Gearbox condition monitoring method and system based on single-class normal data

Abstract

The application discloses a gearbox state monitoring method and system based on single-class normal data, and belongs to the field of fault diagnosis, and comprises the following steps: acquiring normal working condition data of a gearbox, performing nonlinear cross-bispectrum processing on the normal working condition data to obtain a training data set; constructing an interpretable differential diagnosis model, training and optimizing the interpretable differential diagnosis model through the training data set, acquiring running data of the gearbox under unknown states, performing nonlinear cross-bispectrum processing on the running data, and judging whether the running data is fault data according to the interpretable differential diagnosis model after processing the running data; and when the judgment result is fault data, performing feature enhancement on the fault data according to the optimized interpretable differential diagnosis model to obtain an interpretable fault diagnosis result. The above scheme can judge only by using normal working condition data, and overcomes the limitation of a lack of fault samples, and exhibits high precision, strong noise resistance and good generalization capability in gearbox fault diagnosis.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of gear box state monitoring, and particularly relates to a gear box state monitoring method and system based on single-class normal data. BACKGROUND

[0002] Gear boxes are widely used in mechanical transmission systems such as wind power generation, aviation and hoisting transportation. Long-term operation under complex working conditions can easily cause key component wear and fatigue cracks and other faults, and state monitoring is needed to realize fault diagnosis and reliability evaluation to ensure system safety and reduce operation and maintenance costs. Existing technologies mainly fall into two categories:

[0003] (1) Signal processing-based methods (such as fast spectral kurtosis method and spectral amplitude modulation method), which can realize feature extraction by restoring repeated fault transient waveforms, and are highly reliable but depend on professional knowledge;

[0004] (2) Artificial intelligence-based methods (such as deep multi-scale weighted modulation network and adversarial multi-wavelet convolutional neural network), which can handle cross-domain diagnosis problems, but the black box characteristics of the model make the decision-making process untraceable.

[0005] To address the interpretability issue, existing research includes vector quantization guided latent denoising diffusion model, uncertainty-aware deep learning, and two-sample set convex optimization model Shapley value-fault frequency mapping method, which have the common limitation of relying on both normal and fault sample data, making it difficult to apply to actual engineering scenarios where fault samples are scarce, i.e., it is difficult to monitor the fault state of the gear box in the case of scarce fault samples. SUMMARY

[0006] To solve the above technical problems, the application provides a gear box state monitoring method and system based on single-class normal data to solve the problems existing in the prior art.

[0007] To achieve the above purpose, the application provides a gear box state monitoring method based on single-class normal data, comprising:

[0008] Obtaining normal working condition data of the gear box, performing nonlinear cross-bispectrum processing on the normal working condition data to obtain a training data set;

[0009] Constructing an interpretable differential diagnosis model, training and optimizing the interpretable differential diagnosis model through the training data set, wherein the interpretable differential diagnosis model adopts a decision boundary model;

[0010] Obtaining running data of the gear box under unknown state, performing nonlinear cross-bispectrum processing on the running data, and judging the processed running data according to the optimized interpretable differential diagnosis model to obtain a judgment result of whether the running data is fault data.

[0011] When the judgment result is fault data, the feature of the fault data is enhanced according to the optimized explainable differential diagnosis model, and a fault diagnosis explainable result is obtained.

[0012] Optionally, the obtaining process of the training data set comprises:

[0013] Normal working condition data of a gear box transmission part is collected by a vibration acceleration sensor, wherein the normal working condition data is normal working condition operation data of the gear box in a rated speed and load range, and the operation data is vibration data;

[0014] The normal working condition data is processed by a nonlinear interactive bispectrum to extract high-dimensional bispectrum features to construct a training data set.

[0015] Optionally, the process of processing the normal working condition data by a nonlinear interactive bispectrum comprises:

[0016] ;

[0017] wherein, represents a nonlinear interactive bispectrum of the operation data, represents a mathematical expectation, represents a Fourier transform result of the operation data , and a superscript represents a complex conjugate symbol, represents a carrier frequency of the operation data, represents a modulation frequency of the operation data.

[0018] Optionally, the process of training and optimizing the explainable differential diagnosis model by the training data set comprises:

[0019] An explainable differential diagnosis model is constructed, wherein the explainable differential diagnosis model performs fault judgment and fault explanation by a hyper-sphere boundary decision method;

[0020] The hyper-sphere is trained by the training data set to obtain a center and a radius of the hyper-sphere, wherein the hyper-sphere is used to represent a health state feature space boundary.

[0021] Optionally, the process of training the hyper-sphere comprises:

[0022] According to the training data set, an optimization objective function is solved to obtain the center and the radius of the hyper-sphere;

[0023] wherein the optimization objective function is:

[0024] ;

[0025] wherein, As slack variables, This is a penalty factor used to balance the hypersphere radius and the number of data points outside the boundary. Indicates the center, , The numbers represent different training data, where i represents the label of the training data and n represents the total number of training data.

[0026] Optionally, the objective function can be solved using the Lagrange function.

[0027] Optionally, the process of judging the processed running data includes:

[0028] Based on the center and radius of the hypersphere in the optimized interpretable differential diagnostic model, the processed operating data is judged. Specifically, the distance between the processed operating data and the center of the hypersphere is calculated, and the distance is judged based on the radius of the hypersphere. When the distance is less than or equal to the radius of the hypersphere, the operating data is considered normal; otherwise, the operating data is considered faulty.

[0029] Optionally, the process of feature enhancement for fault data includes:

[0030] The fault data is processed using nonlinear interactive bispectral processing to obtain a nonlinear interactive bispectral spectrum. The nonlinear interactive bispectral spectrum of the fault data and the center of the hypersphere are normalized, and the normalization result is subjected to differential calculation to obtain a differential result. The amplitude and probability density curve fitting function of the nonlinear interactive bispectral spectrum of the fault data are extracted. A two-dimensional weight matrix is ​​obtained based on the amplitude and probability density curve fitting function of the nonlinear interactive bispectral spectrum of the fault data. The nonlinear interactive bispectral spectrum of the fault data is weighted according to the two-dimensional weight matrix. Based on the weighted calculation result, the interpretable result of fault diagnosis is obtained.

[0031] Optionally, a two-dimensional weight matrix is ​​obtained through a weight mapping function, wherein the weight mapping function is:

[0032] ;

[0033] in, This indicates taking the minimum value. This indicates taking the maximum value. This indicates normalization processing. The amplitude of the nonlinear interactive bispectral, According to The value range can be set by the user to ensure... range and Corresponding to the domain of definition, D represents the weight mapping coefficient. Indicates weight, represents a bicoherence coefficient, c represents a weight mapping function center, represents a nonlinear cross-bicoherence amplitude, represents a probability density curve fitting function.

[0034] In another aspect, the application provides a gearbox condition monitoring system based on single-class normal data, which is used to perform the above method.

[0035] Compared with the prior art, the application has the following advantages and technical effects:

[0036] The application provides a gearbox condition monitoring method and system based on single-class normal data, and the core of the application is a state difference monitoring mechanism. The mechanism first processes normal working condition data of the gearbox by using a nonlinear cross-bicoherence algorithm, improves the data dimension, and deeply mines hidden features; then an interpretable differential diagnosis model based on support data description is constructed; finally, differential weights are designed according to the discreteness of target domain features and health model center features, so that noise interference is effectively suppressed and the fault recognition capability is enhanced. The method can be operated only by using normal working condition data, and overcomes the limitation of the scarcity of fault samples, and exhibits high precision, strong noise resistance and good generalization capability in the application of parallel shaft and planetary gearbox. BRIEF DESCRIPTION OF DRAWINGS

[0037] The accompanying drawings, which form a part of this application, are included to provide a further understanding of the application and are incorporated in and constitute a part of this application. The embodiments of this application and their explanations are used to explain this application and do not constitute an improper limitation on this application. In the drawings:

[0038] Figure 1 A flowchart of the gearbox condition monitoring method and system based on single-class normal data of the embodiments of the application;

[0039] Figure 2 A minimum hypersphere diagram generated by the interpretable differential diagnosis model of the embodiments of the application;

[0040] Figure 3 Normal working condition data of the gearbox obtained by the embodiments of the application;

[0041] Figure 4 Fault vibration signal waveforms, frequency spectra, envelope spectra and nonlinear cross-bicoherences of the embodiments of the application;

[0042] Figure 5 A minimum hypersphere center generated by the interpretable differential diagnosis model of the embodiments of the application;

[0043] Figure 6 Differential features of the bicoherence domain of the embodiments of the application;

[0044] Figure 7The nonlinear cross-bispectrum of the feature-enhanced fault data of the embodiment of the application;

[0045] Figure 8 The schematic diagram of the gearbox condition monitoring system based on single-class normal data of the embodiment of the application. DETAILED DESCRIPTION

[0046] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0047] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a group of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that here.

[0048] The present application provides a gearbox condition monitoring method and system based on single-class normal data, and the core is a state difference monitoring mechanism. The mechanism first processes the normal working condition data of the gearbox by using a nonlinear cross-bispectrum algorithm, improves the data dimension, and deeply mines hidden features; then an interpretable differential diagnosis model based on support data description is constructed; finally, according to the discreteness of the target domain features and the health model center features, a differential weight is designed, which effectively suppresses noise interference and enhances the fault recognition ability. This method only needs normal working condition data to run, and overcomes the limitation of the scarcity of fault samples, and shows high precision, strong noise resistance and good generalization ability in parallel shaft and planetary gearbox applications.

[0049] Embodiment one

[0050] According to the embodiment of the present application, a rolling bearing fault diagnosis method is provided, Figure 1 The flowchart of the rolling bearing fault diagnosis method of the embodiment of the present application, according to Figure 1 The rolling bearing fault diagnosis method of the embodiment of the present application includes:

[0051] S1, acquiring normal working condition data of the gearbox, processing the data by using a nonlinear cross-bispectrum to construct a training data set; S1 specifically includes:

[0052] The vibration acceleration sensors installed on the bearing seat measuring points and the gearbox measuring points and the industrial-grade data acquisition card are used to acquire the normal working condition running data of the gearbox in the rated speed and load range in real time;

[0053] The data is processed by using a nonlinear cross-bispectrum, high-dimensional bispectrum features are extracted to construct a training data set.

[0054] Suppose The Fourier transform of the collected vibration data can be expressed as:

[0055] ;

[0056] in, for amplitude spectrum, for phase spectrum, The Fourier transform result of the vibration data is represented by e, which represents the natural constant, f, which represents the original signal, and t, which represents time.

[0057] To fully consider the mapping relationship between modulation components and fault characteristics in the signal, a definition is made. Nonlinear interactive bispectral for:

[0058] ;

[0059] in, It represents the mathematical expectation. for . These are harmonic coefficients, set according to the number of sideband frequencies in the modulation sideband, where i represents the harmonic coefficient designation. Indicates the carrier frequency. Indicates the modulation frequency.

[0060] ;

[0061] S2. Construct an interpretable differential diagnostic model based on supporting data description, and train the model using a training dataset; S2 specifically includes:

[0062] The training dataset is input into an interpretable differential diagnostic model constructed based on support vector data description;

[0063] The model is trained to generate a minimized hypersphere that satisfies the following:

[0064] (a) Contains the largest number of training data samples;

[0065] (b) Establish spatial boundaries for health status characteristics;

[0066] The differentiated diagnostic model described herein achieves interpretability by visualizing the decision-making process of the hypersphere boundary.

[0067] The training dataset constructed based on nonlinear interactive bispectral structures can be represented as:

[0068] ;

[0069] where the superscript H denotes the data is in a healthy state, p denotes the number of data in the input data set, denotes the nonlinear cross-spectrum.

[0070] A hyper-sphere with the minimum radius and the most training data is trained as the final training target using the training data set. The schematic diagram of the hyper-sphere is shown in Figure 2 .

[0071] The hyper-sphere can be represented by a center and a radius , where the data dimension of the center depends on the data dimension of the training data set, and the radius is a scalar. The optimization objective function of the model can be represented as:

[0072] ;

[0073] where is a slack variable, . is a penalty factor for balancing the radius of the hyper-sphere and the number of out-of-bound data.

[0074] By introducing a Lagrange multiplier , the objective function can be converted into a Lagrange function :

[0075] ;

[0076] Let the partial derivative of the Lagrange function be zero, and the optimization objective function can finally be written as:

[0077] ;

[0078] and denote different coefficients, i, j denote different training data labels, and n denotes the number of data of the training data.

[0079] After solving according to the optimization objective function, the center and the radius of the hyper-sphere can be calculated using the Gaussian kernel function. The center of the hyper-sphere summarizes and highlights the feature information of the training data.

[0080] The calculation process of the center and the radius of the hyper-sphere is as follows:

[0081]

[0082] ;

[0083] wherein, denotes a Gaussian kernel function. denotes a support vector, wherein subscript v denotes a corresponding label v of the training data as a support vector in the model, subscript i denotes different training data labels, and n denotes a total number of training data.

[0084] S3, collecting gearbox data in an unknown state, and processing the data using nonlinear cross-bispectrum to determine whether the running data in the unknown state contains a fault; S3 specifically includes:

[0085] Using a vibration acceleration sensor installed at a key transmission part of the gearbox and an industrial-grade data acquisition card, running data of the gearbox in an unknown state is obtained;

[0086] By processing the data using nonlinear cross-bispectrum, high-dimensional bispectrum features are extracted to construct test data;

[0087] According to the distance between the test data and the center of the hypersphere, it is determined whether the running data in the unknown state contains a fault.

[0088] Definition The nonlinear cross-bispectrum of the test data is

[0089] ;

[0090] When , it indicates that the distance between the test data and the center of the hypersphere exceeds the radius of the hypersphere, and it can be determined that the unknown data is fault data;

[0091] When , it indicates that the distance between the test data and the center of the hypersphere does not exceed the radius of the hypersphere, and it can be determined that the unknown data is normal data.

[0092] S4, according to the difference features between the bispectrum of the fault data and the health model, a weight mapping function is designed to establish a two-dimensional weight matrix, to strengthen the features of the fault data and realize fault diagnosis. S4 specifically includes:

[0093] Calculate the difference features between the data in the unknown state and the health model in the bispectrum domain;

[0094] Design a weight mapping function to convert the difference features into a two-dimensional weight matrix to quantify the difference features;

[0095] Use the two-dimensional weight matrix to weight the bispectrum of the fault data to enhance the potential features in the unknown data and realize fault diagnosis.

[0096] Differential calculation of the nonlinear cross-bispectrum of the fault data and the center of the hypersphere:

[0097] ;

[0098] in, Nonlinear interactive bispectral representing fault data, The nonlinear interactive bispectral representation of normalized fault data. The center of the normalized hypersphere. For differentiated results.

[0099] For differentiated results, the maximum marginal spectrum of the differentiated results is defined by f in the nonlinear interactive bispectrum. c The maximum amplitude at each frequency on the axis constitutes the maximum envelope spectrum of the differential results, which is composed of the nonlinear interactive bispectrum f. x It consists of the maximum amplitude of each frequency on the axis. The amplitude of the nonlinear interactive bispectral is given. The probability density curve fitting function for the maximum envelope spectrum. Therefore, the weighting mapping function can be expressed as:

[0100] ;

[0101] in, This indicates taking the minimum value. This indicates taking the maximum value. This indicates normalization processing. According to The value range can be set by the user to ensure... range and The domain of definition corresponds to this. D represents the weight mapping coefficient. Indicates weight, denoted by , where c represents the bispectral range coefficient and c represents the center of the weight mapping function.

[0102] Will By multiplying the nonlinear interactive bispectral data with the fault data, the fault characteristics can be enhanced. The highlighted frequencies have a direct mapping relationship with the location of the fault. By identifying the highlighted frequencies, the location of the fault can be determined, thus achieving fault diagnosis.

[0103] ;

[0104] Specifically, in one implementation of the present invention, a sensitivity of 10 mV / m×s is utilized. -2The acceleration sensor collects the vibration signal, and the installation position is the top of the gearbox. When collecting the vibration signal, first, the normally operating gearbox and the sensor are installed on the gearbox test bench, and the gearbox is driven to operate by the motor. Then, the test data is collected by the data collection device and transmitted to the computer, and the subsequent data processing and analysis are carried out by using the MATLAB software. The rotating speed of the shaft where the bearing is located is 3000 rpm, and the sampling frequency is 16000 Hz. Figure 3 The normal operating condition data of the gearbox are obtained.

[0105] When collecting the fault vibration signal, first, the gearbox containing the outer ring fault bearing and the sensor are installed on the gearbox test bench, and the gearbox is driven to operate by the motor. Then, the test data is collected by the data collection device and transmitted to the computer, and the subsequent data processing and analysis are carried out by using the MATLAB software. The rotating speed of the shaft where the bearing is located is 3000 rpm, and the sampling frequency is 16000 Hz. After calculation, the outer ring fault characteristic frequency of the bearing is 179.24 Hz.

[0106] The collected signal is subjected to Fourier transform to obtain the frequency spectrum of the signal. Then, the envelope demodulation method is used to process the signal to obtain the envelope spectrum of the signal. Then, HN=2 is taken to calculate the nonlinear cross bispectrum. Figure 4 The waveforms, frequency spectra, envelope spectra and nonlinear cross bispectra of the collected fault signals are shown in the table. In the waveforms, frequency spectra and envelope spectra of the signals, the characteristics for diagnosing the fault cannot be seen. In the nonlinear cross bispectrum, although the outer ring fault characteristic frequency of the bearing appears, there are many other interference components, and the persuasiveness of the diagnosis result is not strong.

[0107] The application will be further described below in combination with the drawings and examples.

[0108] The vibration signals shown in Figure 3 are input, and the nonlinear cross bispectrum of each vibration signal is calculated as a training data set. The training data set is input into the interpretable differential diagnosis model constructed based on the support vector data description. After training, the model hypersphere center is as shown in Figure 5 .

[0109] The radius of the hypersphere obtained by training is 5.92x10 -9 , and the distance between the test signal and the center is 0.158, which is greater than the radius of the hypersphere. Therefore, it is considered that the test data is fault data.

[0110] The nonlinear cross bispectrum of the fault signal is subjected to differential feature enhancement by using the model hypersphere center. Figure 6 The differential feature is in the bispectrum domain. Figure 7 The nonlinear cross bispectrum of the fault data after feature enhancement.

[0111] Obviously, the noise in the nonlinear cross-bispectrum of the fault data after feature enhancement is basically eliminated, and the outer ring fault characteristic frequency appears obviously. This result is consistent with the actual situation, thus proving the effectiveness of the application for gear box condition monitoring.

[0112] By adopting the embodiment of the application, the following beneficial effects are achieved:

[0113] 1. Breakthrough of data dependence bottleneck: based on single-class normal working condition data, a diagnostic model can be constructed, completely eliminating the dependence on fault samples, and solving the engineering problem of difficult acquisition of fault data in a strong noise environment.

[0114] 2. Feature analysis depth leap: high-dimensional nonlinear features are extracted through nonlinear cross-bispectrum structure, and information such as faults, interference and noise in the signal is analyzed in a high-dimensional plane characterized by carrier frequency and natural frequency, meaning that hidden fault features can be stripped from a deeper layer.

[0115] 3. Decision-making interpretability breakthrough: the features of the health state space domain are mapped through visual hypersphere, and the deep information of the data is accurately enhanced through a quantitative mechanism.

[0116] 4. Excellent engineering generalization ability: the method performs excellently in complex working conditions such as high speed and heavy load, and can effectively cope with the strong coupling and high noise environment commonly seen in industrial equipment, providing strong technical support for gear box condition monitoring.

[0117] Embodiment two

[0118] According to the embodiment of the application, a gear box condition monitoring system based on single-class normal data is provided, Figure 8 FIG. 1 is a schematic diagram of a gear box condition monitoring system based on single-class normal data according to the embodiment of the application, according to Figure 8 the embodiment of the application, the gear box condition monitoring system based on single-class normal data comprises:

[0119] The data acquisition module 80 is configured to acquire normal working condition data of the gear box, process the data through nonlinear cross-bispectrum to construct a training data set, and collect gear box data in an unknown state and process the data through nonlinear cross-bispectrum.

[0120] The data acquisition module 80 is specifically configured to:

[0121] The vibration acceleration sensor installed at the key transmission part of the gear box and the industrial-grade data acquisition card are used to acquire normal working condition operation data and operation data in an unknown state of the gear box in a real-time manner under rated speed and load range.

[0122] The normal working condition operation data is processed through nonlinear cross-bispectrum to extract high-dimensional bispectrum features to construct a training data set;

[0123] The running data in the unknown state is processed by the nonlinear cross-spectrum interaction, high-dimensional bispectrum features are extracted to construct test data.

[0124] The model construction module 81 is configured to construct an interpretable differential diagnosis model based on support data description, and train the model through a training data set.

[0125] The model construction module 81 is specifically configured to:

[0126] The harmonic spectrum kurtosis HSK and the compound tail index CTI are defined as double-target evaluation indexes to quantitatively measure the significance and distribution characteristics of the fault features, respectively.

[0127] The training data set is input into the interpretable differential diagnosis model based on support vector data description.

[0128] The model is trained to generate a minimum hyper-sphere that satisfies:

[0129] (a) contains the maximum number of training data samples;

[0130] (b) establishes the boundary of the health state feature space;

[0131] The differential diagnosis model achieves interpretability through visualizing the hyper-sphere boundary decision process.

[0132] The anomaly discrimination module 82 is configured to determine whether the running data in the unknown state contains a fault.

[0133] The anomaly discrimination module 82 is specifically configured to:

[0134] The distance between the test data and the center of the hyper-sphere is used to determine whether the running data in the unknown state contains a fault.

[0135] After the time-domain signal is reconstructed through inverse transformation, the envelope demodulation technology is used to extract the envelope spectrum, and finally the characteristic frequency and type of the rolling bearing fault are identified to realize fault diagnosis.

[0136] The fault diagnosis module 83 is configured to design a weight mapping function to establish a two-dimensional weight matrix based on the difference features between the bispectrum of the fault data and the health model, strengthen the features of the fault data, and realize fault diagnosis.

[0137] The fault diagnosis module 83 is specifically configured to:

[0138] The difference features between the fault data and the health model are calculated in the bispectrum domain.

[0139] The weight mapping function is designed to convert the difference features into a two-dimensional weight matrix to quantitatively measure the difference features.

[0140] The bicoherence of the fault data is weighted by using a two-dimensional weight matrix to enhance potential features in the fault data, so as to realize fault diagnosis.

[0141] The above merely describes the preferred embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A gearbox condition monitoring method based on single-class normal data, characterized in that, The method comprises the following steps: obtaining normal working condition data of a gearbox, performing nonlinear cross-bispectrum processing on the normal working condition data to obtain a training data set; constructing an interpretable differential diagnosis model, and training and optimizing the interpretable differential diagnosis model through the training data set, wherein the interpretable differential diagnosis model adopts a decision boundary model; obtaining running data of the gearbox under an unknown state, performing nonlinear cross-bispectrum processing on the running data, and judging the processed running data according to the optimized interpretable differential diagnosis model to obtain a judgment result of whether the running data is fault data; when the judgment result is fault data, performing feature enhancement on the fault data according to the optimized interpretable differential diagnosis model to obtain an interpretable result of fault diagnosis; the process of obtaining the training data set comprises: collecting normal working condition data of a transmission part of the gearbox through a vibration acceleration sensor, wherein the normal working condition data is normal working condition running data of the gearbox within a rated speed and load range, and the running data is vibration data; performing nonlinear cross-bispectrum processing on the normal working condition data to extract high-dimensional bispectrum features to construct the training data set; the process of performing nonlinear cross-bispectrum processing on the normal working condition data comprises: ； wherein denotes the nonlinear cross-spectrum of the operational data, denotes the mathematical expectation, denotes the operational data the Fourier transform result of the operational data, the superscript denotes the complex conjugate symbol, denotes the carrier frequency of the operational data, denotes the modulation frequency of the operational data; the process of training and optimizing the interpretable differential diagnosis model through the training data set comprises: constructing an interpretable differential diagnosis model, wherein the interpretable differential diagnosis model performs fault judgment and fault explanation through a hyper-sphere boundary decision method; training a hyper-sphere through the training data set to obtain a center and a radius of the hyper-sphere, wherein the hyper-sphere is used to represent a boundary of a health state feature space.

2. The method of claim 1, wherein: the process of training the hyper-sphere comprises: solving an optimization objective function according to the training data set to obtain the center and the radius of the hyper-sphere; wherein the optimization objective function is: ； wherein, is a slack variable, is a penalty factor for balancing the hypersphere radius and the number of out-of-bound data, denotes the center, , denotes different training data, i denotes the label of the training data, and n denotes the total number of training data.

3. The method of claim 2, wherein: the optimization objective function is solved through a Lagrange function.

4. The method of claim 1, wherein: the process of judging the processed running data comprises: judging the processed running data according to the center and the radius of the hyper-sphere in the optimized interpretable differential diagnosis model, wherein a distance between the processed running data and the center of the hyper-sphere is calculated, and the distance is judged according to the radius of the hyper-sphere; when the distance is less than or equal to the radius of the hyper-sphere, the running data is normal data, otherwise the running data is fault data.

5. The method of claim 4, wherein: the process of performing feature enhancement on the fault data comprises: The nonlinear cross bispectrum of the fault data is obtained by performing nonlinear cross bispectrum processing on the fault data. The center of the nonlinear cross bispectrum of the fault data and the hypersphere is normalized, and a difference result is obtained by performing difference calculation on the normalized result. The amplitude and the probability density curve fitting function of the nonlinear cross bispectrum of the fault data are extracted. A two-dimensional weight matrix is obtained according to the amplitude and the probability density curve fitting function of the nonlinear cross bispectrum of the fault data. The nonlinear cross bispectrum of the fault data is weighted and calculated according to the two-dimensional weight matrix. The fault diagnosis explainable result is obtained according to the weighted calculation result.

6. The method of claim 5, wherein, The two-dimensional weight matrix is obtained by a weight mapping function, wherein the weight mapping function is: ； wherein, denotes taking the minimum value, denotes taking the maximum value, denotes normalization processing, is an amplitude of a nonlinear cross bispectrum, D denotes a weight mapping coefficient, denotes a weight, denotes a bispectrum value range coefficient, c denotes a weight mapping function center, denotes an amplitude of a nonlinear cross bispectrum, denotes a probability density curve fitting function.

7. A gearbox condition monitoring system based on single class normal data, characterized in that, A computer program product for performing the method of any one of the preceding claims 1-6.

Images