A bolt loosening detection method combining time reversal technology and BP neural network

Abstract

The present invention discloses a bolt loosening detection method combining time reversal technology and BP neural network. A plurality of ultrasonic guided wave sensors are arranged near the bolt group to be tested, and the ultrasonic guided wave is excited and received by using time reversal technology to extract the characteristic value of the guided wave signal under a specific damage state. The BP neural network is used for local damage location and quantitative analysis. The input of a single training sample is the characteristic value of the guided wave signal under multiple propagation paths corresponding to all the excitation and receiving sensors, and the output is the damage position and local damage degree of the bolt group. In addition, a linear relationship between the average characteristic value of the signal under multiple propagation paths and the overall looseness degree of the bolt group is established to realize the prediction of the overall damage state of the bolt group. The proposed detection method realizes the damage location and damage degree prediction of the bolt group structure composed of a large number of bolts under the premise of using fewer sensors. The detection results are reliable and have strong engineering applicability, which provides a guarantee for ensuring the safety of bolt-connected steel structures.

Description

Technical Field

[0001] The invention belongs to the technical field of building engineering, and relates to time reversal technology, BP neural network and structural health monitoring technology. Background Art

[0002] Bolt connection is a typical steel component connection method, which is widely used in many industries including civil engineering, aerospace, mechanical engineering, power engineering, etc. The tightness of bolts is related to the operational safety of equipment and the safety of structures. The tightness of bolts is affected by multiple factors such as environmental loads and traffic loads. It is inevitable that one or more bolts will loosen during use. When the bolts loosen to a certain extent, it will cause sudden damage to the structure, causing huge economic losses and adverse social impacts. Realizing health monitoring of bolted connection structures is of great significance to improving the safety performance of equipment.

[0003] The current detection methods for bolt loosening include strain gauge detection, torque meter manual detection, ultrasonic guided wave detection, impedance method, machine vision technology, etc. Among them, the cost of manual detection is too high, the strain gauge detection method has low detection accuracy and requires a large number of sensors to be installed, the impedance method has a small detection area, and ultrasonic guided wave technology and machine vision technology have become research hotspots in recent years.

[0004] In terms of bolt loosening detection based on ultrasonic guided waves, the detection methods currently used include energy method, time reversal method, frequency method, etc. Due to the uneven surface characteristics of the bolt plate, the preload of the bolt determines the actual contact area between the bolt plates. The ultrasonic guided wave signal is excited from one end of the bolt plate, and the signal is transmitted to the other end of the bolt plate through the contact area. The energy transmitted by the signal is determined by the actual contact area of ​​the bolt plate. The detection technology based on the energy method uses this feature of the signal to establish the relationship between the energy of the transmission signal and the bolt torque. However, the actual contact area of ​​the bolt plate will no longer change after the bolt torque reaches a certain level, resulting in this method being unable to detect early bolt loosening. Similarly, the change in signal frequency based on the frequency method is not obvious enough and is easily affected by environmental noise. The time reversal technology achieves accurate prediction of the degree of bolt loosening by focusing the energy of the guided wave signal at the signal receiving sensor.

[0005] Patent document CN113433214A proposes a damage location method for plate structures, but for bolt structures, the structure is more complex and the propagation path of guided waves is difficult to predict. Furthermore, the method uses too many sensors and cannot be promoted in actual engineering applications. Summary of the invention

[0006] The purpose of the present invention is to address the above technical problems and propose a bolt loosening detection method that combines time reversal technology and BP neural network. A small number of sensors are used to detect the overall and local damage degree of a complex bolt group, locate the damaged area, ensure the safety of the bolted connection structure during operation, and prevent disasters.

[0007] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0008] A bolt loosening detection method combining time reversal technology and BP neural network includes the following steps:

[0009] S1: Arrange multiple ultrasonic guided wave excitation and receiving sensors of the same number near the bolt group to be tested, and use a single-transmit and single-receive method to excite and receive ultrasonic guided waves;

[0010] S2: In a lossless state, all excitation and receiving sensors are combined one by one. Each combination represents a different guided wave propagation path. The ultrasonic guided wave reconstruction signal between the specific excitation and receiving sensors is obtained by using time reversal technology, and the maximum value of the reconstructed signal in the time domain is extracted.

[0011] S3: Under the damaged state, the time reversal technique is used to obtain the ultrasonic guided wave reconstruction signal under all the excitation and receiving sensor combinations, and the ratio of the maximum value of the reconstructed signal under the damaged state to the maximum value of the reconstructed signal under the intact state on the corresponding propagation path is calculated as the characteristic value of the guided wave signal on the propagation path;

[0012] S4: Divide the bolt group into multiple independent areas, each area contains multiple independent bolts;

[0013] S5: Build a BP neural network. The input of a single sample is the characteristic value of the guided wave signal on all propagation paths under the damage state. The number of output nodes is the number of regions divided by the bolt group. The position of the output node represents the area where the bolt is loose, and the size of the output node value represents the degree of damage in the area where the bolt is loose.

[0014] S6: Through a large amount of data training and testing of BP neural network, the location of bolt loosening and the degree of local bolt loosening can be accurately identified;

[0015] S7: The characteristic values ​​of the guided wave signals on all propagation paths under a single damage condition are averaged, and a linear relationship with the overall looseness of the bolt group is established through a large number of damage samples to predict the overall damage degree of the bolt group.

[0016] Furthermore, the bolt group can be divided into multiple independent areas according to structural features.

[0017] Furthermore, the ultrasonic guided wave excitation and receiving sensors are characterized in that they include but are not limited to piezoelectric and magnetostrictive ultrasonic guided wave sensors, which can be flexibly selected according to the actual components being measured;

[0018] Furthermore, the ultrasonic guided wave is characterized in that it includes various forms such as Lamb wave and SH wave, and its excitation frequency is reasonably selected according to the dispersion characteristics of the structure to be measured;

[0019] Furthermore, the plurality of ultrasonic guided wave sensors are characterized in that the number of sensors used for exciting and receiving signals is greater than or equal to two, but much less than the total number of bolts, and the number of exciting and receiving sensors is flexibly selected according to the measured size of the measured component and the total number of bolts;

[0020] Furthermore, the time reversal technology is characterized by comprising the following steps:

[0021] S2-1: The excitation sensor excites a Gaussian pulse signal δ(t), and the waveguide signal is transmitted from one side of the bolt plate through the bolt plate connection area to the signal receiving sensor on the other side. Assuming that the structural impulse response is h(t), the received signal is y(t) = h(t);

[0022] S2-2: Convert the received signal y(t) to 0-t in the time domain 0 The time-reversed segment is then converted into a signal y(t 0 -t)=h(t 0 -t) is re-sent to the excitation sensor as an input signal;

[0023] S2-3: Obtaining reconstructed signal from receiving sensor Extract 0 The reconstructed signal value at time

[0024] Furthermore, the BP neural network input is characterized in that, for the input of the BP neural network, the characteristic value of the waveguide signal on a single waveguide propagation path is between 0 and 1 (lossless).

[0025] Furthermore, the BP neural network output is characterized in that, for the output of the BP neural network, the output value of a single neuron node is between 0 (lossless) and 1, representing the degree of damage in the local area.

[0026] Furthermore, the training data set of the BP neural network can be obtained through a finite element model or through actually collected signals, and the hyperparameters of the neural network are determined through repeated tests.

[0027] The method proposed by the present invention requires a small number of sensors and can be applied to complex bolt connection structures. The damage degree detection can be specific to independent areas with high detection accuracy. The present invention creatively combines BP neural network and time reversal technology, takes the damage characteristic values ​​on different guided wave propagation paths as the judgment basis, accurately identifies the damage location, and realizes high-efficiency detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] 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:

[0029] Figure 1 is a schematic diagram of a bolt connection structure exemplified in an embodiment of the present invention;

[0030] Figure 2 is the spectrum of Gaussian pulse signal excited by excitation sensor A1;

[0031] Figure 3 It is the signal spectrum received by the receiving sensor R1;

[0032] Figure 4 It is the signal spectrum received by R1 in time reversal;

[0033] Figure 5 is the reconstructed signal spectrum obtained by receiving sensor R1;

[0034] Figure 6 It is the structure diagram of BP neural network;

[0035] Figure 7 This is a BP neural network prediction result diagram of an embodiment of the present invention;

[0036] Figure 8 It is a diagram showing the prediction result of the overall damage degree of the bolt group according to an embodiment of the present invention. DETAILED DESCRIPTION

[0037] A bolt loosening detection method combining time reversal technology and BP neural network is described in detail below with reference to the accompanying drawings.

[0038] The embodiments described herein are specific embodiments of the present invention, which are used to illustrate the concept of the present invention, are illustrative and exemplary, and should not be construed as limiting the embodiments of the present invention and the 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 contents disclosed in the specification of this application, including technical solutions that adopt any obvious replacement and modification of the embodiments described herein.

[0039] The present invention provides a bolt loosening detection method combining time reversal technology and BP neural network. Figure 1 The specific structural form shown includes the following steps:

[0040] S1: Figure 1 The bolt group under test shown in the figure includes 8 independent bolts. Two ultrasonic guided wave excitation and receiving sensors are arranged at both ends of the bolt connection area. The ultrasonic guided waves are excited and received in a single-transmit and single-receive manner. The propagation paths of the ultrasonic guided waves include A1-R1, A1-R2, A2-R1, and A2-R2.

[0041] S2: In lossless state, Figure 2 The Gaussian pulse signal shown in the figure is input to excite sensor A1, and the ultrasonic guided wave signal is transmitted from the left bolt plate to the right bolt plate through the bolt connection area, and the receiving sensor R1 obtains Figure 3 The received signal is shown;

[0042] S3: Select Figure 3 Received signal 0-1.5×10 -3 s as the effective signal part and in the range of 0-1.5×10 -3 Invert the time-course signal in the s interval and get Figure 4 The inverted signal shown is re-input into the excitation sensor A1;

[0043] S4: Receive sensor R1 to get Figure 5 The reconstructed signal shown in the figure extracts 1.5×10 -3 Signal amplitude Y at time s intact,A1-R1 ;

[0044] S5: Repeat S2-S4 to obtain Y intact,A1-R2 , Y intact,A2-R1 , Y intact,A2-R2 ;

[0045] S6: Loosen the bolts and repeat S2-S5 to obtain the Y value in the current damage state. damage,A1-R1 , Y damage,A1-R2 , Y damage,A2-R1 , Y damage,A2-R2 ;

[0046] S7: Calculate the characteristic value DI of the waveguide signal on the four propagation paths under the current damage state. Take the propagation path A1-R1 as an example. DI A1-R1 The calculation formula is: In lossless state, DI=1;

[0047] S8: Divide the bolt group into 4 rows, each row contains two independent bolts;

[0048] S9: Different DI values ​​on four waveguide propagation paths can be obtained under a single damage state as a set of training samples for the neural network. The sample labels are four numbers ranging from 0 to 1 (lossless). The position of the sample label number represents the row number where the bolt damage is located, and the size of the sample number represents the local damage degree of the bolt in that row;

[0049] S10: Collect training sample sets of neural network through a large number of damage conditions;

[0050] S11: Input the training sample set into Figure 6 The neural network is trained and the hyperparameters of the neural network are adjusted to improve the accuracy of the training results. Then the test data is put into the trained neural network for prediction. Figure 7 The prediction results are shown. Figure 7 It can be seen that the damage occurs in the 1st and 2nd rows, and the corresponding predicted damage degree values ​​are close to the actual values;

[0051] S12: Calculate the average value of DI on the four propagation paths under a single damage state, and establish a linear relationship between the average value of DI and the overall damage state of the bolt group through a large number of samples. Figure 8 The figure shows the relationship between the average DI value and the overall damage degree of the bolt group under different conditions. Figure 8 Linear relationship is used to predict the overall damage degree of bolt groups.

[0052] The present invention is not limited to the above-mentioned embodiments. Anyone can derive other various forms of products under the inspiration of the present invention. However, no matter what changes are made in the shape or structure, any technical solution that is the same or similar to that of the present application falls within the protection scope of the present invention.

Claims

1. A bolt loosening detection method combining time reversal technology and BP neural network, characterized in that: The steps include: S1: Arrange multiple ultrasonic guided wave excitation and receiving sensors of the same number near the bolt group to be tested, and use a single-transmit and single-receive method to excite and receive ultrasonic guided waves; S2: In a lossless state, all excitation and receiving sensors are combined one by one. Each combination represents a different guided wave propagation path. The ultrasonic guided wave reconstruction signal between the excitation and receiving sensors is obtained by using time reversal technology, and the maximum value of the reconstructed signal in the time domain is extracted. S3: Under the damaged state, the time reversal technique is used to obtain the ultrasonic guided wave reconstruction signal under all the excitation and receiving sensor combinations, and the ratio of the maximum value of the reconstructed signal under the damaged state to the maximum value of the reconstructed signal under the intact state on the corresponding propagation path is calculated as the characteristic value of the guided wave signal on the propagation path; S4: Divide the bolt group into multiple independent areas, each area contains multiple independent bolts; S5: Build a BP neural network. The input of a single sample is the characteristic value of the guided wave signal on all propagation paths under the damage state. The number of output nodes is the number of regions divided by the bolt group. The position of the output node represents the area where the bolt is loose, and the size of the output node value represents the degree of damage in the area where the bolt is loose. S6: Through a large amount of data training and testing of BP neural network, the location of bolt loosening and the degree of local bolt loosening can be accurately identified; S7: The characteristic values ​​of the guided wave signals on all propagation paths under a single damage condition are averaged, and a linear relationship with the overall looseness of the bolt group is established through a large number of damage samples to predict the overall damage degree of the bolt group.

2. The bolt loosening detection method combining time reversal technology and BP neural network according to claim 1 is characterized in that: The ultrasonic guided wave excitation and receiving sensors include piezoelectric and magnetostrictive ultrasonic guided wave sensors.

3. The bolt loosening detection method combining time reversal technology and BP neural network according to claim 1 is characterized in that: The ultrasonic guided wave types include Lamb waves and SH waves, and the excitation frequency thereof is reasonably selected according to the dispersion characteristics of the structure to be measured.

4. The bolt loosening detection method combining time reversal technology and BP neural network according to claim 1 is characterized in that: The number of excitation or receiving sensors is greater than or equal to two, but much less than the total number of bolts in the bolt group. The number of excitation and receiving sensors is determined according to the size of the measured component and the number of bolts.

5. The bolt loosening detection method combining time reversal technology and BP neural network according to claim 1 is characterized in that: The time reversal technology comprises the following steps: S2-1: The excitation sensor excites a Gaussian pulse signal, and the waveguide signal is transmitted from one side of the bolt plate through the bolt plate connection area to the signal receiving sensor on the other side; S2-2: Time-reverse the received signal in the time domain 0-t0 segment, and resend the reversed signal as the input signal to the excitation sensor; S2-3: Obtain the reconstructed signal from the receiving sensor and extract the signal value at time t0.

6. The bolt loosening detection method combining time reversal technology and BP neural network according to claim 1 is characterized in that: For the input of the BP neural network, the characteristic value of the waveguide signal calculated under a single waveguide propagation path is between 0 and 1, and the characteristic value of the waveguide signal is equal to 1 in the lossless state.

7. The bolt loosening detection method combining time reversal technology and BP neural network according to claim 1 is characterized in that: For the output of the BP neural network, the position of the output neuron node represents the local bolt area, and the output value of a single neuron node is between 0 and 1, representing the degree of damage in the local area.

Images