Mechanical seal fault diagnosis method and device for data imbalance scene

Abstract

The application discloses a mechanical seal fault diagnosis method for a data non-equilibrium scene, and comprises the following steps: collecting a vibration signal sequence at a rotating machine shaft seal, and adaptively denoising the vibration signal sequence; performing a variable step rough granulation operation on the denoised vibration signal sequence, and calculating the normalized permutation entropy of the denoised vibration signal sequence at different step lengths of each rough granulation scale; using an improved safe level synthesis minority class oversampling technology combined with an edited nearest neighbor method to respectively adaptively amplify the number of mechanical seal fault samples of each category, so that the number of the mechanical seal fault samples is leveled with the number of normal mechanical seal samples; training a KAN classifier based on the feature vectors of the amplified training set samples and corresponding fault type labels, to obtain a multi-classification model for mechanical seal fault diagnosis; and using the multi-classification model to diagnose the fault category of the mechanical seal. The application can improve the fault recognition capability and diagnosis reliability of the mechanical seal under the conditions of low signal-to-noise ratio of the signal and non-equilibrium of the training set samples.

Description

Technical Field

[0001] This invention belongs to the field of rotating machinery condition monitoring and fault diagnosis, and in particular relates to a method and device for mechanical seal fault diagnosis in scenarios of data imbalance. Background Technology

[0002] Mechanical seals are one of the key dynamic sealing components in rotating equipment (mixed-flow pumps, nuclear main pumps, etc.). Their core function is to reliably seal and control leakage of the medium while the shaft rotates at high speed. Compared with packing seals, mechanical seals have advantages such as low leakage, low frictional power consumption, long service life, and adaptability to higher pressures and speeds. Therefore, they are widely used in industries such as petrochemicals, seawater transportation, nuclear power, metallurgy, and pharmaceuticals. In engineering sites, the operating status of mechanical seals is directly related to the safety, environmental protection, and economy of equipment. Once the seal fails, it may lead to the leakage of flammable, explosive, or toxic media, causing fires, explosions, or personal injury. Therefore, upgrading mechanical seals from "reactive maintenance" to "online monitoring, diagnosis, and predictive maintenance" is of great significance for ensuring the long-term safe and stable operation of critical equipment.

[0003] It should also be noted that mechanical seal fault diagnosis faces significant sample imbalance in engineering applications: in real production scenarios, normal operating condition data is readily available for long-term continuous acquisition, while the probability of faults (especially early minor failures) is low, available samples are scarce, and the class distribution is extremely uneven; the frequency of different failure modes also varies significantly, resulting in training data exhibiting the typical characteristics of "normal classes accounting for the absolute majority, while fault classes are few and scattered." Under these conditions, directly using conventional data-driven models often leads to bias towards the majority class, manifesting as seemingly high overall accuracy but low recall rate for the minority class and a high rate of missed detection for early faults, making it difficult to meet the requirements of "early detection and early warning" for online monitoring. Therefore, for online condition monitoring and intelligent diagnosis of mechanical seals, it is urgent to introduce imbalanced learning and data augmentation strategies on the basis of feature extraction and model construction to improve the minority class fault identification capability and the engineering generalization performance of the model.

[0004] Currently, research on imbalanced sample scenarios mainly focuses on two aspects: data-level research and algorithm-level research. Some of the current mainstream methods are as follows: 1) Data layer resampling method These methods primarily alleviate the majority class dominance problem by altering the class distribution of the training data. These methods include oversampling (such as SMOTE), undersampling (such as RUS), and mixed sampling. Oversampling addresses the problem by increasing the number of minority class samples, while undersampling addresses it by reducing the number of majority class samples. These methods remain among the most reliable, are easy to implement in engineering, and are applicable to almost all models, possessing good universality and versatility. However, they are not very resistant to sample noise.

[0005] 2) Generative data augmentation This type of method mainly uses generative models to learn the minority class distribution and generate "realistic" minority class samples to fill the data gaps, such as generative adversarial networks, convolutional variational autoencoders, and diffusion models. This type of method has been a research hotspot in recent years, but it is still rarely used in the engineering field. This is mainly because the training is unstable, prone to pattern collapse, and the authenticity of the generated samples is difficult to quantify, which poses the risk of "deceiving the real from the fake but not helping with classification".

[0006] 3) Algorithm-level cost sensitivity and loss function modification This type of method mainly focuses on making the model pay more attention to minority class errors by using class weights / cost matrices / loss functions without changing the data distribution, such as class weights and cost-sensitive learning. This type of method is suitable for online updates and continuous learning scenarios, but the cost / weight settings lack a unified standard and are heavily dependent on experience or business constraints.

[0007] In summary, existing methods still have certain shortcomings in the application of mechanical seal fault diagnosis in scenarios with unbalanced data, which restricts their practical implementation effectiveness and urgently needs further improvement and refinement. On the other hand, under the combined influence of fluid excitation and mechanical vibration coupling, as well as external interference, the vibration signal-to-noise ratio of rotating machinery shaft seals is low, making fault identification based on vibration signals a significant challenge, thus making it difficult to fully guarantee the stability and reliability of diagnostic results. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention proposes a method and apparatus for diagnosing mechanical seal faults in scenarios with unbalanced data. The specific technical solution is as follows: A method for diagnosing mechanical seal faults in scenarios with unbalanced data includes the following steps: S1: Acquire the vibration signal sequence at the shaft seal of rotating machinery; S2: Adaptive noise reduction is performed on the acquired vibration signal sequence to obtain the noise-reduced vibration signal sequence; S3: Perform a variable step-size coarsening operation on the denoised vibration signal sequence. Calculate the normalized permutation entropy of the denoised vibration signal sequence at different step sizes of each coarsening scale, thereby compressing the high-dimensional vibration signal sequence. S4: Using an improved safety level synthetic minority class oversampling technique combined with the edited nearest neighbor method, the number of fault samples for each category of mechanical seal is adaptively amplified to make it equal to the number of normal mechanical seal samples; S5: Based on the feature vectors of the amplified training set samples and their corresponding fault type labels, train the KAN classifier to obtain a multi-classification model for mechanical seal fault diagnosis. S6: Input the feature vector extracted from the data sample to be diagnosed after adaptive noise reduction into the multi-classification model to complete the training, and output the specific fault category of the mechanical seal.

[0009] A mechanical seal fault diagnosis device for data imbalance scenarios includes one or more processors for implementing a mechanical seal fault diagnosis method for data imbalance scenarios.

[0010] An electronic device, comprising: One or more processors; A storage device for storing one or more programs, which, when executed by the electronic device, enable the electronic device to implement a mechanical seal fault diagnosis method for data imbalance scenarios.

[0011] A computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements a method for diagnosing mechanical seal faults in scenarios of data imbalance.

[0012] The beneficial effects of this invention are as follows: This invention addresses the characteristics of industrial data characterized by "dominant normal samples, scarce faulty samples, and class imbalance," by synergistically improving the model's ability to identify minority classes through adaptive noise reduction / multi-scale feature extraction and imbalanced data augmentation strategies. Firstly, in the signal analysis and feature extraction stage, adaptive noise reduction and variable step size multi-scale permutation entropy feature extraction are used to fully characterize the dynamic feature differences of strong non-stationarity, low signal-to-noise ratio and strong coupling during the deterioration process of mechanical seals, thereby enhancing the separability between different minority class fault samples. Secondly, by combining data augmentation strategies for imbalanced learning (adaptive oversampling and noise cleaning), the recall rate of minority classes and the performance of macro average indicators are effectively improved, avoiding the problem of "poor fault identification and high risk of missed detection" caused by relying solely on overall accuracy. Third, by further leveraging the advantages of the KAN model in function approximation and expression under small sample conditions, the generalization ability and stability of the model can be improved, making it more suitable for online fault diagnosis of mechanical seals of moving equipment in real industrial scenarios.

[0013] This invention can improve the fault identification capability and diagnostic reliability of mechanical seals under conditions of low signal-to-noise ratio and unbalanced training set samples, providing key technical support for the safe, stable and efficient operation of pump systems. Attached Figure Description

[0014] Figure 1 This is a flowchart of the mechanical seal fault diagnosis method for data imbalance scenarios according to the present invention.

[0015] Figure 2This is a time-domain diagram of vibration signal sequences collected under different mechanical seal health conditions in embodiments of the present invention.

[0016] Figure 3 This is a graph showing the changes in training loss and validation loss of the KAN model during the training process in an embodiment of the present invention, along with the iteration rounds.

[0017] Figure 4 This is a confusion matrix diagram of the refined diagnostic output of the KAN model for each mechanical seal vibration signal feature matrix in this embodiment of the invention (after performing row-wise normalization). Figure 5 This is a confusion matrix diagram (without normalization) of the refined diagnostic output results of the KAN model for the characteristic matrices of each mechanical seal vibration signal in the embodiment of the present invention. Detailed Implementation

[0018] The present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments. The purpose and effects of the present invention will become clearer. It should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0019] Definitions: ESL-SMOTE-ENN: Enhanced Safe-Level-Synthetic Minority Over-sampling Technique-Edited Nearest Neighbors. KAN: Kolmogorov-Arnold Networks.

[0020] The mechanical seal fault diagnosis method for unbalanced data scenarios of the present invention first denoises the acquired original vibration signal sequence and extracts a series of variable step-size multi-scale permutation entropy features from the denoised vibration signal sequence. On this basis, an ESL-SMOTE-ENN data augmentation and cleaning strategy for unbalanced data samples is introduced to adaptively amplify the number of minority class fault samples. Finally, the feature vectors extracted from the denoised data samples are input into a KAN multi-class classification model trained on a balanced training set of amplified samples to achieve refined classification and diagnosis of mechanical seal failure modes.

[0021] On the one hand, the mechanical seal fault diagnosis method for data imbalance scenarios of the present invention includes the following steps one to six.

[0022] Step 1: Collect the vibration signal sequence at the shaft seal of the rotating machinery.

[0023] The vibration signal sequence is denoted as: , N s The total number of sampling points is denoted as , and the sampling frequency is denoted as . f s ; During the data acquisition process, the frequency response parameter of the vibration acceleration sensor used should be no less than 1 kHz, the sampling frequency of the vibration signal sequence should be no less than 2.56 kHz, and the sampling duration should be no less than 0.35 s.

[0024] Step 2: Perform adaptive noise reduction on the acquired vibration signal sequence to obtain the noise-reduced vibration signal sequence.

[0025] Step two includes the following sub-steps: S2.1: Constructing the Hankel matrix of the vibration signal sequence : in, This represents the floor operation.

[0026] S2.2: For the Hankel matrix Perform singular value decomposition as follows: in, Represents a left singular value matrix. Represents a series of left singular vectors; Describes a right singular value matrix. Describe a series of right singular vectors; and, and All are orthogonal matrices. It is a with For diagonal matrices of the same dimension, their diagonal elements (i.e., singular values) Sort in descending order and all values ​​are non-negative. And satisfying rank rankL() is the operator for finding the rank of a matrix.

[0027] S2.3: Calculate the cumulative singular value sequence This forms a cumulative singular value sequence matrix. : S2.4: Calculate the diagonal slope of the cumulative singular value sequence. : S2.5: Calculate the adjacent slope sequence of the cumulative singular value sequence. This forms an adjacency slope sequence matrix. : S2.6: Find the slope closest to the diagonal in the adjacent slope sequence. The number of singular value sequences corresponding to the values The number of valid singular values ​​used for adaptive signal reconstruction is determined as follows: S2.7: Use the effective singular values ​​for signal reconstruction to obtain the reconstruction matrix. : in, Indicates the first tp Reconstructed submatrices of each sub-signal component.

[0028] S2.8: Using the anti-diagonal averaging method, based on the reconstructed matrix Reconstructing the one-dimensional signal yields the denoised vibration signal sequence. .

[0029] Step 3: Perform variable step-size coarsening on each denoised vibration signal sequence, at each coarsening scale. At different step sizes, the normalized permutation entropy of the denoised vibration signal sequence is calculated, thereby compressing the high-dimensional vibration signal sequence into... θ max 3D eigenvectors.

[0030] Step three includes the following sub-steps: S3.1: Perform variable step size coarsening operation on each noise-reduced vibration signal sequence to obtain the corresponding coarsened subsequence.

[0031] Among them, the noise-reduced vibration signal sequence scale factor The variable step size coarsening can be expressed as: in, Indicates step size, scale factor With step size All are positive integers and satisfy the following conditions: .

[0032] For a specific scale Due to step size Desirable Therefore, a total of One coarse-grained subsequence: S3.2: Reconstruct the phase space of the coarse-grained subsequence to obtain the embedded subsequence.

[0033] For each given scale and step length Let the embedding dimension be m and the time delay be d, for the sequence Phase space reconstruction is performed to obtain an embedded subsequence of length m. : in, .

[0034] S3.3: For each embedding subsequence at a given scale and step size Sort its elements in ascending order to obtain the sorting order pattern: in, express The rank() operator returns the sequence of indices of the elements in the input vector in ascending order.

[0035] S3.4: Statistically determine the actual frequency of each pattern in the coarse-grained subsequence at a given scale and step size.

[0036] All possible permutation patterns are: The types are respectively as Statistical coarse-grained subsequences The actual frequency of occurrence of each pattern in the text: in, , Count() represents the counting operator.

[0037] S3.5: For a given scale, based on the actual frequency of each mode, calculate the normalized permutation entropy of each coarse-grained subsequence at different step sizes, and take the average of the normalized permutation entropy of all step sizes to obtain the variable step size multi-scale permutation entropy of the signal at a given scale.

[0038] coarse-grained subsequence The permutation entropy (PE) is defined as: To facilitate comparability of different values ​​of m, normalized permutation entropy can be used. : For step size The mean of the normalized permutation entropy is used to obtain the signal at scale. Variable step size multiscale permutation entropy : S3.6: Extract the variable step size multi-scale permutation entropy of the denoised vibration signal sequence at all scales, and use the extracted variable step size multi-scale permutation entropy as a feature to construct the multi-scale feature vector of the denoised vibration signal sequence.

[0039] The multi-scale feature vector is represented as: Considering the diagnostic effectiveness and timeliness of the method implementation in practical applications, the embedding dimension m of the variable step-size multi-scale permutation entropy is set to 3, the time delay d is set to 1, and the maximum scale of the extracted variable step-size multi-scale permutation entropy is determined. θ max Select 30.

[0040] Step 4: Use the ESL-SMOTE-ENN method to adaptively amplify the number of fault samples for each category of mechanical seal, so that it is equal to the number of normal mechanical seal samples.

[0041] Let N1 be the number of normal samples, and the number of fault samples of different categories be respectively... , The total number of fault categories; using N1 as the baseline, the ESL-SMOTE-ENN method is used to classify... , The number of fault samples for each category was expanded to make M ij = N1.

[0042] Among them, the number of fault samples M ij It must not be less than 5% of the normal sample size N1.

[0043] Step four includes the following sub-steps: S4.1: Select any minority class sample h, and randomly select a sample ts from the same class neighborhood of minority class sample h, and calculate the security level sl of h respectively. h and the security level of ts sl ts And according to the security level SL ratio With sl hThe value of is used to determine the tendency factor gap for generating the synthetic sample S from the same line segment of samples h and ts, thus obtaining the synthetic sample.

[0044] Among them, security level SL ratio Represented as: The synthetic sample S is represented as: The specific rules for generating the synthetic sample S are as follows: when And sl h When =0, it means that both h and ts are noise, and no synthetic sample S is generated in this case because noisy instances should be avoided. when and When ts is noise, the synthetic sample S is generated by copying the sample h, that is, the gap value is 0; when When, it means sl h =0, sample h is noise. In this case, the synthetic sample S is generated by copying sample ts, that is, gap value is 1; when When, it indicates sl ts >sl h At this point, the synthesized sample S should be closer to ts, i.e. ; when At the same time, it also means sl ts >sl h At this point, the synthesized sample S should be closer to ts, i.e. ; When the security level is sl ratio When =1, it means sl ts With sl h Similarly, at this point, a synthetic sample S is generated along the line connecting h and ts, that is, at this point... ; when When, it indicates sl h Higher than sl ts At this point, the synthesized sample S should be closer to h, that is, at this time... ; when At that time, it means sl h Higher than sl ts At this point, the synthesized sample S should be closer to h, that is, at this time... .

[0045] S4.2: Find the ky nearest neighbors of the synthetic sample S. If ≥60% of the samples in the ky nearest neighbors have different labels from the minority class sample h, then the synthetic sample S is considered a false sample and deleted.

[0046] S4.3: Repeat S4.1 and S4.2 until the number of fault samples is M. ij It equals the number of normal samples N1; where, .

[0047] Step 5: Based on the feature vectors of the amplified training set samples and their corresponding fault type labels, train the KAN classifier to obtain a multi-classification model for mechanical seal fault diagnosis.

[0048] Step 6: Input the feature vector extracted from the data sample to be diagnosed after adaptive noise reduction into the multi-classification model that has been trained, and output the specific fault category of the mechanical seal.

[0049] On the other hand, corresponding to the aforementioned embodiments of the mechanical seal fault diagnosis method for data imbalance scenarios, the present invention also provides embodiments of a mechanical seal fault diagnosis device for data imbalance scenarios.

[0050] This invention provides a mechanical seal fault diagnosis device for data imbalance scenarios. It includes one or more processors for implementing the mechanical seal fault diagnosis method for data imbalance scenarios described in the above embodiments.

[0051] The embodiments of the mechanical seal fault diagnosis device for data imbalance scenarios of the present invention can be applied to any device with data processing capabilities, such as a computer. The device embodiments can be implemented in software, hardware, or a combination of both. Taking software implementation as an example, as a logical device, it is formed by the processor of the device with data processing capabilities reading the corresponding computer program instructions from non-volatile memory into memory and executing them. From a hardware perspective, in addition to the processor, memory, network interface, and non-volatile memory, the device with data processing capabilities in the embodiments may also include other hardware depending on the actual function of the device, which will not be elaborated further.

[0052] The specific implementation process of the functions and roles of each unit in the above device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0053] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of the present invention according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0054] This invention also provides a computer-readable storage medium storing a program thereon, which, when executed by a processor, implements the mechanical seal fault diagnosis method for data imbalance scenarios described in the above embodiments.

[0055] The computer-readable storage medium can be an internal storage unit of any data processing device described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium can also be an external storage device, such as a plug-in hard disk, smart media card (SMC), SD card, flash card, etc., equipped on the device. Furthermore, the computer-readable storage medium can include both internal storage units of any data processing device and external storage devices. The computer-readable storage medium is used to store the computer program and other programs and data required by the data processing device, and can also be used to temporarily store data that has been output or will be output.

[0056] The following specific implementation example demonstrates the effectiveness of the method and apparatus of the present invention. The specific method flow is as follows: Figure 1 As shown.

[0057] This implementation example focuses on the shaft seal area of ​​a hydraulic rotary mechanical seal, employing vibration sensors to collect and store its vibration response online. The experiment utilizes a centrifugal pump fault simulation test bench. By replacing, combining, and adjusting the parameters of key mechanical seal components, various typical failure scenarios are constructed—the set mechanical seal fault types cover dynamic ring cracking, dynamic ring wear, spring failure, and stationary ring damage, thereby achieving controllable reproduction of multi-state faults in centrifugal pump mechanical seals.

[0058] A wireless vibration accelerometer was installed at the pump end shaft seal to acquire vibration signal sequences, with a sampling frequency of 5120Hz. During the test, the centrifugal pump operated stably at 1500 rpm; the sensor sampled periodically at fixed time intervals, recording data every 5 minutes, with a single sampling duration of 0.6 seconds, resulting in a vibration sequence of 3072 points.

[0059] To closely resemble the data distribution characteristics of "more normal samples and fewer fault samples" in actual engineering scenarios, the specific composition of the samples in the training and test sets is shown in Table 1. Examples of typical vibration sequences under different mechanical seal conditions are shown in [Table 1]. Figure 2 .

[0060] Table 1 Description of Mechanical Seal Failure Dataset Then, adaptive denoising is performed on the acquired original vibration signal sequence to obtain the denoised vibration signal sequence. Then, at each scale factor... The following is achieved by satisfying Different step sizes The signal is coarsened by varying step size. The normalized permutation entropy of each coarsened subsequence is calculated and the average value is taken to obtain the multi-scale permutation entropy of varying step size at each scale. Thus, a series of multi-scale permutation entropy features of varying step size are obtained and used as the multi-scale complexity characterization of the mechanical seal vibration signal sequence.

[0061] The proposed ESL-SMOTE-ENN method is used to adaptively amplify mechanical seal failure samples: First, the number of failure samples for each category of mechanical seal is expanded from 5 to 100 using the rules of the Enhanced Safety Level Oversampling (ESL-SMOTE) method, matching the number of normal samples. Then, "inauthentic" synthetic samples are deleted using the Edit Nearest Neighbor (ENN) mechanism (setting the number of nearest neighbors in the filtering stage to 3 and the minority class proportion threshold in the filtering stage to 0.6). The steps are repeated until the number of normal mechanical seal samples and the number of failure samples for each category are both 100, resulting in the amplified training set.

[0062] Based on the amplified training set sample feature vectors and their corresponding fault type labels, a Kolmogorov-Arnold network (KAN) classifier is trained. The structure and training parameters of the KAN classifier are set as follows: the number of hidden layer nodes H is 32; the number of one-dimensional spline basis functions K per input dimension is 12; the number of training epochs E is 400; the batch size B is 256; the learning rate η is 0.005; and the L2 regularization coefficient λ is... s Pick The activation function for the hidden layer is the hyperbolic tangent function. Furthermore, the spline nodes are adaptively determined based on the quantiles of the training set features.

[0063] Figure 3 The graphs show the changes in training loss and validation loss of the KAN model with the number of iterations during the training process, and finally the trained KAN multi-class classification model is obtained.

[0064] The feature vectors extracted from the test data samples after adaptive noise reduction are input into the trained KAN multi-classification model, which outputs the specific fault category of the mechanical seal—the confusion matrix result is as follows. Figure 4 , Figure 5 As shown, the KAN multi-classification model achieved an average mechanical seal fault identification accuracy of 97.33% on the test dataset, and achieved 100% identification accuracy for normal mechanical seals, rotating ring wear, and stationary ring damage. It also misidentified only one sample in the spring failure category. While there is still room for improvement in the rotating ring crack category, it still achieved an 89% identification accuracy. These results demonstrate that the method of this invention, combining adaptive denoising technology, variable step-size multi-scale permutation entropy features, the ESL-SMOTE-ENN method, and the KAN classification model, can effectively achieve vibration monitoring and fault diagnosis of rotating machinery mechanical seals in data imbalance scenarios.

[0065] It will be understood by those skilled in the art that the above descriptions are merely preferred examples of the invention and are not intended to limit the invention. Although the invention has been described in detail with reference to the foregoing examples, those skilled in the art can still modify the technical solutions described in the foregoing examples or make equivalent substitutions for some of the technical features. All modifications and equivalent substitutions made within the spirit and principles of the invention should be included within the scope of protection of the invention.

Claims

1. A method for diagnosing mechanical seal faults in scenarios with unbalanced data, characterized in that, Includes the following steps: S1: Acquire the vibration signal sequence at the shaft seal of rotating machinery; S2: Adaptive noise reduction is performed on the acquired vibration signal sequence to obtain the noise-reduced vibration signal sequence; S3: Perform a variable step-size coarsening operation on the denoised vibration signal sequence. Calculate the normalized permutation entropy of the denoised vibration signal sequence at different step sizes of each coarsening scale, thereby compressing the high-dimensional vibration signal sequence. S4: Using an improved safety level synthetic minority class oversampling technique combined with the edited nearest neighbor method, the number of fault samples for each category of mechanical seal is adaptively amplified to make it equal to the number of normal mechanical seal samples; S5: Based on the feature vectors of the amplified training set samples and their corresponding fault type labels, train the KAN classifier to obtain a multi-classification model for mechanical seal fault diagnosis. S6: Input the feature vector extracted from the data sample to be diagnosed after adaptive noise reduction into the multi-classification model to complete the training, and output the specific fault category of the mechanical seal.

2. The mechanical seal fault diagnosis method for data imbalance scenarios according to claim 1, characterized in that, S4 includes the following sub-steps: S4.1: Select any minority class sample h, and randomly select a sample ts from the same class neighborhood of minority class sample h, and calculate the security level sl of h respectively. h and the security level of ts sl ts And according to sl h The value of , and sl h With sl ts The ratio of , i.e., security level SL ratio Determine the propensity factor gap for generating the synthetic sample S from the same line segment as samples h and ts, and obtain the synthetic sample S: ； S4.2: Find the ky nearest neighbors of the synthetic sample S. If ≥60% of the samples in the ky nearest neighbors have different labels from the minority class sample h, then the synthetic sample S is considered a false sample and deleted. S4.3: Repeat S4.1 and S4.2 until the number of faulty samples equals the number of normal samples.

3. The mechanical seal fault diagnosis method for data imbalance scenarios according to claim 2, characterized in that, The specific rules for generating the synthetic sample S are as follows: when And sl h When =0, it means that h and ts are both noise, and no synthetic sample S is generated at this time; when and When ts is noise, the synthetic sample S is generated by copying the sample h, that is, the gap value is 0; when When, it means sl h =0, sample h is noise. In this case, the synthetic sample S is generated by copying sample ts, that is, gap value is 1; when When, it indicates sl ts >sl h At this point, the synthesized sample S should be closer to ts, i.e. ; when At the same time, it also means sl ts >sl h At this point, the synthesized sample S should be closer to ts, i.e. ; When the security level is sl ratio When =1, it means sl ts With sl h Similarly, at this point, a synthetic sample S is generated along the line connecting h and ts, that is, at this point... ; when When, it indicates sl h Higher than sl ts At this point, the synthesized sample S should be closer to h, that is, at this time... ; when At that time, it means sl h Higher than sl ts At this point, the synthesized sample S should be closer to h, that is, at this time... .

4. The mechanical seal fault diagnosis method for data imbalance scenarios according to claim 1, characterized in that, S3 includes the following sub-steps: S3.1: Perform variable step size coarsening operation on each noise-reduced vibration signal sequence to obtain the corresponding coarsened subsequence; S3.2: Reconstruct the phase space of the coarse-grained subsequence to obtain the embedded subsequence; S3.3: For each embedded subsequence with a given scale and step size, sort its elements in ascending order to obtain the sorting order pattern; S3.4: Statistically determine the actual frequency of each pattern in the coarse-grained subsequence at a given scale and step size; S3.5: For a given scale, based on the actual frequency of each mode, calculate the normalized permutation entropy of each coarse-grained subsequence at different step sizes, and take the average of the normalized permutation entropy of all step sizes to obtain the variable step size multi-scale permutation entropy of the signal at a given scale. S3.6: Extract the variable step size multi-scale permutation entropy of the denoised vibration signal sequence at all scales, and use the extracted variable step size multi-scale permutation entropy as a feature to construct the multi-scale feature vector of the denoised vibration signal sequence.

5. The mechanical seal fault diagnosis method for data imbalance scenarios according to claim 1, characterized in that, S2 includes the following sub-steps: S2.1: Construct the Hankel matrix of the vibration signal sequence; S2.2: Perform singular value decomposition on the Hankel matrix; S2.3: Calculate the cumulative singular value sequence; S2.4: Calculate the diagonal slope of the cumulative singular value sequence; S2.5: Calculate the adjacent slope sequence of the cumulative singular value sequence; S2.6: The number of singular value sequences corresponding to the values ​​in the adjacent slope sequences that are closest to the diagonal slope is determined as the number of effective singular values ​​for adaptive signal reconstruction; S2.7: Use the effective singular values ​​for signal reconstruction to obtain the reconstruction matrix; S2.8: Using the anti-diagonal averaging method, a one-dimensional signal is reconstructed based on the reconstruction matrix to obtain a noise-reduced vibration signal sequence.

6. The mechanical seal fault diagnosis method for data imbalance scenarios according to claim 1, characterized in that, In S4, the number of fault samples of each category shall not be less than 5% of the number of normal samples.

7. The mechanical seal fault diagnosis method for data imbalance scenarios according to claim 1, characterized in that, In step S1, the frequency response parameter of the vibration acceleration sensor used to collect the vibration signal sequence at the shaft seal of the rotating machinery should be no less than 1 kHz, the sampling frequency of the vibration signal sequence should be no less than 2.56 kHz, and the sampling duration should be no less than 0.35 s.

8. A mechanical seal fault diagnosis device for data imbalance scenarios, characterized in that, It includes one or more processors for implementing the mechanical seal fault diagnosis method for data imbalance scenarios as described in any one of claims 1 to 7.

9. An electronic device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by the electronic device, cause the electronic device to implement the mechanical seal fault diagnosis method for data imbalance scenarios as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores a program that, when executed by a processor, implements the mechanical seal fault diagnosis method for data imbalance scenarios as described in any one of claims 1 to 7.

Images