A valve leakage fault diagnosis classification method and system based on a few labeled samples and LG-LapSVM semi-supervised learning

Abstract

The application provides a valve leakage fault diagnosis classification method and system based on LG-LapSVM semi-supervised learning of a few labeled samples, first, different internal leakage state signals are collected through an acoustic emission sensor, permutation entropy, global node outlink entropy, attention entropy and global ordered pattern attention entropy are extracted, and a combined feature vector is constructed by serial fusion, then a semi-supervised classifier (Lie group Laplace support vector machine) is used, under the condition of only a few labeled samples, a large number of unlabeled sample learning data manifold geometric structure is used to embed Lie group transformation invariance constraint, and high-precision and high-robustness identification of the internal leakage state of the hydraulic valve is realized. The application effectively solves the problems of insufficient single entropy feature representation and weak generalization ability of traditional classifiers for high-dimensional complex signals in the few labeled scene, and significantly improves the accuracy and stability of internal leakage identification.

Description

Technical Field

[0001] This invention relates to the field of hydraulic valve fault diagnosis technology, and in particular to a valve leakage fault diagnosis and classification method and system based on LG-LapSVM semi-supervised learning with few labeled samples. Background Technology

[0002] As the core control component of a hydraulic system, the performance of hydraulic valves directly affects the safety and reliability of the entire system. During long-term operation, hydraulic valves are prone to internal leakage due to wear, corrosion, fatigue, and other factors. If not detected and addressed promptly, this can lead to decreased system pressure, increased energy consumption, and even major safety accidents. Acoustic emission detection technology, due to its high sensitivity and fast dynamic response, has been widely used in diagnosing internal leakage faults in hydraulic valves. However, the acoustic emission signals from internal leakage in hydraulic valves often exhibit nonlinearity, non-stationarity, and strong noise interference, making fault feature extraction difficult and hindering accurate identification.

[0003] Currently, common fault feature extraction methods mainly include time-domain analysis, frequency-domain analysis, and time-frequency-domain analysis. However, a single time-domain or frequency-domain feature is insufficient to fully characterize the complex characteristics of hydraulic valve internal leakage signals. Entropy, as an effective indicator of time series complexity, has been widely used in signal processing. Permutation entropy can effectively quantify the randomness of time series, but it cannot capture the dynamic structural information contained in continuous vectors; global node out-chain entropy can characterize the state transition rules in time series through graph theory modeling; attention entropy has the advantage of computational simplicity by focusing on the repetitive patterns of key patterns; and global ordered pattern attention entropy combines the advantages of ordered pattern frequency and attention entropy. However, a single entropy feature often only describes signal characteristics from a single dimension, making it difficult to fully reflect the multi-scale complex characteristics of hydraulic valve internal leakage signals, resulting in insufficient fault characterization capabilities and limited identification accuracy.

[0004] In classification and recognition, traditional Support Vector Machines (SVMs), as supervised learning methods, heavily rely on a large number of labeled samples to achieve generalization ability. However, in practical applications of hydraulic valve internal leakage fault diagnosis, obtaining labeled fault samples is costly, time-consuming, and requires professional technicians for annotation; conversely, a large number of unlabeled acoustic emission signals are easy to obtain. This scenario of "very few labeled samples and abundant unlabeled samples" poses a severe challenge to traditional supervised classifiers. While semi-supervised methods such as LapSVMs can utilize the manifold structure of unlabeled samples through graph Laplacian regularization, they often ignore the Lie group manifold transformation invariance of the feature space itself, limiting their classification performance under conditions of few labels and strong noise. Therefore, there is an urgent need for a hydraulic valve internal leakage identification method that can fully mine the structural information contained in unlabeled samples and integrate the inherent transformation invariance of signals under conditions of very few labeled samples.

[0005] Therefore, a hydraulic valve internal leakage identification method is needed that can integrate complementary information from multi-dimensional entropy features and simultaneously embed data manifold geometry and Lie group transformation invariance constraints to solve the problems of insufficient representation by single features and weak generalization ability of traditional classifiers for high-dimensional complex signals, thereby improving the accuracy, stability and robustness of internal leakage identification. Summary of the Invention

[0006] In view of this, the purpose of this invention is to provide a valve leakage fault diagnosis and classification method and system based on LG-LapSVM semi-supervised learning with few labeled samples. This method effectively solves the problems of insufficient single entropy feature representation in scenarios with few labels and the weak generalization ability of traditional classifiers for high-dimensional complex signals, and significantly improves the accuracy and stability of internal leakage identification.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a semi-supervised learning method for valve leakage fault diagnosis and classification using LG-LapSVM with few labeled samples, comprising the following steps:

[0008] Step S1: Use an acoustic emission sensor to collect acoustic emission signals of the hydraulic valve under no-leakage, moderate-leakage, and severe-leakage conditions, and obtain its time-domain signal sample set;

[0009] Step S2: Preprocess the collected acoustic emission signal, extract the permutation entropy, global node outgoing chain entropy, attention entropy and global ordered pattern attention entropy respectively, and fuse the four entropy features in series to construct a combined feature vector;

[0010] Step S3: Input the combined feature vector into the LG-LapSVM classifier for training and testing to identify the internal leakage state of the hydraulic valve.

[0011] In a preferred embodiment, the permutation entropy is calculated in step S2 as follows:

[0012] For time series First, the sequence is mapped into vector form through phase space reconstruction:

[0013]

[0014] Where X represents the acoustic emission time series to be analyzed; where x i Let X represent the signal amplitude at the i-th sampling point in the time series X, where i = 1, 2, ..., N; and N represent the total number of sampling points in the time series X. j Let X represent the j-th phase space vector reconstructed from the time series X; j represents the index of the reconstructed vector, j=1,2,…,N-(m-1)τ; N-(m-1)τ represents the number of effective reconstructed vectors that can be obtained from the original time series. For embedded dimensions, For time delay;

[0015] Each reconstruction vector Arrange the elements in ascending order of size to obtain the corresponding sequence pattern. ,in ; Statistical analysis of each order pattern probability of occurrence :

[0016]

[0017] Then the permutation entropy Defined as:

[0018]

[0019] in, Represents the reconstructed vector X j The ordinal pattern corresponding to the elements after ascending order is: ; Indicates the number of elements in the set; Representing ordinal patterns The probability of occurrence in all valid reconstructed vectors; m! represents the total number of ordinal patterns that may occur when the embedding dimension is m.

[0020] In a preferred embodiment, the global node outgoing entropy is calculated in step S2 as follows:

[0021] Time series Reconstructing into ordered patterns, a network for converting ordered patterns is built; each ordered pattern type is defined as a network node, and a directed weighted edge represents the state transition frequency between two consecutive embedded vectors, thus constructing a weighted adjacency matrix. :

[0022]

[0023] in, Indicates from ordinal pattern arrive The transition frequency; where p and q represent the initial and target ordinal patterns, respectively; M represents the number of nodes in the ordered pattern transition network; when all theoretical patterns have appeared, ;

[0024] Eliminating the influence of self-loops, constructing a node-by-node outlink transition matrix. :

[0025]

[0026] This represents the conditional probability of transitioning from ordinal pattern p to ordinal pattern q; This represents the normalized out-chain weight or conditional transition probability of the transition from ordinal pattern p to ordinal pattern k; for each type of ordinal pattern... Calculate the outgoing entropy of the local node. :

[0027]

[0028] This represents the conditional probability of transitioning from ordinal pattern i to ordinal pattern j; combined with pattern frequency. The global node outgoing entropy is obtained as follows:

[0029] .

[0030] In a preferred embodiment, the attention entropy is calculated in step S2 as follows:

[0031] For a given time series With specified key pattern Record key patterns In time series Construct a position sequence from all occurrences. ;

[0032] For position sequence Calculate the first-order difference to obtain the interval sequence. ,in , i=1...K-1; This represents the positional interval between the i-th and (i+1)-th occurrences of the key pattern;

[0033] Calculate the interval sequence Shannon entropy, as a key pattern Lower Time Series attention entropy :

[0034]

[0035] This represents the probability of the interval value d appearing in the interval sequence D; key patterns are defined using four types of intervals: Max-Max, Min-Min, Max-Min, and Min-Max, and the corresponding attention entropy is calculated. Then the final attention entropy is:

[0036] .

[0037] In a preferred embodiment, the global ordered pattern attention entropy is calculated in step S2 as follows:

[0038] Determine the embedding dimension and latency time series Reconstruct into an ordered pattern, and compute each sequential pattern. frequency ,in , ;

[0039] For each sequence pattern Use it as the key pattern to calculate attention entropy. ;

[0040] Then the global ordered mode attention entropy Defined as:

[0041] .

[0042] In a preferred embodiment, in step S2, the method for constructing the combined feature vector is as follows:

[0043] Arrange the entropy feature vector Global node outgoing entropy feature vector Attention entropy feature vector and globally ordered pattern attention entropy feature vector By performing serial concatenation, a high-dimensional comprehensive feature vector is obtained. :

[0044] .

[0045] In a preferred embodiment, step S3, the construction and training of the LG-LapSVM classifier, specifically includes the following sub-steps:

[0046] Step S31, Data Preparation: Divide the combined feature vector set obtained in step S2 into a labeled sample set. and unlabeled sample set , , and usually And perform normalization processing;

[0047] Step S32: Construct the Lie group kernel function: Treat each sample feature as an element on the Lie group manifold, and define any two elements , The geodetic distance between them is ,in It is the Frobenius norm. This is a matrix-logarithmic mapping; based on this, we construct the radial basis kernel function of the Lie group:

[0048]

[0049] in For kernel parameters;

[0050] Step S33: Construct a graph Laplacian regularization: Construct a nearest neighbor graph using all samples, with edge weights... Calculate the graph Laplacian matrix ,in For the angle matrix;

[0051] Step S34: Establish the LG-LapSVM optimization objective function:

[0052]

[0053] The first term is the hinge loss, the second term is the regularization term in the regenerated Hilbert space, and the third term is the manifold regularization term. , For balance parameters;

[0054] in, This represents the classification decision function to be solved. Represent the regenerating kernel Hilbert space induced by the Lie group kernel function; Indicates the number of labeled samples. Indicates the number of unlabeled samples. + This represents the total number of samples participating in manifold learning; This represents the combined feature vector of the i-th sample. This indicates the category label corresponding to the labeled sample; where, This represents the sum of hinge losses over the labeled sample set. The classification decision function represents the classification decision function for the samples. The output value; The squared norm of the decision function f in the reproducing kernel Hilbert space is used to constrain the model complexity; Indicates sample With sample The nearest neighbor graph edge weights between them; This represents the environment space regularization coefficient, used to balance empirical loss and model complexity; This represents the intrinsic manifold regularization coefficient, used to control the strength of the constraint of the unlabeled sample manifold structure on the classification function;

[0055] Step S35, Solve the optimization problem: Use the representation theorem to express the decision function as follows: Substitute the values ​​into the objective function and transform it into a system of linear equations to solve for the classifier parameters.

[0056] Step S36: Use the trained LG-LapSVM model to predict the internal leakage state of the test samples.

[0057] This invention also provides a valve leakage fault diagnosis and classification system with few labeled samples using LG-LapSVM semi-supervised learning, including a processor, a memory, and a bus. The memory stores machine-readable instructions executed by the processor. When the system is running, the processor communicates with the memory via the bus, and when the machine-readable instructions are executed by the processor, the valve leakage fault diagnosis and classification method with few labeled samples using LG-LapSVM semi-supervised learning is described.

[0058] Compared with existing technologies, this invention has the following advantages: This invention combines multi-entropy fusion with LG-LapSVM, and is specifically designed for scenarios with few labeled samples. Compared with traditional supervised SVM (which requires a large number of labeled samples), ordinary LapSVM (which only utilizes manifold regularization but does not consider Lie group invariance) and parameter optimization SVM, this method can still maintain high recognition accuracy and robustness under the condition of few labels with a 20% labeled sample ratio, and is especially suitable for engineering scenarios where it is difficult to obtain labeled samples in the actual measurement of leakage sound emission signals in hydraulic valves. Attached Figure Description

[0059] Figure 1 This is a flowchart illustrating the method implementation of an embodiment of the present invention.

[0060] Figure 2 This is a time-domain waveform diagram of the acoustic emission signal of the hydraulic valve under different internal leakage states in an embodiment of the present invention.

[0061] Figure 3 This is a 3D visualization of the combined feature vectors in an embodiment of the present invention using t-SNE.

[0062] Figure 4 This is a diagram showing the recognition results of LG-LapSVM in an embodiment of the present invention. Detailed Implementation

[0063] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0064] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0065] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations according to this application; as used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise; furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0066] This embodiment provides a valve leakage fault diagnosis system and classification method based on LG-LapSVM semi-supervised learning with few labeled samples. The system is characterized by: firstly, for semi-supervised learning scenarios with very few labeled samples and abundant unlabeled samples, acoustic emission signals of the hydraulic valve under different internal leakage states are first collected using an acoustic emission sensor; then, the permutation entropy, global node outgoing chain entropy, attention entropy, and global ordered pattern attention entropy of the signals are extracted, and these four entropy features are serially fused to construct a combined feature vector; finally, a semi-supervised LG-LapSVM classifier integrating manifold regularization and Lie group kernels is used. With the assistance of few labeled samples, the system utilizes a large number of unlabeled samples to learn the inherent geometric structure of the data, thus completing the identification of the internal leakage state.

[0067] like Figure 1-4 As shown, this embodiment provides a valve leakage fault diagnosis system and its classification method based on LG-LapSVM semi-supervised learning with few labeled samples, specifically including the following steps:

[0068] Step S1: Use an acoustic emission sensor to collect acoustic emission signals from the hydraulic valve under leak-free, moderately leaking, and severely leaking conditions, and obtain its time-domain signal sample set (e.g., ...). Figure 2 (as shown)

[0069] Step S2: Preprocess the collected acoustic emission signal, extract the permutation entropy, global node outgoing chain entropy, attention entropy and global ordered pattern attention entropy respectively, and fuse the four entropy features in series to construct a combined feature vector;

[0070] Step S3: Input the combined feature vector into the LG-LapSVM classifier for training and testing to identify the internal leakage state of the hydraulic valve.

[0071] In step S1, the experimental data used in this embodiment all come from the acoustic emission detection experimental platform for internal leakage of hydraulic valves. The experimental system mainly consists of a DN50 cast steel hard-seal shut-off valve, a pressurized water chamber, an acoustic emission sensor, a preamplifier, and a data acquisition system. The acoustic emission sensor is a PAC R3α type, fixed to both sides of the valve body by magnetic clips. The signal is amplified and filtered by the preamplifier, and then the signal is acquired by the data acquisition card. During the experiment, three typical internal leakage faults, namely perforation leakage, crack leakage, and gap leakage, were set up, and acoustic emission signals under different leakage levels were collected under multiple pressure conditions of 0.2~1.0MPa. According to the leakage amount, the leakage state is divided into three levels: leakage amount ≤0.4L / min is defined as no leakage, 0.4L / min≤leakage amount≤4L / min is defined as general leakage, and leakage amount ≥4L / min is defined as severe leakage.

[0072] In step S2, the permutation entropy is calculated as follows:

[0073] For time series First, the sequence is mapped into vector form through phase space reconstruction:

[0074]

[0075] in, For embedded dimensions, For time delay;

[0076] Each reconstruction vector Arrange the elements in ascending order of size to obtain the corresponding order pattern. ,in ; Statistical analysis of each order pattern Probability of occurrence:

[0077]

[0078] Then the permutation entropy Defined as:

[0079]

[0080] In step S2, the global node outgoing entropy is calculated as follows:

[0081] Time series Reconstructing into ordered patterns, a network for converting ordered patterns is built; each ordered pattern type is defined as a network node, and a directed weighted edge represents the state transition frequency between two consecutive embedded vectors, thus constructing a weighted adjacency matrix. :

[0082]

[0083] in, Indicates from ordinal pattern arrive The transition frequency; eliminating the influence of self-loops, and constructing a node-by-node out-chain transition matrix. :

[0084] For each type of sequence pattern Calculate the outgoing entropy of the local node:

[0085] Combined mode frequency The global node outgoing entropy is obtained as follows:

[0086]

[0087] In step S2, the attention entropy is calculated as follows:

[0088] For a given time series With specified key pattern Record key patterns In sequence Construct a position sequence from all occurrences. ;

[0089] For position sequence Calculate the first-order difference to obtain the interval sequence. ,in ;

[0090] Calculate sequence Shannon entropy, as a key pattern Lower Time Series Attention entropy value:

[0091]

[0092] Key patterns are defined using four types of intervals (Max-Max, Min-Min, Max-Min, and Min-Max), and the corresponding attention entropy is calculated. Then the final attention entropy is:

[0093]

[0094] In step S2, the global ordered pattern attention entropy is calculated as follows:

[0095] Determine the embedding dimension and latency time series Reconstruct into an ordered pattern, and compute each pattern. frequency ,in , For each sequence pattern This is used as a key pattern, and attention entropy is calculated according to the method described in claim 5. Then the global ordered mode attention entropy Defined as:

[0096]

[0097] In step S2, the method for constructing the combined feature vector is as follows:

[0098] Arrange the entropy feature vector Global node outgoing entropy feature vector Attention entropy feature vector and the globally ordered pattern attention entropy feature vector By performing serial concatenation, a high-dimensional comprehensive feature vector is obtained. :

[0099]

[0100] Figure 3 The image shows a 3D visualization of the combined feature vector after t-SNE dimensionality reduction. It can be seen that the combined feature vector, which is formed by the serial fusion of four entropy features, has clear inter-class boundaries and good intra-class clustering, providing high-quality input for subsequent classification and recognition.

[0101] Step S3: Input the combined feature vector into the LG-LapSVM classifier for training and testing to identify the internal leakage status of the hydraulic valve. The specific steps are as follows:

[0102] Step S31: Data Preparation. The combined feature vector set obtained in Step S2 is divided into a training set and a test set (ratio 7:3), where only 20% of the training set consists of labeled samples, and the remaining 80% consists of unlabeled samples, to simulate real-world scenarios with few labels. All samples are normalized.

[0103] Step S32: Construct the Lie group kernel function: Treat each sample feature as an element on the Lie group manifold, and define any two elements , The geodetic distance between them is ,in It is the Frobenius norm. This is a matrix-logarithmic mapping; based on this, we construct the radial basis kernel function of the Lie group:

[0104]

[0105] in For kernel parameters;

[0106] Step S33: Constructing a graph Laplacian regularization: Construct a nearest neighbor graph using all samples (labeled and unlabeled), with edge weights... Calculate the graph Laplacian matrix ,in For the angle matrix;

[0107] Step S34: Establish the LG-LapSVM optimization objective function:

[0108]

[0109] The first term is the hinge loss, the second term is the regularization term in the regenerated Hilbert space, and the third term is the manifold regularization term. , For balance parameters;

[0110] Step S35: Solve the optimization problem: Express the decision function using the representation theorem. Substitute the values ​​into the objective function and transform it into a system of linear equations to solve for the classifier parameters.

[0111] Step S36: Use the trained LG-LapSVM model to predict the internal leakage state of the test samples.

[0112] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A semi-supervised learning method for valve leakage fault diagnosis and classification using LG-LapSVM with few labeled samples, characterized in that, Includes the following steps: Step S1: Use an acoustic emission sensor to collect acoustic emission signals of the hydraulic valve under no-leakage, moderate-leakage, and severe-leakage conditions, and obtain its time-domain signal sample set; Step S2: Preprocess the collected acoustic emission signal, extract the permutation entropy, global node outgoing chain entropy, attention entropy and global ordered pattern attention entropy respectively, and fuse the four entropy features in series to construct a combined feature vector; Step S3: Input the combined feature vector into the LG-LapSVM classifier for training and testing to identify the internal leakage state of the hydraulic valve.

2. The LG-LapSVM semi-supervised learning valve leakage fault diagnosis and classification method with few labeled samples as described in claim 1, characterized in that, In step S2, the permutation entropy is calculated as follows: For time series First, the sequence is mapped into vector form through phase space reconstruction: Where X represents the acoustic emission time series to be analyzed; where x i Let X represent the signal amplitude at the i-th sampling point in the time series X, where i = 1, 2, ..., N; and N represent the total number of sampling points in the time series X. j Let X represent the j-th phase space vector reconstructed from the time series X; j represents the index of the reconstructed vector, j=1,2,…,N-(m-1)τ; N-(m-1)τ represents the number of effective reconstructed vectors that can be obtained from the original time series. For embedded dimensions, For time delay; Each reconstruction vector Arrange the elements in ascending order of size to obtain the corresponding sequence pattern. ,in ; Statistical analysis of each order pattern probability of occurrence : Then the permutation entropy Defined as: in, Represents the reconstructed vector X j The ordinal pattern corresponding to the elements after ascending order is: ; Indicates the number of elements in the set; Representing ordinal patterns The probability of occurrence in all valid reconstructed vectors; m! represents the total number of ordinal patterns that may occur when the embedding dimension is m.

3. The LG-LapSVM semi-supervised learning valve leakage fault diagnosis and classification method with few labeled samples as described in claim 1, characterized in that, In step S2, the global node outgoing entropy is calculated as follows: Time series Reconstructing into ordered patterns, a network for converting ordered patterns is built; each ordered pattern type is defined as a network node, and a directed weighted edge represents the state transition frequency between two consecutive embedded vectors, thus constructing a weighted adjacency matrix. : in, Indicates from ordinal pattern arrive The transition frequency; where p and q represent the initial and target ordinal patterns, respectively; M represents the number of nodes in the ordered pattern transition network; when all theoretical patterns have appeared, ; Eliminating the influence of self-loops, constructing a node-by-node outlink transition matrix. : This represents the conditional probability of transitioning from ordinal pattern p to ordinal pattern q; This represents the normalized out-chain weight or conditional transition probability of the transition from ordinal pattern p to ordinal pattern k; for each type of ordinal pattern... Calculate the outgoing entropy of the local node. : This represents the conditional probability of transitioning from ordinal pattern i to ordinal pattern j; combined with pattern frequency. The global node outgoing entropy is obtained as follows: 。 4. The LG-LapSVM semi-supervised learning valve leakage fault diagnosis and classification method with few labeled samples as described in claim 1, characterized in that, In step S2, the attention entropy is calculated as follows: For a given time series With specified key pattern Record key patterns In time series Construct a position sequence from all occurrences. ; For position sequence Calculate the first-order difference to obtain the interval sequence. ,in , i=1...K-1; This represents the positional interval between the i-th and (i+1)-th occurrences of the key pattern; Calculate the interval sequence Shannon entropy, as a key pattern Lower Time Series attention entropy : This represents the probability that the interval value d appears in the interval sequence D; Key patterns are defined using four types of intervals: Max-Max, Min-Min, Max-Min, and Min-Max, and the corresponding attention entropy is calculated. Then the final attention entropy is: 。 5. The LG-LapSVM semi-supervised learning valve leakage fault diagnosis and classification method with few labeled samples according to claim 1, characterized in that, In step S2, the global ordered pattern attention entropy is calculated as follows: Determine the embedding dimension and latency time series Reconstruct into an ordered pattern, and compute each sequential pattern. frequency ,in , ; For each sequence pattern Use it as the key pattern to calculate attention entropy ; Then the global ordered mode attention entropy Defined as: 。 6. The method for identifying internal leakage of hydraulic valves based on LGLapSVM using entropy combination feature extraction according to claim 1, characterized in that, In step S2, the method for constructing the combined feature vector is as follows: Arrange the entropy feature vector Global node outgoing entropy feature vector Attention entropy feature vector and the globally ordered pattern attention entropy feature vector By performing serial concatenation, a high-dimensional comprehensive feature vector is obtained. : 。 7. The LG-LapSVM semi-supervised learning method for valve leakage fault diagnosis and classification with few labeled samples as described in claim 1, characterized in that, Step S3, the construction and training of the LG-LapSVM classifier specifically includes the following sub-steps: Step S31, Data Preparation: Divide the combined feature vector set obtained in step S2 into a labeled sample set. and unlabeled sample set , ,and And perform normalization processing; Step S32: Construct the Lie group kernel function: Treat each sample feature as an element on the Lie group manifold, and define any two elements , The geodetic distance between them is ,in It is the Frobenius norm. This is a matrix-logarithmic mapping; based on this, we construct the radial basis kernel function of the Lie group: in For kernel parameters; Step S33: Construct a graph Laplacian regularization: Construct a nearest neighbor graph using all samples, with edge weights... Calculate the graph Laplacian matrix ,in For the angle matrix; Step S34: Establish the LG-LapSVM optimization objective function: The first term is the hinge loss, the second term is the regularization term in the regenerated Hilbert space, and the third term is the manifold regularization term. , For balance parameters; in, This represents the classification decision function to be solved. Represent the regenerating kernel Hilbert space induced by the Lie group kernel function; Indicates the number of labeled samples. Indicates the number of unlabeled samples. + This represents the total number of samples participating in manifold learning; This represents the combined feature vector of the i-th sample. This indicates the category label corresponding to the labeled sample; where, This represents the sum of hinge losses over the labeled sample set. The classification decision function represents the classification decision function for the samples. The output value; The squared norm of the decision function f in the reproducing kernel Hilbert space is used to constrain the model complexity; Indicates sample With sample The nearest neighbor graph edge weights between them; This represents the environment space regularization coefficient, used to balance empirical loss and model complexity; This represents the intrinsic manifold regularization coefficient, used to control the strength of the constraint of the unlabeled sample manifold structure on the classification function; Step S35, Solve the optimization problem: Use the representation theorem to express the decision function as follows: Substitute the values ​​into the objective function and transform it into a system of linear equations to solve for the classifier parameters. Step S36: Use the trained LG-LapSVM model to predict the internal leakage state of the test samples.

8. A valve leakage fault diagnosis and classification system based on LG-LapSVM with few labeled samples, comprising a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executed by the processor; characterized in that, When the system is running, the processor communicates with the memory via a bus, and the machine-readable instructions are executed by the processor as described in any one of claims 1 to 7. This is a semi-supervised learning method for valve leakage fault diagnosis and classification using LG-LapSVM with few labeled samples.

Images