Rectangular laminated rubber isolation bearing and its in-situ detection method for aging damage

Abstract

The present invention discloses a rectangular laminated rubber seismic isolation bearing, comprising a bearing body and bolt cover plates arranged on the four sides of the bearing body, wherein the bolt cover plates are provided with bolt holes (parallel grooves), and the bolt cover plates are connected to the bearing body by passing bolts through the bolt holes, and a transducer is provided between the bolt cover plates and the side surfaces of the bearing body; the present invention also discloses a method for realizing in-situ detection of aging damage; the present invention can obtain accurate identification results of the degree of aging damage of the bearing.

Description

Technical Field

[0001] The present invention relates to the technical field of structural damage identification and health monitoring, in particular to a rectangular laminated rubber seismic isolation bearing and a method for realizing in-situ detection of aging damage thereof. Background Art

[0002] The promotion and application of seismic isolation buildings are crucial to reducing casualties and economic and property losses that may be caused by earthquakes. With the formal implementation of the "Regulations on Seismic Management of Construction Projects" (implemented in September 2021), its mandatory regulations on the application of seismic isolation technology will further increase the number of seismic isolation buildings. As the core component for realizing the seismic isolation function of seismic isolation buildings, the performance change of seismic isolation bearings is an important factor leading to performance changes of seismic isolation buildings throughout their life cycle. At present, the most widely used seismic isolation bearings are laminated rubber seismic isolation bearings with the advantages of strong deformation ability, low production cost, and strong seismic isolation ability.

[0003] Laminated rubber seismic isolation bearings are mainly composed of steel plate layers and rubber layers. Among them, the rubber layer is easily affected by environmental factors such as temperature, air, and moisture, and will undergo aging phenomena such as hardening and cracking, thereby reducing the deformation capacity of the bearing and weakening its seismic isolation capacity. The adverse effects of aging on laminated rubber seismic isolation bearings are mainly manifested in four aspects: (1) Strength reduction: During aging, the molecular chain structure of the rubber material will break or lose cross-linking, resulting in a decrease in the strength of the material; (2) Damping characteristics deteriorate: Rubber materials have a certain damping effect, which can absorb seismic energy and reduce structural vibration. After aging, its damping characteristics will deteriorate, reducing its shock absorption effect; (3) Deformation capacity deterioration: Aging weakens the deformation capacity of the rubber material, thereby increasing the seismic force on the structure; (4) Reduced service life: Aging reduces the durability of the rubber material, thereby shortening the service life of the bearing.

[0004] At present, aging has become one of the main problems affecting the service life and isolation performance of laminated rubber isolation bearings. It is very important to accurately and quickly detect and evaluate bearing aging damage.

[0005] The traditional method has the following defects:

[0006] 1. Low accuracy in identifying the degree of aging damage: Changes in axial pressure can affect the bearing's monitoring signal, thereby affecting the identification of the bearing's aging damage degree. Specifically, while the aging degree remains unchanged, changes in axial pressure can alter the characteristics of the monitoring signal, leading to a misinterpretation of the change in bearing aging damage.

[0007] 2. Difficulty in identifying the degree of aging damage: Only through forward analysis and feature extraction of the monitoring signal can we identify changes in the monitoring signal and determine that the bearing status has changed, but we cannot accurately identify the degree of aging damage of the bearing.

[0008] 3. Difficulty in repairing or replacing the transducer: Due to space limitations, it is difficult to repair or replace the transducer if it fails or the connection between the transducer and the support becomes loose. Summary of the Invention

[0009] In order to solve the problems existing in the prior art, the purpose of the present invention is to provide a rectangular laminated rubber seismic isolation bearing and a method for in-situ detection of aging damage thereof, which can obtain accurate identification results of the degree of aging damage of the bearing.

[0010] To achieve the above-mentioned purpose, the technical solution adopted by the present invention is: a rectangular laminated rubber seismic isolation bearing, comprising a bearing body and a bolt cover plate arranged on the four sides of the bearing body, wherein the bolt cover plate is provided with bolt holes, and the bolt cover plate is connected to the bearing body by passing bolts through the bolt holes, and a transducer is provided between the bolt cover plate and the side of the bearing body.

[0011] As a further improvement of the present invention, the support body has no upper and lower rubber layers and the upper and lower steel plate layers are thickened (compared with traditional rectangular laminated rubber bearings); circular bolt holes corresponding to the parallel grooves are arranged on the upper and lower steel plate layers of the support body.

[0012] As a further improvement of the present invention, a groove for fixing the transducer is provided on the bolt cover.

[0013] As a further improvement of the present invention, the bolt holes are parallel grooves and are at a certain distance from the edge of the cover plate.

[0014] As a further improvement of the present invention, the height and width of the bolt cover should be smaller than the height and width of the support body respectively.

[0015] As a further improvement of the present invention, the bolts connecting the bolt cover plate and the support body should be subjected to the same magnitude of pre-tightening force.

[0016] The present invention also provides a method for in-situ detection of aging damage, which is implemented using the rectangular laminated rubber isolation bearing as described above, and the method comprises the following steps:

[0017] Step 1: Install the vibrator and sensor on the side of the support, and install the center of the bottom surface of the transducer at half the height of the support;

[0018] Step 2: Use a sweep frequency signal to stimulate the four vibrators in sequence, and collect monitoring signals through sensors;

[0019] Step 3: Filter the collected monitoring signal and then perform wavelet packet decomposition to obtain the wavelet packet energy spectrum;

[0020] Step 4: Perform calibration testing and establish a deep learning model for bearing aging damage identification;

[0021] Step 5: Substitute the wavelet packet energy spectrum obtained by decomposing the monitoring signal into the deep learning model as an input parameter to obtain the aging damage degree of the bearing.

[0022] As a further improvement of the present invention, in step 3, the collected monitoring signal is filtered as follows:

[0023] (1) First, calculate the monitoring signal length N_fft, set the passband lower limit f_low and upper limit f_high, and input the signal sampling rate FS;

[0024] (2) Perform fast Fourier transform on the monitoring signal to obtain its frequency domain signal:

[0025] X(f)=∫[Signal×e (2jπtf) ]dt

[0026] Where X(f) represents the frequency domain signal, the amplitude and phase information at frequency f; Signal represents the monitoring signal; e (2jπtf) is a complex exponential function, where j is the imaginary unit;

[0027] (3) Calculate the cutoff frequencies N1 and N2 corresponding to the lower and upper limits of the passband range:

[0028] N1=ceil(f_low / (FS / N_fft))

[0029] N2 = floor(f_high / (FS / N_fft))

[0030] Among them, the ceil(x) function returns the smallest integer greater than or equal to x, that is, rounding up; the floor(x) function returns the largest integer less than or equal to x, that is, rounding down;

[0031] (4) The frequency components above the cutoff frequency N2 and below the cutoff frequency N1 in the frequency domain signal are set to 0;

[0032] (5) Perform inverse Fourier transform on the frequency domain signal to achieve the frequency filtering effect:

[0033] Y_Signal=∫[X(f)×e (2jπtf) ]df

[0034] Among them, Y_Signal represents the filtered time domain signal.

[0035] As a further improvement of the present invention, in step 3, wavelet packet decomposition is performed to obtain the wavelet packet energy spectrum as follows:

[0036] (1) Determine the wavelet packet basis db, determine the decomposition level h, perform discrete wavelet transform, and obtain the approximate coefficients and detail coefficients at each level;

[0037] The approximation coefficient A(h) and detail coefficients D(h), D(h-1), …, D(1) are calculated using the following formulas:

[0038] A(h)=x*phi_h(t)+(x*psi_{h,j}(t))_j

[0039] D(h)=(x*psi_{h,j}(t))_j

[0040] D(h-1)=(D(h)*psi_{h-1,j}(t))_j

[0041] D(h-2)=(D(h-1)*psi_{h-2,j}(t))_j

[0042] …

[0043] D(1)=(D(2)*psi_{1,j}(t))_j

[0044] Where * represents the convolution operation, phi_k(t) and psi_{k,j}(t) are the approximate function and detail function of the wavelet packet basis db, and j represents the position in the hth layer;

[0045] (2) Further decompose the detail coefficients of each level to obtain the wavelet packet decomposition results;

[0046] At the kth level, the detail coefficient D(k) is decomposed into 2^k frequency bands. Each frequency band is called a node, numbered from 0 to 2^k-1. For each node i, the node coefficient is calculated by the following formula:

[0047] coef(i)=(D(k)*psi_{k,i}(t))_i

[0048] Among them, psi_{k,i}(t) is the detail function of the wavelet packet basis db of the kth layer, and _i represents the position in the kth layer;

[0049] (3) Solve the norm square of all nodes in the hth layer, that is, the sum of squares:

[0050] E(i)=||wpcoef(w_signal,[h,i-1])||2 2

[0051] Among them, E(i) represents the energy of the i-th node, wpcoef(w_signal,[h,i-1]) represents the coefficient of the i-1-th node in the h-th layer in the wavelet packet decomposition result, and ||.||2 represents the L2 norm, that is, the Euclidean norm. 2 means squaring the result of the norm.

[0052] As a further improvement of the present invention, the step 4 is specifically as follows:

[0053] (1) Design k supports, and apply different axial pressures F in an arithmetic progression to each support during the aging process through the tooling L (Axial pressure during aging), the axial pressure variation range is 0MPa-the maximum allowable stress of the support. The accelerated aging test is carried out in an aging test chamber, the temperature is T℃, the single aging time is D days, and the number of aging times is L;

[0054] (2) Take out the bearing every D days and place it at room temperature for a certain period of time, then conduct compressive elastic modulus and shear elastic modulus tests to obtain the mechanical properties of the bearing at different aging times t; obtain the mechanical properties of all bearings at different aging times t and different axial compression F by wave method test. B The monitoring signal under the axial pressure during the wave method test is filtered and decomposed into wavelet packets to obtain its wavelet packet energy spectrum. The obtained signal matrix is ​​as follows:

[0055]

[0056] Among them, MS IJ Indicates the monitoring signal when the exciter number is I and the sensor number is J.

[0057] When the axial pressure during aging is F L , aging time is t, axial pressure during wave test is F B When the monitoring signal matrix is ​​MS IJ (F L ,t,F B ).

[0058] (3) The wavelet packet energy spectrum of the bearing in different states is used as the input parameter, and the aging damage degree of the bearing (compressive elastic modulus E PR-C , shear modulus E PR-S and aging time t PR ) as the output parameter, and a deep learning model (multi-layer perceptron) was used for modeling. This deep learning model is a multi-task regression model divided into three sub-models, MLP1-3: Sub-model MLP1 is a compressive elastic modulus prediction model, whose output is the compressive elastic modulus; sub-model MLP2 is a shear elastic modulus prediction model, whose output is the shear elastic modulus; and sub-model MLP3 is an aging time prediction model, whose output is the aging time.

[0059] The output parameter matrix is ​​as follows:

[0060]

[0061] The specific steps to build a prediction model are as follows:

[0062] 1) Divide the database into three parts: training set, validation set and test set according to a certain ratio, select the loss function, optimization algorithm and evaluation index, and determine the values ​​of each hyperparameter in the model.

[0063] 2) Train the model on the training set and continuously adjust the weight coefficients through backpropagation and optimization algorithms to minimize the loss function.

[0064] 3) Evaluate the model's predictive performance using the validation set and optimize performance by adjusting hyperparameter values. Stop training after adjusting the model's hyperparameter values ​​several times and save the optimal trained model for later use.

[0065] The beneficial effects of the present invention are:

[0066] 1. The present invention avoids the problem of difficulty in repairing or replacing the transducer by installing the transducer on the side of the support;

[0067] 2. The present invention obtains the changing pattern of the monitoring signal of the bearing under different states by conducting calibration experiments of aging damage and axial pressure on the bearing in advance, and constructs an aging damage identification model of the bearing based on deep learning, which can accurately identify the degree of aging damage of the bearing. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 Schematic diagram of drilling holes in the support body according to an embodiment of the present invention;

[0069] Figure 2 A top view of a rectangular laminated rubber isolation bearing in an embodiment of the present invention;

[0070] Figure 3 The front, left and right views of the rectangular laminated rubber isolation bearing in the embodiment of the present invention;

[0071] Figure 4 This is a top view of a bolt cover plate in an embodiment of the present invention;

[0072] Figure 5 This is a front view of a bolt cover plate in an embodiment of the present invention;

[0073] Figure 6 This is a left view of the bolt cover plate in the embodiment of the present invention;

[0074] Figure 7 for Figure 2 Section 1-1;

[0075] Figure 8 Schematic diagram of axial pressure application in an aging experiment in an embodiment of the present invention.

[0076] Reference numerals:

[0077] 1. Bolt cover, 2. Bolt, 3. Transducer, 4. Bolt hole, 5. Support body, 6. Parallel groove, 7. Groove, 8. Thickened steel plate layer, 9. Non-thickened steel plate layer, 10. Rubber layer, 11. Pressurized steel plate, 12. Large bolt hole. DETAILED DESCRIPTION

[0078] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0079] Example

[0080] like Figure 1-Figure 7 As shown, a rectangular laminated rubber seismic isolation bearing includes a bearing body 5 and a bolt cover plate 1 arranged on the four sides of the bearing body 5, wherein the bolt cover plate 1 is provided with a parallel groove 6, and the bolt cover plate 1 is connected to the bearing body 5 by passing the bolt 2 through the parallel groove 6, and a transducer 3 is provided between the bolt cover plate 1 and the side of the bearing body 5.

[0081] In this embodiment, since the bolt connection method requires drilling holes in the upper and lower steel plates of the support to form bolt holes 4, the conventional laminated rubber seismic isolation support needs to be improved (the upper and lower rubber layers 10 are removed, and the upper and lower steel plate layers 8 are thickened to 10 mm). The bolt connection cover plate 1 used in this embodiment has two grooves 7 with a diameter of 38 mm and a depth of 1 mm, which are used to fix the transducer 3 to ensure that the relative position of the transducer 3 and the support body 5 remains unchanged each time the wave method test is performed. Six parallel grooves 6 are arranged on the bolt connection cover plate 1, so that the transducer 3 can still be arranged at the center position of the side of the support body 5 when the support undergoes axial deformation.

[0082] This embodiment also provides a method for in-situ detection of aging damage, comprising the following steps:

[0083] 1. Preparation: Ensure that the equipment required for the wave method test is operating normally, including: amplifier, collector, exciter and sensor, etc.

[0084] 2. Clean the surface: Clean the side of the support to ensure there is no impurities or dirt to ensure a strong bond between the transducer and the support.

[0085] 3. Transducer Installation: Bolt the transducer (both the exciter and sensor use longitudinal wave transducers with a resonant frequency of 58K, a diameter of 38mm, and a height of 30mm) onto the side of the support. The center of the transducer bottom should be half the height of the support. When bolting, use 12.9-grade M4 bolts (42mm long). Use a torque driver to apply a certain torque value (2.8N·m) to each bolt to ensure consistent pressure on the transducer.

[0086] 4. Monitoring signal acquisition: Use a sweep frequency signal with a duration of 1.0s, an amplitude of ±400V, and a frequency of 1k-100kHz to stimulate the four exciters in sequence, and collect monitoring signals through four sensors (the collection time can be set to 1.5s, and collection starts 0.2s in advance).

[0087] 5. Monitoring signal analysis: First, the collected monitoring signal is filtered and then decomposed into wavelet packets to obtain the wavelet packet energy spectrum (the wavelet packet basis name is 'db5', and the number of decomposition layers is 10).

[0088] 5.1 Filtering

[0089] (1) First, calculate the monitoring signal length N_fft (=1500000), set the passband range lower limit f_low=1000 and upper limit f_high=500000, and the input signal sampling rate FS=1000000;

[0090] (2) Perform fast Fourier transform on the monitoring signal to obtain its frequency domain signal.

[0091] X(f)=∫[Signal×e (2jπtf) ]dt

[0092] Where X(f) represents the frequency domain signal, the amplitude and phase information at frequency f; Signal represents the monitoring signal (time domain signal); e (2jπtf) is a complex exponential function, where j is the imaginary unit.

[0093] (3) Calculate the cutoff frequencies N1 and N2 corresponding to the lower and upper limits of the passband range:

[0094] N1=ceil(f_low / (FS / N_fft))

[0095] N2 = floor(f_high / (FS / N_fft))

[0096] Among them, the ceil(x) function returns the smallest integer greater than or equal to x, that is, rounding up; the floor(x) function returns the largest integer less than or equal to x, that is, rounding down.

[0097] (4) The frequency components in the frequency domain signal that are higher than the cutoff frequency N2 and lower than the cutoff frequency N1 are set to 0.

[0098] (5) Perform inverse Fourier transform on the frequency domain signal to achieve frequency filtering effect.

[0099] Y_Signal=∫[X(f)×e (2jπtf) ]df

[0100] Among them, Y_Signal represents the filtered time domain signal.

[0101] 5.2 Wavelet Packet Decomposition

[0102] (1) The wavelet packet basis is 'db5', the decomposition level is 10, and discrete wavelet transform is performed to obtain the approximate coefficients and detail coefficients at each level.

[0103] The approximation coefficient A(10) and detail coefficients D(10), D(9), …, D(1) can be calculated using the following formulas:

[0104] A(10)=x*phi_10(t)+(x*psi_{10,j}(t))_j

[0105] D(10)=(x*psi_{10,j}(t))_j

[0106] D(9)=(D(10)*psi_{9,j}(t))_j

[0107] D(8)=(D(9)*psi_{8,j}(t))_j

[0108] …

[0109] D(1)=(D(2)*psi_{1,j}(t))_j

[0110] Where * denotes the convolution operation, phi_k(t) and psi_{k,j}(t) are the approximation function and detail function of the 'db5' wavelet, and j denotes the position in the 10th layer.

[0111] (2) Further decompose the detail coefficients of each level to obtain the wavelet packet decomposition results.

[0112] At the kth level, the detail coefficient D(k) is decomposed into 2^k frequency bands. Each frequency band is called a node, numbered from 0 to 2^k-1. For each node i, the node coefficient can be calculated by the following formula:

[0113] coef(i)=(D(k)*psi_{k,i}(t))_i

[0114] Among them, psi_{k,i}(t) is the detail function of the k-th layer 'db5' wavelet, and _i represents the position in the k-th layer.

[0115] (3) Solve the square norm (i.e., sum of squares) of all nodes in the 10th layer

[0116] E(i)=||wpcoef(w_signal,[10,i-1])||2 2

[0117] Where E(i) represents the energy of the i-th node, wpcoef(w_signal,[10,i-1]) represents the coefficient of the i-1-th node in the 10th layer in the wavelet packet decomposition result, and ||.||2 represents the L2 norm (Euclidean norm). 2 means squaring the result of the norm.

[0118] 6. Deep Learning Model Prediction:

[0119] 6.1 Calibration test:

[0120] During the use of the above device, calibration and testing are first required to establish a bearing aging damage identification model. The algorithm establishment process is as follows.

[0121] (1) 14 supports were designed. During the aging process, axial loads of 0 MPa, 2.5 MPa, 5 MPa, 7.5 MPa, 10 MPa, 12.5 MPa, and 15 MPa were applied to each support using a fixture (two supports were used for each axial load). Accelerated aging tests were conducted in an aging test chamber (80°C). Ten aging damage levels were designed (the aging time was 50 days, with each 5-day period representing one aging damage level).

[0122] Schematic diagram of axial pressure application in aging experiment Figure 8 As shown, the middle black part is the support body, and the upper and lower side steel plates 11 are used to apply axial pressure (after the press is pressed, it is fixed with 12.9 grade M24 bolts).

[0123] (2) The bearings were taken out every 5 days and placed at room temperature (23±2℃) for 2 days. Then, the compressive elastic modulus and shear elastic modulus tests (double shear method) were carried out in accordance with the specification "Highway Bridge Plate Rubber Bearings" (JTT 4-2019) to obtain the mechanical properties of the bearings at different aging times. The monitoring signals of all bearings under different axial pressures (1-15MPa, with an interval of 1MPa, a total of 15 axial pressure levels) were obtained through the wave method test. The signals were filtered and wavelet packet decomposition was performed to obtain their wavelet packet energy spectrum. It is worth noting that (1) after the aging test was completed and placed at room temperature for two days, the axial force applied by the tooling was removed and then pressure was applied according to 15 axial pressure levels (1-15MPa, with an interval of 1MPa); (2) the transducer was installed after pressure was applied to the bearing. After the wave method test, the transducer needed to be removed before the axial pressure could be applied again to prevent the bolt cover from restricting the vertical deformation of the bearing.

[0124] The obtained monitoring signal matrix is ​​as follows:

[0125]

[0126] Among them, MS IJ Indicates the monitoring signal when the exciter number is I and the sensor number is J.

[0127] When the axial pressure during aging is F L , aging time is t, axial pressure during wave test is F B When the monitoring signal matrix is ​​MS IJ (F L ,t,F B ).

[0128] (3) The wavelet packet energy spectrum of the bearing under different states is used as the input parameter (one-dimensional vector, 1024×1), and the compressive elastic modulus, shear elastic modulus and aging time of the bearing are used as the output parameters. The output parameter matrix is ​​as follows:

[0129]

[0130] The deep learning model is a multi-task regression model, which is divided into three sub-models MLP1-3: sub-model MLP1 is a compressive elastic modulus prediction model, and its output is the compressive elastic modulus; sub-model MLP2 is a shear elastic modulus prediction model, and its output is the shear elastic modulus; sub-model MLP3 is an aging time prediction model, and its output is the aging time.

[0131] The specific steps are as follows:

[0132] 1) The database was randomly divided into three parts: training set, validation set and test set in the ratio of 0.8:0.1:0.1, and the mean square error MSE was determined as the loss function, Adam as the optimization algorithm and the determination coefficient R2 To evaluate the performance, we determined the values ​​of the hyperparameters in the model (the initial values ​​were: 5 hidden layers, 1024 neurons per layer, 0.001 learning rate, 64 batch size, and 20 training iterations). 2 The calculation formula is:

[0133]

[0134]

[0135] Where N represents the number of samples, Represents the predicted value of the model, y i represents the true value (label) of the model, for.

[0136] 2) Train the model on the training set and continuously adjust the weight coefficients through backpropagation and optimization algorithms to minimize the loss function.

[0137] 3) Evaluate the model's predictive performance using the validation set and continuously adjust hyperparameter values ​​to optimize model performance. After adjusting the model's hyperparameter values ​​several times, stop training and save the optimal trained model for later use.

[0138] 6.2 Identification of bearing aging damage degree: In actual application, the wavelet packet energy spectrum obtained by decomposing the monitoring signal is substituted into the deep learning model as the input parameter to obtain the aging damage degree of the bearing (predicted values ​​such as compressive elastic modulus, shear elastic modulus and aging time).

[0139] The above-described embodiments merely represent specific implementations of the present invention. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, and all such variations and improvements fall within the scope of protection of the present invention.

Claims

1. A method for in-situ detection of aging damage, characterized in that: The invention adopts a rectangular laminated rubber seismic isolation bearing, which includes a bearing body and bolt covers provided on four sides of the bearing body. The bolt covers are provided with bolt holes, and the bolt covers are connected to the bearing body by passing bolts through the bolt holes. A transducer is provided between the bolt cover and the side of the bearing body. The method comprises the following steps: Step 1: Install the vibrator and sensor on the side of the support, and install the center of the bottom surface of the transducer at half the height of the support; Step 2: Use a sweep frequency signal to stimulate the four vibrators in sequence, and collect monitoring signals through sensors; Step 3: Filter the collected monitoring signal and then perform wavelet packet decomposition to obtain the wavelet packet energy spectrum; Step 4: Perform calibration testing and establish a deep learning model for bearing aging damage identification; The step 4 is specifically as follows: (1) Design k supports, and apply different axial pressures F in an arithmetic progression to each support during the aging process through the tooling L The axial pressure variation range is 0MPa-the maximum allowable stress of the support; the accelerated aging test is carried out in an aging test chamber, the temperature is T℃, the single aging time is D days, and the number of aging times is L times; (2) Take out the bearing every D days and place it at room temperature for a certain period of time, then conduct compressive elastic modulus and shear elastic modulus tests to obtain the mechanical properties of the bearing at different aging times t; obtain the mechanical properties of all bearings at different aging times t and different axial compression F by wave method test. B The monitoring signal under the condition of , and the signal is filtered and decomposed by wavelet packet to obtain its wavelet packet energy spectrum; the obtained signal matrix is ​​as follows: Among them, MS IJ Indicates the monitoring signal when the exciter number is I and the sensor number is J; When the axial pressure during aging is F L , aging time is t, axial pressure during wave test is F B When the monitoring signal matrix is ​​MS IJ (F L ,t,F B ); (3) The wavelet packet energy spectrum of the bearing under different states is used as the input parameter, and the aging damage degree of the bearing is used as the output parameter. A deep learning model is used for modeling. The deep learning model is a multi-task regression model, which is divided into three sub-models MLP1-3: sub-model MLP1 is a compressive elastic modulus prediction model, and the output is the compressive elastic modulus; sub-model MLP2 is a shear elastic modulus prediction model, and the output is the shear elastic modulus; sub-model MLP3 is an aging time prediction model, and the output is the aging time. The output parameter matrix is ​​as follows: The specific steps to build a prediction model are as follows: 1) Divide the database into three parts: training set, validation set, and test set according to a certain ratio, select the loss function, optimization algorithm and evaluation index, and determine the values ​​of each hyperparameter in the model; 2) Train the model on the training set and continuously adjust the weight coefficients through backpropagation and optimization algorithms to minimize the loss function; 3) Evaluate the model's predictive performance using the validation set and optimize performance by adjusting hyperparameter values. Stop training after adjusting the model's hyperparameter values ​​multiple times and save the optimal trained model for subsequent use. Step 5: Substitute the wavelet packet energy spectrum obtained by decomposing the monitoring signal into the deep learning model as an input parameter to obtain the aging damage degree of the bearing.

2. The method for in-situ detection of aging damage according to claim 1, characterized in that: In step 3, the collected monitoring signal is filtered as follows: (1) First, calculate the monitoring signal length N_fft, set the passband lower limit f_low and upper limit f_high, and input the signal sampling rate FS; (2) Perform fast Fourier transform on the monitoring signal to obtain its frequency domain signal: X(f)=∫[Signal×e (2jπtf) ]dt Where X(f) represents the frequency domain signal, the amplitude and phase information at frequency f; Signal represents the monitoring signal; e (2jπtf) is a complex exponential function, where j is the imaginary unit; (3) Calculate the cutoff frequencies N1 and N2 corresponding to the lower and upper limits of the passband range: N1=ceil(f_low / (FS / N_fft)) N2 = floor(f_high / (FS / N_fft)) Among them, the ceil(x) function returns the smallest integer greater than or equal to x, that is, rounding up; the floor(x) function returns the largest integer less than or equal to x, that is, rounding down; (4) The frequency components above the cutoff frequency N2 and below the cutoff frequency N1 in the frequency domain signal are set to 0; (5) Perform inverse Fourier transform on the frequency domain signal to achieve the frequency filtering effect: Y_Signal=∫[X(f)×e (2jπtf) ]df Among them, Y_Signal represents the filtered time domain signal.

3. The method for in-situ detection of aging damage according to claim 2, characterized in that: In step 3, wavelet packet decomposition is performed to obtain the wavelet packet energy spectrum as follows: (1) Determine the wavelet packet basis db, determine the decomposition level h, perform discrete wavelet transform, and obtain the approximate coefficients and detail coefficients at each level; The approximation coefficient A(h) and detail coefficients D(h), D(h-1), …, D(1) are calculated using the following formulas: A(h)=x*phi_h(t)+(x*psi_{h,j}(t))_j D(h)=(x*psi_{h,j}(t))_j D(h-1)=(D(h)*psi_{h-1,j}(t))_j D(h-2)=(D(h-1)*psi_{h-2,j}(t))_j … D(1)=(D(2)*psi_{1,j}(t))_j Where * represents the convolution operation, phi_k(t) and psi_{k,j}(t) are the approximate function and detail function of the wavelet packet basis db, and j represents the position in the hth layer; (2) Further decompose the detail coefficients of each level to obtain the wavelet packet decomposition results; At the kth level, the detail coefficient D(k) is decomposed into 2^k frequency bands. Each frequency band is called a node, numbered from 0 to 2^k-1. For each node i, the node coefficient is calculated by the following formula: coef(i)=(D(k)*psi_{k,i}(t))_i Among them, psi_{k,i}(t) is the detail function of the wavelet packet basis db of the kth layer, and _i represents the position in the kth layer; (3) Solve the norm square of all nodes in the hth layer, that is, the sum of squares: E(i)=||wpcoef(w_signal,[h,i-1])||2 2 Where E(i) represents the energy of the i-th node, wpcoef(w_signal,[h,i-1]) represents the coefficient of the i-1-th node in the h-th layer in the wavelet packet decomposition result, ||.||2 represents the L2 norm, i.e., the Euclidean norm, and 2 represents the square of the norm result.

4. The method for in-situ detection of aging damage according to claim 1, wherein: The support body has no upper and lower rubber layers and the upper and lower steel plate layers are thickened; the upper and lower steel plate layers of the support body are provided with circular bolt holes corresponding to the parallel grooves.

5. The method for in-situ detection of aging damage according to claim 1, characterized in that: The bolt cover is provided with a groove for fixing the transducer.

6. The method for in-situ detection of aging damage according to claim 1, characterized in that: The bolt holes on the bolt cover are parallel grooves and are at a certain distance from the edge of the cover.

7. The method for in-situ detection of aging damage according to claim 1, characterized in that: The height and width of the bolt cover should be smaller than the height and width of the support body respectively.

8. The method for in-situ detection of aging damage according to claim 1, characterized in that: The same pre-tightening force should be applied to the bolts of the bolt cover and the support body.

Images