Bearing domain confrontation fault diagnosis method

Abstract

The invention provides a bearing domain confrontation fault diagnosis method, and belongs to the field of dynamics simulation and intelligent fault diagnosis detection, and the method comprises the steps: building a four-degree-of-freedom dynamics simulation rolling bearing system, and obtaining a bearing domain confrontation fault diagnosis result based on the four-degree-of-freedom dynamics simulation rolling bearing system; equivalent contact stiffness K is calculated, and elastic deformation generated when the rolling body moves in the raceway to reach the fault position is calculated; performing simulation based on the gravitational acceleration, the material parameters, the elastic deformation and the equivalent contact stiffness K to obtain a time domain graph of the simulation fault data, and performing fast Fourier transform on the time domain graph of the simulation fault data to obtain a frequency domain graph of the simulation fault data; inputting the frequency domain data of the frequency domain graph of the simulation fault data into a sparse stacked auto-encoder feature extractor to extract fault features; the extracted fault features are classified according to the cross entropy loss function, the method is used for intelligent fault diagnosis of the rotating machinery, and compared with an existing fault diagnosis method, the method has better accuracy and generalization ability.

Description

Technical Field

[0001] The invention relates to the fields of dynamics simulation and intelligent fault diagnosis and detection, and in particular to a bearing domain confrontation fault diagnosis method. Background Art

[0002] In the actual production process, failures of rotating parts such as bearings may cause production accidents. Rotating parts will always maintain high-intensity working conditions during equipment operation. Different pressure environments, different temperatures, different humidity, different lubrication conditions, and different meshing conditions will have an impact on the rotating parts. When one or more factors reach the limit value that the rotating parts can withstand, mechanical failure will occur, which will have a serious impact on the normal operation of the equipment.

[0003] By utilizing the advantages of machine learning and deep learning in processing unstructured data and being able to analyze a large amount of data more deeply to discover the characteristic relationship between data, data processing in unknown fields can be completed. Deep learning also has better advantages in unsupervised learning and high-dimensional data feature processing. In particular, deep learning technology is applied to intelligent fault diagnosis of machinery, which improves the efficiency and accuracy of traditional machinery diagnosis. In the field of rotating component fault diagnosis, deep learning technology is most widely used in fault extraction and fault classification. Although relying on massive data can support the model to complete tasks, repeated experiments are required between different tasks. Transfer learning can well solve the adaptive migration of data between different tasks. Through data migration between the same or different rotating parts, the existing model can be applied to the fault diagnosis tasks of rotating parts in more fields. This greatly improves the efficiency of model diagnosis and simplifies the process, which is of great significance for the safe and stable operation of equipment, improving generation efficiency and protecting people's production safety. When solving the problems of data scarcity and cross-domain adaptability in bearing fault diagnosis, existing methods often find it difficult to obtain a large amount of fault data, which limits the application and development of data-driven methods. Summary of the invention

[0004] To solve the above problems, the present invention provides a bearing domain adversarial fault diagnosis method, which is based on dynamic modeling and maximum mean difference-guided bearing domain adversarial fault diagnosis, and is used for rotating machinery intelligent fault diagnosis. Compared with the existing fault diagnosis methods, it has better accuracy and generalization ability. Specifically, it includes:

[0005] A bearing domain confrontation fault diagnosis method, comprising:

[0006] S1. Establish a four-degree-of-freedom dynamic simulation rolling bearing system, wherein the definition is: the nonlinear factors of the four-degree-of-freedom dynamic simulation rolling bearing system are composed of nonlinear contact forces between the inner and outer rings and the rolling body in the bearing, the plane of movement of the rolling body is the xoy plane, the inner ring is connected to the shaft and moves with the rotating shaft at an angular velocity ω, and the outer ring is fixed on the bearing seat and does not rotate;

[0007] S2. Calculating the equivalent contact stiffness K based on the four-degree-of-freedom dynamics simulation rolling bearing system;

[0008] S3, based on the four-degree-of-freedom dynamic simulation rolling bearing system, calculate the elastic deformation δ that occurs when the rolling element moves in the raceway to the fault position i ;

[0009] S4, based on gravitational acceleration, material parameters, elastic deformation δ i The equivalent contact stiffness K is simulated to obtain a time domain diagram of the simulated fault data, and the time domain diagram of the simulated fault data is fast Fourier transformed to obtain a frequency domain diagram of the simulated fault data;

[0010] S5, inputting the frequency domain data of the frequency domain graph of the simulated fault data into a sparse stacked autoencoder feature extractor to extract fault features;

[0011] S6. Classify the fault features extracted in S5 according to the cross entropy loss function.

[0012] Optionally, the conditions for establishing a four-degree-of-freedom dynamic simulation rolling bearing system in S1 include:

[0013] When the bearing is working, the contact between the rolling elements and the inner and outer rings of the bearing is Hertzian contact.

[0014] Optionally, the formula for calculating the equivalent contact stiffness K of the rolling bearing system based on the four-degree-of-freedom dynamics simulation of S2 is formula (1):

[0015]

[0016] Among them, K i K is the contact stiffness between the rolling element and the inner ring. o is the contact stiffness between the rolling element and the outer ring.

[0017] Optionally, the rolling bearing system based on the four-degree-of-freedom dynamics simulation calculates the elastic deformation δ occurring when the rolling element moves in the raceway to the fault position. i include:

[0018] The elastic deformation δ that occurs when the rolling element moves in the raceway and reaches the fault position i The formula is formula (2):

[0019] δ i =(x1-x2)sinθ i +(y1-y2)cosθ i -c r -H; (2)

[0020] θ i Indicates the position of the i-th rolling element at time t, in rad;

[0021] c r is the radial clearance, in m;

[0022] x1 is the displacement of the inner ring of the bearing in the x direction;

[0023] x2 is the displacement of the bearing outer ring in the x direction;

[0024] y1 is the displacement of the inner ring of the bearing in the y direction;

[0025] y2 is the displacement of the outer ring of the bearing in the y direction;

[0026] H is the amount of deformation when the failure occurs.

[0027] Optionally, the value of the deformation amount H when the fault occurs includes formula (3):

[0028] When the outer ring fails, the value of H is as follows:

[0029]

[0030] Among them, H0 is the change in deformation when the outer ring fails;

[0031] φ o is the constant outer ring fault position, in rad;

[0032] Ψ o is the central angle of the outer ring fault, in rad,

[0033] L is the fault width;

[0034] D o is the outer ring diameter;

[0035] in,

[0036]

[0037] r b is the rolling element radius;

[0038] h is the fault depth.

[0039] Optionally, the value of the deformation amount H when the fault occurs also includes formula (4):

[0040]

[0041] Among them, H i is the change in deformation when the outer ring fails;

[0042] φ i is the constant outer ring fault position, in rad;

[0043] Ψ i is the central angle of the inner circle fault, in rad,

[0044] D i is the inner ring diameter;

[0045] in,

[0046]

[0047] Optionally, performing a fast Fourier transform on the time domain graph of the simulated fault data to obtain a frequency domain graph of the simulated fault data includes:

[0048] The data types of the time domain graph of the simulated fault data include: a source domain and a target domain;

[0049] The source domain and the target domain are respectively subjected to fast Fourier transform to obtain the source domain data and the target domain data after fast Fourier transform, thereby forming a frequency domain diagram of the simulated fault data.

[0050] Optionally, the step of inputting the frequency domain data of the frequency domain graph of the simulated fault data into a sparse stacked autoencoder feature extractor to extract fault features in S5 includes:

[0051] The constraint condition based on the sparsity parameter of the sparse stacked autoencoder feature extractor is formula (5):

[0052]

[0053] Among them, D KL (p||q) represents the difference between probability p and probability q;

[0054] N is the sample size;

[0055] x i is the sample point in the sample space;

[0056] p(x i ) represents the target distribution;

[0057] q(xi ) represents the predicted distribution;

[0058] The formula for measuring the similarity between the source domain data and the target domain data after the fast Fourier transform is formula (6):

[0059]

[0060] Among them, MMD(F,D,T) is the maximum mean difference used to measure the similarity between two distributions D and T;

[0061] F is a set of functions;

[0062] D is the distribution of the source domain;

[0063] T is the distribution of the target domain;

[0064] f represents a continuous function on the sample space,

[0065] E is expectation;

[0066] x1 is a sample collected from the source domain distribution D;

[0067] x2 is a sample collected from the target domain distribution T.

[0068] Optionally, the classifying of the fault features extracted in S5 according to the cross entropy loss function in S6 includes:

[0069] The overall loss function of the domain adversarial model of the cross entropy loss function is formula (7):

[0070]

[0071] Among them, n s is the number of source domain samples, n t is the number of samples in the target domain; θ f is the parameter of the feature extractor, θ y is the parameter of the label classifier, θ d is the parameter of the discriminator; λ is a trade-off parameter used to balance the weight of the source domain classification loss and the domain discrimination loss; is the domain discrimination loss function.

[0072] Optionally, the total classification loss function is formula (8):

[0073]

[0074] G f (x) is a feature extractor that accepts samples and θ f Parameters, output extracted features;

[0075] G y (x) is the label predictor, which accepts the features extracted by the feature extractor and the parameters θ y , output the predicted label;

[0076] is the i-th sample in the source domain;

[0077] θ f are the parameters of the feature extractor;

[0078] is the true label of the i-th sample in the source domain;

[0079] θ y are the parameters of the label predictor;

[0080] is the classification loss function;

[0081] The domain discrimination loss function is formula (9):

[0082]

[0083] d i is the domain label of the i-th sample in the source domain;

[0084] θ d is the parameter of the domain discriminator, the domain discriminator G d Used to determine whether the sample comes from the source domain or the target domain;

[0085] is the i-th sample in the target domain;

[0086] G d (x) is the domain discriminator, which accepts the features extracted by the feature extractor and the parameters θ d , the probability that the output sample belongs to a certain domain;

[0087] It is the domain discrimination loss function, which is used to measure the difference between the discrimination result of the domain discriminator and the true domain label.

[0088] Compared with the prior art, the above technical solution has at least the following beneficial effects:

[0089] (1) Enhanced fault data generation

[0090] By constructing a dynamic simulation model of bearing faults, the present invention can generate diverse and comprehensive fault simulation data, which greatly enriches the data set used to train the diagnosis model and solves the limitation of the availability of real-world data.

[0091] (2) Improving cross-domain generalization

[0092] The combination of the maximum mean difference transfer strategy and the domain adversarial mechanism promotes effective cross-domain transfer between simulated data and real-world data, ensuring that the diagnostic model trained on simulated data maintains high performance when applied to real-world data, reducing domain gaps and improving generalization capabilities.

[0093] (3) Higher fault diagnosis accuracy

[0094] By leveraging advanced domain adaptation techniques, the proposed method significantly improves fault diagnosis accuracy compared to traditional methods. The combination of simulation-based training data and guided domain adversarial learning reduces the noise and inconsistency that is typically present in domain transfer.

[0095] (4) Better adaptability to various conditions

[0096] The use of dynamic modeling enables the system to adapt to different operating and fault conditions, making it smooth and reliable in applications in different scenarios. This dynamic adaptability ensures accurate diagnosis of a range of bearing systems.

[0097] (5) Reduce reliance on large amounts of real-world data

[0098] Generating fault simulation data reduces the reliance on collecting large real-world fault datasets, which can be time-consuming, expensive, or impractical. This makes the proposed method more efficient and more suitable for industrial applications.

[0099] (6) Scalability and application potential

[0100] The cross-domain generalization ability of the proposed method enhances its scalability, making it suitable for wider applications in industries that require powerful fault diagnosis systems, such as manufacturing, transportation, and power generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0101] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0102] Figure 1 It is a schematic diagram of the process of the present invention;

[0103] Figure 2 The schematic diagram of the established four-degree-of-freedom dynamic simulation rolling bearing system is shown in FIG.

[0104] Figure 3 This is a schematic diagram of the outer ring failure of a rolling bearing;

[0105] Figure 4 To construct a dynamic simulation model of bearing faults, a time domain plot of the simulated fault data was generated;

[0106] Figure 5 Frequency domain plot of simulated fault data generated to construct a dynamic simulation model of bearing faults;

[0107] Figure 6 This is the analysis result of the fault by the method described in the present invention. DETAILED DESCRIPTION

[0108] In order to make the purpose, technical solution and advantages of the embodiment of the present invention clearer, the technical solution of the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings of the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all of the embodiments. Based on the described embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0109] Unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. Similarly, words such as "one", "one" or "the" do not indicate a quantitative limitation, but indicate the existence of at least one. Words such as "include" or "comprise" mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Words such as "connect" or "connected" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect.

[0110] It should be noted that the terms "up", "down", "left", "right", "front" and "back" used in the present invention are only used to indicate relative position relationships. When the absolute position of the object being described changes, the relative position relationship may also change accordingly.

[0111] The dynamic simulation method can be used to construct a dynamic simulation model of bearing failure and generate a large amount of simulation data, which effectively alleviates the problem of data scarcity. It is not restricted by actual working conditions and has lower experimental costs.

[0112] The transfer learning fault diagnosis method has excellent feature learning and transfer capabilities, which can improve the knowledge transfer capabilities of data or models in actual scenarios. Existing similar technologies mainly include multi-layer multi-core fault diagnosis methods based on maximum mean difference, transmission network diagnosis methods based on class weighted alignment, and Fourier transform-based generative adversarial network diagnosis methods. In order to make the fault features of the source domain and the target domain more similar in the Hilbert space, the method based on maximum mean difference multi-layer multi-core uses the features from different domains to be close to each other in the reproducing kernel Hilbert space. Based on the class weighted alignment transmission network, the transferable knowledge learned in the source domain is applied to the target domain through partial domain adaptation. Based on the Fourier transform-based generative adversarial network diagnosis method, it is applied to the unbalanced bearing fault diagnosis. The multi-domain distribution of this method is controlled by learning a general mapping from the time domain to the frequency domain through a generative adversarial network during the training process. With the development of technology, the method of fault diagnosis using neural networks has achieved good research results in the diagnosis of mechanical fault features. The above methods are mainly applied to experimental bearings. If the data is scarce or difficult to obtain, such methods will face difficulties. Combining dynamic simulation equations, sparse stacked autoencoder networks, maximum mean difference migration strategy and domain adversarial mechanism during domain adaptation can effectively solve such problems.

[0113] The present invention provides a bearing domain resistance fault diagnosis method, comprising:

[0114] S1. Establish a four-degree-of-freedom dynamic simulation rolling bearing system, wherein the definition is: the nonlinear factors of the four-degree-of-freedom dynamic simulation rolling bearing system are composed of nonlinear contact forces between the inner and outer rings and the rolling body in the bearing, the plane of movement of the rolling body is the xoy plane, the inner ring is connected to the shaft and moves with the rotating shaft at an angular velocity ω, and the outer ring is fixed on the bearing seat and does not rotate;

[0115] S2. Calculating the equivalent contact stiffness K based on the four-degree-of-freedom dynamics simulation rolling bearing system;

[0116] S3, based on the four-degree-of-freedom dynamic simulation rolling bearing system, calculate the elastic deformation δ that occurs when the rolling element moves in the raceway to the fault position i ;

[0117] S4, based on gravitational acceleration, material parameters, elastic deformation δ i The equivalent contact stiffness K is simulated to obtain a time domain diagram of the simulated fault data, and the time domain diagram of the simulated fault data is fast Fourier transformed to obtain a frequency domain diagram of the simulated fault data;

[0118] The material parameters include: inner ring-shaft mass, outer ring-bearing seat mass, bearing seat-base connection damping, shaft connection stiffness, bearing seat-base connection stiffness and the number of rolling elements;

[0119] S5, inputting the frequency domain data of the frequency domain graph of the simulated fault data into a sparse stacked autoencoder feature extractor to extract fault features;

[0120] S6. Classify the fault features extracted in S5 according to the cross entropy loss function.

[0121] In a specific implementation manner, S1, establishing a four-degree-of-freedom dynamic simulation rolling bearing system includes:

[0122] The nonlinear factors in the system only consider the nonlinear contact force between the inner and outer rings and the rolling elements in the bearing, the periodic stiffness changes caused by the rolling elements during movement, and the radial clearance of the rolling elements in the channel. The movement only runs in one plane, that is, the x plane. The inner ring is connected to the shaft and moves with the rotating shaft at an angular velocity, while the outer ring is fixed on the bearing seat and does not rotate. When the bearing is working, it is assumed that the contact between the rolling elements and the inner and outer rings of the bearing is Hertzian contact. It is assumed that the mass and rotational inertia of the rolling elements and the cage will not affect the operation of the system, so their related parameters are not introduced for the time being. At the same time, the rolling elements are simplified to nonlinear springs, and the elastic force generated by them is distributed in a z-uniform state relative to the axis.

[0123] A specific implementation manner, S2, calculating the equivalent contact stiffness K based on the four-degree-of-freedom dynamics simulation rolling bearing system;

[0124] The formula for calculating the equivalent contact stiffness K is formula (1):

[0125]

[0126] Among them, K i K is the contact stiffness between the rolling element and the inner ring. o is the contact stiffness between the rolling element and the outer ring.

[0127] In this step, the equivalent contact stiffness K is calculated to obtain an accurate value. Assume that the rolling element in the bearing is the contact element A, and the outer ring or inner ring is the contact element B. The main contact radius of the rolling element and the inner and outer raceways can be described as:

[0128]

[0129] Among them, D i ,D o and r i ,r o Respectively represent the diameter (m) and radius (m) of the inner and outer ring raceways; k i ,k o Represent the curvatures of the inner and outer circles, respectively, and are set to 0.515 and 0.525; K d Represents the height coefficient of the ring edge, and its value is 0.35.

[0130] The contact stiffness between the rolling element and the inner and outer rings can be defined as:

[0131]

[0132] Among them, E i ,E o is the equivalent elastic modulus, and the elastic modulus and Poisson's ratio of the two contact bodies are E A ,E B ,μ A ,μ B , the equivalent modulus E is:

[0133]

[0134] Based on the above formula, the formula for calculating the equivalent contact stiffness K is derived as formula (1):

[0135] A specific implementation method, S3, based on the four-degree-of-freedom dynamic simulation rolling bearing system, calculates the elastic deformation δ occurring when the rolling element moves in the raceway to the fault position i ; Specifically include the following contents:

[0136] Based on the simulation rolling bearing system described above, local fault conditions are simulated, which are divided into outer ring fault, inner ring fault and normal condition. When the rolling element moves in the raceway and reaches the fault position, it is divided into the following conditions:

[0137] Case 1: When the outer ring fails, the elastic deformation δ occurs when the rolling element moves in the raceway to the failure position i The formula is formula (2):

[0138] δ i =(x1-x2)sinθ i +(y1-y2)cosθ i -c r -H; (2)

[0139] θ i Indicates the position of the i-th rolling element at time t, in rad;

[0140] c r is the radial clearance, in m;

[0141] x1 is the displacement of the inner ring of the bearing in the x direction;

[0142] x2 is the displacement of the bearing outer ring in the x direction;

[0143] y1 is the displacement of the inner ring of the bearing in the y direction;

[0144] y2 is the displacement of the outer ring of the bearing in the y direction;

[0145] H is the amount of deformation when the failure occurs.

[0146] The value of the deformation amount H when the fault occurs includes formula (3):

[0147] When the outer ring fails, the value of H is as follows:

[0148]

[0149] Among them, H0 is the change in deformation when the outer ring fails;

[0150] φ o is the constant outer ring fault position, in rad;

[0151] Ψ o is the central angle of the outer ring fault, in rad,

[0152] L is the fault width;

[0153] D o is the outer ring diameter;

[0154] in,

[0155]

[0156] r b is the rolling element radius;

[0157] h is the fault depth.

[0158] Case 2: When the inner ring fails, the value of the deformation amount H when the failure occurs also includes formula (4):

[0159]

[0160] Among them, H i is the change in deformation when the outer ring fails;

[0161] φ i is the constant outer ring fault position, in rad;

[0162] Ψ i is the central angle of the inner circle fault, in rad,

[0163] D i is the inner ring diameter;

[0164] in,

[0165]

[0166] A specific implementation method, S4, based on gravity acceleration, material parameters, elastic deformation δ i The equivalent contact stiffness K is simulated to obtain the time domain diagram of the simulated fault data, and the time domain diagram of the simulated fault data is fast Fourier transformed to obtain the frequency domain diagram of the simulated fault data, which specifically includes the following contents:

[0167] The data types of the time domain graph of the simulated fault data include: a source domain and a target domain;

[0168] Perform fast Fourier transform on the source domain and the target domain respectively, obtain the source domain data and the target domain data after fast Fourier transform, and form a frequency domain diagram of the simulated fault data;

[0169] The material parameters include: inner ring-shaft mass, outer ring-bearing seat mass, bearing seat-base connection damping, shaft connection stiffness, bearing seat-base connection stiffness and the number of rolling elements. The specific parameters are shown in Table 1:

[0170] Table-1

[0171] Simulation parameters Numeric Simulation parameters Numeric Inner ring-shaft mass 50kg Fault depth h 1mm Outer ring-bearing seat quality 5kg Fault width L 0.18 / 0.36mm <![CDATA[Axis connection damping c i > 1376.8Ns / m <![CDATA[Radial clearance c r > <![CDATA[2×10 -6 m]]> <![CDATA[Bearing housing - Base connection damping c o > 2210.7Ns / m Eccentricity <![CDATA[50×10 -6 m]]> Equivalent contact stiffness K <![CDATA[8.753×10 9 N / m]]> <![CDATA[Rolling element radius r b > 3.373mm <![CDATA[Axial connection stiffness k i > <![CDATA[7.42×10 7 N / m]]> <![CDATA[Rolling element diameter D b > 6.746mm <![CDATA[Bearing housing - Base connection stiffness k o > <![CDATA[1.51×10 7 N / m]]> <![CDATA[Pitch diameter D of bearing p > 28.500mm Gravitational acceleration g <![CDATA[9.8m / s 2 ]]> <![CDATA[Inner raceway diameter D i > 17mm Number of rolling elements 8 <![CDATA[Outer raceway diameter D o > 39.999mm

[0172] In a specific implementation manner, S5, the frequency domain data of the frequency domain diagram of the simulated fault data is input into a sparse stacked autoencoder feature extractor to extract fault features, specifically comprising:

[0173] The constraint condition based on the sparsity parameter of the sparse stacked autoencoder feature extractor is formula (5):

[0174]

[0175] Among them, D KL (p||q) represents the difference between probability p and probability q;

[0176] N is the sample size;

[0177] x i is the sample point in the sample space;

[0178] p(x i ) represents the target distribution;

[0179] q(x i ) represents the predicted distribution;

[0180] The formula for measuring the similarity between the source domain data and the target domain data after the fast Fourier transform is formula (6):

[0181]

[0182] Among them, MMD(F,D,T) is the maximum mean difference used to measure the similarity between two distributions D and T;

[0183] F is a set of functions;

[0184] D is the distribution of the source domain;

[0185] T is the distribution of the target domain;

[0186] f represents a continuous function on the sample space,

[0187] E is expectation;

[0188] x1 is a sample collected from the source domain distribution D;

[0189] x2 is a sample collected from the target domain distribution T.

[0190] In a specific implementation manner, the S6 classifies the fault features extracted in S5 according to the cross entropy loss function, including:

[0191] The overall loss function of the domain adversarial model of the cross entropy loss function is formula (7):

[0192]

[0193] Among them, n s is the number of source domain samples, n t is the number of samples in the target domain; θ f is the parameter of the feature extractor, θ y is the parameter of the label classifier, θ d is the parameter of the discriminator; λ is a trade-off parameter used to balance the weight of the source domain classification loss and the domain discrimination loss; is the domain discrimination loss function.

[0194] The total classification loss function is formula (8):

[0195]

[0196] G f (x) is a feature extractor that accepts samples and θ f Parameters, output extracted features;

[0197] G y (x) is the label predictor, which accepts the features extracted by the feature extractor and the parameters θ y , output the predicted label;

[0198] is the i-th sample in the source domain;

[0199] θ f are the parameters of the feature extractor;

[0200] is the true label of the i-th sample in the source domain;

[0201] θ y are the parameters of the label predictor;

[0202] is the classification loss function;

[0203] The domain discrimination loss function is formula (9):

[0204]

[0205] d i is the domain label of the i-th sample in the source domain;

[0206] θ d is the parameter of the domain discriminator, the domain discriminator G d Used to determine whether the sample comes from the source domain or the target domain;

[0207] is the i-th sample in the target domain;

[0208] G d (x) is the domain discriminator, which accepts the features extracted by the feature extractor and the parameters θ d , the probability that the output sample belongs to a certain domain;

[0209] It is the domain discrimination loss function, which is used to measure the difference between the discrimination result of the domain discriminator and the true domain label.

[0210] The specific implementation methods are as follows:

[0211] Establish as Figure 2The four-degree-of-freedom dynamic simulation rolling bearing system shown in the figure only considers the nonlinear contact force between the inner and outer rings and the rolling elements in the bearing, the periodic stiffness change caused by the rolling elements during movement, and the radial clearance of the rolling elements in the channel. The movement only runs in one plane, namely the xo plane. The inner ring is connected to the shaft and moves with the rotating shaft at an angular velocity, while the outer ring is fixed on the bearing seat and does not rotate. When the bearing is working, it is assumed that the contact between the rolling elements and the inner and outer rings of the bearing is Hertzian contact. It is assumed that the mass and moment of inertia of the rolling elements and the cage will not affect the operation of the system, so their related parameters are not introduced for the time being. At the same time, the rolling elements are simplified into nonlinear springs, and the elastic force generated by them is distributed in a z-uniform state relative to the axis. The equivalent contact stiffness is calculated to obtain an accurate value. Assume that the rolling element in the bearing is contact body A, and the outer ring or inner ring is contact body B. Obtain the main contact radius between the rolling element and the inner and outer raceways and calculate the equivalent contact stiffness between the rolling element and the inner and outer rings. Based on the simulation rolling bearing system described above, the following simulation is performed: Figure 3 The local fault conditions shown are divided into outer ring fault, inner ring fault and normal condition. When the rolling element moves in the raceway to the fault position, the elastic deformation δ of the inner and outer rings when the fault occurs is calculated respectively. i , based on the specific parameters of the bearing (including the elastic deformation of the inner and outer rings when failure occurs i ), the bearing material parameters are set to: E A =E B =207GPa,μ A =μ B =0.3. The sampling frequency is set to 12kHz, the fault type is set to inner and outer race faults, and the fault depths are set to 0.18mm and 0.36mm respectively. The five faults are normal state NC, inner race faults IF18 and IF36, and outer race faults OF18 and OF36. Figure 4 The simulated bearing fault time domain data is shown in Figure 1. Set four different speed conditions, namely 1200rpm, 1500rpm, 1800rpm and 2100rpm. Take 200 segments of data for each fault, each segment of data includes 2048 points, and transform them into the following through fast Fourier transform: Figure 5The frequency domain data shown. The frequency domain data is input into the sparse autoencoder feature extractor to extract features. By stacking multiple encoders and decoders, a complex and deep network structure is formed to extract features. Finally, the maximum mean difference metric is added as a regular term to the training objective of the sparse autoencoder network. Through the joint loss function, the feature extractor is forced to learn feature representations that can both reconstruct the input data and reduce inter-domain differences. In domain adversarial training, the gradient reversal layer realizes gradient reversal during back propagation, which maximizes the iteration of the domain classifier loss function, and performs fault analysis and diagnosis based on the final information of the iteration of the overall loss function of the domain adversarial model. The final information obtained is compared with fault diagnosis methods such as multi-layer perceptron networks, convolutional neural networks, and residual convolutional neural networks, as shown in the figure. Figure 6 It is shown that the present invention has better accuracy and generalization ability.

[0212] By constructing a dynamic simulation model of bearing faults, the present invention can generate diversified and comprehensive fault simulation data, which greatly enriches the data set used to train the diagnosis model and solves the limitation of the availability of real-world data.

[0213] The present invention promotes effective cross-domain migration between simulated data and real-world data by combining the maximum mean difference migration strategy with the domain adversarial mechanism, ensuring that the diagnostic model trained on simulated data maintains high performance when applied to real-world data, reducing domain gaps and improving generalization capabilities.

[0214] By utilizing advanced domain adaptation techniques, the proposed method significantly improves the fault diagnosis accuracy compared to traditional methods. The combination of simulation-based training data and guided domain adversarial learning reduces the noise and inconsistency that is usually present in domain transfer.

[0215] The present invention uses dynamic modeling to enable the system to adapt to different operating and fault conditions, making it stable and reliable in applications in different scenarios. This dynamic adaptability ensures accurate diagnosis of a range of bearing systems.

[0216] The invention generates fault simulation data, which reduces the reliance on collecting large amounts of real-world fault data sets, which may be time-consuming, expensive or impractical. This makes the proposed method more efficient and more suitable for industrial applications.

[0217] The cross-domain generalization capability of our approach enhances its scalability, making it suitable for wider applications in industries that require powerful fault diagnosis systems, such as manufacturing, transportation, and power generation.

[0218] The following points need to be explained:

[0219] (1) The drawings of the embodiments of the present invention only relate to the structures related to the embodiments of the present invention, and other structures may refer to the general design.

[0220] (2) For the sake of clarity, in the drawings used to describe the embodiments of the present invention, the thickness of the layers or regions is exaggerated or reduced, that is, these drawings are not drawn according to the actual scale. It is understood that when an element such as a layer, film, region or substrate is referred to as being "on" or "under" another element, the element may be "directly" "on" or "under" the other element or there may be intermediate elements.

[0221] (3) In the absence of conflict, the embodiments of the present invention and the features therein may be combined with each other to obtain new embodiments.

[0222] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. The protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A bearing domain confrontation fault diagnosis method, characterized in that: include: S1. Establish a four-degree-of-freedom dynamic simulation rolling bearing system, wherein the definition is: the nonlinear factors of the four-degree-of-freedom dynamic simulation rolling bearing system are composed of nonlinear contact forces between the inner and outer rings and the rolling body in the bearing, the plane of movement of the rolling body is the xoy plane, the inner ring is connected to the shaft and moves with the rotating shaft at an angular velocity ω, and the outer ring is fixed on the bearing seat and does not rotate; S2. Calculating the equivalent contact stiffness K based on the four-degree-of-freedom dynamics simulation rolling bearing system; S3, based on the four-degree-of-freedom dynamic simulation rolling bearing system, calculate the elastic deformation δ that occurs when the rolling element moves in the raceway to the fault position i ; S4, based on gravitational acceleration, material parameters, elastic deformation δ i The equivalent contact stiffness K is simulated to obtain a time domain diagram of the simulated fault data, and the time domain diagram of the simulated fault data is fast Fourier transformed to obtain a frequency domain diagram of the simulated fault data; The material parameters include: inner ring-shaft mass, outer ring-bearing seat mass, bearing seat-base connection damping, shaft connection stiffness, bearing seat-base connection stiffness and the number of rolling elements; S5, inputting the frequency domain data of the frequency domain graph of the simulated fault data into a sparse stacked autoencoder feature extractor to extract fault features; S6. Classify the fault features extracted in S5 according to the cross entropy loss function.

2. The bearing domain confrontation fault diagnosis method according to claim 1 is characterized in that: The conditions for establishing the four-degree-of-freedom dynamic simulation of the rolling bearing system in S1 include: When the bearing is working, the contact between the rolling elements and the inner and outer rings of the bearing is Hertzian contact.

3. The bearing domain confrontation fault diagnosis method according to claim 2 is characterized in that: The formula for calculating the equivalent contact stiffness K of the rolling bearing system of S2 based on the four-degree-of-freedom dynamic simulation is formula (1): Among them, K i K is the contact stiffness between the rolling element and the inner ring. o is the contact stiffness between the rolling element and the outer ring.

4. The bearing domain resistance fault diagnosis method according to claim 3 is characterized in that: The rolling bearing system is simulated based on the four-degree-of-freedom dynamics, and the elastic deformation δ occurring when the rolling element moves in the raceway to the fault position is calculated. i include: The elastic deformation δ that occurs when the rolling element moves in the raceway and reaches the fault position i The formula is formula (2): δ i =(x1-x2)sinθ i +(y1-y2)cosθ i -c r -H; (2) θ i Indicates the position of the i-th rolling element at time t, in rad; c r is the radial clearance, in m; x1 is the displacement of the inner ring of the bearing in the x direction; x2 is the displacement of the bearing outer ring in the x direction; y1 is the displacement of the inner ring of the bearing in the y direction; y2 is the displacement of the bearing outer ring in the y direction; H is the amount of deformation when the failure occurs.

5. The bearing domain confrontation fault diagnosis method according to claim 4 is characterized in that: The value of the deformation amount H when the fault occurs includes formula (3): When the outer ring fails, the value of H is as follows: Among them, H0 is the change in deformation when the outer ring fails; φ o is the constant outer ring fault position, in rad; Ψ o is the central angle of the outer ring fault, in rad, L is the fault width; D o is the outer ring diameter; in, r b is the rolling element radius; h is the fault depth.

6. The bearing domain confrontation fault diagnosis method according to claim 5 is characterized in that: The value of the deformation amount H when the fault occurs also includes formula (4): Among them, H i is the change in deformation when the outer ring fails; φ i is the constant outer ring fault position, in rad; Ψ i is the central angle of the inner circle fault, in rad. D i is the inner ring diameter; in, 7. The bearing domain resistance fault diagnosis method according to claim 6 is characterized in that: The performing of a fast Fourier transform on the time domain graph of the simulated fault data to obtain a frequency domain graph of the simulated fault data comprises: The data types of the time domain graph of the simulated fault data include: a source domain and a target domain; The source domain and the target domain are respectively subjected to fast Fourier transform to obtain the source domain data and the target domain data after fast Fourier transform, thereby forming a frequency domain diagram of the simulated fault data.

8. The bearing domain resistance fault diagnosis method according to claim 7 is characterized in that: The step of inputting the frequency domain data of the frequency domain graph of the simulated fault data into the sparse stacked autoencoder feature extractor to extract fault features in S5 includes: The constraint condition based on the sparsity parameter of the sparse stacked autoencoder feature extractor is formula (5): Among them, D KL (p||q) represents the difference between probability p and probability q; N is the sample size; x i is the sample point in the sample space; p(x i ) represents the target distribution; q(x i ) represents the predicted distribution; The formula for measuring the similarity between the source domain data and the target domain data after the fast Fourier transform is formula (6): Among them, MMD(F,D,T) is the maximum mean difference used to measure the similarity between two distributions D and T; F is a set of functions; D is the distribution of the source domain; T is the distribution of the target domain; f represents a continuous function on the sample space, E is expectation; x1 is a sample collected from the source domain distribution D; x2 is a sample collected from the target domain distribution T.

9. The bearing domain resistance fault diagnosis method according to claim 8 is characterized in that: The S6 classifies the fault features extracted in S5 according to the cross entropy loss function, including: The overall loss function of the domain adversarial model of the cross entropy loss function is formula (7): Among them, n s is the number of source domain samples, n t is the number of samples in the target domain; θ f is the parameter of the feature extractor, θ y is the parameter of the label classifier, θ d is the parameter of the discriminator; λ is a trade-off parameter used to balance the weight of the source domain classification loss and the domain discrimination loss; is the domain discrimination loss function.

10. The bearing domain resistance fault diagnosis method according to claim 9, characterized in that: The total classification loss function is formula (8): G f (x) is a feature extractor that accepts samples and θ f Parameters, output extracted features; G y (x) is the label predictor, which accepts the features extracted by the feature extractor and the parameters θ y , output the predicted label; is the i-th sample in the source domain; θ f are the parameters of the feature extractor; is the true label of the i-th sample in the source domain; θ y are the parameters of the label predictor; is the classification loss function; The domain discrimination loss function is formula (9): d i is the domain label of the i-th sample in the source domain; θ d is the parameter of the domain discriminator, the domain discriminator G d Used to determine whether the sample comes from the source domain or the target domain; is the i-th sample in the target domain; G d (x) is the domain discriminator, which accepts the features extracted by the feature extractor and the parameters θ d , the probability that the output sample belongs to a certain domain; It is the domain discrimination loss function, which is used to measure the difference between the discrimination result of the domain discriminator and the true domain label.

Images