the swash plate is prone to wear in the plunger pump. The slipper is tightly attached to the swash plate and performs both rotation and revolution. The high-pressure oil discharged from the plunger cavity creates an oil film with a thickness of 10 to 30 microns between the slipper and the swash plate, making the two in a floating non-contact static pressure unloading form. This oil film uses the oil pressure to bear the load force applied by the plunger to prevent the slipper from directly contacting the swash plate. There is an active gap between the plunger ball head and the slipper ball socket. When the gap exceeds the limit value, the high-pressure oil in the plunger cavity flows out through the gap, and the oil film between the slipper and the swash plate becomes thinner. In severe cases, the static pressure support protection function fails, resulting in direct contact between the two structures and adhesive wear, and the slipper is burned and falls off.

[0004] The current method focuses on the identification of the wear of the plunger pump under rated conditions and single conditions, and the application scenarios are limited. When the hydraulic pump works under time-varying conditions, the monitoring signal will continue to change, and the wear of the plunger pump slipper pair cannot be accurately estimated. Summary of the invention

[0005] In view of this, the present application provides a method and device for estimating the time-varying wear amount of a plunger pump slipper pair, which can achieve accurate estimation of the time-varying wear amount of a plunger pump slipper pair.

[0006] Specifically, the present application is implemented through the following technical solutions:

[0007] The first aspect of the present application provides a method for estimating the time-varying wear amount of a plunger pump slipper pair, the method comprising:

[0008] A wear amount table is formed based on the wear amount of the plunger pump sliding shoe pair at different times, and a wear amount sample is formed based on the wear amount table and the operating parameter signal;

[0009] Calculating the wear amount characterization features and fusion weights of the operating parameter signal, and generating fusion data according to the fusion weights and a time-frequency diagram; wherein the time-frequency diagram is formed based on the wear amount samples;

[0010] Analyzing the correlation of the fused data based on an unsupervised network, and constructing an adjacency matrix based on the correlation;

[0011] An alternating integrated network is formed with the adjacency matrix, and the wear amount samples are used to train the alternating integrated network, wherein the trained alternating integrated network is used to estimate the wear amount of the plunger pump slipper pair in real time.

[0012] A second aspect of the present application provides a device for estimating the time-varying wear of a plunger pump slipper pair, the device comprising a construction module, a calculation module, an analysis module and a processing module; wherein:

[0013] The construction module is used to form a wear amount table based on the wear amount of the plunger pump sliding shoe pair at different times, and to form a wear amount sample based on the wear amount table and the operating parameter signal;

[0014] The calculation module is used to calculate the wear amount characterization features and fusion weights of the operating parameter signal, and generate fusion data according to the fusion weights and the time-frequency diagram; wherein the time-frequency diagram is formed based on the wear amount samples;

[0015] The analysis module is used to analyze the correlation of the fused data based on an unsupervised network, and to construct an adjacency matrix based on the correlation;

[0016] The processing module is used to form an alternating integrated network with the adjacency matrix and train the alternating integrated network using the wear amount samples, wherein the trained alternating integrated network is used to estimate the wear amount of the plunger pump slipper pair in real time.

[0017] The present application provides a method and device for estimating the time-varying wear of a plunger pump slipper pair. By constructing a wear sample and comprehensively considering the wear of the plunger pump slipper pair reflected by multi-source operating parameter signals under variable working conditions, the wear of the slipper pair can be more accurately characterized. The wear characterization features and fusion weights calculated by the wear table and the operating parameter signal can dynamically reflect the wear of the slipper pair under different working conditions and time points, thereby improving the actual wear prediction capability. The use of time-frequency graphs and unsupervised networks for analysis can capture complex patterns and changes in the signal, improve the recognition of features, and do not rely on the annotation of large amounts of data, which helps to estimate the wear more accurately. The wear of the slipper pair can be accurately estimated in real time based on the alternating integrated network. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A flow chart of Embodiment 1 of the method for estimating the time-varying wear amount of the plunger pump slipper pair provided in the present application;

[0019] Figure 2 This is a schematic diagram of the structure of the plunger pump sliding shoe pair shown in this application;

[0020] Figure 3 The pressure pulsation signal diagram shown for this application;

[0021] Figure 4 A vibration signal diagram shown for this application;

[0022] Figure 5 The oil return signal diagram shown in this application;

[0023] Figure 6 A schematic diagram of the wear sensitivity of different statistical characteristics of the pressure pulsation signal shown in this application;

[0024] Figure 7 A schematic diagram of the wear sensitivity of different statistical characteristics of the vibration signal shown in this application;

[0025] Figure 8 This is a schematic diagram of the wear sensitivity of different statistical characteristics of the oil return signal shown in this application;

[0026] Fig. 9 This is a graph showing the characteristic variation range of the wear amount of the operating parameter signal shown in this application;

[0027] Fig.10 This is a confusion matrix diagram for estimating the wear amount of the plunger pump slipper pair shown in this application;

[0028] Fig.11 This is a visualization diagram of the wear characteristics of the plunger pump slipper pair shown in this application;

[0029] Fig.12 This is a schematic structural diagram of Embodiment 2 of the device for estimating the time-varying wear of the plunger pump slipper pair provided in the present application. DETAILED DESCRIPTION

[0030] Here, exemplary embodiments are described in detail, and examples thereof are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application.

[0031] The terms used in this application are only for the purpose of describing specific embodiments and are not intended to limit this application. The singular forms of "a", "said" and "the" used in this application are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used in this article refers to and includes any or all possible combinations of one or more associated listed items.

[0032] It should be understood that although the terms first, second, third, etc. may be used in the present application to describe various information, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0033] Specific embodiments are given below to introduce the technical solution of the present application in detail.

[0034] Figure 1 This is a flow chart of Embodiment 1 of the method for estimating the time-varying wear of the plunger pump slipper pair provided in this application. Figure 1 , the method provided in this embodiment may include:

[0035] S101, forming a wear amount table based on the wear amount of the plunger pump slipper pair at different times, and forming a wear amount sample based on the wear amount table and an operating parameter signal.

[0036] In order to better understand the plunger pump slipper pair, the plunger pump slipper pair is first explained below. Figure 2 This is a schematic diagram of the structure of the plunger pump slipper pair shown in this application, please refer to Figure 2 , Figure 2 51 is the plunger pump slipper pair, 10 is the housing assembly, 20 is the transmission assembly, 21 is the transmission shaft, 30 is the oil distribution plate, 40 is the plunger pump assembly, 41 is the plunger cylinder, 42 is the plunger rod, 43 is the oil return channel, 50 is the variable assembly, 52 is the variable head, and 53 is the adjustment component. The slipper pair in the plunger pump is a key component in the plunger pump hydraulic system. The slipper pair consists of a slipper and a ball head that cooperates with it. They work together to realize the reciprocating motion and adaptive adjustment of the plunger. Among them, the design and performance of the slipper pair directly affect the working efficiency, reliability and service life of the plunger pump.

[0037] It should be noted that a wear amount table is formed based on the wear amount of the plunger pump slipper pair at different times, and a wear amount sample is formed based on the wear amount table and the operating parameter signal, including:

[0038] (1) Measuring the wear amount of the plunger pump slipper pair at different times, forming a wear amount table based on the correspondence between the time information and the wear amount, and establishing a correlation between the wear amount table and the operating parameter signal.

[0039] It should be noted that the amount of wear refers to the degree of material loss caused by friction and other mechanical factors during the operation of the plunger pump slipper pair. The amount of wear can be determined by directly measuring the thickness change of the slipper pair or by calculating indirect parameters (such as pressure, vibration, etc.).

[0040] Specifically, during the operation of the plunger pump, the wear of the slipper pair is measured regularly or continuously. The specific measurement time point can be determined according to the operation cycle or maintenance plan of the equipment. For example, the wear measurement is performed every 100 hours of operation. When measuring the wear, the specific time point or operation time of each measurement is recorded. These moment information can be specific timestamps (such as date and time) or accumulated operating hours. Associate the wear of each measurement with the corresponding moment information to form a one-to-one correspondence. Create a wear table based on the measured wear and the corresponding moment information, for example, as shown in Table 1:

[0041] Table 1 Wear of the plunger pump slipper pair at different times

[0042] time Label Wear 1 S1 0.10mm 2 S2 0.15mm 3 S3 0.20mm 4 S4 0.25mm 5 S5 0.30mm 6 S6 0.35mm 7 S7 0.40mm 8 S8 0.45mm 9 S9 0.50mm

[0043] Please refer to Table 1. In addition to the corresponding relationship between the time information and the wear amount, Table 1 also stores the corresponding tag information (represented by the symbol S). The corresponding time information and wear amount information can be obtained through the tag information. Furthermore, the association relationship between the wear amount table and the operating parameter signal can be established by establishing an associated field or creating an intermediate table.

[0044] (2) Collecting the operating parameter signal, and determining the wear information corresponding to each type of operating parameter signal at different times according to the wear table.

[0045] It should be noted that the operating parameter signal may include a pressure pulsation signal, a vibration signal, an oil return signal, a temperature signal, a noise signal, a current signal, and a speed signal. Specifically, a suitable operating parameter signal may be selected for judgment according to actual needs. For example, in this embodiment, the operating parameter signals are taken as pressure pulsation signals, vibration signals, and oil return signals. Specifically, the collected pressure pulsation signals, vibration signals, and oil return signals are respectively as follows: Figure 3 , Figure 4 and Figure 5 shown.

[0046] Specifically, please refer to the wear amount table shown in Table 1 to match the wear amount at the corresponding moment to the pressure pulsation signal, vibration signal, and oil return signal at a certain moment. Indicates pressure pulsation signal, Indicates vibration signal, It indicates the oil return signal, and t indicates the working time of the plunger pump slipper pair, where: ; The pressure pulsation signal can be obtained at time t (using The wear amount of the piston pump slipper pair measured by Indicated by), vibration signal at time t (available The wear amount of the piston pump slipper pair and the return oil amount signal measured at time t (can be used The wear amount of the piston pump slipper pair measured by (indicates).

[0047] (3) Summarize all types of operating parameter signals and the wear information corresponding to each type of operating parameter signal at different times to form a wear sample.

[0048] Combined with the above examples, the wear volume sample can be obtained , where the input of the wear volume sample is the operating parameter signal, and the output is the label information corresponding to the operating parameter signal. Furthermore, the wear volume sample can be divided into a training set by data preprocessing and resampling. and test set For the description of preprocessing and resampling, please refer to the introduction of related technologies, which will not be repeated here.

[0049] S102, calculating the wear characterization features and fusion weights of the operating parameter signal, and generating fusion data according to the fusion weights and the time-frequency diagram; wherein the time-frequency diagram is formed based on the wear sample.

[0050] It should be noted that the wear volume characterization feature is selected from multiple statistical features of the operating parameter signal (such as pressure pulsation signal, vibration signal, oil return signal, etc.), and can most effectively reflect the wear volume change of the plunger pump slipper pair. The wear volume sensitivity of different statistical features of each operating parameter signal is calculated, and then the maximum value of these sensitivities is taken as the wear volume characterization feature of the signal. For example, for the pressure pulsation signal, it is found through calculation that the statistical feature of the energy index is most sensitive to the wear volume change, so the energy index is determined as the wear volume characterization feature of the pressure pulsation signal.

[0051] The fusion weight is used to determine the relative importance of different operating parameter signals in generating fusion data. The calculation method is to first calculate the wear characterization feature variation range of each operating parameter signal based on the wear table (such as the difference between the maximum and minimum values ​​of the pressure pulsation signal wear characterization feature at different times), and then divide the variation range of each signal's characterization feature by the sum of the variation ranges of all signal characterization features. The obtained ratio is the fusion weight of the signal. For example, if the variation range of the pressure pulsation signal's characterization feature accounts for a large proportion of all signals, then its weight in the fusion data will be higher, and its impact on the final fusion result will be greater. The determination of the fusion weight can enable different operating parameter signals to reasonably contribute their respective information during the fusion process, avoid excessive or insufficient influence of certain signals on the results, and thus more accurately reflect the wear of the slipper pair.

[0052] In addition, the time-frequency diagram is a graphical representation of the operating parameter signals (such as pressure pulsation signals, vibration signals, and oil return signals) in the wear sample converted through short-time Fourier transform (STFT). Short-time Fourier transform can simultaneously display the distribution information of the signal in time and frequency. It obtains the spectrum information corresponding to different moments by Fourier transforming the signal within a certain time window, so that the frequency components of the signal can be intuitively observed to change over time. For example, in the time-frequency diagram of the pressure pulsation signal, the frequency composition and energy distribution of the pressure pulsation signal at different moments can be seen, which helps to analyze the time-varying characteristics of the signal, dig out the wear-related information hidden in the signal, and provide a data basis for subsequent analysis based on the time-frequency domain.

[0053] By calculating the wear characterization features and fusion weights, the information about the wear of the slipper pair contained in different types of operating parameter signals (pressure pulsation, vibration, oil return, etc.) can be integrated to describe the wear state of the slipper pair in a comprehensive way, avoiding the limitations of single signal analysis and improving the accuracy of wear judgment. The generation of time-frequency diagrams can capture the time-varying characteristics of the operating parameter signals. Since the working conditions of the hydraulic pump are constantly changing, the signal characteristics also change accordingly. The time-frequency diagram can better reflect this dynamic change, thus providing a basis for accurately estimating the wear amount. In addition, determining the wear characterization features and fusion weights can screen out the most sensitive and representative features and signals to the wear change, remove redundant information, improve the effectiveness and recognition of data features, and more efficiently and accurately estimate the wear of the slipper pair at different times, realizing effective monitoring and prediction of the wear state.

[0054] Specifically, calculating the wear amount characterization feature and fusion weight of the operating parameter signal, and generating fusion data according to the fusion weight and the time-frequency diagram, including:

[0055] (1) calculating the wear amount characterization characteristics corresponding to each signal based on the statistical characteristics of the operating parameter signal;

[0056] Specifically, the wear amount characterization features corresponding to each signal are calculated based on the statistical features of the operating parameter signal, including:

[0057] (i) Obtaining the variation range of the statistical characteristics corresponding to each of the operating parameter signals.

[0058] It should be noted that the statistical characteristics of the operating parameter signal may include mean, absolute mean, maximum value, minimum value and variance, etc. Specifically, in one embodiment, the statistical characteristics of the operating parameter signal are shown in Table 2:

[0059] Table 2 Statistical characteristics of operating parameter signals

[0060] Serial number feature Serial number feature 1 Mean 11 Maximum 2 Absolute mean 12 Minimum 3 RMS 13 Peak Index 4 Square root amplitude 14 Margin index 5 Peak-to-Peak 15 Waveform indicators 6 variance 16 Pulse Indicator 7 Standard Deviation 17 Center of gravity frequency 8 Skewness 18 RMS frequency 9 Kurtosis 19 Average frequency 10 Energy indicators 20 Frequency variance

[0061] Referring to Table 2, it can be seen that in this embodiment, there are 20 statistical features. Combining the content of Table 2, for example, the number of statistical features of the pressure pulsation signal can be obtained as follows: , the number of statistical features of the vibration signal is , the number of statistical features of the oil return signal is , and then obtain the variation range of the statistical characteristics corresponding to each operating parameter signal. Specifically, for each operating parameter signal, the variation range of its statistical characteristics refers to the difference between the maximum and minimum values ​​of the corresponding statistical characteristics of the signal at different times.

[0062] (ii) calculating the wear sensitivity of each operating parameter signal based on the statistical characteristic variation range corresponding to each operating parameter signal and the wear variation range in the wear table.

[0063] Specifically, it can be as follows:

[0064] ;

[0065] in, and They are the wear sensitivity of pressure pulsation signal, vibration signal and oil return signal respectively; and are the pressure pulsation signals Statistical features , vibration signal Statistical features and oil return signal Statistical features The range of changes; The range of changes in label information.

[0066] (iii) performing a maximum calculation on the wear amount sensitivity of each statistical feature among the multiple statistical features of each operating parameter signal to obtain a wear amount characterization feature corresponding to each signal.

[0067] Specifically, as follows:

[0068] ;

[0069] in, Indicates the maximum value operation.

[0070] Figure 6 A schematic diagram of the wear sensitivity of different statistical characteristics of the pressure pulsation signal shown in this application; Figure 7 A schematic diagram of the wear sensitivity of different statistical characteristics of the vibration signal shown in this application; Figure 8This is a schematic diagram of the wear sensitivity of different statistical characteristics of the oil return signal shown in this application. Please also refer to Figure 6 , Figure 7 and Figure 8 Combining Table 2 with the above maximum calculation results, the wear characteristics of the pressure pulsation signal, vibration signal, and oil return signal are shown in Table 3:

[0071] Table 3 Wear characteristics

[0072] Serial number Multi-source heterogeneous signals Characterization characteristics 1 Pressure pulsation signal Energy indicators 2 Vibration signal RMS 3 Oil return signal variance

[0073] Please refer to Table 3, it can be seen that the wear characteristics of the pressure pulsation signal, vibration signal, and oil return signal are energy index, root mean square, and variance, respectively.

[0074] (2) Calculating a fusion weight of the operating parameter signal according to the wear amount table and the wear amount characterization feature.

[0075] Specifically, calculating the fusion weight of the operating parameter signal according to the wear amount table and the wear amount characterization feature includes:

[0076] (i) Calculating the characteristic variation range of the wear amount of each operating parameter signal based on the wear amount table.

[0077] Specifically, as follows:

[0078] ;

[0079] in, , and Characterize the wear characteristics of pressure pulsation signal, vibration signal and oil return signal , and Corresponds to the label information in the wear table (Table 1) The range of changes; Indicates that the calculation corresponds to the label information range of change.

[0080] Fig. 9 Please refer to the diagram showing the characteristic variation range of the operating parameter signal wear amount shown in this application. Fig. 9 , the wear characterization characteristic variation range of the pressure pulsation signal, vibration signal and oil return signal can be obtained.

[0081] (ii) The ratio of the variation range of the wear characterization feature of each operating parameter signal to the sum of the variation ranges of the wear characterization features of all operating parameter signals is used as the fusion weight of each operating parameter signal.

[0082] Specifically,

[0083] ;

[0084] in, , and are the fusion weights of pressure pulsation signal, vibration signal and oil return signal respectively; , and Characterize the wear characteristics of pressure pulsation signal, vibration signal and oil return signal , and Corresponds to label information in the wear table range of change.

[0085] (3) forming a time-frequency diagram based on the wear amount samples, and generating fusion data by combining the time-frequency diagram with the fusion weight.

[0086] It should be noted that forming a time-frequency graph based on the wear amount samples and generating fusion data by combining the time-frequency graph with the fusion weights includes:

[0087] (i) Obtaining the wear volume sample and converting the wear volume sample into a time-frequency diagram through short-time Fourier transform.

[0088] Specifically, it can be as follows:

[0089] ;

[0090] in, represents the short-time Fourier transform, , and They are the time-frequency diagrams corresponding to the pressure pulsation signal samples, vibration signal samples and oil return signal samples respectively.

[0091] (ii) performing a combination operation based on the time-frequency diagram and the fusion weight to generate fusion data.

[0092] For example, in one embodiment, the combined operation may be:

[0093] ;

[0094] in, Indicates a combined operation; , and They are the time-frequency diagrams corresponding to the pressure pulsation signal samples, vibration signal samples and oil return signal samples respectively; , and are the fusion weights of pressure pulsation signal, vibration signal and oil return signal respectively.

[0095] S103, analyzing the correlation of the fused data based on an unsupervised network, and constructing an adjacency matrix based on the correlation.

[0096] It should be noted that unsupervised networks are a type of neural network that learns without explicit labels or target outputs; unlike supervised learning, unsupervised learning does not rely on pre-labeled training data sets, but instead learns and analyzes by finding the intrinsic structure, patterns, or relationships of the data. In unsupervised network-based analysis, the network automatically discovers hidden patterns or structures in the data by analyzing the characteristics and distribution of input data (such as fused data), which can reflect the relationships and correlations between different signals.

[0097] The adjacency matrix is ​​used to describe the connection relationship between nodes in a graph. In a graph, nodes represent entities and edges represent the relationship or connection between nodes. The adjacency matrix is ​​a square matrix in which each row and column corresponds to a node in the graph. If the element in the i-th row and j-th column in the matrix is ​​1 (or non-zero value), it means that there is an edge between nodes i and j; if it is 0, it means that there is no direct connection between the nodes.

[0098] Specifically, analyzing the correlation of the fused data based on an unsupervised network and constructing an adjacency matrix based on the correlation includes:

[0099] (1) An unsupervised network is formed based on an encoder and a decoder; the encoder includes a preset number of convolutional layers; and the decoder includes a preset number of deconvolutional layers.

[0100] It should be noted that the number of convolutional layers of the encoder and the number of deconvolutional layers of the decoder can be set according to actual needs, and are not limited in this embodiment. For example, in one embodiment, the encoder includes 3 convolutional layers and the decoder includes 3 deconvolutional layers.

[0101] (2) Obtain the fused data and input it into the unsupervised network to obtain reconstructed fused data.

[0102] It should be noted that the input of the unsupervised network is the fused data, and the output is the reconstructed fused data. Specifically, the reconstruction operation in the unsupervised network refers to the network trying to regenerate data similar to the original input by learning the intrinsic structure and features of the input fused data. In this process, the unsupervised network uses the structure of the encoder and decoder to achieve this. Among them, the reconstruction operation enables the unsupervised network to learn the hidden patterns, structures, and feature representations in the fused data, which may not be obvious or difficult to directly discover in the original data. By mapping the fused data from the original space to a feature space (completed by the encoder) and then mapping it back to the original space (completed by the decoder), the network understands the intrinsic structure and feature relationship of the data in this process, so that more meaningful features can be extracted for subsequent analysis.

[0103] The reconstruction process achieves the effect of data compression and enhancement to a certain extent. By extracting key features and removing redundant information, the reconstructed data generated by the network may have a more compact representation while retaining the important information of the original data, which helps to reduce data storage space and computation and improve data processing efficiency. At the same time, the reconstruction process may also enhance the useful features in the data, making these features more prominent in subsequent analysis, and improving the accuracy and reliability of tasks such as wear estimation based on these features.

[0104] (3) Perform feature learning on the reconstructed fused data based on the pre-trained loss function, and use the learned feature representation to perform correlation calculation.

[0105] Specifically, the pre-trained loss function is as follows:

[0106] ;

[0107] in, represents the internal parameters of the unsupervised network, Represents the loss function The adjustment coefficient, is the reconstructed fusion data of the unsupervised network, To fuse data.

[0108] If the features learned from the reconstructed fusion data are expressed as , then the correlation is calculated as follows:

[0109] ;

[0110] in, Indicates and The reconstructed fusion data features and The relevance of and Respectively represent and The reconstructed fusion data features and The average value of represents the 2-norm.

[0111] (4) Constructing an adjacency matrix based on the correlation calculation results.

[0112] Combined with the above description, the adjacency matrix can be constructed as follows:

[0113] ;

[0114] In addition, the construction of the adjacency matrix based on the correlation also includes: obtaining the features of the reconstructed fused data according to the unsupervised network; calculating the feature mean of the reconstructed fused data; calculating the correlation between the two features based on the difference between the feature and the feature mean between the two features; constructing the adjacency matrix based on the correlation, the row number and column number of the adjacency matrix are used to determine the two features corresponding to the correlation, and the value of the adjacency matrix is ​​the correlation value. It should be noted that the unsupervised network obtains the reconstructed fused data by reconstructing the input fused data, and extracts its features from the reconstructed data. For each extracted feature, its mean is calculated. The calculation of the mean is to subsequently measure the degree of deviation of each feature from the average level, so as to judge the relative relationship between the features.

[0115] Furthermore, the difference between two features and the feature mean is used to calculate the correlation between them. The specific calculation method may involve some mathematical formulas or algorithms, such as measuring the similarity or correlation between two features by calculating the product of the difference between the feature value and the mean. This correlation calculation method is based on the fluctuation of the feature relative to the mean. If the change trends of the two features relative to the mean are similar (that is, the product of the difference is large), the correlation between them is high; conversely, if the change trends of the two features are very different (the product of the difference is small or negative), the correlation is low. In this way, the correlation values ​​between each pair of features can be obtained, which reflect the degree of internal connection between different features in the reconstructed fusion data.

[0116] The adjacency matrix is ​​constructed based on the correlation value. The adjacency matrix is ​​a square matrix whose row and column numbers correspond to different features. For example, if there are n features, then the adjacency matrix is ​​an n×n matrix. Each element in the matrix (i row and j column) represents the correlation value between the i-th feature and the j-th feature. If the correlation value is high (exceeding a certain threshold or meeting specific conditions), it means that there is a strong association between the two features, which can be regarded as a strong connection between the nodes corresponding to the two features in the graph structure; if the correlation value is low, it means that the connection between the nodes is weak or there is no direct connection.

[0117] The correlation information of the reconstructed fused data obtained from unsupervised network analysis is converted into a graph structure representation such as an adjacency matrix, realizing the transformation from the data feature level to the graph structure level. This transformation enables the relationship between the data to be presented in an intuitive and easy-to-process manner, facilitating further analysis using graph-related algorithms and techniques.

[0118] S104, forming an alternating integrated network with the adjacency matrix, and training the alternating integrated network with the wear amount samples, wherein the trained alternating integrated network is used to estimate the wear amount of the plunger pump slipper pair in real time.

[0119] Specifically, forming an alternating integrated network with the adjacency matrix and training the alternating integrated network with the wear amount samples includes:

[0120] (1) Obtaining instance node features and instance edge features of a graph based on an adjacency matrix to generate an instance graph; the instance node features are obtained by a convolution operation, and the instance edge features are obtained by a Pearson correlation function.

[0121] Specifically, build an example diagram ,in, Indicates Nodes in layer graph convolution Instance node features of Indicates Nodes in layer graph convolution and nodes Instance edge features of is the instance graph adjacency matrix; and They will be calculated as follows:

[0122] ;

[0123] in, represents the convolution operation used for feature extraction, Represents the node corresponding to Input samples of

[0124] ;

[0125] in, represents the Pearson correlation function, Indicates Nodes in layer graph convolution Instance node features of .

[0126] (2) Constructing a graph information embedding module based on the instance node features of the instance graph.

[0127] It should be noted that the graph information embedding module constructed based on the instance node features of the instance graph may be a Transformer module, wherein the Transformer module is a neural network architecture based on an attention mechanism.

[0128] Specifically, we construct graph information embedding based on the instance node features of the instance graph as follows:

[0129] ;

[0130] in, Represents a splicing operation, Representation Node The location information is embedded, For the Nodes in layer graph convolution Instance node features of .

[0131] When it is a Transformer module, the output of the Transformer module is represented as follows:

[0132] ;

[0133] in, represents the self-attention function, represents the normalization operation, Represents the graph information embedding of the c-th Transformer module.

[0134] (3) Based on the adjacency matrix, the distribution node features and distribution edge features of the graph are obtained to generate a distribution graph.

[0135] Specifically, build a distribution map ,in, Indicates Nodes in layer graph convolution The distribution node characteristics of Indicates Nodes in layer graph convolution and nodes The distribution edge characteristics of is the adjacency matrix of the distribution graph; and They will be calculated as follows:

[0136] ;

[0137] in, Indicates Nodes in layer graph convolution Distribution node characteristics; represents the convolution operation used for feature extraction; represents the Pearson correlation function;

[0138] (4) The instance graph, the graph information embedding module and the distribution graph are sequentially combined to form an alternating integrated network.

[0139] Specifically, a graph convolution-Transformer alternating integration network is constructed in the order of instance graph, Transformer module (graph information embedding module) and distribution graph.

[0140] Preferably, after constructing the graph convolution-Transformer alternating integration network, the above training set can also be used and test set Train the graph convolution-Transformer alternating integration network and estimate the wear of the plunger pump slipper pair at different times; the instance graph and distribution graph are co-optimized, and the designed training loss function as follows:

[0141] ;

[0142] in, , and are the weight coefficients corresponding to the instance graph loss function, distribution graph loss function and classifier function, respectively. , and They represent the instance graph loss function, distribution graph loss function and classifier function respectively, which satisfy the following:

[0143] ;

[0144] Among them, L represents the number of layers of the graph convolutional network, represents the number of samples in the training set, represents the cross entropy loss function, Represents graph convolution-Transformer alternating integration operation; represents the mathematical expectation, It represents the prediction result of the classifier, that is, the estimated wear amount of the plunger pump slipper pair at different times.

[0145] Fig.10 This is the confusion matrix diagram for estimating the wear amount of the piston pump slipper pair shown in this application. Fig.11 This is a visualization diagram of the wear characteristics of the plunger pump slipper pair shown in this application. Please also refer to Fig.10 and Fig.11 , the wear amount of the plunger pump slipper pair at different times can be estimated.

[0146] The time-varying wear estimation method of the plunger pump slipper pair provided in this embodiment forms a wear table based on the wear of the plunger pump slipper pair at different times, and combines the pressure pulsation signal, vibration signal, return oil volume signal and other multi-source operating parameter signals to form a wear sample, comprehensively considering the wear conditions reflected by multiple signals under variable working conditions, avoiding the one-sidedness that may be caused by a single signal, and being able to more comprehensively and accurately characterize the actual wear state of the slipper pair under different working conditions and time points. The wear characterization features and fusion weights are calculated by the wear table and the operating parameter signal, and the importance of each signal can be dynamically adjusted according to the wear data at different times, reflecting the changes in the wear conditions of the slipper pair under different working conditions in real time, making the estimation of the wear more in line with the actual situation and improving the actual prediction ability of the wear. In addition, time-frequency diagrams are used to analyze the wear samples, and short-time Fourier transform converts the signal into time-frequency domain representation, which can simultaneously present the signal's change information in time and frequency, and helps to capture the complex time-varying patterns and feature changes in the signal; and an unsupervised network is used to analyze the correlation of the fused data. The unsupervised network does not require a large amount of pre-labeled data, and can automatically discover the inherent structure and patterns in the data. By performing feature learning on the reconstructed fused data, the hidden correlation information in the data is mined, which reduces dependence on manual labeling, improves the efficiency and accuracy of the analysis, and helps to estimate the wear amount more accurately.

[0147] It should also be noted that constructing a graph convolution-Transformer alternating integration network based on the adjacency matrix and using wear samples for training can effectively integrate the graph convolution network's ability to extract local structural information and the Transformer module's ability to capture global information, thereby achieving real-time and accurate estimation of wear. During the training process, by collaboratively optimizing the instance graph and distribution graph and designing a reasonable training loss function, the network can better learn the features and patterns in the data, thereby improving the generalization ability and stability of the model.

[0148] Corresponding to the aforementioned embodiment of a method for estimating the time-varying wear of a plunger pump slipper pair, the present application also provides an embodiment of a device for estimating the time-varying wear of a plunger pump slipper pair.

[0149] Fig.12This is a schematic diagram of the structure of the second embodiment of the device for estimating the time-varying wear of the plunger pump slipper pair provided by this application. Fig.12 The device provided in this embodiment includes a construction module 1210, a calculation module 1220, an analysis module 1230 and a processing module 1240; wherein:

[0150] The construction module 1210 is used to form a wear amount table based on the wear amount of the plunger pump sliding shoe pair at different times, and to form a wear amount sample based on the wear amount table and the operating parameter signal;

[0151] The calculation module 1220 is used to calculate the wear amount characterization features and fusion weights of the operating parameter signal, and generate fusion data according to the fusion weights and the time-frequency diagram; wherein the time-frequency diagram is formed based on the wear amount samples;

[0152] The analysis module 1230 is used to analyze the correlation of the fused data based on an unsupervised network, and construct an adjacency matrix based on the correlation;

[0153] The processing module 1240 is used to form an alternating integrated network with the adjacency matrix, and train the alternating integrated network with the wear amount samples, wherein the trained alternating integrated network is used to estimate the wear amount of the plunger pump slipper pair in real time.

[0154] The device of this embodiment can be used to perform Figure 1 The steps, specific implementation principles and implementation processes of the method embodiment shown are similar and will not be repeated here.

[0155] The implementation process of the functions and effects of each unit in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, and will not be repeated here.

[0156] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present application scheme. A person of ordinary skill in the art can understand and implement it without paying any creative work.

[0157] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A method for estimating the time-varying wear of a plunger pump slipper pair, characterized in that: The method comprises: A wear amount table is formed based on the wear amount of the plunger pump sliding shoe pair at different times, and a wear amount sample is formed based on the wear amount table and the operating parameter signal; Calculating the wear amount characterization features and fusion weights of the operating parameter signal, and generating fusion data according to the fusion weights and a time-frequency diagram; wherein the time-frequency diagram is formed based on the wear amount samples; Analyzing the correlation of the fused data based on an unsupervised network, and constructing an adjacency matrix based on the correlation; An alternating integrated network is formed with the adjacency matrix, and the wear amount samples are used to train the alternating integrated network, wherein the trained alternating integrated network is used to estimate the wear amount of the plunger pump slipper pair in real time.

2. The method according to claim 1, characterized in that The calculating of the wear characterization features and fusion weights of the operating parameter signals, and generating fusion data according to the fusion weights and the time-frequency diagram, includes: Calculating wear amount characterization features corresponding to each signal based on the statistical features of the operating parameter signal; Calculating a fusion weight of the operating parameter signal according to the wear amount table and the wear amount characterization feature; A time-frequency diagram is formed based on the wear amount samples, and fusion data is generated by combining the time-frequency diagram with the fusion weight.

3. The method according to claim 1, characterized in that The forming of a wear amount table based on the wear amount of the plunger pump slipper pair at different times, and forming a wear amount sample based on the wear amount table and the operating parameter signal include: Measuring the wear amount of the plunger pump slipper pair at different times, forming a wear amount table based on the correspondence between the time information and the wear amount, and establishing a correlation between the wear amount table and the operating parameter signal; Collecting the operating parameter signal, and determining the wear amount information corresponding to each type of operating parameter signal at different times according to the wear amount table; All types of operating parameter signals and the wear amount information corresponding to each type of operating parameter signal at different times are summarized to form a wear amount sample.

4. The method according to claim 2, characterized in that: The calculating the wear amount characterization feature corresponding to each signal based on the statistical feature of the operating parameter signal includes: Obtaining the variation range of the statistical characteristics corresponding to each of the operating parameter signals; Calculating the wear sensitivity of each operating parameter signal based on the statistical characteristic variation range corresponding to each operating parameter signal and the variation range of the wear amount in the wear amount table; A maximum value operation is performed on the wear amount sensitivity of each statistical feature among the multiple statistical features of each operating parameter signal to obtain the wear amount characterization feature corresponding to each signal.

5. The method according to claim 2, characterized in that: The calculating the fusion weight of the operating parameter signal according to the wear amount table and the wear amount characterization feature comprises: Calculate the characteristic variation range of the wear amount of each operating parameter signal based on the wear amount table; The ratio of the variation range of the wear characterization feature of each operating parameter signal to the sum of the variation ranges of the wear characterization features of all operating parameter signals is used as the fusion weight of each operating parameter signal.

6. The method according to claim 2, characterized in that The forming of a time-frequency graph based on the wear amount sample and generating fusion data by combining the time-frequency graph with the fusion weight comprises: Acquire the wear amount sample, and convert the wear amount sample into a time-frequency diagram through short-time Fourier transform; Based on the time-frequency diagram and the fusion weight, a combination operation is performed to generate fusion data.

7. The method according to claim 1, characterized in that Analyzing the correlation of the fused data based on an unsupervised network and constructing an adjacency matrix based on the correlation includes: An unsupervised network is formed based on an encoder and a decoder; the encoder comprises a preset number of convolutional layers; the decoder comprises a preset number of deconvolutional layers; Acquire the fused data and input it into the unsupervised network to obtain reconstructed fused data; Perform feature learning on the reconstructed fusion data based on the pre-trained loss function, and use the learned feature representation to perform correlation calculation; An adjacency matrix is ​​constructed based on the correlation calculation results.

8. The method according to claim 1, characterized in that The forming of an alternating integrated network with the adjacency matrix and training the alternating integrated network with the wear amount samples comprises: Obtaining instance node features and instance edge features of a graph based on an adjacency matrix to generate an instance graph; the instance node features are obtained by a convolution operation, and the instance edge features are obtained by a Pearson correlation function; Constructing a graph information embedding module based on instance node features of the instance graph; Obtain the distribution node features and distribution edge features of the graph based on the adjacency matrix and generate a distribution graph; According to the instance graph, the graph information embedding module and the distribution graph, they are sequentially combined to form an alternating integrated network.

9. The method according to claim 1, characterized in that: The step of constructing an adjacency matrix based on the correlation comprises: Obtaining features of the reconstructed fusion data according to the unsupervised network; Calculate the feature mean of the reconstructed fusion data; Calculate the correlation between two features based on the difference between the feature and the feature mean; An adjacency matrix is ​​constructed based on the correlation, the row number and column number of the adjacency matrix determine two features corresponding to the correlation, and the value of the adjacency matrix is ​​the correlation value.

10. A device for estimating the time-varying wear of a plunger pump shoe pair, characterized in that: The device comprises a building module, a computing module, an analyzing module and a processing module; wherein: The construction module is used to form a wear amount table based on the wear amount of the plunger pump sliding shoe pair at different times, and to form a wear amount sample based on the wear amount table and the operating parameter signal; The calculation module is used to calculate the wear amount characterization features and fusion weights of the operating parameter signal, and generate fusion data according to the fusion weights and the time-frequency diagram; wherein the time-frequency diagram is formed based on the wear amount samples; The analysis module is used to analyze the correlation of the fused data based on an unsupervised network, and to construct an adjacency matrix based on the correlation; The processing module is used to form an alternating integrated network with the adjacency matrix and train the alternating integrated network using the wear amount samples, wherein the trained alternating integrated network is used to estimate the wear amount of the plunger pump slipper pair in real time.

Images