Bolt group damage identification device and method combining ultrasonic guided wave normalized energy transmittance and neural network

Abstract

The present invention discloses a device and method for identifying bolt group damage that combines ultrasonic guided wave normalized energy transmittance and a neural network. The method comprises the following steps: 1) installing an appropriate number of guided wave sensors on both sides of the bolt connection area to form multiple guided wave propagation paths that cover all bolts under test; 2) stimulating and receiving guided wave signals, and calculating the normalized energy transmittance based on the direct wave and transmitted wave signals of different propagation paths as the input of the neural network; 3) simulating damage conditions to create sample training sets and test sets for the neural network; 4) building a neural network, training the training set samples, and then predicting the damage conditions of the test set samples. The method proposed by the present invention can achieve damage location of the bolt group and prediction of local and overall damage levels under the premise of using a relatively small number of sensors, and has high damage identification accuracy. In addition, the method can reduce the impact of environmental factors on guided wave propagation, making it adaptable to complex environmental conditions.

Description

Technical Field

[0001] The present invention belongs to the technical field of construction engineering, and relates to ultrasonic guided wave nondestructive testing technology, neural network and structural health monitoring technology. Background Art

[0002] Bolted connections are widely used across various industries due to their ease of construction and strong load-bearing capacity. Common applications include bolted plate joints on steel bridges, power transmission towers, railway tracks, and aircraft wings. The degree of bolt tightening affects the structural load-bearing capacity and, in turn, its operational safety. However, these applications often operate in harsh environmental conditions, where bolts can easily loosen due to external factors such as environmental loads, temperature, humidity, rain, and snow. Numerous catastrophic accidents have been caused by loose bolts, including the tower collapse of a wind turbine in Gansu in 2010 due to continuous high winds and the nuclear waste leak at a Japanese nuclear power plant in 2022. Such accidents can have severe social impacts and even threaten human life. Therefore, developing effective bolt loosening monitoring methods to provide timely early warnings of loose bolts is crucial for ensuring the overall safety of structures.

[0003] Current nondestructive testing methods for bolt loosening can be categorized as direct and indirect. Direct testing methods include strain gauge-based screw prestress measurement, torque meter-based bolt preload measurement, and embedded fiber Bragg grating (FBG) preload measurement. Indirect measurement methods include marker observation, a combination of machine vision and deep learning, structural feature recognition based on vibration signals, damage identification based on percussion signals, electromechanical impedance analysis, and ultrasonic guided wave (UGW) testing. Among these, UGW technology is currently one of the most promising methods for bolt loosening detection.

[0004] Bolt loosening monitoring based on ultrasonic guided waves can be categorized into five categories: damage identification based on guided wave energy variations, wave velocity measurement based on the acoustic-elastic effect, damage identification based on nonlinear ultrasound, damage identification based on time reversal, and early bolt loosening detection based on coda wave interferometry. Current research focuses on bolted connections consisting of a single bolt, while limited research has been conducted on bolted connections, which are common in engineering. As the number of bolts increases, the propagation of guided waves in the bolted connection region becomes more complex, making bolt loosening detection methods based on single-bolt connections ineffective. Furthermore, damage detection methods based on ultrasonic guided waves for bolted connections only provide information on the overall damage extent of the bolt group, but are unable to localize damage or predict local damage extent. Given the large number of bolts in structural connections in engineering, identifying loose bolts solely based on the overall damage extent of the bolt group is time-consuming and labor-intensive, resulting in extremely low detection efficiency. Therefore, it is necessary to perform damage localization and damage extent identification for the bolt group.

[0005] The surface of artificially machined steel plates is not completely flat, but rather contains minute irregularities. When two plates are bolted together, the actual contact area between the plates is smaller than the nominal contact area. The actual contact area between the plates can be simulated using sinusoidal surfaces and Hertzian contact theory. Studies have found that the actual contact area is positively correlated with the bolt torque, thus allowing the tightness of the bolts to be determined based on the contact area between the plates. The aforementioned guided wave energy-based bolt loosening detection method typically involves mounting sensors on two steel plates. Guided waves are transmitted from one plate to the other through the bolted joint area. Ultrasonic signals experience energy attenuation as they propagate through the bolted joint interface, with energy attenuation varying across different contact areas. Therefore, the actual contact area between the bolt plates, and therefore the tightness of the bolts, can be determined based on changes in the signal amplitude or energy received by the receiver. However, guided wave energy-based detection methods are susceptible to environmental factors, resulting in varying guided wave energy under the same operating conditions, thus affecting damage identification accuracy.

[0006] The patent document "Bolt Loosening Monitoring Device and Method (Publication No.: CN108507609A)" proposes a bolt loosening detection device for flanges. However, this method requires the installation of magnetic washers on each bolt, making it less suitable for structural projects with a large number of bolts. The present invention, on the other hand, achieves the same function using only a smaller number of sensors.

[0007] The patent document "A Bolt Looseness Detection and Alarm Device for Rail Joint Connection Parts (Publication No.: CN108303240A)" proposes a method for detecting rail bolt loosening. However, this method only detects the degree of bolt loosening, does not address damage location, and does not consider the impact of environmental factors on the detection results. The present invention can simultaneously display the damage degree of the bolt group as a whole and locally, and through the arrangement of sensors, it reduces the impact of environmental factors on the detection results. Summary of the Invention

[0008] The present invention addresses the aforementioned technical issues by proposing a device and method for bolt group damage identification that combines the normalized energy transmittance of ultrasonic guided waves with a neural network. Using only a small number of sensors, the method uses the energy transmittance calculated from guided wave signals along different propagation paths to locate bolt loosening and predict the extent of local and global damage. Furthermore, through appropriate sensor placement and damage index definition, the impact of ambient temperature fluctuations on damage identification results can be minimized, making the method adaptable to complex application environments.

[0009] To achieve the above object, according to the first aspect of the present invention, the present invention adopts the following technical solutions:

[0010] A bolt group damage identification device that combines ultrasonic guided wave normalized energy transmittance and a neural network. The bolt group is used in a bolt connection structure and includes multiple bolts. The bolt group damage identification device is characterized in that the bolt group damage identification device also includes multiple sensors, which are respectively installed on both sides of the bolt connection area. The sensors on the first side of the bolt connection area include one or more first sensors for signal excitation and one or more second sensors for signal reception. The sensors on the second side of the bolt connection area include one or more third sensors for signal reception. The first and second sensors are located on the same connected component in the bolt connection structure, and the third sensor is located on another connected component in the bolt connection structure. The signal received by the second sensor is a direct wave signal, the amplitude of which represents the energy of the excitation signal. The guided wave signal excited by the first sensor does not pass through the bolt connection area on the way to the second sensor, so the amplitude of the direct wave does not change with the change in the degree of damage to the bolt group. The signal received by the third sensor is a transmitted wave signal, the amplitude of which is related to the preload level of the bolt group. The greater the preload, the greater the amplitude of the transmitted wave.

[0011] Furthermore, the bolt connection forms include but are not limited to overlap, splice, truss connection, node plate connection, shear connection, double angle connection, etc.

[0012] Furthermore, the number of the excitation and receiving sensors should be flexibly set according to the number of bolts. The number of sensors should not be too many, but it is necessary to ensure that the propagation path of the ultrasonic guided wave covers all bolts. The number of sensors used to receive direct wave signals should be less than or equal to the number of sensors receiving transmitted waves.

[0013] Furthermore, the horizontal distance between the excitation sensor and the receiving sensors on both sides should be kept consistent to ensure that the excitation signal reaches the receiving sensors on both sides almost simultaneously, thereby eliminating the influence of energy attenuation of the guided wave during propagation on signal processing.

[0014] Furthermore, the types of the excitation and receiving sensors include but are not limited to piezoelectric sensors, magnetostrictive waveguide sensors and electromagnetic ultrasonic sensors, and the excitation methods also include laser ultrasound and the like.

[0015] Furthermore, the ultrasonic guided wave mode types include but are not limited to symmetrical and antisymmetrical Lamb wave modes, shear wave modes, etc. The number of cycles and frequency of the excitation signal are reasonably selected according to the dispersion curve of the bolt plate structure to be measured.

[0016] To achieve the above object, according to the second aspect of the present invention, the present invention adopts the following technical solutions:

[0017] A bolt group damage identification method combining ultrasonic guided wave normalized energy transmittance and neural network includes the following steps:

[0018] S1: Build a finite element model of the bolt connection structure under test, simulate the propagation of ultrasonic guided waves, and obtain guided wave signals on different propagation paths under different damage degrees. The finite element model can be built using commercial finite element software including ABAQUS, ANSYS, etc., and the material properties and boundary conditions need to be accurately simulated. Damage is simulated by changing the preload of the bolt. Through a large number of simulations, a sample training set and test set are constructed. The training set needs to include samples with different damage locations and damage degrees. The damage location and damage degree of the test set samples cannot be repeated with those in the training set;

[0019] S2: Calculate the energy of the direct wave and the transmitted wave on different propagation paths to stimulate the sensor T i To receiving sensor T j The calculation formula of the energy of the waveguide signal on the propagation path between is:

[0020]

[0021] Among them, t s and t f is the calculation interval of signal energy, f s is the sampling frequency, y ji (t) represents the waveguide signal under the current path. The waveguide signal used for energy calculation can be subjected to correlation filtering, and the filtering methods include but are not limited to Butterworth filter, wavelet transform, etc.

[0022] S3: Calculate the energy transmittance of different propagation paths under a single damage condition, and stimulate the sensor T i To receiving sensor T j The calculation formula for the energy transmittance of the waveguide signal on the propagation path between is:

[0023]

[0024] Taking the propagation path where the sensors (T1, T3, and T5) are located as an example, the energy transmittance is calculated as follows:

[0025]

[0026] The calculation method for the damage index of other propagation paths is the same. The initial energy transmittance obtained on the same propagation path in the lossless state is used. The energy transmittance I obtained in the damaged state BL Perform normalization to obtain the normalized energy transmittance As a damage index, it is expressed as:

[0027]

[0028] Multiple propagation paths are formed between multiple sensors, resulting in multiple damage indices, which are then used as a sample. Within a single sample, the change in each damage index reflects the change in the bolt preload level along the current propagation path. Through extensive numerical simulations, sample training and test sets are generated.

[0029] S4: Build a neural network and train and test the samples. Neural networks include but are not limited to BP neural networks and convolutional neural networks. Taking a BP neural network as an example, it consists of an input layer, hidden layers, and an output layer. The input is the damage index obtained from different propagation paths for a single damage sample, and the output is the local damage extent of the bolt group. Because the energy of the transmitted wave decreases with bolt loosening, the damage index ranges from 0 to 1, with 1 representing no damage. The local damage extent is defined by the ratio of the local bolt preload loosening value to the initial preload value. Therefore, the neural network output also ranges from 0 to 1, with 1 representing complete local bolt loosening. The number of hidden layers and the number of neurons in each hidden layer are adjusted based on test results, and the neural network hyperparameters are also determined through extensive testing. The location of the neural network output represents the damage location, and the magnitude of the output represents the local damage extent. The sum and average of all output results is the predicted value for the overall damage extent of the bolt group. Through continuous training and testing, the error between the neural network prediction and the actual value meets the damage identification accuracy requirements.

[0030] Furthermore, the excitation signal may be a pulse signal excitation or a sine wave signal modulated by a Hanning window;

[0031] Furthermore, for different damage samples, the calculation range of the guided wave signal energy needs to remain consistent;

[0032] Furthermore, in addition to obtaining the guided wave signal through finite element simulation, the measured signal in actual engineering can also be used.

[0033] Furthermore, the calculation results obtained from finite element simulation can be used to train the neural network, and the training parameters can be adjusted to obtain a higher damage identification accuracy. The trained neural network can then be used to identify damage on the measured data, provided that the accuracy of the finite element model is guaranteed.

[0034] Beneficial effects of the present invention:

[0035] The proposed method uses only a small number of sensors to locate damage and predict both local and global damage levels in bolted joints composed of multiple bolts. Because it's difficult to determine the general pattern of how damage index changes with damage severity along different propagation paths, a neural network is creatively incorporated. Using a large number of training samples, the damage index is then simply fed into the neural network for efficient damage assessment.

[0036] Furthermore, sensor performance can vary slightly due to factors such as ambient temperature fluctuations, causing slight changes in the signal along the same guided wave propagation path under the same damage condition, thus affecting damage identification accuracy. However, the damage index used in this invention is the normalized ratio of the transmitted wave to the direct wave. This effectively eliminates or mitigates the effects of environmental changes on both the transmitted and direct waves, making this method adaptable to complex environmental conditions.

[0037] The present invention uses energy transmittance as a characteristic value for damage identification, minimizing the impact of environmental changes on damage identification. By setting up multiple guided wave propagation paths to cover all bolts under test and inputting the energy transmittance of different propagation paths into a neural network, damage localization can be achieved for a group of bolts, predicting both the overall and local damage extent. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The above advantages of the present invention will become clearer and easier to understand through the detailed description made in conjunction with the following drawings, which are only exemplary and do not limit the present invention, wherein:

[0039] FIG1 (a) and FIG1 (b) are respectively a cross-sectional view and a top view of a bolt connection structure according to an embodiment of the present invention; they also show the arrangement of the excitation sensor and the receiving sensor;

[0040] Figure 2 is the sinusoidal wave modulation signal that excites sensor T1;

[0041] Figure 3 is the transmitted wave signal received by the receiving sensor T3;

[0042] Figure 4 is the transmitted wave signal received by the receiving sensor T4;

[0043] Figure 5 It is the direct wave signal received by the receiving sensor T5;

[0044] Figure 6 It is the damage index of each propagation path calculated under a certain damage condition;

[0045] Figure 7 is the variation of the transmitted wave energy with the residual rate of the bolt group preload at different temperatures;

[0046] Figure 8 is the change of normalized energy transmittance with the residual rate of bolt group preload at different temperatures (this method)

[0047] Figure 9 is the structure of the BP neural network used;

[0048] Figure 10is the prediction result of the neural network. DETAILED DESCRIPTION

[0049] The following describes in detail a bolt group damage identification device and method combining ultrasonic guided wave normalized energy transmittance and neural network with the accompanying drawings.

[0050] The embodiments described herein are specific embodiments of the present invention and are used to illustrate the concept of the present invention. They are illustrative and exemplary and should not be construed as limiting the embodiments and scope of the present invention. In addition to the embodiments described herein, those skilled in the art can also adopt other obvious technical solutions based on the claims and the disclosure of the specification, including technical solutions that adopt any obvious substitutions and modifications to the embodiments described herein.

[0051] The present invention provides a method for identifying bolt group damage by combining normalized energy transmittance of ultrasonic guided waves and a neural network. Taking the bolt connection structure shown in FIG1 as an example, the method includes the following steps:

[0052] S1: Use ABAQUS to establish the finite element model of the bolt connection structure in Figure 1. The structure consists of two bolt plates and eight high-strength bolts. The size of a single plate is 1.3×0.25×0.008m, and the length of the overlap area is 0.15m. Five magnetostrictive shear wave sensors T1-T5 are set up, among which T1, T2 and T5 are installed on the left plate 2, and T3 and T4 are installed on the right plate 3. The horizontal distance between T1, T2 and T3, T4 is 0.35m, and the horizontal distance between T1, T2 and T5 is also 0.35m. Four transmission wave propagation paths (T1-T3, T1-T4, T2-T3, T2-T4) and two direct wave propagation paths (T1-T5, T2-T5) are formed;

[0053] S2: Set the preload of the eight bolts to the initial value required by the specification, which represents a lossless state. First, stimulate the sensor T1. Figure 2 The center frequency of the 2.5-cycle sine wave modulation signal is 80kHz, and the signals received by sensors T3, T4 and T5 are as follows: Figure 3-5 Then, the excitation signal is generated from sensor T2, and the guided wave signals from sensors T3, T4 and T5 are also received.

[0054] S3: Calculate the energy of the waveguide signal on each transmission wave and direct wave propagation path respectively. The calculation formula is:

[0055]

[0056] Among them, t s and t fis the calculation interval of signal energy, f s is the sampling frequency, y ji (t) represents the waveguide signal received on the current path. The energy under the six propagation paths is obtained as follows:

[0057]

[0058] The energy transmittances on the four transmission wave propagation paths are:

[0059] I BL,1 =E1 / E5,I BL,2 =E2 / E5,I BL,3 =E3 / E6,I BL,4 =E4 / E6

[0060] S4: Set the damage level of the neural network training set and test set samples, adjust the preload level of the bolts in the finite element model according to the damage level, and then repeat the signal acquisition and energy transmittance calculation process in steps S2-S3 to finally obtain the energy transmittance of a large number of damaged samples on different propagation paths;

[0061] S5: The energy transmittance on the propagation path i in the lossless state is The energy transmittance on the propagation path i in a single damage state is I BL,i ,use to I BL,i Perform normalization processing to obtain the normalized energy transmission coefficient on the propagation path i under a single damage condition The four propagation paths will result in four normalized energy transmission coefficients, which are used as a set of damage indices and expressed as:

[0062]

[0063] At this time, each sample in the training set and test set contains four damage indices. Figure 6 The distribution of normalized energy transmittance of each propagation path under a certain damage condition is shown.

[0064] S6: The relationship between the energy of the transmitted wave and the normalized energy transmittance and the residual rate of the bolt group preload at different temperatures was tested. The results are shown in Figure 7 and Figure 8 It can be found that the normalized energy transmittance has good stability when the ambient temperature changes, which shows the advantage of using normalized energy transmittance as the input of the neural network.

[0065] S7: Divide the eight bolts into four groups according to the number of rows, with each group containing two bolts. Establish labels for each sample in the training set and test set. The input for each sample is the four damage indices obtained above. The sample label is a vector consisting of four numbers between 0 and 1. The size of the jth number in the label vector represents the local damage level of the two bolts in the jth row. The label reflects the location and local damage level of the loose bolts. For example: when the label is {z} = {0,0,1,0}, it means that the bolts in the third row are completely loose;

[0066] S8: Taking BP neural network as an example, set the structure of the neural network. Figure 9 The test set samples are input into the neural network training, and the neural network parameters are adjusted to reduce the error curve to the ideal range; then, the trained neural network is used to identify damage on the test set samples, and the output value of the neural network is compared with the label value of the test set samples ( Figure 10 ), calculate the neural network local damage degree prediction error and damage location accuracy; finally, sum and average the four output values ​​of the neural network to obtain the prediction result of the overall damage degree of the bolt group, and verify the accuracy of the overall damage degree of the bolt group by comparing the label values.

[0067] The present invention is not limited to the above-mentioned embodiments. Anyone can derive other forms of products under the inspiration of the present invention. However, no matter what changes are made in the shape or structure, any product with the same or similar technical solutions as the present application falls within the scope of protection of the present invention.

Claims

1. A bolt group damage identification method combining normalized energy transmittance of ultrasonic guided waves and neural network, characterized in that The steps include: S1: Build a finite element model of the bolted connection structure under test, simulate the propagation of ultrasonic guided waves, and obtain guided wave signals along different propagation paths under different damage degrees. Damage is simulated by changing the preload of the bolts. Through simulation, a sample training set and test set are constructed. The training set needs to include samples with different damage locations and damage degrees. The damage locations and damage degrees of the test set samples must not be repeated in the training set. S2: Calculate the energy of the direct wave and the transmitted wave on different propagation paths, and stimulate the sensor T i To receiving sensor T j The calculation formula of the energy of the waveguide signal on the propagation path between is: Among them, t s and t f is the calculation interval of signal energy, f s is the sampling frequency, y ji (t) represents the waveguide signal under the current path; S3: Calculate the energy transmittance of different propagation paths under a single damage condition, and stimulate the sensor T i To receiving sensor T j The calculation formula for the energy transmittance of the waveguide signal on the propagation path between is: The initial energy transmittance obtained on the same propagation path in the lossless state is used The energy transmittance I obtained in the damaged state BL Perform normalization to obtain the normalized energy transmittance As a damage index, it is expressed as: Under a single damage condition, multiple propagation paths will be formed between multiple sensors, resulting in multiple damage indices, which are then used as a sample. Within a single sample, the change in each damage index reflects the change in the bolt preload level under the current propagation path. Through a large number of numerical simulations, sample training and test sets are formed. S4: Build a neural network and train and test the samples. The input of the neural network is the normalized energy transmittance obtained from different propagation paths under a single damage sample, that is, the damage index, and the output is the local damage degree of the bolt group. Because the energy of the transmitted wave decreases continuously with the degree of bolt loosening, the value range of the normalized energy transmittance is 0 to 1, and 1 represents no damage. The local damage degree is defined by the ratio of the local bolt preload loosening value to the initial preload value. Therefore, the output value of the neural network also ranges from 0 to 1, and 1 represents that the local bolt is completely loose. The position of the neural network output result represents the damage location, and the size of the output value represents the local damage degree. The sum and average of all output results is the predicted value of the overall damage degree of the bolt group. Through continuous training and testing, the error between the neural network prediction value and the true value meets the damage identification accuracy requirements.

2. The bolt group damage identification method combining ultrasonic guided wave normalized energy transmittance and neural network according to claim 1 is characterized in that The excitation signal can be a pulse signal or a sine wave signal modulated by a Hanning window.

3. The bolt group damage identification method combining ultrasonic guided wave normalized energy transmittance and neural network according to claim 1 is characterized in that For different damage samples, the calculation range of the guided wave signal energy needs to remain consistent.

4. The bolt group damage identification method combining ultrasonic guided wave normalized energy transmittance and neural network according to claim 1 is characterized in that The guided wave signal is obtained through finite element simulation or using the measured signal in actual engineering.

5. A bolt group damage identification device combining ultrasonic guided wave normalized energy transmittance and neural network, wherein the bolt group is used for a bolt connection structure and comprises a plurality of bolts (1), characterized in that The bolt group damage identification device further comprises a plurality of sensors, which are respectively installed on both sides of the bolt connection area. The sensors on the first side of the bolt connection area comprise one or more first sensors (T1, T2) for signal excitation and one or more second sensors (T5) for signal reception, and the sensors on the second side of the bolt connection area comprise one or more third sensors (T3, T4) for signal reception; the first sensors (T1, T2) and the second sensors (T5) are located on the same connected component (2) in the bolt connection structure, and the third sensors (T3, T4) are located on another connected component (3) in the bolt connection structure; the signal received by the second sensor (T5) is a direct wave signal, the amplitude of which represents the energy of the excitation signal; the guided wave signal excited by the first sensors (T1, T2) does not pass through the bolt connection area on the way to being transmitted to the second sensor (T5), so the amplitude of the direct wave does not change with the change of the degree of damage to the bolt group; the signal received by the third sensors (T3, T4) is a transmitted wave signal, the amplitude of which is related to the preload level of the bolt group, and the greater the preload, the greater the amplitude of the transmitted wave; The identification device adopts the identification method described in claim 1.

6. The bolt group damage identification device combining ultrasonic guided wave normalized energy transmittance and neural network according to claim 5 is characterized in that Bolted connection forms include lap joints, splices, truss connections, node plate connections, shear connections, and double angle connections.

7. The bolt group damage identification device combining ultrasonic guided wave normalized energy transmittance and neural network according to claim 5 is characterized in that The number of excitation sensors and receiving sensors should be flexibly set according to the number of bolts. The number of sensors ensures that the propagation path of the ultrasonic guided wave covers all bolts that need to be damaged. The number of sensors used to receive direct wave signals should be less than or equal to the number of sensors receiving transmitted waves.

8. The bolt group damage identification device combining ultrasonic guided wave normalized energy transmittance and neural network according to claim 5 is characterized in that The horizontal distance from the excitation sensor to the receiving sensors on both sides should be kept consistent to ensure that the excitation signal reaches the receiving sensors on both sides almost at the same time, thereby eliminating the impact of energy attenuation during the propagation of the guided wave on signal processing.

9. The bolt group damage identification device combining ultrasonic guided wave normalized energy transmittance and neural network according to claim 5 is characterized in that The types of excitation sensors and receiving sensors include piezoelectric sensors, magnetostrictive waveguide sensors and electromagnetic ultrasonic sensors, and the excitation method also includes laser ultrasound.

10. The bolt group damage identification device combining ultrasonic guided wave normalized energy transmittance and neural network according to claim 5 is characterized in that The ultrasonic guided wave mode types include symmetric and antisymmetric Lamb wave modes and shear wave modes. The number of cycles and frequency of the excitation signal are reasonably selected according to the dispersion curve of the bolt plate structure to be tested.

Images