Punch press transmission shaft bearing fault diagnosis method and system

Abstract

The application provides a punch press transmission shaft bearing fault diagnosis method and system, relates to the bearing fault diagnosis field, aims at the problem that the fault diagnosis model precision is difficult to meet the demand due to the lack of punch press fault data and few labels, considers the equivalence of the test bench device and the punch press in the data level, builds the test bench device consistent with the feature frequency of the punch press, obtains source domain data rich in fault data and labeled, combines the target domain data obtained by the punch press which is lack of fault data and few labels, carries out pretreatment and feature amplitude normalization on the vibration signal, constructs a transfer learning bearing fault diagnosis model, reduces the distribution difference and feature difference between the target domain data and the source domain data, and realizes accurate identification and early diagnosis of the punch press transmission shaft bearing fault.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of bearing fault diagnosis, in particular to a stamping machine transmission shaft bearing fault diagnosis method and system. BACKGROUND

[0002] As the core component of many rotating machines, bearings are prone to various faults in complex working environments and variable working conditions. In the stamping machine, the transmission shaft bearing, as one of the core components, directly affects the stability of the stamping machine. Due to long-term work in extreme conditions such as high load, strong impact and high speed, the stamping machine transmission shaft bearing may fail or degrade due to poor lubrication, high temperature and other problems, resulting in internal wear, fatigue spalling and cracking.

[0003] The main technical means of transmission shaft bearing fault diagnosis usually includes vibration monitoring, temperature monitoring and non-destructive monitoring. Among them, the more commonly used is vibration monitoring, the main principle is that when a fault occurs, it will cause abnormal vibration of the transmission shaft bearing, and by using signal processing technology and artificial intelligence technology, the fault type can be accurately classified and identified. However, the fault diagnosis of the transmission shaft bearing of the stamping machine has the problem of rich health state data, lack of fault data or even missing. By replacing the fault bearing to collect fault data, the label data acquisition needs to consume a lot of manpower and material resources, the cost is too high, and the lack of bearing fault sample data seriously affects the accuracy of fault diagnosis.

[0004] To solve the above problems, a small test bench device can be built to obtain source domain data with rich fault data and labels, and target domain data with lack of fault data and few labels obtained from the stamping machine, and target domain fault diagnosis is realized through transfer learning. However, the traditional test bench device is usually designed according to the geometric scaling scheme, and the equivalence of the test bench device and the stamping machine in the data level and the feature level is not considered. There is a large difference in distribution and features between the target domain data and the source domain data, which leads to the fact that the precision of the current transfer fault diagnosis model still cannot meet the actual demand. SUMMARY

[0005] The purpose of the present application is to solve the defects existing in the prior art, and to provide a stamping machine transmission shaft bearing fault diagnosis method and system. By building a test bench device with the same characteristic frequency as the stamping machine, source domain data with rich fault data and labels is obtained, and target domain data with lack of fault data and few labels is obtained from the stamping machine. Through transfer learning, target domain fault diagnosis is realized.

[0006] The first purpose of the present application is to provide a stamping machine transmission shaft bearing fault diagnosis method, which adopts the following scheme:

[0007] It comprises:

[0008] obtaining a vibration signal of a stamping machine bearing to be diagnosed;

[0009] extracting high-dimensional features of the vibration signal of the stamping machine bearing to be diagnosed by using the target domain coding network optimized by training, to obtain stamping machine bearing feature information;

[0010] inputting the stamping machine bearing feature information into a pre-trained bearing fault diagnosis model, and outputting a fault diagnosis result of the stamping machine bearing to be diagnosed;

[0011] The pre-trained bearing fault diagnosis model is: a test bench device consistent with the natural frequency, fault frequency, rotating frequency and impact load frequency of the stamping machine transmission shaft is constructed, the test bench device is configured with multi-class fault test bearings, vibration signal data sets of healthy states and various fault states of the test bench device are obtained, and vibration signal data sets of unknown states of the stamping machine are obtained, and after preprocessing and model training, the pre-trained bearing fault diagnosis model is obtained.

[0012] Further, the test bench device consistent with the natural frequency of the stamping machine transmission shaft comprises:

[0013] The test bench device comprises a main shaft module composed of a test main shaft, a test counterweight, a test bearing and a test bearing seat, and the multi-order natural frequency of the stamping machine transmission shaft is obtained by modal simulation analysis;

[0014] The error between the multi-order natural frequency of the test main shaft of the test bench device and the multi-order natural frequency of the stamping machine transmission shaft is taken as an objective function, and the counterweight parameter is optimized so that the natural frequencies of the test main shaft of the test bench device and the stamping machine transmission shaft are consistent.

[0015] Further, the test bench device consistent with the fault frequency of the stamping machine transmission shaft comprises:

[0016] According to the principle that the fault frequency of the stamping machine bearing is consistent with that of the test bearing of the test bench device, the inner diameter, outer diameter and number of rolling elements of the test bearing are selected.

[0017] Further, the preprocessing comprises denoising, filtering and decomposition of the vibration signal, and then feature amplitude normalization processing. The feature amplitude normalization processing comprises:

[0018] The static stiffness of the stamping machine bearing seat and the first four order modal stiffness of the transmission shaft, and the static stiffness of the bearing seat of the test bench device and the first four order modal stiffness of the main shaft are obtained by simulation analysis;

[0019] The feature amplitude proportion coefficient between the corresponding static stiffness and modal stiffness of the stamping machine and the test bench device is calculated, and the feature amplitude normalization processing is performed on the corresponding frequency components of the preprocessed vibration signal of the test bench device. Further, the training of the bearing fault diagnosis model comprises:

[0020] The source domain coding network is constructed for the source domain data, and the target domain coding network is constructed for the target domain data.

[0021] A model migration strategy is introduced, and the low-layer model parameters of the source domain coding network are fixed based on the pre-training of the source domain coding network.

[0022] The target domain coding network is optimized through the adversarial training, and the training-optimized target domain coding network is obtained.

[0023] The bearing fault diagnosis model is iteratively updated.

[0024] Further, the source domain coding network and the target domain coding network are established based on the deep convolutional neural network.

[0025] Further, the source domain coding network and the target domain coding network correspond to the first classifier and the second classifier respectively, and are used for outputting the probability score of the source domain and the target domain fault diagnosis.

[0026] The labeled source domain training group data and the partially labeled target domain training group data are respectively input into the source domain coding network and the target domain coding network, the features are extracted and input into the second classifier, the parameters of the target domain coding network and the second classifier are alternately updated by using the adversarial training technology.

[0027] Further, the unknown state vibration signal data set of the punch press is taken as the target domain test data, is input into the training-optimized target domain coding network to extract high-dimensional features, and the diagnosis result is given by the first classifier.

[0028] The second object of the present application is to provide a punch press transmission shaft bearing fault diagnosis system as described in the first object, comprising:

[0029] The data acquisition module is configured to acquire the vibration signal of the punch press bearing to be diagnosed.

[0030] The feature extraction module is configured to perform high-dimensional feature extraction on the vibration signal of the punch press bearing to be diagnosed by using the training-optimized target domain coding network, and obtain the punch press bearing feature information.

[0031] The diagnosis module is configured to input the punch press bearing feature information into the pre-trained bearing fault diagnosis model, and output the fault diagnosis result of the punch press bearing to be diagnosed.

[0032] The pre-trained bearing fault diagnosis model is: a test bench device consistent with inherent frequencies of a transmission shaft of a punch press, fault frequencies, rotating frequencies and impact load frequencies is constructed, the test bench device is configured with multiple types of fault test bearings, vibration signal data sets of healthy states and various fault vibration signal data sets of the test bench device and vibration signal data sets of unknown states of the punch press are acquired, and the pre-trained bearing fault diagnosis model is obtained through preprocessing and model training.

[0033] Further, the preprocessing includes denoising, filtering and decomposition of the vibration signal, and then feature amplitude normalization processing.

[0034] Compared with the prior art, the present application has the advantages and positive effects that:

[0035] In view of the problem that the fault data of the punch press is lack and the labels are few, which leads to difficulty in meeting the requirement of the precision of the fault diagnosis model, the test bench device and the punch press are considered to be equivalent in the data level and the feature level, a test bench device consistent with the feature frequencies of the punch press is built, and source domain data rich in fault data and labels are acquired, the dynamic response of the device and the dynamic response of the punch press are consistent in inherent frequencies of the transmission shaft, fault frequencies of the bearing, impact load frequencies and rotating frequencies, which can effectively improve the reliability of the fault data of the test bench device, the test bench device can simulate several working conditions and multiple bearing faults of the transmission shaft on the same test bench device by adjusting the size and distribution of the counterweight, effectively reducing the experimental cost and being highly adaptable.

[0036] In combination with the health state data and the fault data of the punch press and the test bench device, the vibration signals measured by the two are preprocessed to remove components irrelevant to the bearing fault features, and the health state and fault vibration signals acquired by the test bench device are subjected to feature amplitude normalization according to the simulation analysis of the punch press and the test bench device. Through the test bench device design consistent with the feature frequencies and the feature amplitude normalization processing, in combination with the bearing fault diagnosis model based on the domain adversarial transfer network and the transfer learning, the distribution difference and the feature difference between the target domain data and the source domain data are reduced, the target domain state recognition is performed under the condition of lack of fault data and few label data, and accurate identification and early diagnosis of the bearing fault of the transmission shaft of the punch press are realized. BRIEF DESCRIPTION OF DRAWINGS

[0037] The drawings accompanying the specification of the present application form a part thereof, serve to provide further understanding of the present application, and together with the description, explain the present application. The present application is shown by way of example in the drawings and their descriptions serve the purpose of explaining the present application, and do not constitute an improper limitation thereof.

[0038] Figure 1 A flowchart of a punch press transmission shaft bearing fault diagnosis method in one or more embodiments of the present application.

[0039] Figure 2 Fig. 1 is a structural schematic diagram of a test bench device constructed in one or more embodiments of the present application.

[0040] Figure 3 Fig. 2 is a schematic diagram of the first four order frequency response functions of a punch transmission shaft in one or more embodiments of the present application, Figure 3 (a) in Fig. 2 is a schematic diagram of the first order frequency response function of the punch transmission shaft, Figure 3 (b) in Fig. 2 is a schematic diagram of the second order frequency response function of the punch transmission shaft, Figure 3 (c) in Fig. 2 is a schematic diagram of the third order frequency response function of the punch transmission shaft, Figure 3 (d) in Fig. 2 is a schematic diagram of the fourth order frequency response function of the punch transmission shaft.

[0041] Figure 4 Fig. 3 is a schematic diagram of the first four order frequency response functions of a test spindle of a test bench device in one or more embodiments of the present application, Figure 4 (a) in Fig. 3 is a schematic diagram of the first order frequency response function and the second order frequency response function of the test spindle of the test bench device, Figure 4 (b) in Fig. 3 is a schematic diagram of the third order frequency response function of the test spindle of the test bench device, Figure 4 (c) in Fig. 3 is a schematic diagram of the fourth order frequency response function of the test spindle of the test bench device.

[0042] Figure 5 Fig. 4 is a time-domain graph, a frequency-domain graph and a time-frequency representation of a vibration signal normalized by a characteristic amplitude in one or more embodiments of the present application, Figure 5 (a) in Fig. 4 is a time-domain graph, a frequency-domain graph and a time-frequency representation of a punch healthy bearing vibration signal, Figure 5 (b) in Fig. 4 is a time-domain graph, a frequency-domain graph and a time-frequency representation of a test bench device healthy test bearing vibration signal, Figure 5 (c) in Fig. 4 is a time-domain graph, a frequency-domain graph and a time-frequency representation of a test bench device fault test bearing vibration signal.

[0043] Figure 6 Fig. 5 is a schematic diagram of a bearing fault diagnosis model structure based on a domain pair adversarial migration network in one or more embodiments of the present application.

[0044] In the figure, 1 is a test motor; 2 is a test coupling; 3 is a test spindle; 4 is a test bearing; 5 is a test bearing seat; 6 is a test counterweight; 7 is a test acceleration sensor; 8 is a test eddy current sensor; 9 is a test impact loading mechanism; 10 is a test magnetic powder brake; and 11 is a test base. DETAILED DESCRIPTION

[0045] Embodiment 1

[0046] In one typical embodiment of the present application, as Figures 1-6As shown, a stamping machine transmission shaft bearing fault diagnosis method is given.

[0047] At present, the stamping machine fault data is lack of labels, and the ordinary test bench device built does not consider the equivalence of the test bench device and the stamping machine in the data layer, resulting in a large distribution difference and feature difference between the target domain data and the source domain data, and the precision of the transferred fault diagnosis model still cannot meet the actual demand.

[0048] Based on this, as Figure 1 shown, the embodiment provides a stamping machine transmission shaft bearing fault diagnosis method, comprising:

[0049] obtaining the vibration signal of the stamping machine bearing to be diagnosed;

[0050] using the target domain coding network optimized by training to extract high-dimensional features of the vibration signal of the stamping machine bearing to be diagnosed, to obtain the stamping machine bearing feature information;

[0051] the stamping machine bearing feature information is input into the pre-trained bearing fault diagnosis model, and the fault diagnosis result of the stamping machine bearing to be diagnosed is output;

[0052] Wherein, the bearing fault diagnosis model is: a test bench device consistent with the inherent frequency, fault frequency, rotating frequency and impact load frequency of the stamping machine transmission shaft is constructed, the test bench device is configured with multi-class fault test bearings 4, the vibration signal data set of the health state of the test bench device and the vibration signal data set of each type of fault are obtained, and the vibration signal data set of the unknown state of the stamping machine, after the vibration signal is preprocessed, the pre-trained bearing fault diagnosis model is obtained by training.

[0053] The stamping machine in the embodiment is a large equipment, and the common models are 12 tons, 16 tons, 25 tons, etc. The test bench device consistent with the characteristic frequency of the stamping machine is established, the length of the stamping machine transmission shaft of the stamping machine is 3000mm, and the structure of the stamping machine transmission shaft, the stamping machine bearing and the stamping machine bearing seat adopts the common bearing support assembly structure form.

[0054] As Figure 2As shown, the test bench device is composed of a test motor 1, a test coupling 2, a test main shaft 3, a test bearing 4, a test bearing seat 5, a test counterweight 6, a test acceleration sensor 7, a test eddy current sensor 8, a test impact loading mechanism 9, a test magnetic powder brake 10 and a test base 11. Among them, the test motor 1 can adjust the rotating speed through the upper control system. The output shaft of the test motor 1 is connected with the test main shaft 3 through the test coupling 2. The test main shaft 3 is supported by two test bearings 4, the test bearing 4 adopts a rolling bearing and is fixed on the test bearing seat 5. Three annular test counterweights 6 are assembled on the test main shaft 3. The test bearing 4 far away from the test motor 1 is a test bearing 4 to be tested. The test acceleration sensor 7 is installed on the test bearing seat 5 to measure the acceleration signal of the test bearing seat 5. The test eddy current sensor 8 is installed at the middle position of the test main shaft 3 to measure the displacement signal of the test main shaft 3. The test impact loading mechanism 9 applies impact load to the test bearing seat 5. The test magnetic powder brake 10 is used as a load device at the end of the test bench device and is connected with the test main shaft 3 through the test coupling 2. The test motor 1, the test bearing seat 5 and the load device are connected with the test base 11 through T-shaped sliding grooves and are fixed by bolts and nuts.

[0055] The characteristic frequencies mainly include bearing fault frequency, transmission shaft natural frequency, impact load frequency and rotating frequency. The test bench device includes a main shaft module, a signal acquisition module, an impact loading module, a motor module and a load module. The main shaft module is composed of the test main shaft 3, the test counterweight 6, the test bearing 4 and the test bearing seat 5, and includes:

[0056] Through modal simulation analysis, the first four order natural frequencies of the transmission shaft of the punch press are obtained . Among them, CY represents the punch press.

[0057] The size and spatial distribution position of the test counterweight are parameterized.

[0058] The error between the first four order natural frequencies of the test main shaft 3 of the test bench device and the first four order natural frequencies of the transmission shaft of the punch press is taken as the objective function, and the adaptive multi-objective optimization algorithm is used to optimize the parameters of the test counterweight 6. Among them, SY represents the test bench device.

[0059] According to the principle that the fault frequencies of the bearings of the punch press and the test bearings 4 are consistent, the inner diameter, the outer diameter and the number of rolling bodies of the test bearings 4 are selected.

[0060] Specifically, the design steps of the test bench are as follows:

[0061] S1: According to the actual size of the punch transmission shaft, punch bearing and punch bearing seat system, a three-dimensional model is established, and the material is 42CrMo steel.

[0062] S2: Modal simulation analysis is adopted, and elastic support is added at the support of the punch bearing, and the basic stiffness is set to N / m. Hexahedral mesh division is adopted, and the element size is less than 0.5 mm. The first four order natural frequencies of the punch transmission shaft are extracted , as shown in Figure 3 , Figure 4 .

[0063] S3: The test spindle 3 of the test bench device is a long straight shaft, and the shaft diameter is 35 mm.

[0064] S4: The length, outer diameter and distribution position of the test weight 6 are set as independent variable parameters, and the test bench device is subjected to modal simulation analysis under the same conditions, and the first four order natural frequencies of the test spindle 3 are obtained .

[0065] S5: The error between and is taken as the objective function, the error constraint is set, that is, the relative error of the first two order natural frequencies is less than 3%, and the relative error of the remaining order natural frequencies is less than 5%, the reasonable constraint range of the test weight 6 parameters is set, and the adaptive multi-objective optimization algorithm is adopted to optimize the test weight 6 parameters.

[0066] S6: After executing the optimization algorithm, if the first two order natural frequencies are out of tolerance, the test bench device shaft diameter can be modified and the optimization is repeated, if the remaining order natural frequencies are out of tolerance, the test weight 6 parameters can be adjusted near the candidate point and the candidate point simulation result is updated. In embodiment 1, the simulation optimization result of the dynamic characteristics of the test bench device is shown in Figure 3 , Figure 4 , it can be seen that the first four order natural frequencies meet the error requirements.

[0067] S7: According to the principle of consistent fault frequency of punch bearing and test bearing 4, the inner diameter, outer diameter and number of rolling bodies of test bearing 4 are selected, and the calculation method of various fault frequencies is as follows:

[0068] (1);

[0069] Wherein, is the fault frequency of the outer ring raceway of the inner ring of test bearing 4, is the fault frequency of the inner ring raceway of the outer ring of test bearing 4, is the fault frequency of the cage of test bearing 4, is the fault frequency of the roller of test bearing 4, is the rotating frequency, The outer diameter of the test bearing 4 is The inner diameter of the test bearing 4 is Z The number of rolling elements is

[0070] In the embodiment, the stamping machine bearing parameters are = 300 mm and = 210 mm, and the material is G20Cr2Ni4A. According to the test bearing fault frequency calculation method, the material and the number of rolling elements are maintained unchanged, and the test bearing 4 parameters are determined as = 50 mm and = 35 mm.

[0071] The signal acquisition module includes a test bearing seat 5 vibration contact type acquisition module and a stamping machine transmission shaft and test spindle 3 vibration non-contact type acquisition module. Among them, the test acceleration sensor 7 is responsible for collecting the test bearing seat 5 vibration signal, and the test eddy current sensor 8 is responsible for collecting the stamping machine transmission shaft and test spindle 3 vibration signal.

[0072] An impact loading module is added to directly excite the test bearing seat 5, and the impact loading module is composed of a piezoelectric actuator and a controller.

[0073] The test bearing seat 5 is directly excited to keep consistent with the impact frequency of the stamping machine working.

[0074] A motor module and a load module are configured to ensure that the rotational speed of the test spindle of the test bench device and the transmission shaft of the stamping machine is the same.

[0075] A plurality of fault test bearings 4 are configured on the test bench device to construct the test bench device health state vibration signal dataset and the various fault vibration signal datasets, as well as the stamping machine unknown state vibration signal dataset.

[0076] The vibration signal preprocessing and feature amplitude normalization include:

[0077] The collected vibration signal is preprocessed, including:

[0078] De-noising: wavelet filtering is used to remove noise in the vibration signal, so as to eliminate the interference of noise on the fault diagnosis result.

[0079] Filtering: through a comb filter, the rotating frequency component and its multiple frequency components contained in the vibration signal are removed, so as to eliminate the interference of the rotating frequency on the fault diagnosis result.

[0080] ​Decomposition: extract the impact load frequency components and their multiple components in the vibration signals of the press bearing chock and the test bearing chock 5 by synchronous compression transform (SST) and , the press bearing and the test bearing 4 fault frequency components and their multiple components and . Among them, c represents the impact load frequency component, zz represents the bearing fault frequency component of the bearing chock vibration signal. Extract the first four order natural frequency components in the vibration signals of the press drive shaft and the test main shaft 3 by SST and , the press bearing and the test bearing 4 fault frequency components and their multiple components and . Among them, g represents the natural frequency component, zc represents the bearing fault frequency component of the vibration signals of the press drive shaft and the test main shaft 3.

[0081] The preprocessed vibration signal is subjected to feature amplitude normalization processing, including:

[0082] The static stiffness of the press bearing chock is obtained by static and harmonic response simulation analysis and the first four order modal stiffness of the press drive shaft , and the static stiffness of the test bearing chock 5 of the test bench device and the first four order modal stiffness of the test main shaft 3 , as shown in Figure 3 , Figure 4 . Among them, the static stiffness of the press bearing chock, the static stiffness of the test bearing chock 5, the first four order natural frequency of the press drive shaft, the first four order natural frequency of the test main shaft 3, the first four order modal stiffness of the press drive shaft, the first four order modal stiffness of the test main shaft 3 are shown in Table 1:

[0083]

[0084] The feature amplitude proportionality coefficient between the corresponding static stiffness and modal stiffness of the press and the test bench device is calculated, and the feature amplitude normalization processing is performed on each frequency component in the preprocessed vibration signal of the test bench device.

[0085] The short-time Fourier transform (STFT) is performed on each single component (including a certain feature frequency component and its multiple components) after preprocessing and feature amplitude normalization, to obtain the time-frequency representation of the single component. The time-frequency representations of all single components are combined to obtain the time-frequency representation of the vibration signal.

[0086] The vibration signals are collected from the test rig device, preprocessed and normalized in feature amplitude to construct the health state and fault data sets of the test rig device and the unknown state data set of the punch.

[0087] Specifically, in the embodiment, the preprocessing and feature amplitude normalization include:

[0088] The wavelet filtering is used to remove the noise in the vibration signal, and the implementation steps are as follows:

[0089] S1: Multi-scale wavelet decomposition is performed on the noisy vibration signal to obtain wavelet coefficients of different frequency bands and approximation coefficients . Wherein, , L is the maximum decomposition layer, and L = 5.

[0090] S2: Calculate the noise variance estimate value of the wavelet coefficient, and the calculation method is .

[0091] S3: Generate a threshold value according to the characteristics of the sub-band . Wherein, is the number of the layer sub-band coefficient.

[0092] S4: Soft threshold processing is performed on the wavelet coefficient , and the calculation method is as follows:

[0093] (2);

[0094] S5: Use and to perform inverse wavelet transform to obtain the denoised vibration signal.

[0095] The rotating frequency component and its multiple frequency components contained in the vibration signal are removed by using a comb filter, and the notch bandwidth of the comb filter is 50 Hz, and the notch frequency is equal to the rotating frequency / 60 of the punch transmission shaft and the test main shaft 3. Wherein, is the rotating speed of the punch transmission shaft and the test main shaft 3.

[0096] The impact load frequency component and its multiple frequency components and in the vibration signal of the punch bearing seat / test bearing seat 5, the fault frequency component and its multiple frequency components and of the punch bearing and the test bearing 4, and the first four order natural frequency components and in the vibration signal of the punch transmission shaft and the test main shaft 3 are extracted by SST., the failure frequency components of the press bearing and test bearing 4 and their multiple frequency components and , the implementation steps are as follows:

[0097] S1: Perform short-time Fourier transform (STFT) on the vibration signal to obtain time-frequency representation S ( t , ω ).

[0098] S2: Estimate the instantaneous frequency of by the phase difference method, the calculation method is as follows:

[0099] (3);

[0100] S3: According to the instantaneous frequency, the energy of S ( t , ω ) is redistributed to generate compressed time-frequency representation T ( t , ω ), the calculation method is as follows:

[0101] (4);

[0102] S4: Separate each component using T ( t , ω ), the calculation method is as follows:

[0103] (5);

[0104] wherein, and are the frequency range of the component.

[0105] The preprocessed vibration signal is subjected to characteristic amplitude normalization processing, and the implementation steps are as follows:

[0106] Static simulation analysis is adopted, standard earth gravity is applied, static force is applied on the press bearing seat and test bearing seat 5, and the static stiffness and of the press bearing seat and test bearing seat 5 are calculated. Among them, and are the maximum displacement of the press bearing seat and test bearing seat 5.

[0107] Harmonic response simulation analysis is adopted, simple harmonic excitation force is applied on the press transmission shaft and test main shaft 3, and modal stiffness and are calculated. Among them, and The first order modal maximum displacement.

[0108] The corresponding characteristic amplitude ratio coefficient between the static stiffness and the modal stiffness of the press and the test rig device is calculated , and the calculation method is as follows:

[0109] (6);

[0110] The characteristic amplitude normalization processing is performed on each frequency component in the vibration signal of the pretreated test rig device, and the calculation method is as follows:

[0111] (7);

[0112] The short-time Fourier transform (STFT) is performed on each single component (including a certain characteristic frequency component and its multiple frequency component) after pretreatment and characteristic amplitude normalization, to obtain the time-frequency representation of the single component. The time-frequency representations of all single components are combined to obtain the time-frequency representation of the vibration signal, and the calculation method is as follows:

[0113] (8);

[0114] (9);

[0115] After pretreatment and characteristic amplitude normalization, the energy of the obtained time-frequency feature is mainly concentrated in the fault frequency, the impact frequency and the natural frequency. As shown in Figure 5 (a) and Figure 5 (b), for the time-frequency features of the test bearing 4 and the press bearing in the healthy state, the energy level remains basically the same; as shown in Figure 5 (c), for the time-frequency feature of the test rig fault test bearing 4, the energy is more concentrated in the bearing fault frequency.

[0116] Tests are carried out on the test rig device and the press built, and according to the collected vibration signal data set, through pretreatment, characteristic amplitude normalization and time-frequency conversion, a data set for supporting subsequent fault diagnosis models is constructed, including:

[0117] Time-frequency representation data set of unknown state vibration signal of press .

[0118] Time-frequency representation data set of healthy state vibration signal of test rig device .

[0119] Time-frequency representation data set of fault vibration signal of test rig device .

[0120] As Figure 6 shown, the bearing fault diagnosis model based on domain adversarial transfer network includes the following contents:

[0121] Considering the distribution difference between source domain data and target domain data, a corresponding source domain encoding network and a target domain encoding network are respectively constructed based on a deep convolutional neural network.

[0122] Among them, the constructed deep convolutional neural network is composed of multiple convolutional layers, batch normalization layers, pooling layers, fully connected layers and a Softmax classification layer, and the output layers of which correspond to classifiers C and D respectively, used to output the probability scores of source domain and target domain fault diagnosis.

[0123] The model migration strategy is introduced, and the pre-trained is used as the basis to fix the low-level model parameters of . Using a small amount of labeled target domain data, combined with a small learning rate, the high-level model parameters of are fine-tuned using the stochastic gradient descent method to obtain the optimized target domain encoding network , realizing the preliminary migration and reuse of source domain knowledge.

[0124] The adversarial training technique is introduced to optimize , reducing the distribution difference 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, and by iteratively updating the model parameters, the output features of and are made indistinguishable.

[0125] Specifically, the bearing fault diagnosis model based on domain adversarial transfer network is implemented as follows:

[0126] Step 1: The labeled source domain data obtained by the test bench device and the small amount of labeled target domain data obtained by the punch press are randomly divided into training and testing groups according to 50% each.

[0127] Step 2: The labeled source domain training group data is used to supervise the training of and C, and a parallel training method is used, with the average value of Softmax cross-entropy as the training loss, and the adaptive learning rate optimization algorithm is used for gradient update and loss value minimization, with a maximum training number of 150 and a learning rate of 0.0001.

[0128] Step 3: The model migration strategy is used to initialize the model parameters of ​ By using the method of freezing specific layers and tuning, diagnostic knowledge of the source domain is migrated to , and the specific steps are as follows:

[0129] S1: Replace the B6 layer with the number of categories of target domain fault diagnosis, and train the target domain training group data with a small amount of labels by freezing the B1-B5 layer parameters, using a small learning rate and a stochastic gradient descent method to tune the B6 layer parameters.

[0130] S2: Fix the parameters of the previous B1-B4 layers, and tune the B5 and B6 layer parameters using a small learning rate to find the optimal number of transferable layers.

[0131] S3: Repeat the model migration process in S2 until the set classification accuracy or iteration number is reached. Determine the optimal number of transferable layers suitable for target domain fault diagnosis, and obtain the tuned migration fault diagnosis model.

[0132] Step 4: Input the labeled source domain training group data and the target domain training group data with a small amount of labels into and , extract features and input them into D. Use the adversarial training technique to update the parameters of and D alternately.

[0133] Step 5: Repeat the training of step 4 so that D gradually cannot distinguish between source domain and target domain data, gradually reduce the data distribution difference between the source domain and the target domain, until the set iteration threshold or the set diagnostic accuracy threshold is reached.

[0134] Step 6: In the test phase, input the target domain test data into G t to extract features, and directly give the diagnostic results of each state category by C.

[0135] Using deep convolutional neural networks to extract high-dimensional features, migrating labeled source domain data knowledge for target domain fault diagnosis, and using an adversarial transfer learning strategy to make the network learn features suitable for each domain to reduce the data distribution difference between the source domain and the target domain, and improve the fault diagnosis accuracy.

[0136] Embodiment 2

[0137] In another typical embodiment of the present application, as shown in Figures 1-6 , a stamping machine transmission shaft bearing fault diagnosis system is given, comprising:

[0138] A data acquisition module configured to acquire a vibration signal of a stamping machine bearing to be diagnosed.

[0139] The feature extraction module is configured to perform high-dimensional feature extraction on the vibration signal of the stamping machine bearing to be diagnosed by using the target domain coding network optimized by training, to obtain bearing feature information.

[0140] The diagnosis module is configured to input the bearing feature information into a pre-trained bearing fault diagnosis model, and output a fault diagnosis result of the stamping machine bearing to be diagnosed.

[0141] The pre-trained bearing fault diagnosis model is constructed by constructing a test bench device consistent with the inherent frequency, fault frequency, rotating frequency and impact load frequency of the stamping machine transmission shaft, configuring multiple types of fault test bearings 4 on the test bench device, obtaining a health state vibration signal data set and various fault vibration signal data sets of the test bench device, and an unknown state vibration signal data set of the stamping machine, and obtaining the pre-trained bearing fault diagnosis model by training after pre-processing the vibration signal.

[0142] The pre-processing includes denoising, filtering and decomposition, and the feature amplitude of the pre-processed vibration signal is normalized.

[0143] The working method of the stamping machine transmission shaft bearing fault diagnosis system is referred to the stamping machine transmission shaft bearing fault diagnosis method in the embodiment 1, and will not be repeated here.

[0144] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Those skilled in the art can make various modifications and changes to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A press drive shaft bearing fault diagnostic method characterized by, The method comprises the following steps: acquiring a vibration signal of a stamping machine bearing to be diagnosed; extracting high-dimensional features of the vibration signal of the stamping machine bearing to be diagnosed by using a target domain coding network optimized by training, to obtain stamping machine bearing feature information; inputting the stamping machine bearing feature information into a pre-trained bearing fault diagnosis model, and outputting a fault diagnosis result of the stamping machine bearing to be diagnosed; wherein the pre-trained bearing fault diagnosis model is: constructing a test rig device consistent with the inherent frequency, fault frequency, rotating frequency and impact load frequency of the stamping machine transmission shaft, configuring the test rig device with multiple types of fault test bearings, acquiring a health state vibration signal dataset and various fault vibration signal datasets of the test rig device, and a vibration signal dataset of the stamping machine bearing with a small amount of labels and a vibration signal dataset of the stamping machine in an unknown state, and obtaining the pre-trained bearing fault diagnosis model through preprocessing and model training; constructing a test rig device consistent with the inherent frequency of the stamping machine transmission shaft comprises: the test rig device comprises a main shaft module composed of a test main shaft, a test counterweight, a test bearing and a test bearing seat, and the multi-order inherent frequencies of the stamping machine transmission shaft are acquired through modal simulation analysis; taking the error between the multi-order inherent frequencies of the test main shaft of the test rig device and the multi-order inherent frequencies of the stamping machine transmission shaft as an objective function, and optimizing the counterweight parameters, so that the inherent frequencies of the test main shaft of the test rig device and the stamping machine transmission shaft are consistent; constructing a test rig device consistent with the fault frequency of the stamping machine transmission shaft comprises: selecting the inner diameter, outer diameter and number of rolling elements of the test bearing according to the fault frequency consistency principle of the stamping machine bearing and the test bearing of the test rig device; training the bearing fault diagnosis model comprises: constructing a source domain coding network for source domain data and a target domain coding network for target domain data; introducing a model migration strategy, fixing the low-level model parameters of the pre-trained source domain coding network, adjusting the high-level model parameters of the source domain coding network using target domain data, and obtaining an optimized target domain coding network; optimizing the target domain coding network through adversarial training to obtain a target domain coding network optimized by training; iteratively updating the bearing fault diagnosis model; the source domain coding network and the target domain coding network correspond to a first classifier and a second classifier respectively, the first classifier is used to output probability scores of source domain and target domain fault diagnosis, and the second classifier is a discriminator used to distinguish data of the source domain and the target domain; inputting the labeled source domain training group data and the partially labeled target domain training group data into the source domain coding network and the target domain coding network respectively, extracting features and inputting them into the second classifier, and alternately updating the parameters of the target domain coding network and the second classifier by using adversarial training technology; inputting the stamping machine unknown state vibration signal dataset as target domain test data into the target domain coding network optimized by training to extract high-dimensional features, and giving a diagnosis result by the first classifier.

2. The press drive shaft bearing fault diagnostic method of claim 1, wherein, The preprocessing comprises denoising, filtering and decomposition of the vibration signal, and then feature amplitude normalization processing.

3. The press machine drive shaft bearing fault diagnostic method of claim 1, wherein, Based on a deep convolutional neural network, a source domain coding network and a target domain coding network are established.

4. A press drive shaft bearing fault diagnostic system characterized by, The method comprises the following steps: The data acquisition module is configured to acquire a vibration signal of a stamping machine bearing to be diagnosed. The feature extraction module is configured to perform high-dimensional feature extraction on the vibration signal of the stamping machine bearing to be diagnosed by using a target domain coding network optimized by training, to obtain stamping machine bearing feature information. The diagnosis module is configured to input the stamping machine bearing feature information into a pre-trained bearing fault diagnosis model, and output a fault diagnosis result of the stamping machine bearing to be diagnosed. The pre-trained bearing fault diagnosis model is constructed by: constructing a test bench device consistent with the natural frequency, fault frequency, rotating frequency and impact load frequency of the stamping machine transmission shaft, configuring the test bench device with multi-class fault test bearings, acquiring a healthy state vibration signal dataset and various fault vibration signal datasets of the test bench device, and a vibration signal dataset of the stamping machine bearing with a small amount of labels, a vibration signal dataset of the stamping machine in an unknown state, and pre-processing and model training, to obtain the pre-trained bearing fault diagnosis model. The test bench device consistent with the natural frequency of the stamping machine transmission shaft comprises: The test bench device comprises a main shaft module composed of a test main shaft, a test counterweight, a test bearing and a test bearing seat, and the multi-order natural frequency of the stamping machine transmission shaft is obtained by modal simulation analysis; The error between the multi-order natural frequency of the test main shaft of the test bench device and the multi-order natural frequency of the stamping machine transmission shaft is taken as an objective function, and the counterweight parameters are optimized so that the natural frequencies of the test main shaft of the test bench device and the stamping machine transmission shaft are consistent; The test bench device consistent with the fault frequency of the stamping machine transmission shaft comprises: The inner diameter, outer diameter and number of rolling elements of the test bearing are selected according to the principle that the fault frequency of the stamping machine bearing is consistent with that of the test bearing of the test bench device; The training of the bearing fault diagnosis model comprises: A source domain coding network is constructed for source domain data, and a target domain coding network is constructed for target domain data; A model migration strategy is introduced, the low-level model parameters of the pre-trained source domain coding network are fixed, the high-level model parameters of the source domain coding network are adjusted using target domain data, and an optimized target domain coding network is obtained; The target domain coding network is optimized by adversarial training, and a training-optimized target domain coding network is obtained; The bearing fault diagnosis model is iteratively updated; The source domain coding network and the target domain coding network correspond to a first classifier and a second classifier respectively, the first classifier is used to output probability scores of source domain and target domain fault diagnosis, and the second classifier is a discriminator used to distinguish source domain and target domain data; The labeled source domain training group data and the partially labeled target domain training group data are input into the source domain coding network and the target domain coding network respectively, features are extracted and input into the second classifier, and the parameters of the target domain coding network and the second classifier are alternately updated by using adversarial training technology; The stamping machine unknown state vibration signal dataset is input into the training-optimized target domain coding network as target domain test data to extract high-dimensional features, and the first classifier gives a diagnosis result.

5. The press drive shaft bearing fault diagnostic system of claim 4, wherein, The preprocessing comprises denoising, filtering and decomposition of the vibration signal, and then feature amplitude normalization processing.

Images