Bolt loosening monitoring method and system based on deep learning and ultrasonic guided waves under the influence of ambient temperature

Abstract

The application discloses a bolt loosening monitoring method and system based on deep learning and ultrasonic guided waves under the influence of ambient temperature. Q The bolt loosening monitoring method comprises the following steps: acquiring guided wave signals transmitted through bolt groups based on an ultrasonic guided wave transducer; acquiring principal components related to ambient temperature in the received signals through principal component analysis; calculating an index value based on residual signals from which the principal components related to the ambient temperature are removed, as a basis for judging whether bolt loosening exists; searching for independent signal components most related to bolt loosening in the residual signals based on independent component analysis, and constructing a sample training set of a deep neural network; and finally, building the deep neural network to locate the bolt loosening. The application not only overcomes the influence of ambient temperature changes on guided wave monitoring signals, but also realizes accurate positioning of bolt loosening under the premise of using a smaller number of transducers, and has high engineering application value.

Description

Technical Field

[0001] This invention belongs to the field of bolt condition monitoring technology, and relates to a bolt loosening monitoring method and system based on deep learning and ultrasonic guided waves under the influence of ambient temperature. Background Technology

[0002] Bolted connections offer advantages such as convenient construction, low cost, and good overall integrity, and are widely used in steel component connection nodes in civil engineering, machinery, and other fields. However, under the influence of cyclic traffic loads and changes in ambient temperature, bolt loosening frequently occurs, seriously affecting the overall load-bearing capacity of the structure. Bolt loosening can lead to localized structural failures, causing significant economic losses and social impacts. Therefore, developing effective bolt loosening monitoring methods and providing timely warnings are crucial for ensuring structural safety.

[0003] With the development of non-destructive testing (NDT) technologies, such as ultrasonic guided wave testing, electromechanical impedance testing, and computer vision, have been applied to bolt loosening monitoring. Electromechanical impedance testing determines the loosening state of a bolt by capturing changes in the impedance signal of a piezoelectric sensor near the bolt, but its monitoring range is limited. Compared to computer vision, ultrasonic guided wave testing offers higher accuracy in identifying early-stage bolt loosening and can monitor bolt loosening in concealed locations.

[0004] Domestic and international researchers have proposed various methods for monitoring bolt loosening based on ultrasonic guided waves, including transmission energy methods, time reversal techniques, and wake wave interferometry. Their monitoring effectiveness under specific conditions has been verified through finite element simulations or experiments. These methods acquire guided wave signals using piezoelectric or magnetostrictive guided wave transducers. Piezoelectric transducers are attached to the surface of the structure under test using a coupling agent; however, the piezoelectric ceramic itself is significantly affected by ambient temperature, and the coupling agent also undergoes morphological changes under varying temperatures. Magnetostrictive guided wave transducers include both contact and non-contact types, both of which are affected by ambient temperature, resulting in unstable received signals. Therefore, existing monitoring methods suffer from unstable monitoring results, false alarms, or missed alarms in practical applications. Overcoming the influence of ambient temperature on guided wave monitoring signals and achieving accurate positioning of loose bolts is a crucial technical problem that needs to be addressed.

[0005] Patent document CN120947886A discloses an acoustic non-destructive testing method and system for bolt tightness, which simultaneously excites longitudinal and transverse waves in the structure and utilizes characteristics such as the ratio of the propagation times of the two types of sound waves to overcome the interference of temperature changes on the signal. However, in actual operation, it has been found that the change in the propagation time of the sound waves is very small and is easily affected by measurement noise or traffic loads, and the monitoring efficiency of this method is not high.

[0006] Patent document CN118706302A discloses an ultrasonic-based method for measuring bolt axial stress. This method pre-obtains a dataset of received signals at different temperatures and then corrects the received signals based on the measured temperature. Clearly, this method has poor versatility, requiring the creation of multiple datasets for different bolt types, resulting in a significant amount of pre-calibration work. Furthermore, this method cannot be used in in-service structures.

[0007] Based on this, the present invention proposes a bolt loosening monitoring method that eliminates the influence of ambient temperature on guided wave monitoring through principal component analysis and combines independent component analysis with deep neural networks, effectively solving the above problems. Summary of the Invention

[0008] The purpose of this invention is to address the aforementioned technical problems by proposing a bolt loosening monitoring method and system based on deep learning and ultrasonic guided waves under the influence of ambient temperature. First, principal component analysis is used to remove signal components related to ambient temperature from the monitoring signal. Then, independent component analysis is used to search for the independent signal components in the monitoring signal that are most relevant to bolt loosening. Finally, a bolt loosening monitoring neural network based on deep learning is built to locate the loose bolt.

[0009] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0010] A bolt loosening monitoring method based on deep learning and ultrasonic guided waves under the influence of ambient temperature includes: acquiring guided wave signals passing through a bolt group using an ultrasonic guided wave transducer; using guided wave monitoring signals obtained at different temperatures in a non-destructive state as reference signals; obtaining signal components related to ambient temperature through principal component analysis of the reference signals; and calculating the residual signal after removing the principal components related to ambient temperature. Q The index value serves as the basis for determining whether there is bolt loosening in the test signal;

[0011] Independent component analysis was used to search for the most relevant independent signal components to bolt loosening in the residual signal, and a sample training set for a deep neural network was constructed. The deep neural network was then used to locate bolt loosening.

[0012] Specifically, the method includes the following steps:

[0013] S1: Ultrasonic guided wave transducers are installed on both sides of the bolted connection area of ​​the structure under test in a single-transmitter, single-receiver manner for guided wave signal excitation and guided wave signal reception.

[0014] S2: Place the entire structure under test in a temperature chamber to obtain ultrasonic guided wave signals of the bolt group under different temperature conditions and under different damage conditions;

[0015] S3: Construct a reference signal matrix X using guided wave signals obtained at different temperatures under non-destructive conditions, perform principal component analysis, and extract the principal components in the guided wave received signal that are most correlated with temperature.

[0016] S4: Remove temperature-related principal components from the received signal and calculate to determine whether damage exists based on the residual signal under lossless conditions. Q Indicator value threshold;

[0017] S5: Based on test signal Q The index value is used to determine whether there is bolt loosening under test conditions;

[0018] S6: Combine the residual signals under both undamaged and damaged conditions to conduct independent component analysis and obtain the independent components in the residual signals that are most relevant to bolt loosening;

[0019] S7: Based on the independent components of the residual signal, establish a training sample set for deep neural networks under different damage states;

[0020] S8: Based on the deep neural network trained by the training sample set, with independent components as input and bolt loosening position as output, accurate bolt loosening location is achieved.

[0021] Furthermore, the steps for constructing the reference signal matrix X described in S3 are as follows:

[0022] S3-1: Perform bandpass filtering on the i-th received signal, and the filtered signal is denoted as xi;

[0023] S3-2: Normalize xi;

[0024] S3-3: Arrange the normalized xi from top to bottom according to the temperature from smallest to largest to obtain an m×n matrix. , where m represents the number of reference signals, n represents the number of sampling points in a single reference signal, and the matrix element xij represents the value of signal xi at the j-th sampling point;

[0025] S3-4: From the matrix Columns ts to tf are selected to form the reference signal matrix X; the value of ts is equal to the distance between the two guided wave transducers divided by the group velocity of the currently excited guided wave, and the value of tf is flexibly selected according to the number of wave packets of the received signal.

[0026] Furthermore, in S3, after performing principal component analysis on the reference signal matrix X, the scores of the guided wave signal under the k-th principal component are compared. The temperature change trends of each signal in the reference signal matrix are used to determine the top r principal components most correlated with ambient temperature changes; the selected r values ​​must be sufficient to calculate the values ​​under both undamaged and damaged conditions of the bolt group.Q The index values ​​do not overlap, and the residual signal described in S4 This refers to the signal after removing the first r principal component information from the source signal, and the reference signal. Q The index value is obtained by the following formula: .

[0027] Furthermore, the threshold value mentioned in S4 is set as follows: Based on the 3σ criterion, the threshold value for determining whether bolt loosening exists is set to... ,in Calculate for all reference signals in the lossless state Q The average value of the indicator The mean square error is denoted as ; if the test signal's mean square error is . Q If the indicator value is greater than the threshold, it is determined that the bolt is loose.

[0028] Furthermore, in S6 and S7 specifically:

[0029] The residual reference signal matrix after removing principal components related to ambient temperature Residual test signals with known damage The data is assembled to form a new observation signal matrix. Extracting the observation signal matrix based on the fast independent component analysis algorithm Given the source signal matrix S and the mixing matrix A, calculate the correlation coefficient between each column of data in the mixing matrix A and the step vector b. m is the number of reference signals; then, all correlation coefficients are sorted from largest to smallest, and the principal components corresponding to the larger correlation coefficients are most correlated with bolt loosening. The top q independent components in the source signal matrix S that are most correlated with bolt loosening are selected and multiplied with the weight values ​​corresponding to the last row in the mixing matrix A to obtain a training sample Xinput for the deep neural network.

[0030] The above operation is repeated for the guided wave received signals obtained under all damage conditions at different temperatures in S2, thereby constructing a training sample set for the deep neural network.

[0031] Furthermore, the observation signal matrix In the structure, rows 1 to m are the signals after removing the first r principal components from the original reference signal matrix X; the last row is... , representing the residual signal after removing the first r principal components of a test signal with known damage; the number of sampling points for a single signal is . ;

[0032] Observation signal matrix The relationship between the source signal matrix S and the mixing matrix A is as follows: .

[0033] Furthermore, in S7, the label value of the training sample is the bolt loosening status related to the location. In S8, the number of output values ​​of the deep neural network corresponds to the number of bolts being tested. Each output value is 0 or 1, which respectively represent that the bolt at the current output position is not loose or that the bolt at the current output position is loose.

[0034] The present invention also provides an electronic device, comprising:

[0035] One or more processors;

[0036] Memory, used to store one or more programs;

[0037] When the one or more programs are executed by the one or more processors, the one or more processors perform the method as described in any of the preceding methods.

[0038] A computer-readable storage medium storing computer-executable instructions, which, when executed, are used to implement the method described in any of the preceding claims.

[0039] The beneficial effects of this invention are:

[0040] 1) This invention creatively proposes a bolt loosening monitoring method based on deep learning and ultrasonic guided waves under the influence of ambient temperature. Principal component analysis is performed on guided wave monitoring signals obtained at different temperatures in a non-destructive state to obtain signal components related to ambient temperature. The method then uses these signal components, after removing the influence of ambient temperature, to calculate… Q The indicator value allows for a quick assessment of bolt looseness.

[0041] 2) This invention uses independent component analysis to obtain the most relevant independent components to bolt loosening in the residual guided wave signal after removing the influence of ambient temperature. It proposes using these independent components as input to a deep neural network to predict the location of loose bolts. The proposed method is applicable to common bolted connection structures in engineering, has high detection efficiency, and provides accurate prediction results. It can locate the loosening of multiple bolts in a single node using only a small number of transducers, and is expected to be widely used in engineering projects. Attached Figure Description

[0042] Figure 1 This is a flowchart of a bolt loosening monitoring method based on deep learning and ultrasonic guided waves under the influence of ambient temperature.

[0043] Figure 2 It is a common lap bolt connection structure;

[0044] Figure 3 It is a test system for acquiring guided wave signals under different temperatures and different bolt loosening conditions in the laboratory;

[0045] Figure 4The results are before and after bandpass filtering of the received signal under lossless condition at 5℃;

[0046] Figure 5 These are the first four principal component vectors of the reference signal matrix;

[0047] Figure 6 It is the score distribution of the reference signal under the first three principal components;

[0048] Figure 7 It is the residual signal after the first three principal components are proposed;

[0049] Figure 8 It is the distribution of the Q index values ​​of the test signal;

[0050] Figure 9 These are the first four independent components of the observed signal matrix corresponding to a certain damage condition;

[0051] Figure 10 It is the distribution of correlation coefficients corresponding to the 7 independent components;

[0052] Figure 11 It consists of the results of a deep neural network and some of its hyperparameters;

[0053] Figure 12 This is the prediction result of the deep neural network on the location of the loose bolt. Detailed Implementation

[0054] The present invention will be further described below with reference to embodiments. The description of the embodiments below is only for the purpose of helping to understand the present invention. It should be noted that those skilled in the art can make several modifications to the present invention without departing from the principle of the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

[0055] like Figure 1 As shown, the bolt loosening monitoring method based on deep learning and ultrasonic guided waves under the influence of ambient temperature provided by the present invention includes the following steps:

[0056] S1: Install ultrasonic guided wave transducers on both sides of the bolted connection area of ​​the structure under test. One transducer is used for guided wave signal excitation, and the other is used for guided wave signal reception. The ultrasonic guided wave transducers include, but are not limited to, piezoelectric transducers, magnetostrictive transducers, and laser ultrasound. The excited ultrasonic guided wave signal modes include, but are not limited to, Lamb waves and in-plane shear mode guided waves. S2: Place the entire structure under test in a temperature chamber. Using the service environment of the bolted connection structure as a reference, set the temperature change range and temperature change gradient to obtain ultrasonic guided wave signals of the bolt group under non-destructive and different damage conditions at different temperatures. For large structures, local nodes with the same bolt arrangement as those in the structure can be fabricated for laboratory calibration and placed in a temperature chamber for testing under different temperatures and damage conditions.

[0057] S3: Construct a reference signal matrix X using guided wave signals obtained at different temperatures under non-destructive conditions. Perform principal component analysis on the reference signal matrix, and determine the principal component in the received guided wave signal that is most correlated with temperature by using the score values ​​of the guided wave signal under different principal components; specifically:

[0058] The steps for constructing the reference signal matrix X are as follows:

[0059] S3-1: Perform bandpass filtering on the i-th received signal, and the filtered signal is denoted as xi;

[0060] S3-2: Normalize xi;

[0061] S3-3: Arrange the normalized xi from top to bottom according to the temperature from smallest to largest to obtain an m×n matrix. , where m represents the number of reference signals, n represents the number of sampling points in a single reference signal, and the matrix element xij represents the value of signal xi at the j-th sampling point;

[0062] S3-4: From the matrix Columns ts to tf are selected to form the reference signal matrix X; the signal between columns ts and tf is the effective signal component, containing complete direct wave information. The value of ts is equal to the distance between the two guided wave transducers divided by the group velocity of the currently excited guided wave, and the value of tf is flexibly selected according to the number of wave packets of the received signal.

[0063] S4: Remove the first r-order principal components related to ambient temperature from the original guided wave received signal to obtain the residual signal. Then according to Calculate all reference signals in the lossless state. Q Based on the 3σ criterion, the threshold value for determining whether bolt loosening exists is set as follows: ,in Calculate for all reference signalsQ The average value of the indicator For all Q The mean square error of the index value. The value of r should not be too large to avoid the residual signal losing the characteristic information of bolt loosening. It is necessary to ensure that the calculated values ​​under both non-destructive and damaged conditions can be obtained based on the selected r value. Q The index values ​​do not overlap; the aforementioned Q The index value describes the difference of the test signal outside the principal component plane formed by the reference signal matrix. The reference signal matrix consists of signals at different temperatures when the bolt is not loose. When the test signal corresponds to the bolt being loose, its... Q The indicator value must exceed the threshold set under the condition of no looseness. );

[0064] S5: Calculate the test signal in the same way as S3-S4. Q The index value, if the test signal Q If the indicator value is less than the threshold, it is determined that no bolts are loose; if the test signal... Q If the indicator value is greater than the threshold, it is determined that there is a loose bolt, and then the bolt loosening location is carried out based on deep learning, i.e., S6-S8;

[0065] S6: Combine the residual signals from the undamaged and damaged states to form a new observation signal matrix, and extract the independent signal components most relevant to bolt loosening based on the fast independent component analysis algorithm to construct a training sample for a deep neural network; the specific scheme is as follows:

[0066] The residual reference signal matrix after removing principal components related to ambient temperature Residual test signals with known damage The data is assembled to form a new observation signal matrix. Extracting the observation signal matrix based on the fast independent component analysis algorithm Given the source signal matrix S and the mixing matrix A, calculate the correlation coefficient between each column of data in the mixing matrix A and the step vector b. Here, m represents the number of reference signals. Then, all correlation coefficients are sorted from largest to smallest. The principal components corresponding to the largest correlation coefficients are most correlated with bolt loosening. The top q independent components in the source signal matrix S that are most correlated with bolt loosening are selected, and multiplied by the weight values ​​corresponding to the last row of the mixing matrix A to obtain a training sample Xinput for the deep neural network. The value of q can be adjusted based on the subsequent training results of the neural network. If the training effect of the neural network is not good, the number of independent components can be increased.

[0067] S7: Repeat the operation of S6 for the guided wave received signals obtained under all damage conditions at different temperatures in S2, thereby constructing a training sample set for the deep neural network.

[0068] S8: Build a deep neural network structure. The input of the neural network is a signal matrix composed of the top q independent components most relevant to bolt loosening, and the output is the location of bolt loosening. Train and test the neural network using the training sample set. 80% of the data can be randomly selected from the training sample set for training the neural network, and the remaining 20% ​​of the data can be used to test the prediction accuracy of the neural network. By continuously testing and optimizing the structure and hyperparameters of the neural network, accurate bolt loosening can be located.

[0069] As one example, such as Figure 2 The lap bolted connection structure shown here consists of two steel plates and four M10 high-strength bolts. The relevant dimensions are as follows: Figure 2 As shown. Two magnetostrictive shear-mode guided wave transducers (MS1 and MS2) are installed on both sides of the bolted connection area. MS1 is used for signal excitation, and MS2 is used for signal reception. The excitation signal is a 3-cycle sinusoidal modulated signal with a center frequency of 80 kHz, and the propagation speed of the shear-mode guided wave at the current frequency is 3230 m / s. Construction Figure 3 The test system shown is used to acquire guided wave signals under different temperatures and bolt loosening conditions. The system includes a temperature chamber and control system, a torque wrench, a guided wave instrument, an electronic thermometer, and a control computer. The bolted connection structure is placed entirely within the temperature chamber, with a temperature range of 5 to 65°C and a temperature change step of 5°C, totaling 13 temperature points. Electronic thermometers are attached to both sides of the bolted connection nodes to obtain real-time surface temperatures on the board, which are then compared with the temperature values ​​from the temperature chamber control system. At a single temperature test point, five sets of guided wave signals are acquired under undamaged conditions. Then, by changing the bolt loosening condition using a torque wrench, 20 sets of guided wave signals are acquired under damaged conditions. Ten different bolt loosening conditions are set, including four single bolt loosening conditions (B1, B2, B3, B4) and six conditions with two loose bolts (B1 and B2, B1 and B3, B1 and B4, B2 and B3, B2 and B4, B3 and B4). Therefore, there are a total of 65 guided wave signals in the undamaged state and 2600 guided wave signals in the damaged state.

[0070] Figure 4 The results of bandpass filtering on the received signal xi at 5°C under lossless conditions are shown. Since the center frequency of the excitation signal is 80 kHz, the frequency range of the bandpass filter is 60 to 100 kHz. Subsequently, xi is normalized; the specific operation procedure is as follows:

[0071]

[0072]

[0073]

[0074] In the formula, xij represents the value of xi at the j-th sampling point. The same processing is applied to the received signals under all different temperatures and lossless conditions. Then, the normalized xi are arranged from top to bottom according to the temperature from smallest to largest, resulting in the matrix. :

[0075]

[0076] Where m represents the number of reference signals, in this example m = 65. n represents the number of sampling points in a single reference signal. From the matrix Columns 119 to 400 are selected to form a reference signal matrix X, with a size of 65×282.

[0077] The principal component analysis of the reference signal matrix X is performed as follows:

[0078]

[0079]

[0080]

[0081]

[0082] Matrix C is the covariance matrix of matrix X; P is the principal component vector matrix. This represents the k-th principal component vector; This is the eigenvalue matrix corresponding to the principal components. Figure 5 The first four principal component vectors are shown. to The first four eigenvalues ​​are 0.0878, 0.0128, 0.0019, and 7.718e-4, respectively. The score for each signal in the reference signal matrix under the first three principal components is calculated using the following formula:

[0083]

[0084] By comparison By comparing the temperature change trends of each signal in the reference signal matrix, the top r principal components in P that are most correlated with environmental temperature changes can be determined. In the example above, the score distribution results of the 65 sets of reference signals under the first 3 principal components are as follows: Figure 6 As shown. The results showed that, The score distribution under the first principal component showed the highest consistency with the temperature value distribution of the 65 reference signals, indicating that the second principal component is most correlated with changes in ambient temperature. Furthermore, the score distributions under the other two principal components also showed some correlation with temperature. Therefore, in this example, the signal components covered by the first three principal components will be removed from the original received signal. The remaining signal after removing the first three principal components... The calculation formula is:

[0085]

[0086] In the formula, This refers to the matrix composed of the 4th to 65th principal component vectors. Figure 4 Taking the signal in the example, the residual signal after removing the first three principal components is shown in [the image]. Figure 7 The Q index value of the residual reference signal is calculated using the following formula:

[0087]

[0088] Based on the 3σ criterion, the threshold for determining whether bolt loosening exists is set as follows: ,in Calculate for all reference signals Q The average value of the indicator For all Q The mean square deviation of the index value.

[0089] 2600 sets of guided wave test signals under different temperatures and damage conditions were imported. These signals were filtered and normalized using the same processing method as the reference signals, and the signal components in the first three principal components were also removed. Then, the values ​​of the test signals were calculated. Q Indicator values, results are shown in Figure 8 Clearly, the lossless and lossy states can be calculated based on the residual signal after removing the first three principal components. Q The index values ​​are clearly distinguishable and do not overlap. The test signal (damage state) is calculated... Q All indicator values ​​exceed the threshold, indicating that it can be determined based on... Q The threshold value of the indicator is used to determine whether the bolt has become loose, and then bolt loosening is located based on deep learning.

[0090] The residual reference signal matrix after removing principal components related to ambient temperature With one residual test signal that was determined to be damaged The data is assembled to form a new observation signal matrix. , The dimensions are 66×282. The format is as follows:

[0091]

[0092] in, Rows 1 to 65 are the signals after removing the first three principal components from the original reference signal matrix X; The residual signal after removing the first three principal components of the test signal determined to be damaged is used. The observed signal matrix is ​​extracted based on a fast independent component analysis algorithm. The source signal matrix S and the mixing matrix A, the observed signal matrix The relationship between the source signal matrix S and the mixing matrix A is as follows:

[0093]

[0094] Taking the signal under a specific damage state as an example, through independent component analysis, the observed signal matrix composed of this signal and the residual reference signal is obtained. The signal is decomposed into a signal with 7 independent components. The source signal matrix S has a size of 7×282, and the mixing matrix A has a size of 66×7. Figure 9 The first four independent components were shown.

[0095] Calculate the correlation coefficient between each column of data in the mixing matrix A and the step vector b, where The calculation formula is:

[0096]

[0097] In the formula, This represents the covariance between the j-th column of data in A and the step vector b; and They are respectively The standard deviation of the step vector b. Figure 10 The correlation coefficient distribution for the seven independent components is shown. It can be observed that independent components 2, 3, 4, and 7 have the highest correlation coefficients and are most closely related to bolt loosening. Multiplying these four independent components by the weight values ​​corresponding to the last row of the mixture matrix A yields a training sample, Xinput, for the deep neural network. Xinput has the following form:

[0098]

[0099] The size of Xinput is 4×282. The above operation is repeated for the test signals under 2600 damage states to obtain 2600 training samples. The label value of the sample is the state of bolt loosening. For example, if only B1 is loose, the label is [1, 0, 0, 0]; if B1 and B3 are loose, the label is [1, 0, 1, 0].

[0100] A deep neural network was built to locate loose bolts. The structure of the neural network is shown below. Figure 11The neural network consists of an input layer, two convolutional layers, one pooling layer, two fully connected layers, and an output layer. 80% of the data was randomly selected from the training sample set for training the neural network, while the remaining 20% ​​was used to test its prediction accuracy. Through continuous testing, the network's structure and hyperparameters were optimized. Ultimately, the neural network achieved a training accuracy of 99.5%. Figure 12 The results of bolt loosening location under four random working conditions are presented, demonstrating the accuracy of the proposed method.

[0101] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0102] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0103] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0104] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0105] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. The embodiments described above are merely some preferred solutions of the present invention and are not intended to limit the invention. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention. Therefore, all technical solutions obtained by equivalent substitution or equivalent transformation fall within the protection scope of the present invention.

Claims

1. A bolt loosening monitoring method based on deep learning and ultrasonic guided waves under the influence of ambient temperature, characterized in that, Includes the following steps: S1: Ultrasonic guided wave transducers are installed on both sides of the bolted connection area of ​​the structure under test in a single-transmitter, single-receiver manner for guided wave signal excitation and guided wave signal reception. S2: Place the entire structure under test in a temperature chamber to obtain ultrasonic guided wave signals of the bolt group under different temperature conditions and under different damage conditions; S3: Construct a reference signal matrix X using guided wave signals obtained at different temperatures under non-destructive conditions, perform principal component analysis, and extract the principal components in the guided wave received signal that are most correlated with temperature. S4: Remove temperature-related principal components from the received signal and calculate to determine whether damage exists based on the residual signal under lossless conditions. Q Indicator value threshold; S5: Based on test signal Q The index value is used to determine whether there is bolt loosening under test conditions; S6: Combine the residual signals under both undamaged and damaged states, perform independent component analysis to obtain the independent components in the residual signals most relevant to bolt loosening; specifically: extract the residual reference signal matrix after removing principal components related to ambient temperature. Residual test signals with known damage The data is assembled to form a new observation signal matrix. Extracting the observation signal matrix based on the fast independent component analysis algorithm Given the source signal matrix S and the mixing matrix A, calculate the correlation coefficient between each column of data in the mixing matrix A and the step vector b. Let m be the number of reference signals. Then, sort all the correlation coefficients from largest to smallest. The principal components corresponding to the larger correlation coefficients are most correlated with bolt loosening. Select the top q independent components in the source signal matrix S that are most correlated with bolt loosening, and multiply them by the weight values ​​corresponding to the last row of the mixing matrix A to obtain a training sample X for the deep neural network. input ; S7: Repeat the operation of S6 for the guided wave received signals obtained under all damage conditions at different temperatures in S2, thereby constructing a training sample set for the deep neural network. S8: Based on the deep neural network trained by the training sample set, with independent components as input and bolt loosening position as output, accurate bolt loosening location is achieved.

2. The bolt loosening monitoring method based on deep learning and ultrasonic guided waves under the influence of ambient temperature as described in claim 1, characterized in that, The steps for constructing the reference signal matrix X described in S3 are as follows: S3-1: Perform bandpass filtering on the i-th received signal, and denot the filtered signal as x. i ; S3-2: For x i Perform normalization processing; S3-3: Arrange the normalized x values ​​in ascending order of temperature. i Arranged from top to bottom, we obtain an m×n matrix. Where m represents the number of reference signals, n represents the number of sampling points in a single reference signal, and the matrix element x ij Representing signal x i The value at the j-th sampling point; S3-4: From the matrix Select the t-th s To t f The columns form the reference signal matrix X; t s The value of t is equal to the distance between the two waveguide transducers divided by the group velocity of the waveguide currently being generated, t. f The value is flexibly selected based on the number of wave packets of the received signal.

3. The bolt loosening monitoring method based on deep learning and ultrasonic guided waves under the influence of ambient temperature as described in claim 1, characterized in that, In S3, after performing principal component analysis on the reference signal matrix X, the scores of the guided wave signal under the kth principal component are compared. The temperature change trends of each signal in the reference signal matrix are used to determine the top r principal components most correlated with changes in ambient temperature; the selected... r The value must be calculated under both undamaged and damaged bolt group conditions. Q The index values ​​do not overlap, and the residual signal described in S4 This refers to the signal after removing the first r principal component information from the source signal, and the reference signal. Q The index value is obtained by the following formula: .

4. The bolt loosening monitoring method based on deep learning and ultrasonic guided waves under the influence of ambient temperature as described in claim 1, characterized in that, The threshold value described in S4 is set as follows: Based on the 3σ criterion, the threshold value for determining whether bolt loosening exists is set to... ,in Calculate all reference signals in the lossless state of S4 Q The average value of the indicator The mean square error is denoted as ; if the test signal's mean square error is . Q If the indicator value is greater than the threshold, it is determined that the bolt is loose.

5. The bolt loosening monitoring method based on deep learning and ultrasonic guided waves under the influence of ambient temperature as described in claim 1, characterized in that, The observation signal matrix described in S6 In the structure, rows 1 to m are the signals after removing the first r principal components from the original reference signal matrix X; the last row is... , representing the residual signal after removing the first r principal components of a test signal with known damage; the number of sampling points for a single signal is . ; Observation signal matrix The relationship between the source signal matrix S and the mixing matrix A is as follows: 。 6. The bolt loosening monitoring method based on deep learning and ultrasonic guided waves under the influence of ambient temperature as described in claim 1, characterized in that, In S7, the label value of the training sample is the bolt loosening status related to the location. In S8, the number of output values ​​of the deep neural network corresponds to the number of bolts being tested. Each output value is 0 or 1, which respectively represent that the bolt at the current output position is not loose or that the bolt at the current output position is loose.

7. The bolt loosening monitoring method based on deep learning and ultrasonic guided waves under the influence of ambient temperature as described in claim 1, characterized in that, The deep neural network is a convolutional neural network, a recurrent neural network, a long short-term memory neural network, or a Transformer neural network.

8. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-7.

Images