In-situ damage detection method for coupling beam damper

Abstract

The invention discloses a coupling beam damper in-situ damage detection method, which comprises the steps that a piezoelectric element is installed on a coupling beam damper, the piezoelectric element comprises vibration exciters and sensors, and the vibration exciters and the sensors are installed on an upper end plate, a lower end plate, a left flange plate and a right flange plate of the coupling beam damper in pairs; acquiring a monitoring signal by using a piezoelectric element; the monitoring signals are analyzed, firstly, the collected monitoring signals are subjected to filtering processing, and then wavelet packet decomposition is carried out; identifying the damage state of the coupling beam damper by adopting a deep learning model; according to the method, accurate identification results of the damage type and the damage degree of the coupling beam damper can be obtained.

Description

Technical Field

[0001] The invention relates to the technical field of structural damage identification and health monitoring, and in particular to an in-situ damage detection method for a coupling beam damper. Background Art

[0002] As an important component of shock-absorbing building structures, the coupling beam damper plays a key role in improving earthquake resistance, enhancing structural stability and reducing vibration effects. It can effectively absorb and disperse the energy caused by earthquakes or wind loads, reduce the vibration amplitude of the structure, and thus protect the building and its internal equipment from damage.

[0003] In the event of a small or moderate earthquake, the web of the beam damper undergoes plastic deformation and energy loss. Due to defects inside the metal, this deformation will produce microcracks. The expansion of microcracks will cause different types of damage or failure in the damper, resulting in a decrease in the damping effect of the damper or loss of its function. Damage to the beam damper has serious consequences and may lead to the following problems: (1) Decreased damping effect: The damaged beam damper cannot effectively absorb and disperse earthquake energy, and the damping effect will be greatly reduced. This will cause the building structure to suffer greater damage in earthquakes or wind disasters, endangering the lives of people; (2) Reduced structural stability: The damaged beam damper cannot provide sufficient stability support, resulting in a decrease in the overall stability of the building structure. This may lead to increased deformation of the structure and even cause the collapse of the structure; (3) Limited equipment function: The damaged beam damper cannot effectively reduce structural vibration, resulting in greater vibration impact on the equipment inside the building. This will cause damage to the equipment or failure of its function, which will have a serious impact on work and life.

[0004] However, the traditional detection methods have the following defects:

[0005] 1. Inability to perform comprehensive inspection: The existing scheme first determines the dampers to be inspected by numerical calculation, and then installs the dampers and the inspection device. However, during an earthquake, damage or destruction of the connecting beam damper will cause changes in the overall stiffness and elastic properties of the structure, resulting in unpredictable damage. Sampling is not conducive to judging the overall damage of the structure.

[0006] 2. Long-term monitoring is required: The existing solutions for analyzing the energy consumption and ductility performance of the damper require long-term monitoring of the damper, which increases the detection cost. In addition, under long-term monitoring, the impact of environmental factors on the stability of the monitoring system cannot be ruled out.

[0007] 3. Troubleshooting and replacing sensors are complicated: When the monitoring signal of the damper is abnormal, troubleshooting and replacing sensors are complicated and inefficient. Summary of the invention

[0008] In order to solve the problems existing in the prior art, the purpose of the present invention is to provide an in-situ damage detection method for a coupling beam damper, which can obtain accurate identification results of the damage type and damage degree of the coupling beam damper.

[0009] To achieve the above object, the technical solution adopted by the present invention is: a method for in-situ damage detection of a coupling beam damper, comprising the following steps:

[0010] Step 1, installing a piezoelectric element on the connecting beam damper, wherein the piezoelectric element includes an exciter and a sensor, and the exciter and the sensor are installed in pairs on the upper and lower end plates and the left and right flange plates of the connecting beam damper;

[0011] Step 2: Obtain monitoring signals using piezoelectric elements;

[0012] Step 3: Analyze the monitoring signal, first filter the collected monitoring signal, and then perform wavelet packet decomposition;

[0013] Step 4: Use the deep learning model to identify the damage state of the coupling beam damper.

[0014] As a further improvement of the present invention, the resonant frequency of the piezoelectric element is 50kHz-200kHz.

[0015] As a further improvement of the present invention, the exciters and sensors are installed in pairs on the upper and lower end plates and left and right flange plates of the connecting beam damper by gluing, two exciters and sensors are respectively installed at 1 / 4 of the upper and lower end plates and close to the web, and two exciters and sensors are respectively installed at 1 / 4 of the length direction of the right flange plate and in the middle of the width direction.

[0016] As a further improvement of the present invention, the step 2 is specifically as follows:

[0017] The exciter and sensor of the piezoelectric component have the same structure, and the functions of excitation and reception are interchangeable. When the monitoring signal is acquired, the signal is collected once from the exciter and the sensor: (1) A sweep frequency signal with a duration of 2.0s and an amplitude of ±300V is used to excite the exciter, and the monitoring signal is collected through the sensor; (2) The same signal is used to excite the sensor, and the signal is collected through the corresponding exciter to obtain signals that propagate symmetrically.

[0018] As a further improvement of the present invention, in step 3, filtering the collected monitoring signal is specifically performed as follows:

[0019] First, perform a Fourier transform on the collected monitoring signal to convert the time domain signal into a frequency domain signal:

[0020]

[0021] In the formula, N is the length of the transform interval of Fourier transform;

[0022] According to the sampling theorem, the frequency domain signal is filtered: the lowest frequency of the input parameters is f_low=1000Hz, the highest frequency is f_high=500kHz, and the sequence values ​​N1 and N2 corresponding to the lowest and highest frequencies are obtained, and the sequence value corresponding to half of the sampling frequency domain is N3.

[0023] N1=[f_low / (FS / N)]+1

[0024] N2=[f_high / (FS / N)]

[0025] N3=[N / 2]

[0026] In the sampling frequency domain (0, FS / 2), the frequencies corresponding to the sequences n=0,1,···,N1-1 and n=N2-1,N2,···,N3-1 are converted to 0; in the sampling frequency domain (FS / 2, FS), the frequencies corresponding to the sequences n=N3-1,N3,···N2+N3-1 and n=N-N1-1,N-N1,···,N are converted to 0; and the filtered frequency domain signal is inversely Fourier transformed to obtain the filtered time domain signal Y(k):

[0027]

[0028] As a further improvement of the present invention, in step 3, the wavelet packet decomposition is performed as follows:

[0029] Perform wavelet packet transform on the monitoring signal Y(k), and the decomposed subspace signal is y i (n);

[0030] (1) First-layer decomposition: Convolve and downsample the filtered signal Y[n] to obtain low-frequency D[1][n] and high-frequency G[1][n] components:

[0031] D[1][n]=∑ a h[a]·Y[2n-a]

[0032] G[1][n]=∑ a g[a]·Y[2n-a]

[0033] Where a is the index of the filter coefficient, h[a] is a low-pass filter, g[a] is a high-pass filter, and n is the discrete time index;

[0034] (2) Iterative decomposition: The subsequent levels J are obtained by decomposing the previous level J as 1; for the Kth node of any level J, the low-frequency component decomposition formula is as follows:

[0035] D[J+1][n]=∑ a h[a]·D[J][2n-a]

[0036] G[J+1][n]=∑ a g[a]·D[J][2n-a]

[0037] The high frequency component decomposition formula is as follows:

[0038] D[J+1][n]=∑ a h[a]·G[J][2n-a]

[0039] G[J+1][n]=∑ a g[a]·G[J][2n-a]

[0040] Calculate the 2-norm of the signal in each subspace as the energy of the subspace signal and normalize it; signal x i The normalized energy E i The calculation formula is as follows:

[0041]

[0042] where x i represents the signal in the i-th subspace, i represents the sequence number of the subspace;

[0043] The normalized energy of the signal in each subspace is calculated respectively to obtain the normalized wavelet packet energy spectrum; the frequency band of the wavelet packet energy spectrum under different damage states is analyzed to obtain its frequency band characteristics under different damage states of the coupling beam damper.

[0044] As a further improvement of the present invention, the step 4 is specifically as follows:

[0045] The wavelet packet energy spectrum E obtained by decomposing the monitoring signal i (i=1,2,3,…256) is substituted into the deep learning model as the input parameter to obtain the predicted mechanical damage index D. When the predicted mechanical damage index D is greater than the threshold D b When the specimen is considered to have reached a moderate damage level or above, the beam damper should be connected in time.

[0046] As a further improvement of the present invention, the process of establishing the deep learning model is specifically as follows:

[0047] The data set is obtained through calibration experiments and divided into three parts: training set, validation set and test set according to a certain ratio, which are used for training, adjustment and evaluation of deep learning models respectively; each monitoring signal sample includes wavelet packet energy spectrum E i and mechanical damage index D, which are used as input parameters and output parameters of the deep learning model respectively;

[0048] Establish a deep learning model framework consisting of convolutional layers, maximum pooling layers, global average pooling layers, and Re lu layers, using the ReLU function as the activation function;

[0049] The mean square error between the measured value of each sample and the model predicted value is used as the loss function of the model. Where N is the total number of samples, y i is the measured value, Predict values ​​for the model and use the loss function to calculate the gradient of each weight parameter; starting from the output layer, backpropagate through each layer of the network back to the input layer, thereby calculating the gradient of the weight parameters in each layer;

[0050] According to the calculated gradient, the Adam optimizer is used to optimize the weight parameters in the network structure to reduce the loss function value;

[0051] When the loss function value on the test set reaches the threshold or the number of iterations reaches the set value (settable), the model training is stopped and the model at this moment is output as the final model for mechanical damage index prediction.

[0052] As a further improvement of the present invention, the data set is obtained through calibration experiments as follows:

[0053] The coupling beam damper is calibrated by using a load-displacement dual control mode. Before yielding, the load is controlled and each load level is repeated once; after yielding, the displacement is controlled and each load level is repeated three times. The number of tests for each type of coupling beam damper is not less than three.

[0054] Before loading each coupling beam damper, the ultrasonic guided wave signal and the corresponding wavelet packet energy spectrum signal in the healthy state are measured;

[0055] The loading test is carried out, and the ultrasonic guided wave signal under different loading conditions and the corresponding wavelet packet energy spectrum signal are obtained for each loading cycle; this signal is recorded as E ij (k), where i represents the loading level, j represents the number of loading times (j=0, 1, 2, 3), k takes values ​​of 1, 2, 3, ···, n, and n represents the number of wavelet packet energy spectra after filtering.

[0056] As a further improvement of the present invention, the calculation method of the mechanical damage index D is as follows:

[0057] Calculate the area enclosed by each circle of stress-strain curve and the coordinate axis, that is, the energy dissipated by the damper when loading one circle, recorded as E ij , where i is the loading level, j is the number of loading cycles, and the values ​​are 1, 2, and 3; the energy dissipated in each cycle is E ij Add together to get the overall energy consumption capacity of the component The mechanical damage index after the mth loading at the kth level is expressed as Damage index threshold D b It is determined by the earthquake intensity α, the site and foundation type influence coefficient β, the building structure importance coefficient γ, and the damper importance η parameter. D0 is the reference value of the damage coefficient.

[0058] It is of great significance to develop and apply in-situ detection methods to accurately and in real time evaluate the damage of the connecting beam damper. The in-situ detection method of the present invention can obtain the internal information of the connecting beam damper without destroying it, and provide a more comprehensive and objective damage assessment. These methods can collect and analyze data on the connecting beam damper by installing sensors, strain gauges or other monitoring equipment to monitor the changes in its vibration response, frequency characteristics or other physical parameters in real time. By analyzing these data, it can be determined whether the connecting beam damper is damaged, faulty or abnormal, so that maintenance or replacement measures can be taken in time to ensure the safe operation of the structure.

[0059] The beneficial effects of the present invention are:

[0060] 1. Use piezoelectric devices with low price, good stability and easy replacement, and use piezoelectric wave method for detection, which reduces the cost of detection.

[0061] 2. Piezoelectric elements can be installed on all dampers, and the changes in piezoelectric signals can be measured after the earthquake to achieve comprehensive testing of the connecting beam dampers.

[0062] 3. By establishing a deep learning model, the damage type and degree of the connecting beam damper can be judged through the signal changes after the earthquake, thereby improving the detection efficiency.

[0063] 4. Whether the piezoelectric element is damaged can be determined by the signal change. At the same time, due to the small number and small size of piezoelectric elements, it is simple to troubleshoot and replace the sensor. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 is a top view of a connecting beam damper in an embodiment of the present invention;

[0065] Figure 2 is a front view of a connecting beam damper in an embodiment of the present invention;

[0066] Figure 3It is a left side view of the connecting beam damper in the embodiment of the present invention;

[0067] Figure 4 A is a cross-sectional view of the connecting beam damper in an embodiment of the present invention;

[0068] Figure 5 B is a cross-sectional view of the connecting beam damper in the embodiment of the present invention. DETAILED DESCRIPTION

[0069] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0070] Example

[0071] like Figure 1 for Figure 5 As shown, Figure 1 for Figure 5 : is a structural schematic diagram of a connecting beam damper, in which 1 is an end plate, 2 is a flange plate, 3 is a mild steel web plate, 4 is a rib plate, 5 is a piezoelectric element, and 6 is a bolt hole; based on this structure, this embodiment provides an in-situ damage detection method for a connecting beam damper, comprising the following steps:

[0072] 1. Preparation: Ensure that the equipment required for ultrasonic guided wave testing is operating normally, including: signal generator, power amplifier, exciter, sensor, charge amplifier, and signal acquisition instrument. Connect the signal generator to the power amplifier, and the voltage signal generated by the signal generator is converted into a high voltage signal by the power amplifier; connect the power amplifier to the exciter. Due to the inverse piezoelectric effect, the high voltage signal in the power amplifier causes the piezoelectric element in the exciter to deform, and the exciter generates an excitation wave, which is transmitted to the sensor through the damper, causing the piezoelectric element in the sensor to deform. Due to the piezoelectric effect, the deformation of the piezoelectric element in the sensor will cause the voltage of the sensor to change; connect the sensor to the charge amplifier to match the circuit impedance; connect the charge amplifier to the signal acquisition instrument to collect the voltage changes in the sensor, and display the voltage signal collected by the acquisition instrument on the display screen.

[0073] 2. Transducer selection: According to the characteristic frequency of the damper, select the appropriate type from the existing piezoelectric components: piezoelectric transducer, dielectric constant 1700F / m, piezoelectric (strain) constant 10-12C / N, resonant frequency between 50-200kHz. According to the longitudinal wave sound velocity of steel V = 5856m / s, the detected damper buckling influence range is 3~12cm, and the damper detection frequency can be obtained as Taking into account the stability of the piezoelectric components attached to the damper, the matching of the characteristic frequency, and the economy, a piezoelectric transducer with a resonant frequency of 58kHz is used as the exciter and receiver.

[0074] 3. Clean the surface: Clean the surface of the connecting beam damper to ensure that there are no impurities or dirt to ensure the firm bonding of the exciter and the sensor. Since the metal connecting beam damper is welded from steel plates, which often rust, it is necessary to polish the parts where the exciter and the sensor are installed to reduce the impact of non-damaging factors on the signal transmission of the exciter and the sensor.

[0075] 4. Installation of exciters and sensors: Install the exciters and sensors in pairs on the end plates and flanges of the connecting beam damper by gluing. Install the two exciters at 1 / 4 of the upper end plate, close to the web; install the sensors at 1 / 4 of the lower end plate, close to the web. Install the other two exciters at 1 / 4 of the length direction of the right flange plate and in the middle of the width direction; install the other two sensors at the corresponding positions of the exciters. The diffusion angle of the sound wave beam meets High-frequency stress wave signals propagate in a straight line, with a diffusion angle of 26° in steel. The piezoelectric element is installed at 1 / 4 so that the stress wave completely covers the damper, achieving the propagation of the stress wave throughout the damper.

[0076] 5. Acquisition of monitoring signals: The exciter and sensor of the piezoelectric component have the same structure, and the functions of excitation and reception can be interchanged. When acquiring monitoring signals, the exciter and sensor are each collected once: (1) A sweep frequency signal (1kHz-100k Hz) with a duration of 2.0s and an amplitude of ±300V is used to excite the exciter, and the monitoring signal is collected through the sensor. (2) The same signal is used to excite the sensor, and the signal is collected through the corresponding exciter to obtain signals that propagate symmetrically.

[0077] 6. Monitoring signal analysis: First, filter the collected monitoring signal, and then perform wavelet packet decomposition:

[0078] Filtering:

[0079] First, perform a Fourier transform on the collected monitoring signal to convert the time domain signal into a frequency domain signal:

[0080]

[0081] In the formula, N is the length of the transform interval of Fourier transform.

[0082] According to the swept frequency domain signal, the characteristic frequency range of the damper is 1.5kHz-100kHz. According to the sampling theorem, the frequency domain signal is filtered: the lowest frequency of the input parameters is f_l ow=1000Hz, and the highest frequency is f_h i gh=500kHz. The sequence values ​​N1 and N2 corresponding to the lowest frequency and the highest frequency are obtained, and the sequence value corresponding to half of the sampling frequency domain is N3.

[0083] N1=[f_low / (FS / N)]+1

[0084] N2=[f_high / (FS / N)]

[0085] N3=[N / 2]

[0086] In the sampling frequency domain (0, FS / 2), the frequencies corresponding to the sequences n=0,1,···,N1-1 and n=N2-1,N2,···,N3-1 are converted to 0; in the sampling frequency domain (FS / 2, FS), the frequencies corresponding to the sequences n=N3-1,N3,···N2+N3-1 and n=N-N1-1,N-N1,···,N are converted to 0. Then, the filtered frequency domain signal is inversely Fourier transformed to obtain the filtered time domain signal Y(k).

[0087]

[0088] Wavelet packet decomposition:

[0089] The monitoring signal Y is transformed into wavelet packet using the "db5" wavelet basis. The number of layers of wavelet packet decomposition is 8. The subspace signal after decomposition is y i (n).

[0090] (1) First-layer decomposition: Convolve and downsample the filtered signal Y[n] to obtain low-frequency D[1][n] and high-frequency G[1][n] components.

[0091] D[1][n]=Σ a h[a]·Y[2n-a]

[0092] G[1][n]=Σ a g[a]·Y[2n-a]

[0093] Where a is the index of the filter coefficient, h[a] is a low-pass filter, g[a] is a high-pass filter, and n is the discrete time index.

[0094] (2) Iterative decomposition: The subsequent level J is obtained by decomposing the previous level J-1. For the Kth node of any level J, the low-frequency component decomposition formula is as follows:

[0095] D[J+1][n]=∑ a h[a]·D[J][2n-a]

[0096] G[J+1][n]=∑ a g[a]·D[J][2n-a]

[0097] The high frequency component decomposition formula is as follows:

[0098] D[J+1][n]=∑ a h[a]·G[J][2n-a]

[0099] G[J+1][n]=Σ a g[a]·G[J][2n-a]

[0100] Calculate the 2-norm of the signal in each subspace as the energy of the subspace signal and normalize it. i The normalized energy E i The calculation formula is as follows:

[0101]

[0102] where x i Represents the signal in the i-th subspace, and i represents the sequence number of the subspace.

[0103] The normalized energy of the signal in each subspace is calculated to obtain the normalized wavelet packet energy spectrum. The frequency band of the wavelet packet energy spectrum under different damage states is analyzed to obtain its frequency band characteristics under different damage states of the coupling beam damper.

[0104] 7. Deep learning model prediction:

[0105] 7.1. Calibration test:

[0106] (1) Calibration test of the coupling beam damper is carried out. Each coupling beam damper is loaded according to the method described in the Code for Seismic Test Methods of Buildings JGJ 101-96. The load-displacement dual control method is adopted. Before yielding, the load is controlled and each load level is repeated once. After yielding, the displacement is controlled and each load level is repeated three times. The number of tests for each damper shall not be less than three.

[0107] (2) Before loading each coupling beam damper, perform steps 1-5 to measure the ultrasonic guided wave signal and the corresponding wavelet packet energy spectrum signal in a healthy state.

[0108] (3) Carry out the loading test shown in 7.1(1), and obtain the ultrasonic guided wave signal under different loading conditions and the corresponding wavelet packet energy spectrum signal for each loading cycle. This signal is recorded as E ij (k), where i represents the loading level, j represents the number of loading times (j=0, 1, 2, 3), k takes values ​​of 1, 2, 3, ···, n, and n represents the number of wavelet packet energy spectra after filtering.

[0109] (4) The degree of damage of the coupling beam damper can be reflected by the mechanical damage coefficient D. The mechanical damage index is a coefficient related to the accumulated plastic energy of the component, which reflects the damage degree of the structure as a whole and is obtained from the stress-strain curve of the specimen in the calibration test.

[0110] The specific calculation method is as follows: Calculate the area enclosed by each circle of stress-strain curve and the coordinate axis, that is, the energy dissipated by the damper when loading one circle, recorded as E ij , where i is the loading level and j is the number of loading cycles, with values ​​of 1, 2, and 3. The energy dissipated in each cycle is E ij Add together to get the overall energy consumption capacity of the component The mechanical damage coefficient after the mth loading in the kth level loading can be expressed as Damage coefficient threshold D b It is determined by the earthquake intensity α, the site and foundation type influence coefficient β, the building structure importance coefficient γ, and the damper importance η parameter. D0 is the reference value of the damage coefficient.

[0111] (5) Establish a deep learning model:

[0112] 1) Divide the data set obtained in the calibration experiment into three parts: training set, validation set and test set (the division ratio can be 7:2:1), which are used for model training, adjustment and evaluation respectively. Each monitoring signal sample includes the wavelet packet energy spectrum E i (The number of decomposition levels can be 8, then i = 1, 2, 3, ..., 256) and mechanical damage index D are used as the input parameters and output parameters of the model respectively.

[0113] 2) Establish a deep learning model framework consisting of a convolutional layer, a maximum pooling layer, a global average pooling layer, and a Re lu layer (the number of layers of each network structure can be 4 layers, 2 layers, 1 layer, and 1 layer respectively), and use the ReLU function as the activation function.

[0114] 3) The mean square error between the measured value of each sample and the model prediction value is used as the loss function of the model. Where N is the total number of samples, y i is the measured value, Predict values ​​for the model and use the loss function to calculate the gradient of each weight parameter. Starting from the output layer, backpropagate through each layer of the network back to the input layer to calculate the gradient of the weight parameters in each layer.

[0115] 4) According to the calculated gradient, the Adam optimizer is used to optimize the weight parameters in the network structure to reduce the loss function value.

[0116] 5) When the loss function value on the test set reaches the threshold (which can be set to 0.05) or the number of iterations reaches the set value (which can be set to 1000 times), stop training the model and output the model at this moment as the final model for mechanical damage index prediction.

[0117] 7.2. Identification of damage status of coupling beam damper:

[0118] In practical applications, the wavelet packet energy spectrum E obtained by decomposing the monitoring signal is i (i=1,2,3,…256) is substituted into the deep learning model as the input parameter to obtain the predicted mechanical damage index D. When the predicted mechanical damage index D is greater than the threshold D b , it can be considered that the specimen has reached a moderate damage level or above, and the damper should be replaced in time.

[0119] The above-mentioned embodiments only express the specific implementation of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A method for in-situ damage detection of a coupling beam damper, characterized in that: The following steps are involved: Step 1, installing a piezoelectric element on the connecting beam damper, wherein the piezoelectric element includes an exciter and a sensor, and the exciter and the sensor are installed in pairs on the upper and lower end plates and the left and right flange plates of the connecting beam damper; Step 2: Obtain monitoring signals using piezoelectric elements; Step 3: Analyze the monitoring signal, first filter the collected monitoring signal, and then perform wavelet packet decomposition; Step 4: Use the deep learning model to identify the damage state of the coupling beam damper.

2. The in-situ damage detection method for a coupling beam damper according to claim 1, characterized in that: The resonant frequency of the piezoelectric element is 50kHz-200kHz.

3. The in-situ damage detection method for a coupling beam damper according to claim 2, characterized in that: The exciters and sensors are installed in pairs on the upper and lower end plates and left and right flange plates of the connecting beam damper by gluing. Two exciters and sensors are installed at 1 / 4 of the upper and lower end plates and close to the web, and two exciters and sensors are installed at 1 / 4 of the length direction of the right flange plate and in the middle of the width direction.

4. The in-situ damage detection method for a coupling beam damper according to claim 1, characterized in that: The step 2 is specifically as follows: The exciter and sensor of the piezoelectric component have the same structure, and the functions of excitation and reception are interchangeable. When the monitoring signal is acquired, the signal is collected once from the exciter and the sensor: (1) A sweep frequency signal with a duration of 2.0s and an amplitude of ±300V is used to excite the exciter, and the monitoring signal is collected through the sensor; (2) The same signal is used to excite the sensor, and the signal is collected through the corresponding exciter to obtain signals that propagate symmetrically.

5. The in-situ damage detection method for a coupling beam damper according to claim 1, characterized in that: In step 3, the collected monitoring signal is filtered as follows: First, perform a Fourier transform on the collected monitoring signal to convert the time domain signal into a frequency domain signal: In the formula, N is the length of the transform interval of Fourier transform; According to the sampling theorem, the frequency domain signal is filtered: the lowest frequency of the input parameters is f_low=1000Hz, the highest frequency is f_high=500kHz, and the sequence values ​​N1 and N2 corresponding to the lowest and highest frequencies are obtained, and the sequence value corresponding to half of the sampling frequency domain is N3. N1=[f_low / (FS / N)]+1 N2=[f_high / (FS / N)] N3=[N / 2] In the sampling frequency domain (0, FS / 2), the frequencies corresponding to the sequences n=0,1,···,N1-1 and n=N2-1,N2,···,N3-1 are converted to 0; in the sampling frequency domain (FS / 2, FS), the frequencies corresponding to the sequences n=N3-1,N3,···N2+N3-1 and n=N-N1-1,N-N1,···,N are converted to 0; and the filtered frequency domain signal is inversely Fourier transformed to obtain the filtered time domain signal Y(k):

6. The in-situ damage detection method for a coupling beam damper according to claim 5, characterized in that: In step 3, the wavelet packet decomposition is performed as follows: Perform wavelet packet transform on the monitoring signal Y(k), and the decomposed subspace signal is y i (n); (1) First-layer decomposition: Convolve and downsample the filtered signal Y[n] to obtain low-frequency D[1][n] and high-frequency G[1][n] components: D[1][n]=Σ a h[a]·Y[2n-a] G[1][n]=Σ a g[a]·Y[2n-a] Where a is the index of the filter coefficient, h[a] is a low-pass filter, g[a] is a high-pass filter, and n is the discrete time index; (2) Iterative decomposition: The subsequent levels J are obtained by decomposing the previous level J as 1; for the Kth node of any level J, the low-frequency component decomposition formula is as follows: D[J+1][n]=∑ a h[a]·D[J][2n-a] G[J+1][n]0Σ a g[a]·D[J][2n-a] The high frequency component decomposition formula is as follows: D[J+1][n]=∑ a h[a]·G[J][2n-a] G[J+1][n]=∑ a g[a]·G[J][2n-a] Calculate the 2-norm of the signal in each subspace as the energy of the subspace signal and normalize it; signal x i The normalized energy E i The calculation formula is as follows: where x i represents the signal in the i-th subspace, i represents the sequence number of the subspace; The normalized energy of the signal in each subspace is calculated respectively to obtain the normalized wavelet packet energy spectrum; the frequency band of the wavelet packet energy spectrum under different damage states is analyzed to obtain its frequency band characteristics under different damage states of the coupling beam damper.

7. The in-situ damage detection method for a coupling beam damper according to claim 6, characterized in that: The step 4 is specifically as follows: The wavelet packet energy spectrum E obtained by decomposing the monitoring signal i (i=1,2,3,…256) is substituted into the deep learning model as the input parameter to obtain the predicted mechanical damage index D. When the predicted mechanical damage index D is greater than the threshold D b When the test piece is considered to have reached a moderate damage level or above, the coupling beam damper should be replaced in time.

8. The in-situ damage detection method for a coupling beam damper according to claim 7, characterized in that: The process of establishing the deep learning model is as follows: The data set is obtained through calibration experiments and divided into three parts: training set, validation set and test set according to a certain ratio, which are used for training, adjustment and evaluation of deep learning models respectively; each monitoring signal sample includes wavelet packet energy spectrum E i and mechanical damage index D, which are used as input parameters and output parameters of the deep learning model respectively; Establish a deep learning model framework consisting of convolutional layers, maximum pooling layers, global average pooling layers, and ReLU layers, using the ReLU function as the activation function; The mean square error between the measured value of each sample and the model predicted value is used as the loss function of the model. Where N is the total number of samples, y i is the measured value, Predict values ​​for the model; and use the loss function to calculate the gradient of each weight parameter; starting from the output layer, backpropagate through each layer of the network back to the input layer, thereby calculating the gradient of the weight parameters in each layer; According to the calculated gradient, the Adam optimizer is used to optimize the weight parameters in the network structure to reduce the loss function value; When the loss function value on the test set reaches the threshold or the number of iterations reaches the set value (settable), the model training is stopped and the model at this moment is output as the final model for mechanical damage index prediction.

9. The in-situ damage detection method for a coupling beam damper according to claim 8, characterized in that: The data set obtained through the calibration experiment is as follows: The coupling beam damper shall be calibrated by using a load-displacement dual control method. Before yielding, the load control shall be applied once for each load level. After yielding, the displacement control shall be applied three times for each load level. The number of tests for each type of coupling beam damper shall not be less than three. Before loading each coupling beam damper, the ultrasonic guided wave signal and the corresponding wavelet packet energy spectrum signal in the healthy state are measured; The loading test is carried out, and the ultrasonic guided wave signal under different loading conditions and the corresponding wavelet packet energy spectrum signal are obtained for each loading cycle; this signal is recorded as E ij (k), where i represents the loading level, j represents the number of loading times (j=0, 1, 2, 3), k takes values ​​of 1, 2, 3, ···, n, and n represents the number of wavelet packet energy spectra after filtering.

10. The in-situ damage detection method for a coupling beam damper according to claim 9, characterized in that: The mechanical damage index D is calculated as follows: Calculate the area enclosed by each circle of stress-strain curve and the coordinate axis, that is, the energy dissipated by the damper when loading one circle, recorded as E ij , where i is the loading level, j is the number of loading cycles, and the values ​​are 1, 2, and 3; the energy dissipated in each cycle is E ij Add together to get the overall energy consumption capacity of the component The mechanical damage index after the mth loading in the kth loading level is expressed as Damage index threshold D b It is determined by the earthquake intensity α, the site and foundation type influence coefficient β, the building structure importance coefficient γ, and the damper importance η parameter. D0 is the reference value of the damage coefficient.

Images