Fault diagnosis method and system for rotor-bearing system based on virtual-real fusion

Abstract

The invention provides a rotor-bearing system fault diagnosis method and system based on virtual-real fusion, and relates to the technical field of signal detection, and the method comprises the steps: deducing a mixed eccentric unbalanced magnetic pull expression, and building a rotor-bearing system twinborn model in combination with a Hertz contact theory; introducing rotor eccentricity and bearing inner / outer ring faults, and constructing a multi-fault working condition twinborn model; performing multi-parameter identification on the model through an improved genetic algorithm (introducing a feature sensitivity evaluation factor, a self-adaptive crossover mutation probability and simulated annealing) to obtain a corrected twin model; generating a twin fault sample based on the correction model, and training by using a one-dimensional cyclic generative adversarial network (designing a comprehensive loss function) with a classifier to generate a sample close to real distribution; and virtual and real fusion samples form a balanced data set, and fault classification is realized through one-dimensional convolutional neural network training. According to the method, the problems of fault data shortage, large difference between a twin model and real data and the like are solved, and the fault diagnosis precision and reliability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of signal detection, and particularly to a fault diagnosis method and system for a rotor-bearing system based on virtual-real fusion. Background Art

[0002] In the modern industrial system, rotating machinery, as a key basic component for the development of the national economy, is widely used in many fields such as energy, manufacturing, and transportation. The stability of the internal rotor-bearing system is directly related to the safe and reliable operation of the whole machine. Once a fault occurs in the rotor-bearing system, it will not only cause the equipment to stop running, resulting in production interruption and economic losses, but may also trigger safety accidents and threaten the lives of personnel.

[0003] However, in the actual operation process, due to safety considerations, most rotating machinery is in a healthy operation state for a long time, making it extremely difficult to obtain rich fault data. The lack of fault data severely restricts the fault analysis work of the rotor-bearing system. Traditional fault diagnosis methods often rely on a large number of fault samples for model training and feature analysis. The lack of data makes it difficult for these methods to accurately identify and diagnose faults, and they cannot meet the requirements of early warning and accurate diagnosis of equipment faults in actual production.

[0004] To solve the problem of lack of fault data, a digital twin model of the rotor-bearing system based on the analytical method has emerged. By introducing corresponding fault excitations for actual fault types, this model can obtain twin data in different states, providing a new data source for fault diagnosis. However, affected by factors such as simplified numerical models and different environmental noises, there are certain differences between the obtained twin data and real data, and they cannot be directly fused with the virtual data. This makes the fault diagnosis results based on this model deviate, and it is difficult to guarantee the diagnosis accuracy and reliability.

[0005] Although the genetic algorithm can identify multiple parameters of the twin model and improve the quality of the twin model to a certain extent, it still needs to be further optimized to better adapt to complex and changeable actual working conditions. The generative adversarial network can make the generated fault data closer to the real distribution based on the twin data. However, its application in the fault diagnosis of the rotor-bearing system is not yet perfect, and there are problems such as low quality of the generated data and incomplete retention of category information.

[0006] In summary, the current fault diagnosis of the rotor-bearing system faces challenges such as difficult acquisition of fault data, large differences in existing model data, and imperfect application of related algorithms and networks. Therefore, optimizing the genetic algorithm and the generative adversarial network, obtaining high-quality fault data, and improving the fault diagnosis ability of the rotor-bearing system have become urgent problems to be solved. Summary of the Invention

[0007] To this end, the present invention provides a fault diagnosis method and system for a rotor-bearing system based on virtual-real fusion, aiming to solve the problems in the prior art such as unbalanced real sample data during the intelligent diagnosis of the rotor-bearing system, difficulty in obtaining fault data, and large differences between the data of the twin model and the real data and insufficient fault diagnosis ability due to various influencing factors.

[0008] To solve the above problems, an embodiment of the present invention provides a fault diagnosis method for a rotor-bearing system based on virtual-real fusion, and the method includes: S1: Derive a unified expression for the combined eccentric unbalanced magnetic pull force, introduce the bearing nonlinear restoring force based on the Hertz contact theory, and establish a twin model of the rotor-bearing system; S2: Introduce rotor eccentricity faults, inner ring faults of the bearing, and outer ring faults of the bearing into the twin model of the rotor-bearing system to obtain the twin model of the rotor-bearing system under different fault conditions; S3: Establish a fitness function by introducing a feature sensitivity evaluation factor into the genetic algorithm, and combine the adaptive crossover and mutation probabilities and the simulated annealing idea to perform multi-parameter identification on the twin model of the rotor-bearing system under different fault conditions to obtain a corrected twin model; S4: Obtain twin fault samples based on the corrected twin model, add a classifier to the one-dimensional cyclic generative adversarial network and design a comprehensive loss function to train and generate generated samples close to the real sample distribution; S5: Fuse the generated samples and the real samples in a virtual-real manner to obtain a balanced data set, and use a one-dimensional convolutional neural network for training to achieve fault classification.

[0009] Preferably, the process of establishing the twin model of the rotor-bearing system includes: First, based on the electrical machinery theory, derive the air-gap length, magnetic conductance, and magnetomotive force distribution under combined eccentricity, and obtain the unified expression for the unbalanced magnetic pull force in the , direction through the Maxwell stress integral; Then, based on the Hertz contact theory, derive the position and contact deformation of the bearing rolling elements, and accumulate to obtain the bearing nonlinear contact force in the , direction; Finally, combine the unbalanced magnetic pull force and the bearing nonlinear contact force to establish the dynamic equation of the rotor-bearing system.

[0010] Preferably, the unified expression for the unbalanced magnetic pull force in the , direction is: ; ; wherein, is the unbalanced magnetic pull force in the is the unbalanced magnetic pull force in the the rotor radius; the axial length; the amplitude of the air-gap magnetomotive force; the permeability of free space; the constant component of the uniform air-gap permeance without eccentricity; the permeance component caused by static eccentricity; the permeance component caused by dynamic eccentricity; the rotor mechanical angular frequency; the initial phase; time; the dynamic eccentricity angle; the number of pole pairs.

[0011] Preferably, the , direction bearing non-linear contact forces are: ; ; wherein, is the direction bearing non-linear contact force; is the direction bearing non-linear contact force; the number of rolling elements; the equivalent contact stiffness obtained from the local deformation of the contact area; is the total contact deformation of the th rolling element; is the th rolling element position.

[0012] Preferably, the expression of the rotor-bearing system twin model is: ; wherein, is the rotor mass; is the damping coefficient; is the mechanical stiffness of the rotor shaft; and are respectively the exciting forces in the , directions; and are respectively the bearing non-linear contact forces in the , directions; , , are the displacement, velocity, and acceleration of the rotor in the direction respectively; , , are the displacement, velocity, and acceleration of the rotor in the direction respectively.

[0013] Preferably, the method for introducing rotor eccentricity faults, inner race faults of the bearing, and outer race faults of the bearing is as follows: The rotor eccentricity fault is set by the relative static eccentricity and the relative dynamic eccentricity . When , it is a static eccentricity condition, is a dynamic eccentricity condition, and when both and are not zero, it is a mixed eccentricity condition; For the inner race fault of the bearing and the outer race fault of the bearing, rectangular defects are respectively set on the inner race and outer race surfaces, and the total contact deformation of the rolling element is characterized by the depth of the rolling element sinking into the defect. Among them, the inner race defect depth is related to the inner race radius, rolling element radius, and angular difference, and the outer race defect depth is related to the outer race radius and angular difference.

[0014] Preferably, the inner race defect depth is expressed as: ; In the formula, is the inner race defect depth, , are the inner race radius of the bearing and the rolling element radius respectively; is the initial angle when the rolling element enters the defect area; is the angle when the rolling element sinks into the maximum depth of the defect area, that is, the central angle; is the termination angle when leaving the defect area; is the angular difference between the rotation of the inner race of the bearing and the rolling element.

[0015] Preferably, the outer race defect depth is expressed as: ; In the formula, is the inner race defect depth, , are the inner race radius of the bearing and the rolling element radius respectively; is the initial angle when the rolling element enters the defect area; is the angle when the rolling element sinks into the maximum depth of the defect area, that is, the central angle; is the termination angle when leaving the defect area; is the angular difference between the rotation of the inner race of the bearing and the rolling element.

[0016] Preferably, the method for establishing the feature sensitivity evaluation factor is as follows: Taking the ratio of the between-class variance to the within-class variance as the feature sensitivity evaluation factor , where the between-class variance is calculated through the total mean value and the number of samples of the feature values of the real samples and the simulation samples, and the within-class variance is calculated through the variance of the feature values in each sample state. Among them, the feature sensitivity evaluation factor is expressed as: ; In the formula, and are the within-class variances in the sample states and the sample state respectively; is the between-class variance.

[0017] Preferably, the expressions for the adaptive crossover and mutation probabilities are: ; ; In the formula, , are the adaptive crossover probability and the adaptive mutation probability respectively; , are the maximum and minimum values of the predefined crossover probability respectively; , are the maximum and minimum values of the predefined mutation probability respectively; is the fitness value of the individual with a higher fitness value among the two individuals participating in the crossover operation; is the fitness value of the mutant individual; is the average fitness value in the current population; is the maximum fitness value in the current population.

[0018] Preferably, the implementation of the simulated annealing idea includes: Setting the outer loop to control the temperature decay, and the initial temperature gradually decreases to the termination temperature according to the temperature reduction coefficient α; the inner loop iterates multiple times at each temperature, and accepts new individuals through the Metropolis criterion, where the acceptance probability is: ; In the formula, is the acceptance probability, is the current temperature, and are the fitness values of the new individual and the original individual respectively.

[0019] Preferably, the one-dimensional cyclic generative adversarial network includes two generators and and two discriminators and and two classifiers and , the comprehensive loss function of the one-dimensional cyclic generative adversarial network includes: The adversarial loss function between the generator and the discriminator is expressed as: ; The adversarial loss function between the generator and the discriminator is expressed as: ; In the formula, and are one-dimensional signals in the sample spaces and respectively; denotes follows the expectation of the data distribution in the sample space ; denotes follows the expectation of the data distribution in the sample space ; The cycle consistency loss function is expressed as: ; In the formula, denotes the L1 norm; The cross-entropy loss function of the classifier is expressed as: ; The cross-entropy loss function of the classifier is expressed as: ; In the formula, is a one-dimensional signal in the sample space ; is a one-dimensional signal in the sample space ; is the total number of samples for parameter loss calculation in the sample space ; is the total number of samples for parameter loss calculation in the sample space ; The final loss function is expressed as: ​​ ; In the formula, , , are weight coefficients, which are respectively used to control the importance of the cycle consistency loss and the two classification losses.

[0020] Preferably, the network structure of the one-dimensional convolutional neural network is the same as the classifier structure of the one-dimensional recurrent generative adversarial network.

[0021] The embodiment of the present invention also provides a rotor-bearing system fault diagnosis system based on virtual-real fusion, which is used to implement the rotor-bearing system fault diagnosis method based on virtual-real fusion described above, and specifically includes: A model establishment module, which is used to deduce a unified expression of the hybrid eccentric unbalanced magnetic pull force, introduce the bearing nonlinear restoring force based on the Hertz contact theory, and establish a twin model of the rotor-bearing system; A fault introduction module, which is used to introduce rotor eccentricity faults, inner race faults and outer race faults of the bearing into the twin model of the rotor-bearing system to obtain a twin model of the rotor-bearing system under different fault conditions; A model correction module, which is used to establish a fitness function by introducing a feature sensitivity evaluation factor into the genetic algorithm, and combine the ideas of adaptive crossover, mutation probability and simulated annealing to perform multi-parameter identification on the twin model of the rotor-bearing system under different fault conditions to obtain a corrected twin model; A sample generation module, which is used to obtain twin fault samples based on the corrected twin model, add a classifier to the one-dimensional recurrent generative adversarial network and design a comprehensive loss function, and train to generate generated samples close to the real sample distribution; A fault classification module, which is used to fuse the generated samples and the real samples in a virtual-real manner to obtain a balanced data set, and use a one-dimensional convolutional neural network for training to achieve fault classification.

[0022] It can be seen from the above technical solutions that the present invention application has the following beneficial effects: The embodiment of the present invention provides a fault diagnosis method and system for a rotor-bearing system based on virtual-real fusion. The present invention combines a twin model, parameter identification, and a cyclic generative adversarial network to achieve the virtual-real fusion of real samples and simulation samples. By introducing a fault excitation into the twin model of the rotor-bearing system, simulation fault data under different working conditions are obtained; by introducing a feature sensitivity evaluation factor and adaptive crossover and mutation probabilities in parameter identification, the influence of factors such as a simplified numerical model and environmental noise on the twin model is reduced; by adding a classifier and designing a new loss function in the cyclic generative adversarial network, fault simulation samples closer to the real distribution are further obtained. The present invention realizes the fusion of simulation samples and real samples, effectively makes up for the shortage of fault samples of the rotor-bearing system in fault diagnosis, and improves the generalization performance of the classification model. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly describe the drawings required in the embodiments. By referring to the drawings, the features and advantages of the present invention will be more clearly understood. The drawings are schematic and should not be construed as limiting the present invention in any way. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts. Among them: Figure 1 It is a flowchart of a fault diagnosis method for a rotor-bearing system based on virtual-real fusion provided by the present invention; Figure 2 It is a simplified structure diagram of the rotor-bearing system in the present invention, where (a) is the schematic diagram of the rotor-bearing system, and (b) is the coordinate system of the rotor-bearing system; Figure 3 It is a schematic diagram of introducing a fault excitation in the present invention, where (a) is a rotor eccentricity fault, (b) is an inner race fault of the bearing, and (c) is an outer race fault of the bearing; Figure 4 It is a spectrogram of simulation samples and real samples under the rotor eccentricity fault condition in the present invention, where (a) is the spectrum of the real rotor eccentricity fault, and (b) is the spectrum of the simulation rotor eccentricity fault; Figure 5 It is a spectrogram of simulation samples and real samples under the inner race fault condition of the bearing in the present invention, where (a) is the spectrum of the real inner race fault of the bearing, and (b) is the spectrum of the simulation inner race fault of the bearing; Figure 6 It is a spectrogram of simulation samples and real samples under the outer race fault condition of the bearing in the present invention, where (a) is the spectrum of the real outer race fault of the bearing, and (b) is the spectrum of the simulation outer race fault of the bearing; Figure 7 It is a flowchart of the improved genetic algorithm in the present invention; Figure 8This is the structural diagram of the one-dimensional cyclic generative adversarial network in the present invention; Figure 9 This is the block diagram of a rotor-bearing system fault diagnosis system based on virtual-real fusion provided by the present invention. Specific implementation manners

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0025] Embodiment 1: To solve the problems in the prior art that the real sample data is unbalanced during the intelligent diagnosis of the rotor-bearing system, it is difficult to obtain fault data, and the twin model is affected by various factors, resulting in a large difference between the data and the real data and insufficient fault diagnosis ability. As Figure 1 shown, the present invention proposes a rotor-bearing system fault diagnosis method based on virtual-real fusion, and the method includes: S1: Derive the unified expression of the hybrid eccentric unbalanced magnetic pull force, introduce the bearing nonlinear restoring force based on the Hertz contact theory, and establish a twin model of the rotor-bearing system; S2: Introduce rotor eccentricity faults, bearing inner ring faults, and bearing outer ring faults into the twin model of the rotor-bearing system to obtain the twin model of the rotor-bearing system under different fault conditions; S3: Establish a fitness function by introducing a feature sensitivity evaluation factor into the genetic algorithm, and combine the ideas of adaptive crossover, mutation probability, and simulated annealing to perform multi-parameter identification on the twin model of the rotor-bearing system under different fault conditions to obtain a corrected twin model; S4: Obtain twin fault samples based on the corrected twin model, add a classifier to the one-dimensional cyclic generative adversarial network and design a comprehensive loss function to train and generate generated samples close to the real sample distribution; S5: Virtually and really fuse the generated samples and the real samples to obtain a balanced data set, and use a one-dimensional convolutional neural network to train for fault classification.

[0026] As can be seen from the above technical solution, the present invention proposes a fault diagnosis method for a rotor-bearing system based on virtual-real fusion. By deriving a unified expression for the hybrid eccentricity unbalanced magnetic pull force and combining it with the bearing's non-linear restoring force, it aims to establish a twin model that can reflect the fault characteristics. During the parameter identification process of the twin model, a corresponding fitness function is established by introducing a feature sensitivity evaluation factor to maximize the difference between the simulation samples and the real samples. Combining the adaptive crossover and mutation probabilities can improve the ability of the genetic algorithm to find the optimal solution. Adding a classifier to the cyclic generative adversarial network helps to retain the class information in the original signal, which is conducive to solving the problem of unbalanced fault samples in the rotor-bearing system and realizing more accurate fault diagnosis of the rotor-bearing system.

[0027] In step S1, the present invention establishes a twin model of the rotor-bearing system by deriving a unified expression for the hybrid eccentricity unbalanced magnetic pull force and introducing the bearing's non-linear restoring force based on the Hertz contact theory.

[0028] First, based on the theory of electrical machinery, the air-gap length, magnetic conductance, and magnetomotive force distribution under hybrid eccentricity are derived, and the unified expression for the unbalanced magnetic pull force in the 、 direction is obtained through the Maxwell stress integral.

[0029] Specifically, based on the theory of electrical machinery, considering both static eccentricity and dynamic eccentricity, the air-gap length at any dynamic eccentricity angle under hybrid eccentricity between the rotor and the stator is a function related to time and angle, which can be approximately expressed as: ; In the formula, is the average air-gap length without eccentricity; is the relative static eccentricity ratio, is the relative dynamic eccentricity ratio, and are the ratios of the static deviation and the dynamic deviation to the average air-gap respectively; is the angle at any position during the rotor operation; is the mechanical angular frequency of the rotor; is the time; is the dynamic eccentricity angle.

[0030] Furthermore, when the eccentricity ratio is small, the air-gap magnetic conductance can be expanded into a Fourier series. Ignoring the high-order components, it can be expressed as: ; In the formula, is the vacuum magnetic permeability; is the constant component of the uniform air-gap magnetic conductance without eccentricity; is the magnetic conductance component caused by static eccentricity; The permeance component caused by dynamic eccentricity.

[0031] Furthermore, the fundamental magnetomotive force of the air gap is: ; In the formula, is the amplitude of the air-gap magnetomotive force; is the initial phase; is the amplitude of the fundamental magnetomotive force of the rotor; is the amplitude of the fundamental magnetomotive force of the stator armature winding; is the internal power factor angle.

[0032] Furthermore, by using the air-gap magnetomotive force and permeance, the air-gap magnetic flux density distribution can be established and expressed as: .

[0033] Since the radial air-gap magnetic density is much larger than the tangential air-gap magnetic density, the tangential component can be ignored. According to Maxwell's electromagnetic field law, the electromagnetic force acting on the rotor is proportional to the square of its radial magnetic density and can be expressed as: .

[0034] Furthermore, by integrating the Maxwell stress on the rotor surface, a unified expression for the unbalanced magnetic pull under different pole numbers is proposed. The unbalanced magnetic pull in the direction and the unbalanced magnetic pull in the ; ; In the formula, is the rotor radius; is the axial length; is the pole number.

[0035] Then, based on Hertz contact theory, the positions and contact deformation amounts of the bearing rolling elements are deduced, and the , direction bearing nonlinear contact forces are accumulated.

[0036] Specifically, let the rolling element numbered 1 be located on the positive direction of the axis. Then the position of the th rolling element can be expressed by the following formula: ; In the formula, is the initial angular position of the rolling element numbered 1; is the rotational angular velocity of the rolling element; is the number of rolling elements; is the introduced random sliding deviation, which is caused by the existence of the lubricating oil film and the reserved clearance, resulting in a certain slip phenomenon of the rolling elements during movement; is at uniformly distributed random probability function; is the angular velocity of the rotor rotation; and are the inner diameter and outer diameter of the bearing respectively.

[0037] Under the action of the rotor eccentricity, the bearing vibrates. At this time, the total contact deformation amount of the rolling elements is a function related to the horizontal displacement of the bearing inner ring, the vertical displacement of the bearing inner ring, the position of the th rolling element and the bearing clearance , and its expression is as follows: .

[0038] Furthermore, according to the Hertz contact theory, by accumulating the contact forces of each rolling element, the bearing contact force in the direction and the bearing contact force in the direction can be obtained, and can be expressed as: ; ; In the formula, is the number of rolling elements; is the equivalent contact stiffness obtained from the local deformation of the contact area.

[0039] Finally, the simplified structural schematic diagram of the rotor-bearing system is as shown in Figure 2 . Combining the unbalanced magnetic pull force and the bearing nonlinear contact force, the dynamic equation of the rotor-bearing system can be established and can be expressed as: ; In the formula, is the rotor mass; is the damping coefficient; is the mechanical stiffness of the rotor shaft; and are the exciting forces in the , directions respectively; , , are the displacement, velocity, and acceleration of the rotor in the direction respectively; , , are the displacement, velocity, and acceleration of the rotor in the direction, respectively.

[0040] In step S2, the present invention obtains the twin model of the rotor-bearing system under different fault conditions by introducing rotor eccentricity fault, inner race fault, and outer race fault into the twin model of the rotor-bearing system.

[0041] Specifically, the rotor eccentricity fault is as shown in Figure 3 (a). By setting the relative static eccentricity and the relative dynamic eccentricity , when it is the static eccentricity condition, it is the dynamic eccentricity condition when and are both non-zero, it is the mixed eccentricity condition.

[0042] For the inner race fault and outer race fault of the bearing, rectangular defects are set on the inner and outer race surfaces respectively, and the total contact deformation of the rolling element is characterized by the depth of the rolling element sinking into the defect. Among them, the inner race defect depth is related to the inner race radius, rolling element radius, and angular difference, and the outer race defect depth is related to the outer race radius and angular difference.

[0043] Furthermore, the total contact deformation of the rolling element is expressed as: ; In the formula, is the depth that the rolling element sinks into when passing through the defect.

[0044] The inner race defect of the bearing is as shown in Figure 3 (b). When the inner race defect of the bearing is introduced, the inner race defect depth can be expressed as: ; In the formula, is the inner race defect depth, , are the inner race radius and rolling element radius of the bearing respectively; is the initial angle when the rolling element enters the defect area; is the angle when the rolling element sinks into the maximum depth of the defect area, that is, the central angle; is the termination angle when leaving the defect area; is the angular difference between the inner race and the rolling element during rotation, which can be expressed as: .

[0045] The outer race defect of the bearing is as shown in Figure 3As shown in (c), when introducing the defect of the bearing outer ring, the depth of the outer ring defect can be expressed as: ; In the formula, is the depth of the inner ring defect, , are the inner ring radius and the rolling element radius of the bearing respectively; is the initial angle when the rolling element enters the defect area; is the angle when the rolling element sinks into the maximum depth of the defect area, that is, the central angle; is the termination angle when leaving the defect area; is the angle difference between the rotation of the bearing inner ring and the rolling element.

[0046] Normalize the signals simulated by the twin model of the rotor-bearing system under different fault conditions, and perform Fourier transform. The spectrograms of the simulated samples and the real samples under the three fault conditions are as shown in Figure 4 , Figure 5 , Figure 6 . At this time, the spectrogram of the simulated signal basically includes the rotation frequency and the fault characteristic frequency in the real signal.

[0047] In step S3, the present invention establishes a fitness function by introducing a feature sensitivity evaluation factor into the genetic algorithm, combines the adaptive crossover and mutation probabilities and the simulated annealing idea, and performs multi-parameter identification on the twin model of the rotor-bearing system under different fault conditions to obtain a corrected twin model. Figure 7 is the flow chart of the improved genetic algorithm.

[0048] Specifically, first, the ratio of the between-class variance to the within-class variance is used as the feature sensitivity evaluation factor . The between-class variance is calculated through the total mean value and the number of samples of the feature values of the real samples and the simulated samples, and the within-class variance is calculated through the variance of the feature values under each sample state. Among them, the feature sensitivity evaluation factor is expressed as: ; In the formula, and are the within-class variances in the sample states and the sample state respectively; is the between-class variance.

[0049] Among them, the between-class variance can be expressed as: ; In the formula, and are the numbers of the real samples and the simulated samples respectively; and are the values of the feature in the th sample and the th sample in the sample state respectively; is the total mean value of the feature in the two samples. In the present invention, the selected features are peak value, root mean square, kurtosis, skewness, peak factor, waveform factor, center frequency and mean square frequency, and the feature sensitivity evaluation factors

[0050] are calculated respectively, and the feature with the largest value is selected to establish the fitness function. Then, an adaptive crossover and mutation probability are added in the crossover and mutation operations, which can be expressed as:

[0051] ; ; ; In the formula, , are the adaptive crossover probability and the adaptive mutation probability respectively; , are the maximum and minimum values of the predefined crossover probability respectively; , are the maximum and minimum values of the predefined mutation probability respectively; is the fitness value of the individual with the higher fitness value among the two individuals participating in the crossover operation; is the fitness value of the mutant individual; is the average fitness value in the current population; is the maximum fitness value in the current population.

[0052] Finally, the outer loop and the inner loop are added to the genetic algorithm in combination with the simulated annealing idea. The outer loop is set to control the temperature decay, and the initial temperature gradually decreases to the termination temperature according to the cooling coefficient α; the inner loop iterates multiple times at each temperature, and new individuals are accepted through the Metropolis criterion, where the acceptance probability is: ; In the formula, is the acceptance probability, is the current temperature, and are the fitness values of the new individual and the original individual respectively.

[0053] In step S4, the present invention obtains twin fault samples based on the modified twin model, adds a classifier to the one-dimensional cyclic generative adversarial network and designs a comprehensive loss function to train the generated samples close to the distribution of the real samples.

[0054] Specifically, the cyclic generative adversarial network consists of two generators ( and ) and two discriminators ( and ). The generator generates a corrected false signal based on the input simulation signal, aiming to make the generated false signal as close as possible to the real signal so that the discriminator cannot distinguish them; while the discriminator needs to distinguish the generated false signal from the real signal as much as possible. To further make the generated signal similar to the simulation signal in terms of category, two classifiers ( and ) are introduced on this basis to retain the category information in the original signal. The structural diagram of the improved cyclic generative adversarial network is shown in Figure 8 .

[0055] During the game confrontation process of the cyclic generative adversarial network, there are two sample spaces and . The cyclic generative adversarial network aims to learn the general mapping between and so as to convert the simulation signal into an approximation of the real signal. The training process of the network is jointly realized by three types of loss functions, which are specifically described as follows: (1) Adversarial loss function The adversarial loss function between the generator and the discriminator is expressed as: ; The adversarial loss function between the generator and the discriminator is expressed as: ; In the formula, and are one-dimensional signals in the sample spaces and respectively; denotes the expectation that obeys the data distribution of the sample space ; denotes the expectation that obeys the data distribution of the sample space .

[0056] (2) Cyclic consistency loss function The cyclic generative adversarial network uses the data conversion mechanism to introduce cycle consistency loss, so that the signal can retain the original style characteristics after passing through two generators continuously. The cycle consistency loss uses the L1 norm and can be expressed as: ; In the formula, represents the L1 norm.

[0057] (3) Cross Entropy Loss Function Two classifiers are introduced into the basic cycle generative adversarial network to retain category information, and the classifier is trained using the standard cross entropy loss function. and .

[0058] Classifier The cross entropy loss function It is expressed as: ; Classifier The cross entropy loss function It is expressed as: .

[0059] Combining the above three types of loss functions, the final loss function for: ; In the formula, , , are weight coefficients, which are used to control the importance of cycle consistency loss and two classification losses respectively.

[0060] Therefore, the optimization goal of the one-dimensional cyclic generative adversarial network of the present invention is: ; ; ; In the formula, , , , , , To train to reach Nash equilibrium , , , , , status.

[0061] Specifically, 2048 points in the real samples and simulation samples are selected as a group. For each type of fault, 160 training samples and 40 test samples are set. The specific network parameters are shown in Table 1. The number of network iterations is 800, the learning rate is 0.0002, and the batch size is 32.

[0062] Table 1 Related parameters of one-dimensional recurrent generative adversarial network

[0063] In step S5, the present invention obtains a balanced data set by fusing the generated samples and real samples in a virtual-real manner, and uses a one-dimensional convolutional neural network for training to achieve fault classification.

[0064] Specifically, the generated samples and real samples are fused to obtain a data set with the same number of samples under each working condition. Based on this data set, 1D CNN is used to train it to achieve fault classification of the rotor-bearing system, where 1D CNN adopts the same network structure as the classifier in the one-dimensional recurrent generative adversarial network.

[0065] Embodiment 2: As Figure 9 shown, the present invention provides a fault diagnosis system for a rotor-bearing system based on virtual-real fusion. This system is used to implement the fault diagnosis method for a rotor-bearing system based on virtual-real fusion in the above Embodiment 1, and specifically includes: A model establishment module 100, which is used to deduce a unified expression of the combined eccentric unbalanced magnetic pull, introduce the bearing nonlinear restoring force based on the Hertz contact theory, and establish a twin model of the rotor-bearing system; A fault introduction module 200, which is used to introduce rotor eccentricity faults, inner race faults of the bearing, and outer race faults of the bearing into the twin model of the rotor-bearing system to obtain the twin model of the rotor-bearing system under different fault conditions; A model correction module 300, which is used to establish a fitness function by introducing a feature sensitivity evaluation factor into the genetic algorithm, and combine the ideas of adaptive crossover, mutation probability, and simulated annealing to perform multi-parameter identification on the twin models of the rotor-bearing system under different fault conditions to obtain the corrected twin models; A sample generation module 400, which is used to obtain twin fault samples based on the corrected twin model, add a classifier to the one-dimensional recurrent generative adversarial network and design a comprehensive loss function, and train to generate generated samples with a distribution close to that of the real samples; A fault classification module 500, which is used to obtain a balanced data set by fusing the generated samples and real samples in a virtual-real manner, and uses a one-dimensional convolutional neural network for training to achieve fault classification.

[0066] A fault diagnosis system for a rotor-bearing system based on virtual-real fusion in this embodiment is used to implement the aforementioned fault diagnosis method for a rotor-bearing system based on virtual-real fusion. Therefore, the specific implementation manners in the fault diagnosis system for a rotor-bearing system based on virtual-real fusion can be seen in the embodiment part of the aforementioned fault diagnosis method for a rotor-bearing system based on virtual-real fusion. For example, the model establishment module 100, the fault introduction module 200, the model correction module 300, the sample generation module 400, and the fault classification module 500 are respectively used to implement steps S1, S2, S3, S4, and S5 in the aforementioned fault diagnosis method for a rotor-bearing system based on virtual-real fusion. Therefore, the specific implementation manners can refer to the descriptions of the corresponding individual embodiments. To avoid redundancy, they will not be elaborated here.

[0067] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0068] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of the flows and / or blocks in the flowchart and / or block diagram can also be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the specified functions in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0069] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device implements the specified functions in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks. These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Therefore, the instructions executed on the computer or other programmable device provide means for implementing the specified functions in the flowFigure 1 One process or multiple processes and / or boxes Figure 1 Steps of functions specified in one box or multiple boxes.

[0070] Obviously, the above embodiments are merely examples given for clear illustration and are not limitations on the implementation manners. For those of ordinary skill in the art, other different forms of changes or alterations can be made based on the above description. It is not necessary and impossible to list all implementation manners here. And the obvious changes or alterations derived therefrom are still within the protection scope of the present invention.

Claims

1. A fault diagnosis method for a rotor-bearing system based on the fusion of virtual and real, characterized in that, Including: S1: Derive the unified expression of the hybrid eccentric unbalanced magnetic pull force, introduce the bearing nonlinear restoring force based on the Hertz contact theory, and establish the twin model of the rotor-bearing system; S2: Introduce the rotor eccentricity fault, the inner race fault of the bearing, and the outer race fault of the bearing into the twin model of the rotor-bearing system to obtain the twin model of the rotor-bearing system under different fault conditions; S3: Establish a fitness function by introducing a feature sensitivity evaluation factor into the genetic algorithm, and combine the adaptive crossover and mutation probabilities and the simulated annealing idea to perform multi-parameter identification on the twin models of the rotor-bearing system under different fault conditions to obtain the corrected twin model; S4: Obtain twin fault samples based on the corrected twin model, add a classifier to the one-dimensional cyclic generative adversarial network and design a comprehensive loss function to train and generate generated samples close to the real sample distribution; S5: Fuse the generated samples and the real samples in a virtual-real manner to obtain a balanced dataset, and use a one-dimensional convolutional neural network for training to achieve fault classification.

2. The fault diagnosis method for a rotor-bearing system based on virtual-real fusion according to claim 1, wherein The process of establishing the twin model of the rotor-bearing system includes: First, based on the theory of electrical machinery, the air-gap length, permeance, and magnetomotive force distribution under hybrid eccentricity are derived, and the unified expression of the unbalanced magnetic pull in the , direction is obtained through Maxwell stress integration; Then, based on the Hertz contact theory, the positions of the bearing rolling elements and the contact deformation amounts are deduced, and the non-linear contact forces of the bearings in the and directions are obtained by accumulation. and direction Finally, combine the unbalanced magnetic pull force and the bearing nonlinear contact force to establish the dynamic equation of the rotor-bearing system.

3. The fault diagnosis method for a rotor-bearing system based on virtual-real fusion according to claim 2, wherein The said and The unified expression of the directional unbalanced magnetic pull force is as follows: ； ； In the formula, is the unbalanced magnetic pull in the direction; is the unbalanced magnetic pull in the direction; is the amplitude of the air-gap magnetomotive force; is the permeability of free space; is the constant component of the uniform air-gap permeance without eccentricity; is the permeance component caused by static eccentricity; is the permeance component caused by dynamic eccentricity; is the mechanical angular frequency of the rotor; is the initial phase; is the time; is the dynamic eccentricity angle; is the number of pole pairs.

4. The fault diagnosis method for a rotor-bearing system based on virtual-real fusion according to claim 2, wherein, The said and the non-linear contact force of the directional bearing is as follows: ； ； Wherein, is the non-linear contact force of the bearing in the direction; is the non-linear contact force of the bearing in the direction; is the number of rolling elements; is the equivalent contact stiffness obtained from the local deformation of the contact area; is the total contact deformation of the th rolling element; is the bearing clearance; is the position of the th rolling element.

5. The fault diagnosis method for a rotor-bearing system based on virtual-real fusion according to claim 1 or 4, characterized in that The expression of the twin model of the rotor-bearing system is: ； In the formula, is the rotor mass; is the damping coefficient; is the mechanical stiffness of the rotor shaft; and are respectively the exciting forces in the , directions; and are respectively the bearing nonlinear contact forces in the , directions; , , are respectively the displacement, velocity, and acceleration of the rotor in the direction; , , are respectively the displacement, velocity, and acceleration of the rotor in the direction.

6. The fault diagnosis method for a rotor-bearing system based on virtual-real fusion according to claim 1, wherein The method of introducing the rotor eccentricity fault, the inner race fault of the bearing, and the outer race fault of the bearing is: The rotor eccentricity fault is characterized by setting the relative static eccentricity and the relative dynamic eccentricity . When , it is in the static eccentricity condition, , it is in the dynamic eccentricity condition, and are both non-zero, it is in the mixed eccentricity condition; For the inner race fault and the outer race fault of the bearing, rectangular defects are set on the inner and outer race surfaces respectively, and the total contact deformation of the rolling elements is characterized by the depth of the rolling elements sinking into the defects. The depth of the inner race defect is related to the inner race radius, the rolling element radius, and the angular difference, and the depth of the outer race defect is related to the outer race radius and the angular difference.

7. The fault diagnosis method for a rotor-bearing system based on virtual-real fusion according to claim 6, wherein The depth of the inner race defect is expressed as: ； In the formula, is the depth of the inner ring defect, , are the radius of the bearing inner ring and the radius of the rolling element respectively; is the initial angle when the rolling element enters the defect area; is the angle when the rolling element sinks to the maximum depth in the defect area, that is, the central angle; is the termination angle when leaving the defect area; is the angular difference between the rotation of the bearing inner ring and the rolling element.

8. The fault diagnosis method for a rotor-bearing system based on virtual-real fusion according to claim 6, characterized in that The depth of the outer race defect is expressed as: ； In the formula, is the depth of the inner ring defect, , are the radius of the bearing inner ring and the radius of the rolling element respectively; is the initial angle when the rolling element enters the defect area; is the angle when the rolling element sinks to the maximum depth in the defect area, that is, the central angle; is the termination angle when leaving the defect area; is the angular difference between the rotation of the bearing inner ring and the rolling element.

9. The fault diagnosis method for a rotor-bearing system based on virtual-real fusion according to claim 1, wherein The method of establishing the feature sensitivity evaluation factor is: Take the ratio of the between-class variance to the within-class variance as the feature sensitivity evaluation factor , where the between-class variance is calculated based on the total mean and the number of samples of the feature values of the real samples and the simulation samples, and the within-class variance is calculated based on the variance of the feature values in each sample state Where Feature sensitivity evaluation factor It is expressed as: ； In the formula, and are the within-class variances in the sample states and the sample state respectively; is the between-class variance.

10. The fault diagnosis method for a rotor-bearing system based on virtual-real fusion according to claim 1, characterized in that The expressions of the adaptive crossover and mutation probabilities are: ； ； Wherein, and are the adaptive crossover probability and the adaptive mutation probability respectively; and are the maximum and minimum values of the predefined crossover probability respectively; and are the maximum and minimum values of the predefined mutation probability respectively; is the fitness value of the individual with a higher fitness value among the two individuals participating in the crossover operation; is the fitness value of the mutant individual; is the average fitness in the current population; is the maximum fitness in the current population.

11. The method for fault diagnosis of a rotor-bearing system based on virtual-real fusion according to claim 1, wherein The implementation of the simulated annealing idea includes: Set the outer loop to control the temperature decay, and the initial temperature gradually decreases to the termination temperature according to the temperature reduction coefficient α; the inner loop iterates multiple times at each temperature, and accepts new individuals through the Metropolis criterion, where the acceptance probability is: ； In the formula, is the acceptance probability, is the current temperature, and are the fitness values of the new individual and the original individual, respectively.

12. The fault diagnosis method for a rotor-bearing system based on virtual-real fusion according to claim 1, wherein The one-dimensional cyclic generative adversarial network includes two generators and , two discriminators and as well as two classifiers and . The comprehensive loss function of the one-dimensional cyclic generative adversarial network includes: Generator and the discriminator The adversarial loss function between is expressed as: ； Generator and the discriminator The adversarial loss function between is expressed as: ； Wherein, and are respectively one-dimensional signals in the sample spaces and ; denotes the expectation of the data distribution subject to the sample space ; denotes the expectation of the data distribution subject to the sample space ; Cycle consistency loss function It is expressed as: ； In the formula, represents the L1 norm; Classifier cross-entropy loss function is expressed as: ； Classifier cross-entropy loss function is expressed as: ； In the formula, is a one-dimensional signal in the sample space ; is a one-dimensional signal in the sample space ; is the total number of samples for parameter loss calculation in the sample space ; is the total number of samples for parameter loss calculation in the sample space ; Final loss function is expressed as: ； wherein, , , are weight coefficients, respectively used to control the importance of the cycle consistency loss and the two classification losses.

13. The fault diagnosis method for a rotor-bearing system based on virtual-real fusion according to claim 1, wherein The network structure of the one-dimensional convolutional neural network is the same as the classifier structure of the one-dimensional cyclic generative adversarial network.

14. A fault diagnosis system for a rotor-bearing system based on the fusion of virtual and real, characterized in that, The system is used to implement the virtual-real fusion-based fault diagnosis method for the rotor-bearing system described in any one of claims 1 to 13, specifically including: A model establishment module, which is used to derive the unified expression of the hybrid eccentric unbalanced magnetic pull force, introduce the bearing nonlinear restoring force based on the Hertz contact theory, and establish the twin model of the rotor-bearing system; A fault introduction module, which is used to introduce the rotor eccentricity fault, the inner race fault of the bearing, and the outer race fault of the bearing into the twin model of the rotor-bearing system to obtain the twin model of the rotor-bearing system under different fault conditions; A model correction module, which is used to establish a fitness function by introducing a feature sensitivity evaluation factor into the genetic algorithm, and combine the adaptive crossover and mutation probabilities and the simulated annealing idea to perform multi-parameter identification on the twin models of the rotor-bearing system under different fault conditions to obtain the corrected twin model; The sample generation module is used to obtain twin fault samples based on the corrected twin model, add a classifier to the one-dimensional cyclic generative adversarial network and design a comprehensive loss function to train and generate generated samples close to the distribution of real samples; The fault classification module is used to obtain a balanced data set by fusing the generated samples and real samples in a virtual and real manner, and use a one-dimensional convolutional neural network for training to achieve fault classification.

Images