Full-ceramic bearing wear state identification method

Abstract

The invention relates to the technical field of mechanical structure state monitoring, in particular to a full-ceramic bearing wear state recognition method, which comprises the following steps of: acquiring a vibration signal of an outer ring of a full-ceramic bearing to be recognized in the operation process of the full-ceramic bearing under a specified working condition; acquiring specified characteristic parameters in the time domain characteristics of the vibration signal according to the vibration signal; according to specified characteristic parameters and a first function formula, obtaining a wear loss predicted value of the to-be-identified full-ceramic bearing; the first function formula is a simulated vibration signal obtained by simulating the full-ceramic bearing through a preset condition Wasserstein generative adversarial network model, a real vibration signal of the full-ceramic bearing obtained through an experiment, and preset category label information; processing to obtain a function formula between the specified characteristic parameters and the abrasion loss of the full-ceramic bearing; the category label information comprises the vibration frequency of the outer ring and / or the inner ring when the full-ceramic bearing breaks down under the specified working condition.

Description

Technical Field

[0001] The present invention relates to the technical field of mechanical structure condition monitoring, and particularly relates to a method for identifying the wear state of a fully ceramic bearing. Background Art

[0002] Due to its excellent high-temperature resistance, corrosion resistance, low density and high hardness characteristics, fully ceramic bearings have been widely used in high-end equipment fields such as aerospace, ships, metallurgy, chemical engineering and semiconductor manufacturing. In the above special working condition environment, fully ceramic bearings need to withstand the combined action of multiple loads such as high contact stress, wide temperature range and corrosive media for a long time. Their wear behavior directly affects the safety and stability of equipment operation and is one of the key factors affecting the bearing life and accuracy. Among them, wear has been widely recognized as the main failure mode of fully ceramic bearings. Therefore, accurately obtaining their wear state during service is of great significance for extending their service life and reducing maintenance costs.

[0003] At present, bearing wear detection technologies mainly include the following two categories: one is the method based on particle analysis, which indirectly judges the bearing wear state by analyzing the composition and size of wear particles generated in the lubricating medium. However, this method has high detection costs, is cumbersome to operate, has a long cycle, and most of the research objects are metal bearings, so its applicability to ceramic bearings is low and relevant research is scarce. The other is the method based on bearing vibration signal analysis. This method monitors the vibration signals generated during the operation of the bearing and combines signal processing and feature extraction technologies to identify the fault state of the bearing, and has a certain online monitoring ability.

[0004] However, the existing methods mainly focus on the classification and identification of fault types, lack quantitative modeling and prediction of bearing wear amount, and are difficult to meet the requirements of real-time and accurate feedback on the wear state of fully ceramic bearings. Summary of the Invention

[0005] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a method for identifying the wear state of a fully ceramic bearing, which solves the technical problem in the prior art that there is a lack of quantitative modeling and prediction of bearing wear amount and it is difficult to achieve quantitative identification and prediction of wear amount.

[0006] In order to achieve the above object, the main technical solutions adopted by the present invention include:

[0007] An embodiment of the present invention provides a method for identifying the wear state of a fully ceramic bearing, including:

[0008] S100. Obtain the vibration signal of the outer ring of the fully ceramic bearing during the operation of the fully ceramic bearing to be identified under specified working conditions;

[0009] S200. Obtain the specified characteristic parameters in the time-domain characteristics of the vibration signal;

[0010] S300. According to the specified characteristic parameters in the time-domain characteristics of the vibration signal and the pre-constructed first function formula, obtain the predicted wear amount of the all-ceramic bearing to be identified;

[0011] Wherein, the first function formula is a function formula between the specified characteristic parameters obtained by processing the simulated vibration signal obtained by simulating the all-ceramic bearing, the real vibration signal of the all-ceramic bearing obtained by experiment, and the pre-set class label information by the pre-set conditional Wasserstein generative adversarial network model and the wear amount of the all-ceramic bearing;

[0012] The class label information includes the vibration frequencies of the outer ring and / or inner ring when the all-ceramic bearing fails under specified working conditions.

[0013] Preferably, before the S100, it further includes:

[0014] S000-1. Conduct wear experiments on N all-ceramic bearings under specified working conditions to obtain the real vibration signals of the all-ceramic bearings;

[0015] Wherein, N is a positive integer greater than or equal to 2;

[0016] S000-2. Simulate the all-ceramic bearing under specified working conditions to obtain the simulated vibration signal and simulated wear amount of the all-ceramic bearing;

[0017] S000-3. Use the simulated vibration signal, real vibration signal, and pre-set class label information to train the conditional Wasserstein generative adversarial network model to obtain a trained conditional Wasserstein generative adversarial network model;

[0018] S000-4. According to the trained conditional Wasserstein generative adversarial network model and the simulated wear amount, obtain the first function formula.

[0019] Preferably, the S000-1 specifically includes:

[0020] S000-1-1. Select N all-ceramic bearings as experimental objects, clean and dry each all-ceramic bearing, and then use an electronic balance to measure the mass of each all-ceramic bearing 5 times, and take the average value of the 5 measurement values as the initial mass m0;

[0021] S000-1-2. Divide N fully ceramic bearings into two fully ceramic bearing groups, and conduct wear experiments on the two fully ceramic bearing groups respectively under specified working conditions in a fully ceramic bearing life enhancement test machine, specifically including:

[0022] Run each fully ceramic bearing in the first fully ceramic bearing group from the normal state to meet the set conditions respectively under the specified working conditions in the fully ceramic bearing life enhancement test machine to complete a complete life experiment, and collect the real vibration signals of the outer ring of the fully ceramic bearing using an acceleration sensor during the complete life experiment;

[0023] Among them, the two fully ceramic bearing groups include the first fully ceramic bearing group and the second fully ceramic bearing group; the first fully ceramic bearing group includes M fully ceramic bearings; the second fully ceramic bearing group includes Q fully ceramic bearings; N = M + Q; and M is greater than or equal to 2; Q is greater than or equal to 2;

[0024] Among them, the set conditions are that the complete life experiment time reaches the theoretical life h0 calculated according to the bearing life formula or the kurtosis of the real vibration signal during the complete life experiment exceeds 3 times the kurtosis of the real vibration signal in the first time period after the start of the experiment;

[0025] After the complete life experiment, clean and dry the fully ceramic bearings in the first fully ceramic bearing group again, weigh the mass of each fully ceramic bearing 5 times respectively, and take the average value as the mass m1 of the fully ceramic bearing after wear in the complete life experiment, and use the difference between the initial mass m0 and the mass m1 of the fully ceramic bearing after wear in the complete life experiment to represent the complete wear amount m01;

[0026] Conduct segmented life experiments on each fully ceramic bearing in the second fully ceramic bearing group respectively under the specified working conditions in the fully ceramic bearing life enhancement test machine. The time of each experimental stage is h0 / 5. At the end of each experimental stage, clean, dry and measure the mass of each fully ceramic bearing, and obtain the wear amounts of each experimental stage in the previous 4 experimental stages in turn until the cumulative experimental time of the segmented life experiment reaches h0×4 / 5;

[0027] Among them, the wear amounts of each experimental stage include:

[0028] The wear amount ma in the first experimental stage: represents the difference between the initial mass m0 and the average value of the masses of all fully ceramic bearings in the second fully ceramic bearing group after the first experimental stage;

[0029] The wear amount mb in the second experimental stage: represents the difference between the average value of the masses of all fully ceramic bearings in the second fully ceramic bearing group after the first experimental stage and the average value of the masses of all fully ceramic bearings in the second fully ceramic bearing group after the second experimental stage;

[0030] The wear amount mc in the third experimental stage: It represents the difference between the average mass of all the fully ceramic bearings in the second fully ceramic bearing group after the second experimental stage and the average mass of all the fully ceramic bearings in the second fully ceramic bearing group after the third experimental stage;

[0031] The wear amount md in the fourth experimental stage: It represents the difference between the average mass of all the fully ceramic bearings in the second fully ceramic bearing group after the third experimental stage and the average mass of all the fully ceramic bearings in the second fully ceramic bearing group after the fourth experimental stage;

[0032] S000-1-3. Calculate the theoretical maximum contact force and the ball rotation speed of the fully ceramic bearing based on the Hertz contact theory and the kinematic model. Combine with the Archard wear model to calculate the wear coefficient k1 under the complete wear amount m01 and the wear coefficients of each experimental stage under the wear amounts of each experimental stage;

[0033] Among them, the wear coefficient corresponding to the first experimental stage is ka, the wear coefficient corresponding to the second experimental stage is kb, the wear coefficient corresponding to the third experimental stage is kc, and the wear coefficient corresponding to the fourth experimental stage is kd.

[0034] Preferably, the S000-2 specifically includes:

[0035] S000-2-1. Construct a high-fidelity finite element model of the fully ceramic bearing according to the size parameters of the fully ceramic bearing, and perform mesh division on the high-fidelity finite element model to obtain the mesh division result;

[0036] S000-2-2. Apply the specified working condition in the LSDYNA simulation software, perform transient dynamics simulation on the mesh division result, and obtain the simulation data of the fully ceramic bearing;

[0037] The simulation data includes: the simulated vibration signal of the outer ring of the fully ceramic bearing, the simulated contact force between the balls and the inner and outer rings in the fully ceramic bearing, and the simulated rotation speed of the balls in the fully ceramic bearing;

[0038] S000-2-3. According to the Archard wear model and the wear coefficients of each experimental stage under the wear amounts of each experimental stage, calculate the simulated wear amount of the balls and the inner and outer rings in the fully ceramic bearing within h0 / n time, where n is the total number of simulations; n = 10;

[0039] S000-2-4. Update the high-fidelity finite element model of the all-ceramic bearing according to the simulated wear amount obtained each time, then perform mesh division on the updated high-fidelity finite element model of the all-ceramic bearing to obtain a new mesh division result, and repeat steps S000-2-2 to S000-2-5 for n times, and obtain the simulated vibration signals of the all-ceramic bearing for n preset time periods;

[0040] Among them, the preset time period is 0.1 s.

[0041] Preferably, S000-3 specifically includes:

[0042] S000-3-1. The generator in the conditional Wasserstein generative adversarial network model processes the first data to obtain the processing result of the first data;

[0043] The first data is obtained by performing Hadamard product operation on the class label information after Embedding processing and the input data; the input data is obtained by superimposing any segment of the simulated vibration signal and the pre-generated random noise;

[0044] S000-3-2. The fourth fully connected layer in the discriminator of the conditional Wasserstein generative adversarial network model processes the class label information after Embedding processing to obtain the processed class label information;

[0045] The discriminant module in the discriminator of the conditional Wasserstein generative adversarial network model performs Hadamard product operations on the processed class label information with the sample data and the processing result of the first data respectively to obtain the first multiplication result and the second multiplication result, and inputs the first multiplication result and the second multiplication result into the fully connected layer unit;

[0046] The sample data includes the real vibration signals of the outer rings of the first M - 1 all-ceramic bearings in the first all-ceramic bearing group collected by acceleration sensors during the complete life experiment;

[0047] The fully connected layer unit processes the first multiplication result and the second multiplication result respectively to obtain the corresponding first matrix P r and the second matrix P g ;

[0048] S000-3-3. Repeat steps S000-3-1 to S000-3-2 until the loss function in the conditional Wasserstein generative adversarial network model satisfies the first-order Lipschitz function, then the trained conditional Wasserstein generative adversarial network model is obtained.

[0049] Preferably,

[0050] The generator in the conditional Wasserstein generative adversarial network model successively includes: a first fully-connected layer, a second fully-connected layer, a third fully-connected layer, and an output fully-connected layer;

[0051] Among them, the number of neurons in the first fully-connected layer is 256; the number of neurons in the second fully-connected layer is 512 and includes regularization processing; the number of neurons in the third fully-connected layer is 1024 and includes regularization processing;

[0052] Among them, the LeakyReLU activation function is used in the first fully-connected layer, the second fully-connected layer, and the third fully-connected layer to prevent neuron death;

[0053] The number of neurons in the output fully-connected layer is 784, and the output result of the output fully-connected layer is reconstructed into a tensor with a size of 28*1.

[0054] Preferably,

[0055] The fully-connected layer unit includes: Among them, the fully-connected layer unit includes a fifth fully-connected layer connected in sequence;

[0056] Among them, the number of neurons in each fifth fully-connected layer is 512, and each includes regularization processing; the LeakyReLU activation function is used in each fifth fully-connected layer to prevent neuron death.

[0057] Preferably,

[0058] Among them, the loss function is: L(G, D) = W(P r , P g ) - GP;

[0059] GP is a preset penalty function;

[0060] W(P r , P g ) is the Wasserstein distance between the first matrix P r and the second matrix P g .

[0061] Preferably, the S000-4 specifically includes:

[0062] S000-4-1. According to the trained conditional Wasserstein generative adversarial network model, obtain the complete simulated vibration signal generated by the trained conditional Wasserstein generative adversarial network model;

[0063] S000-4-2. Obtain the characteristic parameters in the time domain characteristics of the vibration signal of the complete simulation, and perform time fitting on each characteristic parameter to respectively obtain the function and image of each characteristic parameter with respect to time;

[0064] Among them, the characteristic parameters include: the mean value, variance, root mean square, peak value, kurtosis, waveform factor, and pulse factor of the vibration signal of the simulation in the time domain;

[0065] S000-4-3. Perform curve fitting on the simulated wear amount to obtain the function and image of the wear amount with respect to time;

[0066] S000-4-4. Align the function of each characteristic parameter with respect to time and the function of the wear amount with respect to time in terms of time to obtain the mapping relationship between the characteristic parameter and the wear amount, and perform curve fitting on the characteristic parameter and the wear amount to obtain the function formula between the characteristic parameter and the wear amount;

[0067] S000-4-5. Multiple time points pre-screened during the complete life experiment of the last full-ceramic bearing in the first full-ceramic bearing group, and according to the true wear amount during the complete life experiment corresponding to each screened time point for each characteristic parameter and the function formula between each characteristic parameter and the wear amount, determine the first function formula.

[0068] Preferably, the S000-4-5 includes:

[0069] S000-4-5-1. According to the true wear amount during the complete life experiment corresponding to each screened time point for each characteristic parameter, and the predicted value of the wear amount corresponding to each screened time point for the characteristic parameter, use formula (1) to obtain the cumulative error amount corresponding to the characteristic parameter;

[0070] The formula (1) is:

[0071] Δ i =Δt i1 +…+Δt ih +…Δt iH ;

[0072] Δ i is the cumulative error amount corresponding to the i-th characteristic parameter;

[0073] Δt ih is the absolute value of the difference between the true wear amount and the predicted value of the wear amount during the complete life experiment corresponding to the h-th time point screened for the i-th characteristic parameter;

[0074] Among them, the predicted value of the wear amount during the complete life experiment corresponding to the h-th time point selected for the i-th characteristic parameter is calculated through the function formula between the characteristic parameter and the wear amount;

[0075] H is the total number of pre-screened time points, where H is greater than 1;

[0076] S000-4-5-2. Take the characteristic parameter corresponding to the minimum cumulative error amount as the specified characteristic parameter, and take the function formula between the specified characteristic parameter and the wear amount as the first function formula.

[0077] The beneficial effects of the present invention are:

[0078] A method for identifying the wear state of a fully ceramic bearing according to the present invention, because it adopts the method of collecting the vibration signal of the outer ring of the fully ceramic bearing under specified working conditions, and combines the time-domain characteristic parameters of the vibration signal and the first function formula between the characteristic parameter and the wear amount established by the conditional Wasserstein generative adversarial network model. Compared with the prior art, it can quantitatively predict the wear amount of the fully ceramic bearing without disassembling the bearing, achieving the effect of improving the wear identification accuracy and the intelligent level of state evaluation.

[0079] A method for identifying the wear state of a fully ceramic bearing according to the present invention, because before the formal identification step, it introduces the process of obtaining the real vibration signals and simulation vibration signals of multiple fully ceramic bearings, and trains the conditional Wasserstein generative adversarial network model by combining experiments and simulations. Compared with the prior art, it can improve the generalization ability and stability of the identification model, achieving the effect of constructing a wear identification model closer to the actual working conditions.

[0080] A method for identifying the wear state of a fully ceramic bearing according to the present invention, because during the experiment, the bearing samples are grouped, cleaned, weighed in stages, and the acceleration sensors are used for collection, and the life state is judged by combining the kurtosis change. At the same time, according to the Archard model and the Hertz contact theory, the wear coefficients in different stages are deduced. Compared with the prior art, it can realize the wear feature calibration in multiple time periods within the whole life cycle, achieving the effects of more refined modeling of the wear process and significantly improving the quality of training data.

[0081] A method for identifying the wear state of a fully ceramic bearing according to the present invention, because it adopts a high-fidelity finite element model and applies transient working condition conditions in the LSDYNA simulation software, and cooperates with the wear model to iteratively update the finite element model to obtain the simulation vibration signals in multiple time periods. Compared with the prior art, it can obtain high-precision wear simulation data without relying on a large number of physical test samples, achieving the effects of reducing the experimental cost and enhancing the dimension and diversity of the model training data. Description of the Drawings

[0082] Figure 1 It is a flowchart of a method for identifying the wear state of a fully ceramic bearing according to the present invention;

[0083] Figure 2 It is an image showing the relationship between the simulated linear velocity and time of the balls in the fully ceramic bearing in the second embodiment of the present invention;

[0084] Figure 3 It is an image showing the relationship between the simulated contact force and time between the balls and the outer raceway in the fully ceramic bearing in the second embodiment of the present invention;

[0085] Figure 4 It is a schematic diagram of the high-fidelity finite element model of the fully ceramic bearing after a certain update in the second embodiment of the present invention;

[0086] Figure 5 It is a schematic diagram of the generator structure in the conditional Wasserstein generative adversarial network model in the second embodiment of the present invention. Detailed Embodiments

[0087] In order to better explain the present invention for easy understanding, the present invention will be described in detail below with reference to the drawings through specific embodiments.

[0088] In order to better understand the above technical solutions, the exemplary embodiments of the present invention will be described in more detail below with reference to the drawings. Although the exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present invention can be understood more clearly and thoroughly, and the scope of the present invention can be fully communicated to those skilled in the art.

[0089] Embodiment 1

[0090] Refer to Figure 1 , this embodiment provides a method for identifying the wear state of a fully ceramic bearing, including:

[0091] S100. Obtain the vibration signal of the outer ring of the fully ceramic bearing during operation under specified working conditions of the fully ceramic bearing to be identified;

[0092] S200. Obtain the specified characteristic parameters in the time-domain characteristics of the vibration signal according to the vibration signal;

[0093] S300. Obtain the predicted wear amount of the fully ceramic bearing to be identified according to the specified characteristic parameters in the time-domain characteristics of the vibration signal and a pre-constructed first function formula;

[0094] Among them, the first function formula is a function formula between the specified characteristic parameters obtained by processing the simulated vibration signals of the fully ceramic bearing obtained by simulation, the real vibration signals of the fully ceramic bearing obtained by experiment, and the preset category label information by a preset conditional Wasserstein generative adversarial network model and the wear amount of the fully ceramic bearing;

[0095] The category label information includes the vibration frequencies of the outer ring and / or the inner ring when the fully ceramic bearing fails under specified working conditions.

[0096] In this embodiment, before the S100, it further includes:

[0097] S000-1. Conduct wear experiments on N fully ceramic bearings under specified working conditions to obtain the real vibration signals of the fully ceramic bearings; where N is a positive integer greater than or equal to 2;

[0098] S000-2. Conduct simulations on the fully ceramic bearings under specified working conditions to obtain the simulated vibration signals and simulated wear amounts of the fully ceramic bearings;

[0099] S000-3. Use the simulated vibration signals, real vibration signals, and the preset category label information to train the conditional Wasserstein generative adversarial network model to obtain a trained conditional Wasserstein generative adversarial network model;

[0100] S000-4. According to the trained conditional Wasserstein generative adversarial network model and the simulated wear amount, obtain the first function formula.

[0101] In this embodiment, the S000-1 specifically includes:

[0102] S000-1-1. Select N fully ceramic bearings as experimental objects, clean and dry each fully ceramic bearing, and then use an electronic balance to measure the mass of each fully ceramic bearing 5 times, and take the average of the 5 measurement values as the initial mass m0;

[0103] In this embodiment, by using an electronic balance to measure the mass of the bearing 5 times in each experimental stage and taking the average, the accidental error is significantly reduced, and the stability and reliability of the mass measurement are improved.

[0104] S000-1-2. Divide the N fully ceramic bearings into two fully ceramic bearing groups, and conduct wear experiments on the two fully ceramic bearing groups respectively under specified working conditions in a fully ceramic bearing life enhancement test machine, specifically including:

[0105] Run each fully ceramic bearing in the first fully ceramic bearing group from the normal state to meet the set conditions under the specified working conditions in the fully ceramic bearing life enhancement test machine to complete the full life experiment, and collect the real vibration signals of the outer ring of the fully ceramic bearing using an acceleration sensor during the full life experiment;

[0106] Among them, the two fully ceramic bearing groups include the first fully ceramic bearing group and the second fully ceramic bearing group; the first fully ceramic bearing group includes M fully ceramic bearings; the second fully ceramic bearing group includes Q fully ceramic bearings; N = M + Q; and M is greater than or equal to 2; Q is greater than or equal to 2;

[0107] Among them, the set condition is that the full life experiment time reaches the theoretical life h0 calculated according to the bearing life formula or the kurtosis of the real vibration signal exceeds 3 times the kurtosis of the real vibration signal in the first time period after the start of the full life experiment;

[0108] In this embodiment, the kurtosis is introduced as the change criterion of the vibration signal to introduce a dynamic termination mechanism, effectively avoiding signal acquisition redundancy and ensuring that the vibration characteristics in the effective wear process are collected.

[0109] After the full life experiment is over, clean and dry the fully ceramic bearings in the first fully ceramic bearing group again, and weigh the mass of each fully ceramic bearing 5 times respectively, and take the average value as the mass m1 of the fully ceramic bearing after wear in the full life experiment, and use the difference between the initial mass m0 and the mass m1 of the fully ceramic bearing after wear in the full life experiment to represent the full wear amount m01;

[0110] Carry out segmented life experiments on each fully ceramic bearing in the second fully ceramic bearing group under the specified working conditions in the fully ceramic bearing life enhancement test machine. The time of each experimental stage is h0 / 5. At the end of each experimental stage, clean, dry and measure the mass of each fully ceramic bearing, and obtain the wear amount of each experimental stage in the previous 4 experimental stages in turn until the cumulative experimental time of the segmented life experiment reaches h0×4 / 5;

[0111] Among them, the wear amount of each experimental stage includes:

[0112] The wear amount ma of the first experimental stage: represents the difference between the initial mass m0 and the average value of the masses of all fully ceramic bearings in the second fully ceramic bearing group after the first experimental stage;

[0113] The wear amount mb of the second experimental stage: represents the difference between the average value of the masses of all fully ceramic bearings in the second fully ceramic bearing group after the first experimental stage and the average value of the masses of all fully ceramic bearings in the second fully ceramic bearing group after the second experimental stage;

[0114] The wear amount mc in the third experimental stage: It represents the difference between the average mass of all the fully ceramic bearings in the second fully ceramic bearing group after the second experimental stage and the average mass of all the fully ceramic bearings in the second fully ceramic bearing group after the third experimental stage;

[0115] The wear amount md in the fourth experimental stage: It represents the difference between the average mass of all the fully ceramic bearings in the second fully ceramic bearing group after the third experimental stage and the average mass of all the fully ceramic bearings in the second fully ceramic bearing group after the fourth experimental stage;

[0116] S000-1-3. Calculate the theoretical maximum contact force and the ball rotation speed of the fully ceramic bearing based on the Hertz contact theory and the kinematic model. Combine with the Archard wear model to calculate the wear coefficient k1 under the complete wear amount m01 and the wear coefficients in each experimental stage under the wear amounts in each experimental stage; among them, the wear coefficient corresponding to the first experimental stage is ka, the wear coefficient corresponding to the second experimental stage is kb, the wear coefficient corresponding to the third experimental stage is kc, and the wear coefficient corresponding to the fourth experimental stage is kd.

[0117] In the practical application of this embodiment, the S000-2 specifically includes:

[0118] S000-2-1. Construct a high-fidelity finite element model of the fully ceramic bearing according to the size parameters of the fully ceramic bearing, and perform mesh division on the high-fidelity finite element model to obtain the mesh division result;

[0119] S000-2-2. Apply the specified working condition conditions in the LSDYNA simulation software, and perform transient dynamics simulation on the mesh division result to obtain the simulation data of the fully ceramic bearing;

[0120] The simulation data includes: the simulated vibration signal of the outer ring of the fully ceramic bearing, the simulated contact force between the balls and the inner and outer rings in the fully ceramic bearing, and the simulated rotation speed of the balls in the fully ceramic bearing;

[0121] S000-2-3. According to the Archard wear model and the wear coefficients in each experimental stage under the wear amounts in each experimental stage, calculate the simulated wear amount of the balls and the inner and outer rings in the fully ceramic bearing within h0 / n time, where n is the total number of simulations; n = 10;

[0122] The high-fidelity finite element model combined with the wear coefficients (ka, kb, kc, kd) enables the simulation process to truly restore the changes in material properties and kinematic behaviors under different wear stages. The wear amount is calculated through the Archard wear model, and the wear coefficients used are from actual experiments, which avoids the deviation caused by the "idealized parameters" in the pure simulation model and makes the simulation have higher credibility and engineering application value.

[0123] S000-2-4. Update the high-fidelity finite element model of the all-ceramic bearing according to the simulated wear amount calculated each time, then perform mesh division on the updated high-fidelity finite element model of the all-ceramic bearing to obtain a new mesh division result, and repeat steps S000-2-2 to S000-2-5 for n times, and obtain the simulated vibration signals of the all-ceramic bearing for n preset time periods; wherein, the preset time period is 0.1 s.

[0124] In this embodiment, the geometric morphology of the finite element model is dynamically updated according to the wear amount in each stage, so that the model gradually evolves from an ideal state to a state closer to the failure state, thereby realizing the coupled analysis of structural wear and dynamic vibration response. Each simulation outputs a vibration signal, and finally 10 sets of simulated vibration signals evolving with wear are obtained.

[0125] Specifically, S000-3 specifically includes:

[0126] S000-3-1. The generator in the conditional Wasserstein generative adversarial network model processes the first data to obtain a processing result of the first data;

[0127] The first data is obtained by performing a Hadamard product operation on the category label information after Embedding processing and the input data; the input data is obtained by superimposing any segment of the simulated vibration signal and the pre-generated random noise;

[0128] S000-3-2. The fourth fully connected layer in the discriminator of the conditional Wasserstein generative adversarial network model processes the category label information after Embedding processing to obtain the processed category label information;

[0129] The discriminant module in the discriminator of the conditional Wasserstein generative adversarial network model performs Hadamard product operations on the processed category label information with the sample data and the processing result of the first data respectively to obtain a first multiplication result and a second multiplication result, and inputs the first multiplication result and the second multiplication result into the fully connected layer unit;

[0130] The sample data includes the real vibration signals of the outer ring of the all-ceramic bearing collected by the acceleration sensor during the complete life experiment of the first M-1 all-ceramic bearings in the first all-ceramic bearing group;

[0131] The fully connected layer unit processes the first multiplication result and the second multiplication result respectively to obtain the corresponding first matrix P r and the second matrix P g ;

[0132] S000-3-3. Repeat steps S000-3-1 to S000-3-2 until the loss function in the conditional Wasserstein generative adversarial network model satisfies the first-order Lipschitz function, then the trained conditional Wasserstein generative adversarial network model is obtained.

[0133] In this embodiment, the loss function in the conditional Wasserstein generative adversarial network model satisfying the first-order Lipschitz function enables the gradient change of the discriminator in the sample space to be controlled, which helps the generator learn a more physically reasonable and detail-rich high-dimensional signal structure.

[0134] In this embodiment, the generator in the conditional Wasserstein generative adversarial network model sequentially includes: a first fully connected layer, a second fully connected layer, a third fully connected layer, and an output fully connected layer;

[0135] Among them, the number of neurons in the first fully connected layer is 256; the number of neurons in the second fully connected layer is 512 and includes regularization processing; the number of neurons in the third fully connected layer is 1024 and includes regularization processing. In this embodiment, the number of neurons in the first fully connected layer, the second fully connected layer, and the third fully connected layer in the generator of the conditional Wasserstein generative adversarial network model gradually increases, so that the feature expression ability is gradually improved. The regularization processing in this embodiment can inhibit the over-reliance of some neurons during training, avoid the memory of specific noise features during the training stage, keep the generator still having the ability to generate credible samples when facing new inputs, and improve its generalization ability and generation diversity.

[0136] Among them, the LeakyReLU activation function is used in the first fully connected layer, the second fully connected layer, and the third fully connected layer to prevent neuron death;

[0137] The number of neurons in the output fully connected layer is 784, and the output result of the output fully connected layer is reconstructed into a tensor with a size of 28*1.

[0138] In this embodiment, by designing the generator as a fully connected network with three layers gradually widened, combined with regularization means and the LeakyReLU activation function, the learning and restoration ability of the generator for complex bearing vibration signal features is effectively improved. At the same time, the output layer outputs a size of 784 and is reconstructed into a 28×1 tensor, so that the generated samples are highly consistent with the real samples in structure and can be directly applied to subsequent fault diagnosis and intelligent prediction tasks, with strong applicability, scalability, and practical engineering application value.

[0139] In this embodiment, the fully connected layer unit includes: Among them, the fully connected layer unit includes a fifth fully connected layer connected in sequence;

[0140] Among them, the number of neurons in each fifth fully connected layer is 512, and each includes regularization processing; the LeakyReLU activation function is used in each fifth fully connected layer to prevent neuron death.

[0141] In this embodiment, by setting multiple fifth fully connected layers with 512 neurons and introducing regularization processing and the LeakyReLU activation function in each layer, the feature extraction ability, discrimination accuracy, and training stability of the discriminator can be effectively improved. Among them, the regularization processing can suppress the overfitting risk of the discriminator and improve the generalization ability of the network; the LeakyReLU activation function prevents the phenomenon of neuron death and maintains the gradient flow of each layer of the network, which helps the model to complete the discrimination task stably and quickly when processing complex bearing vibration signals, thereby providing a more reliable optimization direction for the generator and improving the training efficiency and generation quality of the overall conditional Wasserstein generative adversarial network model.

[0142] Among them, the loss function is: L(G, D) = W(P r , P g ) - GP;

[0143] GP is a preset penalty function used to enforce the first-order Lipschitz function. Constraining the gradient norm of the discriminator (about 1) can naturally achieve the Lipschitz condition and improve the theoretical rationality and actual training performance. The GP term controls the network learning rate by constraining the gradient norm ≈ 1 at the interpolation samples between the real samples and the generated samples as the discriminator input; it can prevent the situation of too large or zero gradients and significantly improve the robustness and training stability of the network.

[0144] W(P r , P g ) is the Wasserstein distance between the first matrix P r and the second matrix P g .

[0145] In this embodiment, by designing a loss function with the Wasserstein distance and the gradient penalty term as the core, the training stability and generation effect of the conditional Wasserstein generative adversarial network can be significantly improved. The Wasserstein distance can provide smooth and non-zero gradient information, effectively alleviating the mode collapse and gradient disappearance problems in the training of traditional adversarial networks and ensuring that the difference between the generated samples and the real data distribution can be accurately reflected; the introduced gradient penalty term can enforce the discriminator to satisfy the first-order Lipschitz constraint condition, further ensuring the theoretical convergence and model stability. The design of this loss function enables the generator to effectively learn the deep structural features in the simulated vibration signals.

[0146] Specifically, the S000-4 specifically includes:

[0147] S000-4-1. According to the trained conditional Wasserstein generative adversarial network model, obtain the complete simulated vibration signal generated by the trained conditional Wasserstein generative adversarial network model;

[0148] S000-4-2. According to the complete simulated vibration signal, obtain the characteristic parameters in the time domain characteristics of the complete simulated vibration signal, and perform time fitting on each characteristic parameter to respectively obtain the function and image of each characteristic parameter with respect to time;

[0149] Among them, the characteristic parameters include: the mean, variance, root mean square, peak value, kurtosis, waveform factor, and impulse factor of the simulated vibration signal in the time domain;

[0150] S000-4-3. Perform curve fitting on the simulated wear amount to obtain the function and image of the wear amount with respect to time;

[0151] S000-4-4. Align the function of each characteristic parameter with respect to time with the function of the wear amount with respect to time in time to obtain the mapping relationship between the characteristic parameter and the wear amount, and perform curve fitting on the characteristic parameter and the wear amount to obtain the function formula between the characteristic parameter and the wear amount;

[0152] S000-4-5. Multiple time points pre-screened during the complete life experiment of the last full-ceramic bearing in the first full-ceramic bearing group, and according to the true wear amount during the complete life experiment corresponding to each of the selected time points for each characteristic parameter and the function formula between each characteristic parameter and the wear amount, determine the first function formula.

[0153] In this embodiment, through the specific steps in S000-4, typical time domain characteristic parameters can be extracted based on the simulated vibration signal, and function fitting and mapping relationship modeling are carried out in combination with the wear evolution process to realize the indirect inference from the characteristic parameters to the bearing wear amount. Compared with the traditional method that relies on a complex sensor network or direct wear measurement, this method has stronger universality and engineering adaptability, can significantly reduce the cost of bearing life tests, and at the same time improve the efficiency and accuracy of wear assessment and life prediction.

[0154] In this embodiment, the S000-4-5 includes:

[0155] S000-4-5-1. According to the true wear amount during the complete life experiment corresponding to each selected time point for each characteristic parameter, and the predicted value of the wear amount corresponding to each selected time point for this characteristic parameter, use formula (1) to obtain the cumulative error amount corresponding to this characteristic parameter;

[0156] The said formula (1) is:

[0157] Δ i =Δt i1 +…+Δt ih +…Δt iH ;

[0158] Δ i is the cumulative error amount corresponding to the i-th characteristic parameter;

[0159] Δt ih is the absolute value of the difference between the true wear amount and the predicted value of the wear amount during the complete life experiment corresponding to the h-th time point selected for the i-th characteristic parameter;

[0160] wherein, the predicted value of the wear amount corresponding to the h-th time point selected for the i-th characteristic parameter during the complete life experiment is calculated through the function formula between this characteristic parameter and the wear amount;

[0161] H is the total number of pre-selected time points, where H>1;

[0162] S000-4-5-2. Take the characteristic parameter corresponding to the smallest cumulative error amount as the specified characteristic parameter, and take the function formula between this specified characteristic parameter and the wear amount as the first function formula.

[0163] In this embodiment, through the cumulative error evaluation mechanism between the true wear amount at multiple pre-selected time points and the wear amount predicted by each characteristic parameter, the characteristic parameter and the wear function formula with the optimal prediction performance are systematically screened out to determine the first function formula. This method realizes the objective and quantitative comparison of multiple candidate characteristics and functions, avoids the subjective deviation in the model establishment process, and significantly improves the accuracy, stability and generalization ability of the prediction model.

[0164] In this embodiment, the multiple pre-selected time points include: the first time point, the second time point, and the third time point; the first time point is the moment at 40% of the complete life experiment time; the second time point is the moment at 60% of the complete life experiment time; the third time point is the moment at 80% of the complete life experiment time.

[0165] In this embodiment, during the complete life cycle of the bearing, it generally goes through three stages: the early running-in stage, the intermediate stable wear stage, and the late stage with accelerated deterioration. Selecting the 40% time point can better represent the early stage, the 60% time point is in the intermediate stage, and the 80% time point represents the late stage when wear enters an obvious acceleration phase. Through these three time points, the change trend of the characteristic parameters of the vibration signal during the entire wear process can be effectively captured, ensuring that the relationship established between the parameters and the wear amount can cover the entire process from minor wear to severe deterioration. Selecting 40%, 60%, and 80% of the complete life experiment as the key time points can effectively cover the entire wear process, improve the applicability of the characteristic function modeling, reduce the data acquisition cost, and at the same time enhance the credibility of the error assessment.

[0166] In addition, this embodiment also provides a system for identifying the wear state of a fully ceramic bearing, including: at least one processor; and at least one memory communicatively connected to the processor, wherein the memory stores program instructions executable by the processor, and the processor can execute the method for identifying the wear state of a fully ceramic bearing as described in Embodiment 1 by invoking the program instructions.

[0167] Embodiment 2

[0168] This embodiment provides a method for identifying the wear state of a fully ceramic bearing, including:

[0169] Step 001: Select 8 fully ceramic bearings as the experimental objects, clean and dry each fully ceramic bearing, and then use an electronic balance to measure the mass of each fully ceramic bearing 5 times, and take the average value of the 5 measurement values as the initial mass m0; for example, the initial mass m0 in this embodiment is 65.2375 g.

[0170] In this embodiment, the measurement error is effectively reduced by taking the average value of five weighings of the fully ceramic bearing, and the mass detection is carried out respectively in the complete life and segmented life stages, effectively improving the evaluation accuracy of the bearing wear amount.

[0171] Step 002: Divide the 8 fully ceramic bearings into two fully ceramic bearing groups, and conduct wear experiments on the two fully ceramic bearing groups respectively under specified working conditions in a fully ceramic bearing life strengthening test machine, specifically including:

[0172] Run each fully ceramic bearing in the first fully ceramic bearing group from the normal state to meet the set conditions under the specified working conditions in the fully ceramic bearing life strengthening test machine to complete the complete life experiment, and collect the real vibration signals of the outer ring of the fully ceramic bearing using an acceleration sensor during the complete life experiment;

[0173] When conducting wear experiments on a fully ceramic bearing life enhancement testing machine, the setting of specified working conditions is crucial for the reliability and repeatability of the experiments. Wear experiments are carried out on two fully ceramic bearing groups under the same specified working conditions (4000N load, 5000rpm rotation speed) to evaluate the wear resistance, life performance, and failure mechanism of the fully ceramic bearings.

[0174] In this embodiment, the kurtosis of the vibration signal is used as the judgment basis to set dynamic termination conditions, avoiding the acquisition of a large amount of redundant vibration data, thereby improving the data processing efficiency and focusing on the key feature information in the significant bearing wear stage.

[0175] Among them, the two fully ceramic bearing groups include the first fully ceramic bearing group and the second fully ceramic bearing group; the first fully ceramic bearing group includes 4 fully ceramic bearings; the second fully ceramic bearing group includes 4 fully ceramic bearings;

[0176] Among them, the set conditions are that the complete life experiment time reaches the theoretical life h0 calculated according to the bearing life formula or the kurtosis of the real vibration signal during the complete life experiment exceeds 3 times the kurtosis of the real vibration signal in the first time period after the start of the experiment;

[0177] For example, in the actual calculation process, the theoretical life h0 of the fully ceramic bearings in this embodiment is 155.52h.

[0178] In this embodiment, a dynamic termination mechanism is introduced by using kurtosis as the change criterion of the vibration signal, effectively avoiding redundant signal acquisition and ensuring that the vibration characteristics in the effective wear process are collected.

[0179] After the complete life experiment, the fully ceramic bearings in the first fully ceramic bearing group are cleaned and dried again, and the mass of each fully ceramic bearing is weighed 5 times, and the average value is taken as the mass m1 of the fully ceramic bearing after wear in the complete life experiment. The difference between the initial mass m0 and the mass m1 of the fully ceramic bearing after wear in the complete life experiment is used to represent the complete wear amount m01; in this embodiment, the complete wear amount m01 obtained after the complete life experiment is 0.05g.

[0180] Each fully ceramic bearing in the second fully ceramic bearing group is respectively subjected to a segmented life experiment under the specified working conditions in the fully ceramic bearing life enhancement testing machine. The time of each experimental stage is h0 / 5. At the end of each experimental stage, the fully ceramic bearings are cleaned, dried, and their masses are measured, and the wear amounts in the previous 4 experimental stages are obtained in turn until the cumulative experimental time of the segmented life experiment reaches h0×4 / 5;

[0181] Among them, the wear amounts in each experimental stage include:

[0182] The wear amount ma in the first experimental stage: It represents the difference between the initial mass m0 and the average mass of all the all-ceramic bearings in the second all-ceramic bearing group after the first experimental stage; in this embodiment, the wear amount ma in the first experimental stage is 0.005 g.

[0183] The wear amount mb in the second experimental stage: It represents the difference between the average mass of all the all-ceramic bearings in the second all-ceramic bearing group after the first experimental stage and the average mass of all the all-ceramic bearings in the second all-ceramic bearing group after the second experimental stage; in this embodiment, the wear amount mb in the second experimental stage is 0.013 g.

[0184] The wear amount mc in the third experimental stage: It represents the difference between the average mass of all the all-ceramic bearings in the second all-ceramic bearing group after the second experimental stage and the average mass of all the all-ceramic bearings in the second all-ceramic bearing group after the third experimental stage; in this embodiment, the wear amount mc in the third experimental stage is 0.023 g.

[0185] The wear amount md in the fourth experimental stage: It represents the difference between the average mass of all the all-ceramic bearings in the second all-ceramic bearing group after the third experimental stage and the average mass of all the all-ceramic bearings in the second all-ceramic bearing group after the fourth experimental stage; in this embodiment, the wear amount ma in the fourth experimental stage is 0.037 g.

[0186] Step 003: Calculate the theoretical maximum contact force and the ball rotation speed of the all-ceramic bearing based on the Hertz contact theory and the kinematic model, and combine with the Archard wear model to calculate the wear coefficient k1 under the complete wear amount m01 and the wear coefficients in each experimental stage under the wear amounts in each experimental stage; among them, the wear coefficient corresponding to the first experimental stage is ka, the wear coefficient corresponding to the second experimental stage is kb, the wear coefficient corresponding to the third experimental stage is kc, and the wear coefficient corresponding to the fourth experimental stage is kd.

[0187] Step 004: Construct a high-fidelity finite element model of the all-ceramic bearing according to the size parameters of the all-ceramic bearing, and perform mesh division on the high-fidelity finite element model to obtain the mesh division result;

[0188] Step 005: Apply the specified working condition conditions in the LSDYNA simulation software, perform transient dynamic simulation on the mesh division result, and obtain the simulation data of the all-ceramic bearing;

[0189] The simulation data includes: the simulated vibration signal of the outer ring of the all-ceramic bearing, the simulated contact force between the balls and the inner and outer rings in the all-ceramic bearing, and the simulated rotation speed of the balls in the all-ceramic bearing;

[0190] Figure 3It is an image of the relationship between the simulated contact force and time of the ball and the outer raceway in the all-ceramic bearing. In this embodiment, the simulated rotational speed of the ball in the all-ceramic bearing is reflected by the relationship between the simulated linear velocity and time of the ball in the all-ceramic bearing (see Figure 2 ), where the linear velocity refers to the velocity of a point on an object moving along the circumferential direction; and the rotational speed represents the number of rotations completed by an object per unit time. The linear velocity is equal to the radius of rotation of the object multiplied by the angle of rotation per second (angular velocity), and can also be expressed as: the linear velocity is proportional to the rotational speed and also proportional to the radius of rotation.

[0191] Step 006: Calculate the simulated wear amount of the ball and the inner and outer rings in the all-ceramic bearing within h0 / n time according to the Archard wear model and the wear coefficients in each experimental stage under the wear amounts in each experimental stage, where n is the total number of simulations; n = 10;

[0192] The high-fidelity finite element model combined with the wear coefficients (ka, kb, kc, kd) enables the simulation process to truly reproduce the changes in material properties and kinematic behaviors under different wear stages. The wear amount is calculated through the Archard wear model, and the used wear coefficients are derived from actual experiments, which avoids the deviation caused by the "idealized parameters" in the pure simulation model and makes the simulation have higher credibility and engineering application value. A total of 10 groups of high-quality vibration signals reflecting different wear states are obtained, providing sufficient and balanced data support for subsequent machine learning modeling.

[0193] Step 007: Update the high-fidelity finite element model of the constructed all-ceramic bearing according to the simulated wear amount calculated each time (see Figure 4 ), then perform mesh division on the updated high-fidelity finite element model of the all-ceramic bearing to obtain a new mesh division result, and repeat steps 005 to 007 10 times, and obtain the simulated vibration signals of the all-ceramic bearing for 10 preset time periods; where the preset time period is 0.1 s.

[0194] In this embodiment, the geometric shape of the finite element model is dynamically updated according to the wear amount in each stage, so that the model gradually evolves from an ideal state to a state closer to the failure state, thereby realizing the coupled analysis of structural wear and dynamic vibration response. Each simulation outputs a vibration signal, and finally 10 groups of simulated vibration signals evolving with wear are obtained.

[0195] Step 008: The generator in the pre-set Wasserstein generative adversarial network model processes the first data to obtain the processing result of the first data;

[0196] The first data is obtained by performing a Hadamard product operation on the class label information after Embedding processing and the input data; the input data is obtained by superimposing any segment of simulated vibration signal and pre-generated random noise.

[0197] In this embodiment, the Embedding processing is a process of converting discrete labels into continuous vectors, which is convenient for deep learning models to identify, fuse, and utilize class information, and is one of the very core operations in the conditional generation model.

[0198] In this embodiment, a conditional Wasserstein generative adversarial network model is used to model and generate vibration signals. By introducing class label information through Embedding processing, the network has conditional constraint ability when generating signals in different wear stages, thereby enhancing the discriminant accuracy of the model in identifying wear states.

[0199] See Figure 5 , in this embodiment, the generator in the conditional Wasserstein generative adversarial network model sequentially includes: a first fully connected layer, a second fully connected layer, a third fully connected layer, and an output fully connected layer.

[0200] Among them, the number of neurons in the first fully connected layer is 256; the number of neurons in the second fully connected layer is 512 and includes regularization processing; the number of neurons in the third fully connected layer is 1024 and includes regularization processing. In this embodiment, the number of neurons in the first fully connected layer, the second fully connected layer, and the third fully connected layer in the generator of the conditional Wasserstein generative adversarial network model gradually increases, so that the feature expression ability is gradually improved. The regularization processing in this embodiment can inhibit the over-dependence of some neurons during training, avoid the memory of specific noise features during the training stage, keep the generator still capable of generating credible samples when facing new inputs, and improve its generalization ability and generation diversity.

[0201] Among them, the LeakyReLU activation function is used in the first fully connected layer, the second fully connected layer, and the third fully connected layer to prevent neuron death.

[0202] The number of neurons in the output fully connected layer is 784, and the output result of the output fully connected layer is reconstructed into a tensor with a size of 28*1.

[0203] In this embodiment, by designing the generator as a fully connected network with three layers gradually widening, combining regularization means and the LeakyReLU activation function, the learning and restoration ability of the generator for the characteristics of complex bearing vibration signals is effectively improved. At the same time, the output layer has an output size of 784 and is reconstructed into a 28×1 tensor, making the generated samples highly consistent with the real samples in structure, which can be directly applied to subsequent fault diagnosis and intelligent prediction tasks, and has strong applicability, scalability and practical engineering application value.

[0204] Step 009: The fourth fully connected layer in the discriminator of the conditional Wasserstein generative adversarial network model processes the class label information after Embedding processing to obtain the processed class label information.

[0205] The discriminant module in the discriminator of the conditional Wasserstein generative adversarial network model performs Hadamard product operations on the processed class label information with the sample data and the processing result of the first data respectively to obtain the first multiplication result and the second multiplication result, and inputs the first multiplication result and the second multiplication result into the fully connected layer unit.

[0206] The sample data includes the real vibration signals of the outer rings of the first 3 full-ceramic bearings in the first full-ceramic bearing group collected by an acceleration sensor during the complete life experiment.

[0207] The fully connected layer unit processes the first multiplication result and the second multiplication result respectively to obtain the corresponding first matrix P r and the second matrix P g ;

[0208] Step 010: Repeat steps 008 to 010 until the loss function in the conditional Wasserstein generative adversarial network model satisfies the 1st-order Lipschitz function, then the trained conditional Wasserstein generative adversarial network model is obtained.

[0209] In this embodiment, the loss function in the conditional Wasserstein generative adversarial network model satisfying the 1st-order Lipschitz function makes the gradient change of the discriminator in the sample space controlled, which helps the generator learn a more physically reasonable and detail-rich high-dimensional signal structure.

[0210] In this embodiment, the fully connected layer unit includes: Among them, the fully connected layer unit includes a fifth fully connected layer connected in sequence;

[0211] Among them, the number of neurons in each fifth fully connected layer is 512, and all of them include regularization processing; the LeakyReLU activation function is used in each fifth fully connected layer to prevent neuron death.

[0212] In this embodiment, by setting multiple fifth fully-connected layers with 512 neurons and introducing regularization processing and the LeakyReLU activation function in each layer, the feature extraction ability, discrimination accuracy, and training stability of the discriminator can be effectively improved. Among them, the regularization processing can suppress the overfitting risk of the discriminator and improve the generalization ability of the network; the LeakyReLU activation function prevents the phenomenon of neuron death and maintains the gradient flow of each layer of the network, which helps the model to complete the discrimination task stably and quickly when processing complex bearing vibration signals, thereby providing a more reliable optimization direction for the generator and improving the training efficiency and generation quality of the overall conditional Wasserstein generative adversarial network model.

[0213] Among them, the loss function is: L(G, D) = W(P r , P g ) - GP;

[0214] GP is a preset penalty function used to enforce the first-order Lipschitz function.

[0215] W(P r , P g ) is the Wasserstein distance between the first matrix P r and the second matrix P g .

[0216] In this embodiment, by designing a loss function with the Wasserstein distance and the gradient penalty term as the core, the training stability and generation effect of the conditional Wasserstein generative adversarial network can be significantly improved. The Wasserstein distance can provide smooth and non-zero gradient information, effectively alleviating the mode collapse and gradient disappearance problems in the training of traditional adversarial networks, and ensuring that the difference between the generated samples and the real data distribution can be accurately reflected; the introduced gradient penalty term can enforce the discriminator to satisfy the first-order Lipschitz constraint condition, further ensuring the theoretical convergence and the stability of the model. This loss function design enables the generator to effectively learn the deep structural features in the simulated vibration signals.

[0217] Step 011: According to the trained conditional Wasserstein generative adversarial network model, obtain the complete simulated vibration signals generated by the trained conditional Wasserstein generative adversarial network model;

[0218] Step 012: According to the complete simulated vibration signals, obtain the characteristic parameters in the time domain characteristics of the complete simulated vibration signals, and perform time fitting on each characteristic parameter to obtain the function and image of each characteristic parameter with respect to time respectively;

[0219] Among them, the characteristic parameters include: the mean value, variance, root mean square, peak value, kurtosis, waveform factor, and impulse factor of the simulated vibration signal in the time domain;

[0220] Step 013: Curve fit the simulated wear amount to obtain the function and image of wear amount versus time;

[0221] Step 014: Align the function of each characteristic parameter versus time with the function of wear amount versus time in terms of time to obtain the mapping relationship between the characteristic parameter and the wear amount, and curve fit the characteristic parameter and the wear amount to obtain the function formula between the characteristic parameter and the wear amount;

[0222] Step 015: Multiple time points pre-screened during the complete life experiment of the last full-ceramic bearing in the first full-ceramic bearing group, and determine the first function formula according to the true wear amount during the complete life experiment corresponding to each screened time point for each characteristic parameter and the function formula between each characteristic parameter and the wear amount.

[0223] In this embodiment, through the specific steps in S000-4, typical time-domain characteristic parameters can be extracted based on the simulated vibration signal, and function fitting and mapping relationship modeling can be carried out in combination with the wear evolution process to realize the indirect inference from the characteristic parameters to the bearing wear amount. Compared with the traditional methods that rely on complex sensor networks or direct wear measurement, this method has stronger universality and engineering adaptability, can significantly reduce the cost of bearing life tests, and at the same time improve the efficiency and accuracy of wear assessment and life prediction.

[0224] In this embodiment, the step 015 includes:

[0225] Step 015-1: According to the true wear amount during the complete life experiment corresponding to each screened time point for each characteristic parameter, and the predicted value of the wear amount corresponding to each screened time point for this characteristic parameter, use formula (1) to obtain the cumulative error amount corresponding to this characteristic parameter;

[0226] The formula (1) is:

[0227] Δ i =Δt i1 +…+Δt ih +…Δt iH ;

[0228] Δ i is the cumulative error amount corresponding to the i-th characteristic parameter;

[0229] Δt ihis the absolute value of the difference between the true wear amount and the predicted value of the wear amount corresponding to the selected h-th time point for the i-th characteristic parameter during the complete life experiment;

[0230] wherein, the predicted value of the wear amount corresponding to the selected h-th time point for the i-th characteristic parameter during the complete life experiment is calculated by a function formula between this characteristic parameter and the wear amount; H is the total number of pre-selected time points, where H is greater than 1;

[0231] Step 015-2: Take the characteristic parameter corresponding to the minimum cumulative error amount as the specified characteristic parameter, and take the function formula between this specified characteristic parameter and the wear amount as the first function formula.

[0232] Step 016: Obtain the vibration signal of the outer ring of the all-ceramic bearing during the operation of the all-ceramic bearing to be identified under the specified working conditions;

[0233] Step 017: According to the vibration signal, obtain the specified characteristic parameter in the time domain characteristics of the vibration signal;

[0234] Step 018: According to the specified characteristic parameter in the time domain characteristics of the vibration signal and the pre-constructed first function formula, obtain the predicted value of the wear amount of the all-ceramic bearing to be identified;

[0235] wherein, the first function formula is a function formula between the specified characteristic parameter obtained by processing the simulated vibration signal obtained by simulating the all-ceramic bearing, the true vibration signal of the all-ceramic bearing obtained by experiment, and the pre-set class label information by the pre-set conditional Wasserstein generative adversarial network model and the wear amount of the all-ceramic bearing;

[0236] The class label information includes the vibration frequencies of the outer ring and / or inner ring when the all-ceramic bearing fails under the specified working conditions.

[0237] In a method for identifying the wear state of an all-ceramic bearing in this embodiment, since the process of obtaining the true vibration signal and the simulated vibration signal of multiple all-ceramic bearings is introduced before the formal identification step, and the conditional Wasserstein generative adversarial network model is trained by combining experiment and simulation, compared with the prior art, it can improve the generalization ability and stability of the identification model, achieving the effect of constructing a wear identification model closer to the actual working conditions.

[0238] In the description of the present invention, it should be understood that the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present invention, "a plurality of" means two or more unless otherwise specifically defined.

[0239] In the present invention, unless otherwise clearly defined and limited, the terms "mounted", "connected", "coupled", "fixed", etc. shall be construed in a broad sense. For example, it may be a fixed connection, a detachable connection, or integrated; it may be a mechanical connection or an electrical connection; it may be directly connected or indirectly connected through an intermediate medium; it may be the internal communication of two components or the interaction relationship between two components. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0240] In the present invention, unless otherwise clearly defined and limited, when the first feature is "on" or "under" the second feature, it may be that the first and second features are in direct contact, or the first and second features are in indirect contact through an intermediate medium. Moreover, when the first feature is "above", "over" and "on top of" the second feature, it may be that the first feature is directly above or obliquely above the second feature, or merely indicates that the first feature has a higher horizontal height than the second feature. When the first feature is "under", "beneath" and "underneath" the second feature, it may be that the first feature is directly below or obliquely below the second feature, or merely indicates that the first feature has a lower horizontal height than the second feature.

[0241] In the description of this specification, the descriptions of the terms "one embodiment", "some embodiments", "embodiment", "example", "specific example" or "some examples", etc. mean that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in a suitable manner in any one or more embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0242] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for identifying the wear state of a fully ceramic bearing, characterized in that, Including: S100. Obtain the vibration signal of the outer ring of the all-ceramic bearing during operation under specified working conditions of the all-ceramic bearing to be identified; S200. Obtain the specified characteristic parameters in the time-domain characteristics of the vibration signal according to the vibration signal; S300. Obtain the predicted wear amount of the all-ceramic bearing to be identified according to the specified characteristic parameters in the time-domain characteristics of the vibration signal and the pre-constructed first function formula; Wherein, the first function formula is a function formula between the specified characteristic parameters obtained by processing the simulated vibration signal obtained by simulating the all-ceramic bearing, the real vibration signal of the all-ceramic bearing obtained by experiment, and the pre-set class label information by the pre-set conditional Wasserstein generative adversarial network model and the wear amount of the all-ceramic bearing; The class label information includes the vibration frequencies of the outer ring and / or inner ring of the all-ceramic bearing during failure under specified working conditions.

2. The method for identifying the wear state of an all-ceramic bearing according to claim 1, wherein, Before the S100, it further includes: S000-1. Conduct wear experiments on N all-ceramic bearings under specified working conditions to obtain the real vibration signals of the all-ceramic bearings; Wherein, N is a positive integer greater than or equal to 2; S000-2. Simulate the all-ceramic bearing under specified working conditions to obtain the simulated vibration signal and simulated wear amount of the all-ceramic bearing; S000-3. Train the conditional Wasserstein generative adversarial network model using the simulated vibration signal, real vibration signal, and pre-set class label information to obtain the trained conditional Wasserstein generative adversarial network model; S000-4. Obtain the first function formula according to the trained conditional Wasserstein generative adversarial network model and the simulated wear amount.

3. The method for identifying the wear state of the all-ceramic bearing according to claim 2, characterized in that, The S000-1 specifically includes: S000-1-1. Select N all-ceramic bearings as experimental objects, clean and dry each all-ceramic bearing, then use an electronic balance to measure the mass of each all-ceramic bearing 5 times, and take the average value of the 5 measurement values as the initial mass m0; S000-1-2. Divide the N all-ceramic bearings into two all-ceramic bearing groups, and conduct wear experiments on the two all-ceramic bearing groups respectively under specified working conditions in an all-ceramic bearing life enhancement test machine, specifically including: Operate each all-ceramic bearing in the first all-ceramic bearing group from the normal state to meet the set conditions to complete a complete life experiment under the specified working conditions in the all-ceramic bearing life enhancement test machine, and collect the real vibration signal of the outer ring of the all-ceramic bearing using an acceleration sensor during the complete life experiment; Wherein, the two all-ceramic bearing groups include the first all-ceramic bearing group and the second all-ceramic bearing group; the first all-ceramic bearing group includes M all-ceramic bearings; the second all-ceramic bearing group includes Q all-ceramic bearings; N = M + Q; and M is greater than or equal to 2; Q is greater than or equal to 2; Wherein, the set condition is that the complete life test time reaches the theoretical life h0 calculated according to the bearing life formula or the kurtosis of the true vibration signal during the complete life test exceeds 3 times the kurtosis of the true vibration signal in the first time period after the start of the experiment; After the complete life test, the all-ceramic bearings in the first all-ceramic bearing group are cleaned and dried again, and the mass of each all-ceramic bearing is weighed 5 times respectively, and the average value is taken as the mass m1 of the all-ceramic bearing after wear in the complete life test. The complete wear amount m01 is represented by the difference between the initial mass m0 and the mass m1 of the all-ceramic bearing after wear in the complete life test; Each all-ceramic bearing in the second all-ceramic bearing group is respectively subjected to a segmented life test under specified working conditions in an all-ceramic bearing life enhancement test machine. The time of each test stage is h0 / 5. At the end of each test stage, each all-ceramic bearing is cleaned, dried and its mass is measured, and the wear amounts in the previous 4 test stages are obtained in sequence until the cumulative test time of the segmented life test reaches h0×4 / 5; Wherein, the wear amounts in each test stage include: The wear amount ma in the first test stage: It represents the difference between the initial mass m0 and the average value of the masses of all the all-ceramic bearings in the second all-ceramic bearing group after the first test stage; The wear amount mb in the second test stage: It represents the difference between the average value of the masses of all the all-ceramic bearings in the second all-ceramic bearing group after the first test stage and the average value of the masses of all the all-ceramic bearings in the second all-ceramic bearing group after the second test stage; The wear amount mc in the third test stage: It represents the difference between the average value of the masses of all the all-ceramic bearings in the second all-ceramic bearing group after the second test stage and the average value of the masses of all the all-ceramic bearings in the second all-ceramic bearing group after the third test stage; The wear amount md in the fourth test stage: It represents the difference between the average value of the masses of all the all-ceramic bearings in the second all-ceramic bearing group after the third test stage and the average value of the masses of all the all-ceramic bearings in the second all-ceramic bearing group after the fourth test stage; S000-1-3: Calculate the theoretical maximum contact force and the ball rotation speed of the all-ceramic bearing based on the Hertz contact theory and the kinematic model, and combine with the Archard wear model to calculate the wear coefficient k1 under the complete wear amount m01 and the wear coefficients in each test stage under the wear amounts in each test stage respectively; Wherein, the wear coefficient corresponding to the first test stage is ka, the wear coefficient corresponding to the second test stage is kb, the wear coefficient corresponding to the third test stage is kc, and the wear coefficient corresponding to the fourth test stage is kd.

4. The method for identifying the wear state of the all-ceramic bearing according to claim 3, wherein The specific content of S000-2 includes: S000-2-1: Construct a high-fidelity finite element model of the all-ceramic bearing according to the dimensional parameters of the all-ceramic bearing, and perform mesh division on the high-fidelity finite element model to obtain the mesh division result; S000-2-2: Apply the specified working condition conditions in the LSDYNA simulation software, and perform transient dynamics simulation on the mesh division result to obtain the simulation data of the all-ceramic bearing; The simulation data includes: the simulated vibration signal of the outer ring of the all-ceramic bearing, the simulated contact force between the balls and the inner and outer rings in the all-ceramic bearing, and the simulated rotational speed of the balls in the all-ceramic bearing; S000-2-3. Calculate the simulated wear amount of the balls and the inner and outer rings in the all-ceramic bearing within h0 / n time according to the Archard wear model and the wear coefficients in each experimental stage under the wear amounts in each experimental stage, where n is the total number of simulations; n = 10; S000-2-4. Update the high-fidelity finite element model of the all-ceramic bearing constructed according to the simulated wear amount obtained each time, and then perform mesh division on the updated high-fidelity finite element model of the all-ceramic bearing to obtain a new mesh division result, and repeat steps S000-2-2 to S000-2-5 for n times, and obtain the simulated vibration signals of the all-ceramic bearing in n preset time periods; Among them, the preset time period is 0.1 s.

5. The method for identifying the wear state of an all-ceramic bearing according to claim 4, characterized in that, The specific steps of S000-3 are as follows: S000-3-1. The generator in the conditional Wasserstein generative adversarial network model processes the first data to obtain the processing result of the first data; The first data is obtained by performing a Hadamard product operation on the class label information after Embedding processing and the input data; the input data is obtained by superimposing any segment of the simulated vibration signal and the pre-generated random noise; S000-3-2. The fourth fully connected layer in the discriminator of the conditional Wasserstein generative adversarial network model processes the class label information after Embedding processing to obtain the processed class label information; The discriminant module in the discriminator of the conditional Wasserstein generative adversarial network model performs Hadamard product operations on the processed class label information with the sample data and the processing result of the first data respectively to obtain a first multiplication result and a second multiplication result, and inputs the first multiplication result and the second multiplication result into the fully connected layer unit; The sample data includes the real vibration signals of the outer rings of the first M - 1 all-ceramic bearings in the first all-ceramic bearing group collected by an acceleration sensor during the complete life experiment; The fully connected layer units process the first multiplication result and the second multiplication result respectively to obtain corresponding first matrix P r and second matrix P g ; S000-3-3. Repeat steps S000-3-1 to S000-3-2 until the loss function in the conditional Wasserstein generative adversarial network model satisfies the first-order Lipschitz function, then the trained conditional Wasserstein generative adversarial network model is obtained.

6. The method for identifying the wear state of an all-ceramic bearing according to claim 5, characterized in that The generator in the conditional Wasserstein generative adversarial network model sequentially includes: a first fully connected layer, a second fully connected layer, a third fully connected layer, and an output fully connected layer; Among them, the number of neurons in the first fully connected layer is 256; the number of neurons in the second fully connected layer is 512 and includes regularization processing; the number of neurons in the third fully connected layer is 1024 and includes regularization processing; Among them, the LeakyReLU activation function is used in the first fully connected layer, the second fully connected layer, and the third fully connected layer to prevent neuron death; The number of neurons in the output fully connected layer is 784, and the output result of the output fully connected layer is reconstructed into a tensor with a size of 28 * 1.

7. The full-ceramic bearing wear state recognition method according to claim 5, characterized in that The fully connected layer unit includes: Among them, the fully connected layer unit includes a fifth fully connected layer connected in sequence; Among them, the number of neurons in each fifth fully connected layer is 512, and each includes regularization processing; the LeakyReLU activation function is used in each fifth fully connected layer to prevent neuron death.

8. The full-ceramic bearing wear state recognition method according to claim 5, characterized in that Among them, The loss function is: L(G, D) = W(P r , P g ) - GP; GP is a preset penalty function; W(P r , P g ) is the Wasserstein distance between the first matrix P r and the second matrix P g .

9. The method for identifying the wear state of an all-ceramic bearing according to claim 6, characterized in that, The S000-4 specifically includes: S000-4-1. According to the trained conditional Wasserstein generative adversarial network model, obtain the complete simulated vibration signal generated by the trained conditional Wasserstein generative adversarial network model; S000-4-2. According to the complete simulated vibration signal, obtain the characteristic parameters in the time domain features of the complete simulated vibration signal, and perform time fitting on each characteristic parameter to respectively obtain the function and image of each characteristic parameter with time; Among them, the characteristic parameters include: the mean value, variance, root mean square, peak value, kurtosis, waveform factor, and pulse factor of the simulated vibration signal in the time domain; S000-4-3. Perform curve fitting on the simulated wear amount to obtain the function and image of the wear amount with time; S000-4-4. Align the function of each characteristic parameter with time and the function of the wear amount with time in time to obtain the mapping relationship between the characteristic parameter and the wear amount, and perform curve fitting on the characteristic parameter and the wear amount to obtain the function formula between the characteristic parameter and the wear amount; S000-4-5. Multiple time points pre-screened during the complete life experiment of the last full-ceramic bearing in the first full-ceramic bearing group, and according to the true wear amount during the complete life experiment corresponding to each time point screened out for each characteristic parameter and the function formula between each characteristic parameter and the wear amount, determine the first function formula.

10. The method for identifying the wear state of the all-ceramic bearing according to claim 9, characterized in that, The S000-4-5 includes: S000-4-5-1. According to the true wear amount during the complete life experiment corresponding to each time point screened out for each characteristic parameter, and the predicted value of the wear amount corresponding to each time point screened out for each characteristic parameter, use formula (1) to obtain the cumulative error amount corresponding to the characteristic parameter; The formula (1) is: Δ i = Δt i1 + … + Δt ih + … Δt ih ; Δ i is the cumulative error amount corresponding to the i-th characteristic parameter; Δt ih is the absolute value of the difference between the true wear amount and the predicted value of the wear amount corresponding to the h-th time point selected for the i-th characteristic parameter during the complete life experiment; Among them, the predicted value of the wear amount during the complete life experiment corresponding to the h-th time point screened out for the i-th characteristic parameter is calculated by the function formula between the characteristic parameter and the wear amount; H is the total number of pre-screened time points, where H is greater than 1; S000-4-5-2. Use the characteristic parameter corresponding to the smallest cumulative error amount as the specified characteristic parameter, and use the function formula between the specified characteristic parameter and the wear amount as the first function formula.

Images