Punching machine transmission shaft bearing fault diagnosis method and system

Abstract

The invention provides a punching machine transmission shaft bearing fault diagnosis method and system, relates to the field of bearing fault diagnosis, and aims to solve the problem that the precision of a fault diagnosis model is difficult to meet the requirement due to lack of fault data and few labels of a conventional punching machine. A test bed device consistent with the characteristic frequency of a punching machine is built, source domain data with rich fault data and labels are obtained, in combination with target domain data with short fault data and few labels obtained by the punching machine, vibration signals are preprocessed and subjected to characteristic amplitude normalization, and a transfer learning bearing fault diagnosis model is built. And the distribution difference and the characteristic difference of the target domain data and the source domain data are reduced, and accurate identification and early diagnosis of the punch transmission shaft bearing fault are realized.

Description

Technical Field

[0001] This invention relates to the field of bearing fault diagnosis, and specifically to a method and system for diagnosing bearing faults in a stamping press drive shaft. Background Technology

[0002] Bearings, as core components of many rotating machines, are prone to various failures under complex working environments and changing operating conditions. In stamping presses, the drive shaft bearing, as one of the core components, directly affects the stability of the press operation. Due to prolonged operation under extreme conditions such as high load, strong impact, and high speed, the drive shaft bearing of the stamping press may fail or degrade due to problems such as poor lubrication and excessive temperature, resulting in internal wear, fatigue spalling, and cracks.

[0003] The main technical means for fault diagnosis of drive shaft bearings typically include vibration monitoring, temperature monitoring, and non-destructive testing. Vibration monitoring is the most commonly used, based on the principle that a fault will cause abnormal vibration in the drive shaft bearing. Signal processing and artificial intelligence technologies can be used to accurately classify and identify the fault type. However, fault diagnosis of press drive shaft bearings suffers from a problem of abundant health status data but scarce or even missing fault data. Collecting fault data by replacing faulty bearings is affected by objective collection conditions and regular maintenance strategies. Acquiring labeled data requires significant manpower and resources, resulting in high costs. The lack of bearing fault sample data severely impacts the accuracy of fault diagnosis.

[0004] To address the aforementioned issues, a small-scale test bench can be built to acquire abundant, labeled source domain data on faults. This data can then be combined with target domain data from the stamping machine, which lacks sufficient fault data and labels. Transfer learning can then be used to diagnose faults in the target domain. However, traditional test benches are typically designed using a geometric scaling approach, failing to consider the equivalence between the test bench and the stamping machine at the data and feature levels. Significant distribution and feature differences exist between the target and source domain data, resulting in current transfer learning fault diagnosis models failing to meet practical requirements in terms of accuracy. Summary of the Invention

[0005] The purpose of this invention is to address the deficiencies of existing technologies by providing a method and system for diagnosing faults in the transmission shaft bearings of a stamping press. By constructing a test bench device that matches the characteristic frequency of the stamping press, rich and labeled source domain data with fault data is acquired. Combined with target domain data with scarce fault data and few labels acquired from the stamping press, transfer learning is used to achieve fault diagnosis in the target domain.

[0006] The first objective of this invention is to provide a method for diagnosing bearing faults in a stamping press drive shaft, employing the following approach: include: Obtain the vibration signal of the bearing of the press to be diagnosed; The target domain coding network, after training and optimization, is used to extract high-dimensional features from the vibration signal of the press bearing to be diagnosed, thereby obtaining the bearing feature information. The characteristic information of the stamping press bearing is input into the pre-trained bearing fault diagnosis model, and the fault diagnosis result of the stamping press bearing to be diagnosed is output. The pre-trained bearing fault diagnosis model is as follows: a test bench device is constructed with the natural frequency, fault frequency, rotation frequency and impact load frequency of the press drive shaft. The test bench device is equipped with multiple types of fault test bearings. The test bench device healthy state vibration signal dataset, various fault vibration signal datasets, and unknown state vibration signal dataset of the press are obtained. After preprocessing and model training, the pre-trained bearing fault diagnosis model is obtained.

[0007] Furthermore, the test bench device that matches the natural frequency of the press drive shaft includes: The test bench device includes a spindle module consisting of a test spindle, a test counterweight, a test bearing, and a test bearing housing. The multi-order natural frequencies of the press drive shaft are obtained through modal simulation analysis. Using the error between the multi-order natural frequencies of the test spindle of the test bench and the multi-order natural frequencies of the press drive shaft as the objective function, the counterweight parameters are optimized to make the natural frequencies of the test spindle of the test bench and the press drive shaft consistent.

[0008] Furthermore, the test bench setup, which replicates the failure frequency of the press drive shaft, includes: Based on the principle that the failure frequency of the press bearing and the test bearing of the test bench are consistent, the inner diameter, outer diameter and number of rolling elements of the test bearing are selected.

[0009] Furthermore, the preprocessing includes denoising, filtering, and decomposition of the vibration signal, followed by characteristic amplitude normalization. The characteristic amplitude normalization process includes: The static stiffness of the press bearing housing and the first four modal stiffnesses of the transmission shaft are obtained through simulation analysis, as well as the static stiffness of the test bench bearing housing and the first four modal stiffnesses of the main shaft. The characteristic amplitude proportionality coefficients between the static stiffness and modal stiffness of the stamping press and the test bench are calculated. Based on this, the characteristic amplitudes of the corresponding frequency components of the vibration signal of the preprocessed test bench are normalized. Furthermore, training the bearing fault diagnosis model includes: Construct a source domain encoding network for the source domain data and a target domain encoding network for the target domain data; A model transfer strategy is introduced, which is based on a pre-trained source domain encoding network. The low-level model parameters of the source domain encoding network are fixed, and the high-level model parameters of the source domain encoding network are adjusted using target domain data to obtain an optimized target domain encoding network. The target domain encoding network is optimized through adversarial training, resulting in a trained and optimized target domain encoding network. Iteratively update the bearing fault diagnosis model.

[0010] Furthermore, based on deep convolutional neural networks, source domain encoding networks and target domain encoding networks are established.

[0011] Furthermore, the source domain encoding network and the target domain encoding network correspond to the first classifier and the second classifier, respectively, and are used to output the probability scores for source domain and target domain fault diagnosis. The labeled source domain training data and the partially labeled target domain training data are input into the source domain encoding network and the target domain encoding network, respectively. Features are extracted and input into the second classifier. Adversarial training technique is used to alternately update the parameters of the target domain encoding network and the second classifier.

[0012] Furthermore, the vibration signal dataset of the stamping machine under unknown conditions is used as the target domain test data and input into the trained and optimized target domain encoding network to extract high-dimensional features, and the first classifier provides the diagnostic results.

[0013] A second objective of this invention is to provide a fault diagnosis system for a press drive shaft bearing as described in the first objective, comprising: The data acquisition module is configured to acquire the vibration signal of the bearing of the press to be diagnosed. The feature extraction module is configured to: use the trained and optimized target domain coding network to perform high-dimensional feature extraction on the vibration signal of the stamping machine bearing to be diagnosed, and obtain the feature information of the stamping machine bearing; The diagnostic module is configured to: input the characteristic information of the stamping press bearing into the pre-trained bearing fault diagnosis model, and output the fault diagnosis result of the stamping press bearing to be diagnosed; The pre-trained bearing fault diagnosis model is as follows: a test bench device is constructed with the natural frequency, fault frequency, rotation frequency and impact load frequency of the press drive shaft. The test bench device is equipped with multiple types of fault test bearings. The test bench device healthy state vibration signal dataset, various fault vibration signal datasets, and unknown state vibration signal dataset of the press are obtained. After preprocessing and model training, the pre-trained bearing fault diagnosis model is obtained.

[0014] Furthermore, the preprocessing includes denoising, filtering, and decomposition of the vibration signal, followed by characteristic amplitude normalization.

[0015] Compared with the prior art, the advantages and positive effects of this invention are: To address the current problem of insufficient fault data and limited labels for stamping presses, which makes it difficult for fault diagnosis models to meet accuracy requirements, this paper considers the equivalence of the test bench and the stamping press at the data and feature levels. A test bench with the same characteristic frequencies as the stamping press is constructed to acquire rich, labeled source domain data on faults. The dynamic response of this test bench is consistent with the natural frequencies of the drive shaft, bearing failure frequencies, impact load frequencies, and rotational frequencies contained in the dynamic response of the stamping press. This effectively improves the reliability of fault data from the test bench. By adjusting the counterweight size and its distribution, the test bench can simulate several operating conditions and multiple drive shaft bearing failures on the same test bench, effectively reducing experimental costs and demonstrating strong adaptability.

[0016] By combining the health status data of the stamping press and the health status and fault data of the test bench, the vibration signals measured by both are preprocessed to remove components unrelated to bearing fault characteristics. Based on simulation analysis of the stamping press and the test bench, the characteristic amplitudes of the health status and fault vibration signals acquired by the test bench are normalized. Through the design of a test bench with consistent characteristic frequencies and characteristic amplitude normalization processing, combined with a transfer learning bearing fault diagnosis model based on domain adversarial transfer networks, the distribution and characteristic differences between the target domain data and the source domain data are reduced. This enables target domain state identification even when fault data is missing or label data is scarce, achieving accurate identification and early diagnosis of stamping press drive shaft bearing faults. Attached Figure Description

[0017] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0018] Figure 1 This is a flowchart illustrating a method for diagnosing bearing faults in a press drive shaft according to one or more embodiments of the present invention.

[0019] Figure 2 This is a schematic diagram of the structure of the test bench device in one or more embodiments of the present invention.

[0020] Figure 3 This is a schematic diagram of the first four order frequency response functions of the press drive shaft in one or more embodiments of the present invention. Figure 3 (a) in the diagram is a schematic diagram of the first-order frequency response function of the press drive shaft. Figure 3 (b) in the diagram is a schematic diagram of the second-order frequency response function of the press drive shaft. Figure 3 (c) in the diagram is a schematic diagram of the third-order frequency response function of the press drive shaft. Figure 3 (d) in the figure is a schematic diagram of the fourth-order frequency response function of the press drive shaft.

[0021] Figure 4 This is a schematic diagram of the first four order frequency response functions of the test spindle of the test bench device in one or more embodiments of the present invention. Figure 4 (a) in the diagram is a schematic diagram of the first and second order frequency response functions of the test spindle of the test bench. Figure 4 (b) in the diagram is a schematic diagram of the third-order frequency response function of the test spindle of the test bench. Figure 4 (c) in the figure is a schematic diagram of the fourth-order frequency response function of the test spindle of the test bench.

[0022] Figure 5 This refers to the time-domain plot, frequency-domain plot, and time-frequency characterization of the vibration signal after characteristic amplitude normalization in one or more embodiments of the present invention. Figure 5 (a) shows the time-domain plot, frequency-domain plot, and time-frequency characterization of the vibration signal of the healthy bearing of the stamping machine. Figure 5 (b) shows the time-domain plot, frequency-domain plot, and time-frequency characterization of the bearing vibration signal during the health test of the test bench. Figure 5 (c) in the figure represents the time-domain plot, frequency-domain plot, and time-frequency characterization of the bearing vibration signal during the fault test of the test bench.

[0023] Figure 6 This is a schematic diagram of the bearing fault diagnosis model based on domain adversarial migration network in one or more embodiments of the present invention.

[0024] In the figure, 1. Test motor; 2. Test coupling; 3. Test spindle; 4. Test bearing; 5. Test bearing housing; 6. Test counterweight; 7. Test acceleration sensor; 8. Test eddy current sensor; 9. Test impact loading mechanism; 10. Test magnetic powder brake; 11. Test base. Detailed Implementation

[0025] Example 1 In a typical embodiment of the present invention, such as Figures 1-6 As shown, a method for diagnosing bearing faults in a stamping press drive shaft is presented.

[0026] Currently, there is a lack of fault data and few labels for stamping machines. Furthermore, the ordinary test bench setup does not consider the equivalence between the test bench setup and the stamping machine at the data level, resulting in significant distribution and feature differences between the target domain data and the source domain data. Consequently, the accuracy of the transfer fault diagnosis model still cannot meet the actual needs.

[0027] Based on this, such as Figure 1 As shown, this embodiment provides a method for diagnosing bearing faults in a stamping press drive shaft, including: Obtain the vibration signal of the bearing of the press to be diagnosed; The target domain coding network, after training and optimization, is used to extract high-dimensional features from the vibration signal of the press bearing to be diagnosed, thereby obtaining the bearing feature information. The characteristic information of the stamping press bearing is input into the pre-trained bearing fault diagnosis model, and the fault diagnosis result of the stamping press bearing to be diagnosed is output. The bearing fault diagnosis model is as follows: a test bench device is constructed with the natural frequency, fault frequency, rotation frequency and impact load frequency of the press drive shaft. The test bench device is equipped with multiple types of fault test bearings 4. The test bench device healthy state vibration signal dataset and various fault vibration signal datasets, as well as the press unknown state vibration signal dataset are obtained. After the vibration signals are preprocessed, a pre-trained bearing fault diagnosis model is obtained through training.

[0028] The stamping press in this embodiment is a large piece of equipment, with common models being 12 tons, 16 tons, 25 tons, etc. A test bench device with the characteristic frequency of the stamping press is established. The length of the stamping press drive shaft is 3000mm. The structure of the stamping press drive shaft, the stamping press bearing, and the stamping press bearing housing adopts the common bearing support component structure.

[0029] like Figure 2 As shown, the test bench assembly consists of a test motor 1, a test coupling 2, a test spindle 3, test bearings 4, test bearing housings 5, test counterweights 6, a test accelerometer 7, a test eddy current sensor 8, a test impact loading mechanism 9, a test magnetic powder brake 10, and a test base 11. The test motor 1's speed can be adjusted by a host control system. The output shaft of the test motor 1 is connected to the test spindle 3 via the test coupling 2. The test spindle 3 is supported by two test bearings 4, which are rolling bearings and fixed to the test bearing housings 5. Three annular test counterweights 6 are mounted on the test spindle 3. The test bearing 4 furthest from the test motor 1 is the bearing to be tested. The test bearing housing 5 is equipped with a test accelerometer 7 to measure the acceleration signal of the test bearing housing 5. A test eddy current sensor 8 is installed in the middle of the test spindle 3 to measure the displacement signal of the test spindle 3. The test impact loading mechanism 9 applies an impact load to the test bearing housing 5. At the end of the test bench, a test magnetic powder brake 10 is used as a load device, and it is connected to the test spindle 3 through a test coupling 2. The test motor 1, the test bearing seat 5, and the load device are connected to the test base 11 through a T-slot and are fixed with bolts and nuts.

[0030] Characteristic frequencies mainly include bearing failure frequency, drive shaft natural frequency, impact load frequency, and rotational frequency. The test bench device includes a spindle module, signal acquisition module, impact loading module, motor module, and load module. The spindle module is designed to consist of a test spindle 3, a test counterweight 6, a test bearing 4, and a test bearing housing 5, including: Modal simulation analysis was used to obtain the first four natural frequencies of the press drive shaft. .in, CY It refers to a stamping machine.

[0031] The dimensions and spatial distribution of the experimental counterweights are parameterized.

[0032] The first four natural frequencies of the test spindle 3 of the test bench device. With the first four natural frequencies of the press drive shaft The error between the parameters is used as the objective function, and an adaptive multi-objective optimization algorithm is employed to optimize the six parameters of the experimental weight. SY This refers to the test bench setup.

[0033] Based on the principle that the failure frequency of the press bearing and the test bearing 4 are consistent, the inner diameter, outer diameter and number of rolling elements of the test bearing 4 are selected.

[0034] Specifically, the design steps for the test bench are as follows: S1: Based on the actual dimensions of the press drive shaft, press bearings, and press bearing housing system, a three-dimensional model is established, using 42CrMo steel as the material.

[0035] S2: Modal simulation analysis is used to add an elastic support at the bearing support of the stamping machine, and the foundation stiffness is set to... N / m. A hexahedral mesh was used, with element size less than 0.5 mm. The first four natural frequencies of the press drive shaft were extracted. ,like Figure 3 , Figure 4 As shown.

[0036] S3: The test spindle 3 of the test bench device is a long straight shaft with a diameter of 35 mm.

[0037] S4: Set the length, outer diameter, and distribution position of the test counterweight 6 as independent parameters, and perform modal simulation analysis on the test bench under the same conditions to obtain the first four natural frequencies of the test spindle 3. .

[0038] S5: with and The error between the two natural frequencies is used as the objective function. Error constraints are set, namely, the relative error of the first two natural frequencies is less than 3%, and the relative error of the remaining natural frequencies is less than 5%. Reasonable constraint ranges are set for the six parameters of the experimental weight, and an adaptive multi-objective optimization algorithm is used to optimize the six parameters of the experimental weight.

[0039] S6: After executing the optimization algorithm, if the first two natural frequencies are out of tolerance, the shaft diameter of the test bench can be modified and the optimization repeated. If the remaining natural frequencies are out of tolerance, the test counterweight parameters 6 can be adjusted near the candidate point and the simulation results of the candidate point can be updated. In Example 1, the simulation optimization results of the relevant dynamic characteristics of the test bench are as follows: Figure 3 , Figure 4 As shown, it can be seen that its first four natural frequencies all meet the error requirements.

[0040] S7: Based on the principle that the failure frequencies of the press bearing and the test bearing 4 are consistent, the inner diameter, outer diameter, and number of rolling elements of the test bearing 4 are selected. The calculation methods for various failure frequencies are as follows: (1); in, To test the failure frequency of the outer raceway of the inner ring of bearing 4, To test the failure frequency of the inner raceway of the outer ring of bearing 4, To test the failure frequency of bearing 4 cage, To test the failure frequency of the 4 rollers in the bearing, For rotation frequency, To test the outer diameter of bearing 4, To test the inner diameter of bearing 4, Z This represents the number of rolling elements.

[0041] S8: In this embodiment, the parameters of the stamping press bearing are as follows: = 300mm and = 210mm, material is G20Cr2Ni4A. According to the method for calculating the failure frequency of test bearings, while keeping the material and number of rolling elements constant, it is necessary to maintain... The parameters of the test bearing 4 remain unchanged. = 50mm and = 35mm.

[0042] The signal acquisition module includes a contact-type vibration acquisition module for the test bearing housing 5 and a non-contact vibration acquisition module for the press drive shaft and test spindle 3. Specifically, the test accelerometer 7 is responsible for acquiring the vibration signal of the test bearing housing 5, and the test eddy current sensor 8 is responsible for acquiring the vibration signal of the press drive shaft and test spindle 3.

[0043] An impact loading module was added to directly excite the test bearing housing 5. The impact loading module consists of a piezoelectric actuator and a controller.

[0044] The test bearing housing 5 is directly excited to maintain the same impact frequency as the press.

[0045] Configure the motor module and load module to ensure that the test spindle of the test bench device rotates at the same speed as the transmission shaft of the stamping machine.

[0046] Multiple types of fault test bearings 4 are configured on the test bench device to construct a healthy state vibration signal dataset of the test bench device, a vibration signal dataset of various faults, and a vibration signal dataset of the stamping machine under unknown state.

[0047] The preprocessing and characteristic amplitude normalization of vibration signals include: The collected vibration signals are preprocessed, including: Noise reduction: Wavelet filtering is used to remove noise from the vibration signal, thereby eliminating the interference of noise on the fault diagnosis results.

[0048] Filtering: The rotational frequency component and its harmonics in the vibration signal are removed by a comb filter, thereby eliminating the interference of rotational frequency on the fault diagnosis results.

[0049] Decomposition: The impact load frequency component and its harmonics are extracted from the vibration signals of the stamping press bearing housing and test bearing housing 5 using Synchronous Compression Transform (SST). and 4. Fault frequency components and their harmonics of press bearings and test bearings and .in, c Represents the frequency component of the impact load. zz This represents the bearing fault frequency component of the bearing housing vibration signal. The first four natural frequency components of the vibration signals from the press drive shaft and test spindle 3 are extracted using SST. and 4. Fault frequency components and their harmonics of press bearings and test bearings and .in, g Represents the natural frequency component. zc This represents the bearing fault frequency component of the vibration signal of the press drive shaft and test spindle 3.

[0050] The preprocessed vibration signal undergoes characteristic amplitude normalization processing, including: The static stiffness of the press bearing housing was obtained through static and harmonic response simulation analysis. The first four modal stiffnesses of the press drive shaft And the static stiffness of the test bearing housing 5 of the test bench device. The first four modal stiffnesses of the test principal axis 3 ,like Figure 3 , Figure 4As shown in Table 1, the static stiffness of the press bearing housing, the static stiffness of the test bearing housing 5, the first four natural frequencies of the press drive shaft, the first four natural frequencies of the test spindle 3, the first four modal stiffnesses of the press drive shaft, and the first four modal stiffnesses of the test spindle 3 are all present.

[0051] Calculate the characteristic amplitude ratio coefficients between the static stiffness and modal stiffness of the press and the test bench, and normalize the characteristic amplitudes of each frequency component in the vibration signal of the preprocessed test bench.

[0052] Short-time Fourier Transform (STFT) is performed on each individual component (including a certain characteristic frequency component and its harmonics) after preprocessing and characteristic amplitude normalization to obtain the time-frequency characterization of that individual component. The time-frequency characterizations of all individual components are combined to obtain the time-frequency characterization of the vibration signal.

[0053] Vibration signals were collected from the test bench, and after preprocessing and characteristic amplitude normalization, a dataset of the health status and fault status of the test bench, as well as a dataset of the unknown status of the stamping machine, were constructed.

[0054] Specifically, in this embodiment, preprocessing and feature amplitude normalization include: The following steps are taken to remove noise from vibration signals using wavelet filtering: S1: Perform multi-scale wavelet decomposition on the noisy vibration signal to obtain wavelet coefficients in different frequency bands. and approximation coefficients .in, , L To determine the maximum number of decomposition levels, take... L = 5.

[0055] S2: Calculate the noise variance estimate of the wavelet coefficients. The calculation method is as follows .

[0056] S3: Generate threshold based on subband characteristics .in, For the first The number of sub-band coefficients.

[0057] S4: For wavelet coefficients The soft thresholding process is performed using the following calculation method: (2); S5: Use and The denoised vibration signal is obtained by performing inverse wavelet transform.

[0058] A comb filter is used to remove the rotational frequency component and its harmonics from the vibration signal. The notch bandwidth of the comb filter is 50 Hz, and the notch frequency is equal to the rotational frequency of the press drive shaft and the test spindle 3. / 60. Among them, The rotational speeds of the press drive shaft and the test spindle 3 are given.

[0059] The impact load frequency component and its harmonics are extracted from the vibration signal of the stamping machine bearing housing / test bearing housing 5 using SST. and 4. Fault frequency components and their harmonics of press bearings and test bearings and The first four natural frequency components in the vibration signals of the press drive shaft and the test spindle 3. and 4. Fault frequency components and their harmonics of press bearings and test bearings and The implementation steps are as follows: S1: For vibration signals Perform a short-time Fourier transform (STFT) to obtain the time-frequency representation. S ( t , ω ).

[0060] S2: Estimation via phase difference method The instantaneous frequency is calculated as follows: (3); S3: According to instantaneous frequency S ( t , ω The energy is redistributed to generate a compressed time-frequency representation. T ( t , ω The calculation method is as follows: (4); S4: Utilize T ( t , ω Separate the components and calculate them as follows: (5); in, and This represents the frequency range of the component.

[0061] The characteristic amplitude of the preprocessed vibration signal is normalized. The steps are as follows: Static simulation analysis was employed, applying standard Earth gravity to the press bearing housing and test bearing housing 5, with static forces applied. Calculate the static stiffness of the press bearing housing and the test bearing housing 5. and .in, and This represents the maximum displacement of the press bearing housing and the test bearing housing 5.

[0062] Harmonic response simulation analysis was used to apply a simple harmonic excitation force to the press drive shaft and the test spindle 3. Calculate modal stiffness and .in, and For the press drive shaft and test spindle 3 Maximum displacement of the first mode.

[0063] Calculate the characteristic amplitude proportionality coefficients between the static stiffness and modal stiffness of the stamping press and the test bench. The calculation method is as follows: (6); The characteristic amplitude of each frequency component in the vibration signal of the preprocessed test bench is normalized. The calculation method is as follows: (7); Short-time Fourier Transform (STFT) is performed on each individual component (including a certain characteristic frequency component and its harmonics) after preprocessing and characteristic amplitude normalization to obtain the time-frequency characterization of that individual component. The time-frequency characterizations of all individual components are then combined to obtain the time-frequency characterization of the vibration signal. The calculation method is as follows: (8); (9); After preprocessing and characteristic amplitude normalization, the energy of the obtained time-frequency features is mainly concentrated in the fault frequency, impulse frequency, and natural frequency. For example... Figure 5 (a) and Figure 5 As shown in (b), the energy levels of the test bearing 4 and the press bearing are basically consistent under healthy conditions in terms of time-frequency characteristics; Figure 5 As shown in (c), for the time-frequency characteristics of bearing 4 in the test bench fault test, its energy is more concentrated at the bearing fault frequency.

[0064] Tests were conducted on the constructed test bench and stamping press. Based on the collected vibration signal dataset, a dataset supporting the subsequent fault diagnosis model was constructed through preprocessing, characteristic amplitude normalization, and time-frequency conversion. This dataset includes: Time-frequency characterization dataset of vibration signals of stamping press under unknown conditions .

[0065] Time-frequency characterization dataset of vibration signals in the health state of the test bench .

[0066] Time-frequency characterization dataset of fault vibration signals of test bench equipment .

[0067] like Figure 6 As shown, the bearing fault diagnosis model based on domain adversarial migration network includes the following: Considering the distribution differences between source domain data and target domain data, corresponding source domain encoding networks are constructed for the source domain data and target domain data respectively based on deep convolutional neural networks. and target domain coding network .

[0068] The constructed deep convolutional neural network consists of multiple convolutional layers, batch normalization layers, pooling layers, fully connected layers, and a softmax classification layer. and The output layers correspond to classifiers C and D, respectively, and are used to output the probability scores for fault diagnosis in the source and target domains.

[0069] Introduce a model transfer strategy to utilize pre-trained... Based on, fixed The low-level model parameters. Using target domain data with a small number of labels and a small learning rate, stochastic gradient descent is employed to refine the parameters. Fine-tuning the parameters of the high-level model yields the optimized target domain encoding network. This enables the initial transfer and reuse of knowledge from the source domain.

[0070] Introducing adversarial training techniques Optimize to reduce the distribution differences between source domain data and target domain data. High-dimensional features are extracted from the target domain data and input into the discriminator D to perform adversarial training. The model parameters are then updated iteratively to achieve... and The output features are inseparable.

[0071] Specifically, the implementation steps of the bearing fault diagnosis model based on domain adversarial migration networks are as follows: Step 1: The labeled source domain data acquired by the test bench device and the target domain data with a small number of labels acquired by the stamping machine are randomly divided into a training group and a test group, each accounting for 50%.

[0072] Step 2: Use labeled source domain training group data to... Supervised training was performed on C using a parallel training method. The average cross-entropy of Softmax was used as the training loss. An adaptive learning rate optimization algorithm was used to update the gradient and minimize the loss value. The maximum number of training epochs was 150, and the learning rate was set to 0.0001.

[0073] Step 3: Employ a model transfer strategy to... The model parameters are used for initialization. By freezing specific layers and optimizing the process, Diagnostic knowledge transfer to The specific steps are as follows: S1: Replace layer B6 with the number of categories for fault diagnosis in the target domain. For training data of the target domain with a small number of labels, fine-tune the parameters of layer B6 by freezing the parameters of layers B1-B5 and using a small learning rate and stochastic gradient descent.

[0074] S2: Fix the parameters of the first B1-B4 layers, use a small learning rate to fine-tune the parameters of B5 and B6 layers, and find the optimal number of transferable layers.

[0075] S3: Repeat the model transfer process in S2 until the set classification accuracy or number of iterations is achieved. Determine the optimal number of transferable layers suitable for fault diagnosis in the target domain to obtain a tuned transferable fault diagnosis model.

[0076] Step 4: Input the labeled source domain training data and the target domain training data with a small number of labels into the database respectively. and Features are extracted and input into D. Adversarial training techniques are used to alternately update... And the parameters of D.

[0077] Step 5: Repeat the training in Step 4 until D gradually loses the ability to distinguish between source and target domain data. Gradually reduce the data distribution difference between the source and target domains until the set iteration threshold or the set diagnostic accuracy threshold is reached.

[0078] Step 6: During the testing phase, input the target domain test data into G. t Features are extracted from the data, and C directly provides the diagnostic results for each state category.

[0079] A deep convolutional neural network is used to extract high-dimensional features, and knowledge from labeled source domain data is transferred to the target domain for fault diagnosis. At the same time, an adversarial transfer learning strategy is adopted to enable the network to learn features suitable for each domain, thereby reducing the data distribution differences between the source and target domains and improving the accuracy of fault diagnosis.

[0080] Example 2 In another typical embodiment of the present invention, such as Figures 1-6As shown, a fault diagnosis system for a press drive shaft bearing is provided, comprising: The data acquisition module is configured to acquire the vibration signal of the bearing of the press to be diagnosed. The feature extraction module is configured to: use the trained and optimized target domain coding network to perform high-dimensional feature extraction on the vibration signal of the stamping machine bearing to be diagnosed, and obtain the bearing feature information; The diagnostic module is configured to: input bearing feature information into a pre-trained bearing fault diagnosis model, and output the fault diagnosis results of the stamping press bearing to be diagnosed; The pre-trained bearing fault diagnosis model is as follows: a test bench device is constructed with the natural frequency, fault frequency, rotation frequency and impact load frequency of the press drive shaft. The test bench device is equipped with multiple types of fault test bearings 4. The test bench device healthy state vibration signal dataset and various fault vibration signal datasets, as well as the press unknown state vibration signal dataset are obtained. After the vibration signals are preprocessed, the pre-trained bearing fault diagnosis model is obtained through training.

[0081] The preprocessing includes denoising, filtering, and decomposition, and the preprocessed vibration signal is normalized for its characteristic amplitude.

[0082] The working method of the press drive shaft bearing fault diagnosis system is the same as that in Example 1, and will not be repeated here.

[0083] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for diagnosing bearing faults in a stamping press drive shaft, characterized in that, include: Obtain the vibration signal of the bearing of the press to be diagnosed; The target domain coding network, after training and optimization, is used to extract high-dimensional features from the vibration signal of the press bearing to be diagnosed, thereby obtaining the bearing feature information. The characteristic information of the stamping press bearing is input into the pre-trained bearing fault diagnosis model, and the fault diagnosis result of the stamping press bearing to be diagnosed is output. The pre-trained bearing fault diagnosis model is as follows: a test bench device is constructed with the natural frequency, fault frequency, rotation frequency and impact load frequency of the press drive shaft. The test bench device is equipped with multiple types of fault test bearings. The test bench device healthy state vibration signal dataset, various fault vibration signal datasets, and unknown state vibration signal dataset of the press are obtained. After preprocessing and model training, the pre-trained bearing fault diagnosis model is obtained.

2. The method for diagnosing faults in the transmission shaft bearings of a stamping machine as described in claim 1, characterized in that, The test bench setup that matches the natural frequency of the press drive shaft includes: The test bench device includes a spindle module consisting of a test spindle, a test counterweight, a test bearing, and a test bearing housing. The multi-order natural frequencies of the press drive shaft are obtained through modal simulation analysis. Using the error between the multi-order natural frequencies of the test spindle of the test bench and the multi-order natural frequencies of the press drive shaft as the objective function, the counterweight parameters are optimized to make the natural frequencies of the test spindle of the test bench and the press drive shaft consistent.

3. The method for diagnosing faults in the transmission shaft bearings of a stamping machine as described in claim 1, characterized in that, The test bench setup that simulates the failure frequency of the press drive shaft includes: Based on the principle that the failure frequency of the press bearing and the test bearing of the test bench are consistent, the inner diameter, outer diameter and number of rolling elements of the test bearing are selected.

4. The method for diagnosing faults in the transmission shaft bearings of a stamping machine as described in claim 1, characterized in that, The preprocessing includes denoising, filtering, and decomposition of the vibration signal, followed by characteristic amplitude normalization.

5. The method for diagnosing faults in the transmission shaft bearings of a stamping machine as described in claim 1, characterized in that, Training the bearing fault diagnosis model includes: Construct a source domain encoding network for the source domain data and a target domain encoding network for the target domain data; A model transfer strategy is introduced, which is based on a pre-trained source domain encoding network. The low-level model parameters of the source domain encoding network are fixed, and the high-level model parameters of the source domain encoding network are adjusted using target domain data to obtain an optimized target domain encoding network. The target domain encoding network is optimized through adversarial training, resulting in a trained and optimized target domain encoding network. Iteratively update the bearing fault diagnosis model.

6. The method for diagnosing faults in the transmission shaft bearings of a stamping machine as described in claim 5, characterized in that, Based on deep convolutional neural networks, source domain encoding networks and target domain encoding networks are established.

7. The method for diagnosing faults in the transmission shaft bearings of a stamping machine as described in claim 6, characterized in that, The source domain coding network and the target domain coding network correspond to the first classifier and the second classifier, respectively, and are used to output the probability scores for source domain and target domain fault diagnosis. The labeled source domain training data and the partially labeled target domain training data are input into the source domain encoding network and the target domain encoding network, respectively. Features are extracted and input into the second classifier. Adversarial training technique is used to alternately update the parameters of the target domain encoding network and the second classifier.

8. The method for diagnosing faults in the transmission shaft bearings of a stamping machine as described in claim 5, characterized in that, The dataset of vibration signals of the stamping machine under unknown conditions is used as the test data for the target domain. It is then input into the trained and optimized target domain encoding network to extract high-dimensional features, and the first classifier provides the diagnostic results.

9. A fault diagnosis system for a stamping press drive shaft bearing, characterized in that, include: The data acquisition module is configured to acquire the vibration signal of the bearing of the press to be diagnosed. The feature extraction module is configured to: use the trained and optimized target domain coding network to perform high-dimensional feature extraction on the vibration signal of the stamping machine bearing to be diagnosed, and obtain the feature information of the stamping machine bearing; The diagnostic module is configured to: input the characteristic information of the stamping press bearing into the pre-trained bearing fault diagnosis model, and output the fault diagnosis result of the stamping press bearing to be diagnosed; The pre-trained bearing fault diagnosis model is as follows: a test bench device is constructed with the natural frequency, fault frequency, rotation frequency and impact load frequency of the press drive shaft. The test bench device is equipped with multiple types of fault test bearings. The test bench device healthy state vibration signal dataset, various fault vibration signal datasets, and unknown state vibration signal dataset of the press are obtained. After preprocessing and model training, the pre-trained bearing fault diagnosis model is obtained.

10. The fault diagnosis system for the transmission shaft bearing of a stamping machine as described in claim 9, characterized in that, The preprocessing includes denoising, filtering, and decomposition of the vibration signal, followed by characteristic amplitude normalization.

Images