Bolt pre-tightening force prediction method based on ultrasonic echo time-frequency characteristics

Abstract

The invention discloses a bolt pre-tightening force prediction method based on ultrasonic echo time-frequency characteristics, and the method comprises the steps: S1, carrying out the ultrasonic excitation of a target bolt, obtaining an ultrasonic echo signal generated by the bolt, and carrying out the time-frequency analysis, and forming an initial time-frequency characteristic matrix; s2, according to a bolt material characteristic database, determining characteristic parameters of ultrasonic echo propagation corresponding to a target bolt material, and performing weighting processing on the initial time-frequency characteristic matrix to obtain a weighted time-frequency characteristic matrix; s3, extracting a time-frequency characteristic index through the weighted time-frequency characteristic matrix, and forming a target characteristic vector; and S4, inputting the target feature vector into the trained double-branch prediction model, outputting a pre-tightening force energy feature value and a pre-tightening force state change value, inputting the pre-constructed neural network model, and outputting a bolt pre-tightening force prediction value. According to the method, the bolt pre-tightening force can be predicted with high precision and high reliability, and the technical problems of low prediction precision and great environmental influence are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of bolt preload prediction, and in particular to a bolt preload prediction method based on ultrasonic echo time-frequency characteristics. Background Art

[0002] In modern industry, bolted connections, as a common mechanical connection method, are widely used in various equipment and structures. The magnitude of the bolt preload has a crucial impact on the reliability, stability, and service life of the connection structure. If the preload is too low, the bolts will loosen during operation, causing vibration, displacement, or even failure in the connection, seriously impacting the normal operation of the equipment and leading to safety accidents. Excessive preload, on the other hand, subjects the bolts to excessive stress, leading to tensile deformation or even fracture, also threatening the safe and stable operation of the equipment. Precise control of bolt preload is particularly important in key industries such as automotive manufacturing and large-scale bridge construction.

[0003] Traditional methods for measuring bolt preload force mainly include the torque method, the angle method, and the strain gauge method. The torque method indirectly controls the preload force by controlling the torque applied to the bolt. Its operation is relatively simple and widely used in practical production. However, the relationship between torque and preload force is not strictly linear. It is affected by factors such as the bolt surface friction coefficient and thread accuracy, resulting in low measurement accuracy and difficulty meeting the requirements for high-precision preload force control. The angle method, based on the torque method, estimates the preload force by measuring the angle of rotation during the bolt tightening process. While this method improves measurement accuracy to a certain extent, it still involves numerous uncertainties, such as elastic deformation of the bolt and material inhomogeneities, which can lead to measurement deviations. The strain gauge method directly measures the bolt strain by attaching a strain gauge to the bolt and then calculating the preload force. This method offers high measurement accuracy, but requires attaching the strain gauge to the bolt, which is complex and costly. Furthermore, the strain gauge is susceptible to environmental factors such as temperature and humidity, resulting in unstable measurement results. This method has significant limitations in practical applications.

[0004] With the continuous development of science and technology, existing technologies have made certain progress in bolt preload prediction. For example, preload measurement methods based on ultrasonic technology have gradually emerged, which use the relationship between the propagation characteristics of ultrasound in the bolt and the preload to predict the preload. Compared with traditional methods, ultrasonic technology has the advantages of non-contact, fast detection speed, and little impact on equipment operation. However, existing preload prediction methods based on ultrasonic technology also have some shortcomings. On the one hand, the information contained in the ultrasonic echo signal is complex and diverse, and how to accurately extract the feature information related to the preload is a key issue. The current time-frequency analysis method needs to be improved in terms of the accuracy and completeness of feature extraction; on the other hand, the existing prediction models are often simple in structure, making it difficult to fully explore the complex nonlinear relationship between ultrasonic echo characteristics and preload, resulting in less than ideal prediction accuracy and unable to meet the requirements of industrial production for high-precision preload prediction.

[0005] In summary, both traditional methods and existing technologies have certain limitations in bolt preload prediction technology. Therefore, to address the above issues, this application proposes a bolt preload prediction method based on ultrasonic echo time-frequency characteristics, which solves the above technical problems. Summary of the Invention

[0006] Based on the above content, the present invention provides a method for predicting bolt preload based on ultrasonic echo time-frequency characteristics, including the following steps:

[0007] S1. Ultrasonic excitation is performed on the target bolt to obtain the ultrasonic echo signal generated by the bolt; the ultrasonic echo signal is subjected to time-frequency analysis using wavelet transform to obtain a time-frequency matrix containing information on the time variation of different frequency components as the initial time-frequency feature matrix;

[0008] S2. Based on a pre-built database of bolt material properties and the material type of the target bolt, characteristic parameters corresponding to ultrasonic echo propagation of the target bolt material are determined. Based on the characteristic parameters, the initial time-frequency characteristic matrix is ​​weighted to highlight the time-frequency characteristics related to the bolt preload, thereby obtaining a weighted time-frequency characteristic matrix.

[0009] S3, extracting time-frequency feature indicators through the weighted time-frequency feature matrix to form a target feature vector;

[0010] S4. Input the target feature vector into the trained dual-branch prediction model, output the preload energy feature value and the preload state change value, input the pre-built neural network model, and output the bolt preload prediction value.

[0011] Preferably, in S1, the target bolt is ultrasonically excited by a multimodal ultrasonic excitation device, and the multimodal ultrasonic excitation device includes a piezoelectric transducer array and an electromagnetic ultrasonic transducer. The piezoelectric transducer array emits ultrasonic signals of different frequencies and phases to generate multi-frequency composite ultrasonic excitation; the electromagnetic ultrasonic transducer induces a Lorentz force on the surface of the bolt to excite non-contact ultrasound, and by controlling the excitation timing and parameter combination of the two transducers, an ultrasonic echo signal of multimodal information is obtained.

[0012] Preferably, the excitation timing and parameter combination of the two transducers are controlled by constructing a control model, and the formula is: Where f is the ultrasonic signal frequency, Δf is the adjustment step, n z is the iteration number, E is the energy function of the difference between the time-frequency characteristics of the echo signal and the multimodal characteristics, and the formula is: is the i-th echo signal z time-frequency feature vectors, is the i-th ideal multimodal z feature vectors, For the i z feature component weights, m z is the total number of eigenvectors;

[0013] Set the time interval to T, at time The piezoelectric transducer array is driven by the piezoelectric transducer ultrasonic signal frequency f p (n z ) emits an ultrasonic signal, and after T time, the electromagnetic ultrasonic transducer is at time According to the ultrasonic signal frequency f of the electromagnetic ultrasonic transducer e (n z ) transmits ultrasonic signals and controls the excitation timing and parameter combination of the transducer.

[0014] Preferably, in S1, wavelet transform is used to perform time-frequency analysis on the ultrasonic echo signal. According to the material, shape and size characteristics of the target bolt, the amplitude change, phase offset and echo energy distribution characteristic information in the frequency band of the bolt ultrasonic echo signal are obtained through wavelet basis function. The ultrasonic echo signal is decomposed into the frequency range of the influence of the bolt's own structure and material on ultrasonic propagation by multi-scale layering. Each scale corresponds to a different frequency range. The wavelet coefficients of each scale are rearranged and combined to form a two-dimensional matrix. The rows of the matrix represent different frequency components and the columns represent different time points, which serves as the initial time-frequency feature matrix.

[0015] Preferably, the bolt material property database pre-constructed in S2 includes basic information of the bolt, ultrasonic propagation characteristic parameters under different working conditions and material microstructure information; by obtaining basic information of the bolt, including but not limited to the material and size of the bolt, ultrasonic propagation experiments are performed, and the propagation characteristics of the ultrasonic echo, including the sound speed, attenuation coefficient and scattering characteristics, the data are classified, sorted and correlated, and the processed data is structured and stored according to the material properties and ultrasonic propagation characteristics of the bolt to construct a bolt material property database.

[0016] Preferably, in S2, the initial time-frequency feature matrix is ​​weighted to obtain a weighted time-frequency feature matrix, specifically:

[0017] The initial time-frequency feature matrix is ​​M i =[m ij ] p×q , where m ij is the time-frequency eigenvalue at the i-th frequency and j-th time point, i=1,2,…,p is the row of the time-frequency matrix, corresponding to the frequency, j=1,2,…,q is the column of the time-frequency matrix, corresponding to the time, and the characteristic parameter vector is F=(f1,f2,…,f r ), where r is the number of characteristic parameters, construct the weight generation function where α k is the weight coefficient of the kth feature parameter, g ij (f k ) is the characteristic parameter f k The influence function of the time-frequency position (i, j) is: β is the adjustment influence coefficient; through the weight matrix W = [W ij ] p×q Weight the initial time-frequency feature matrix to obtain the weighted time-frequency feature matrix M w =[m ij W ij ] p×q .

[0018] Preferably, the time-frequency characteristic indicators in S3 include frequency band energy distribution, time-frequency peak change rate and phase mutation characteristics; the frequency band energy distribution is divided into frequency bands according to the bolt material and ultrasonic propagation characteristics, and the energy proportion in each frequency band is obtained; the time-frequency peak change rate, in the weighted time-frequency characteristic matrix, identifies the time-frequency peak position, calculates the time-frequency peak offset of the ultrasonic echo, and obtains the dynamic offset of the time-frequency peak in time and frequency as the time-frequency peak change rate; the phase mutation characteristic of the characteristic frequency analyzes the phase change of the frequency at different time points, monitors the position, amplitude and frequency of the phase mutation, and obtains the phase mutation characteristic; the time-frequency characteristic indicators of the frequency band energy distribution, the time-frequency peak change rate and the phase mutation characteristic are combined to form a target characteristic vector.

[0019] Preferably, the dual-branch prediction model in S4 includes a frequency-phase joint branch module and a multi-scale time-frequency energy feature branch module; the frequency-phase joint branch module obtains the coupling characteristics of the ultrasonic echo frequency and phase in the target feature vector at different time points through the frequency-phase correlation operator, and obtains the bolt preload state change value by fusing the data information of the ultrasonic echo frequency and phase; the multi-scale time-frequency energy feature branch module performs multi-scale decomposition on the ultrasonic echo signal, obtains the energy distribution characteristics in the time-frequency domain, outputs the preload energy characteristic value of the ultrasonic energy transferred under the bolt stress state, inputs the preload energy characteristic value and the bolt preload state change value output by the frequency-phase joint branch module into the pre-built neural network model, and outputs the preload prediction value.

[0020] Preferably, the frequency-phase joint branch module obtains the bolt preload state change value through the frequency-phase correlation operator, specifically:

[0021] Get the target feature vector V and extract the frequency of the i-th time point Phase Constructing frequency-phase correlation operators The formula is: in, It is the rate of change of frequency at adjacent time points. By calculating the coupling characteristic values ​​at different time points, the coupling characteristic sequence is obtained. Establish the preload force change prediction function ΔF(C), the formula is in is the weight of the coupled eigenvalue, δ is a constant bias term, and the state change value of the bolt preload is obtained.

[0022] Preferably, the neural network model is a fusion attention residual network model, which receives the preload energy eigenvalue and the preload state change value respectively through the input layer, enters the adaptive fusion layer, and performs adaptive weighted fusion on the preload energy eigenvalue and the preload state change value through the weight matrix to form a fusion feature vector, which enters the attention residual block. The feature information in the fusion feature vector that is strongly correlated with the bolt preload prediction is enhanced through the attention mechanism. The input information is directly transmitted across layers to the output end through the residual connection structure, and after deep feature extraction and conversion by multiple layers of attention residual blocks, it is input to the fully connected output layer to output the preload prediction value.

[0023] Compared with the prior art, the technical solution of this application has the following technical effects:

[0024] The present invention uses a multimodal ultrasonic excitation device to ultrasonically excite the target bolt, and combines wavelet transform to perform time-frequency analysis to obtain the initial time-frequency feature matrix. This technical solution solves the technical problem that traditional methods are difficult to accurately obtain characteristic information related to the preload force in the bolt ultrasonic echo. It comprehensively and accurately reflects the effect of the time-varying information of different frequency components of the bolt ultrasonic echo, and provides a reliable data basis for the subsequent accurate prediction of the preload force.

[0025] The present invention solves the technical problem that the information related to the bolt preload force is not prominent in the time-frequency features by determining the characteristic parameters based on a pre-constructed database of bolt material properties and performing weighted processing on the initial time-frequency characteristic matrix. It achieves the effect of strengthening the time-frequency features related to the preload force, making subsequent feature extraction more targeted and improving the quality of the prediction model input data.

[0026] The present invention solves the technical problem of incomplete feature extraction and inability to effectively characterize the bolt preload state by extracting time-frequency feature indicators such as frequency band energy distribution, time-frequency peak change rate and phase mutation characteristics from the weighted time-frequency feature matrix to form a target feature vector. It comprehensively and accurately reflects the effects of bolt preload-related characteristics, provides a more representative input for the prediction model, and improves the accuracy of the prediction.

[0027] The present invention adopts a dual-branch prediction model, namely the frequency-phase joint branch module and the multi-scale time-frequency energy feature branch module working together, and combines the fusion attention residual network model for prediction. The technical problem that the existing prediction model is difficult to mine the complex nonlinear relationship between ultrasonic echo characteristics and preload and has low prediction accuracy is solved. The bolt preload is predicted and the confidence interval of the predicted value is output to characterize the reliability of the prediction result, meeting the demand of industrial production for high-precision bolt preload prediction.

[0028] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application so that it can be implemented in accordance with the contents of the specification, and to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following is a detailed description of the preferred embodiment of the present application in conjunction with the accompanying drawings.

[0029] Based on the detailed description of the specific embodiments of the present application in conjunction with the accompanying drawings below, those skilled in the art will become more aware of the above and other objects, advantages and features of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following will briefly introduce the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings without inventive work. In all drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn according to the actual scale.

[0031] Figure 1 This is a flow chart of the bolt preload prediction method based on ultrasonic echo time-frequency characteristics of the present invention;

[0032] Figure 2 This is a structural diagram of a dual-branch prediction model and a neural network model of a bolt preload prediction method based on ultrasonic echo time-frequency characteristics of the present invention;

[0033] Figure 3 This is a graph showing the percentage of bolts with pre-tightening forces exceeding the specified range according to the present invention;

[0034] Figure 4 A comparison diagram of preload deviation fluctuations of the three methods of the present invention; Figure 5 This is the preload force deviation fluctuation diagram of the present invention. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. In the following description, specific details such as specific configurations and components are provided only to help fully understand the embodiments of the present application. Therefore, it should be clear to those skilled in the art that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present application. In addition, for clarity and brevity, the description of known functions and structures has been omitted in the embodiments.

[0036] It should be understood that references throughout this specification to "one embodiment" or "this embodiment" mean that a particular feature, structure, or characteristic associated with the embodiment is included in at least one embodiment of the present application. Therefore, the appearance of "one embodiment" or "this embodiment" throughout this specification does not necessarily refer to the same embodiment. Furthermore, these particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0037] In addition, the present application may repeat reference numerals and / or letters in different examples. This repetition is for the purpose of simplicity and clarity and does not in itself indicate the relationship between the various embodiments and / or settings discussed.

[0038] The term "and / or" in this article is only a description of the association relationship of associated objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist at the same time. The term " / and" in this article describes another type of association object relationship, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " in this article generally indicates that the previous and subsequent associated objects are in an "or" relationship.

[0039] The term "at least one" in this article is merely a description of the association relationship between associated objects, indicating that three relationships may exist. For example, at least one of A and B can mean: A exists alone, A and B exist at the same time, and B exists alone.

[0040] It should also be noted that, in this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include," "comprises," or any other variations thereof are intended to cover non-exclusive inclusion.

[0041] Example 1

[0042] This embodiment mainly describes a method for predicting bolt preload based on the time-frequency characteristics of ultrasonic echoes. Figure 1 As shown, the following steps are included:

[0043] S1. Ultrasonic excitation is performed on the target bolt to obtain the ultrasonic echo signal generated by the bolt; the ultrasonic echo signal is subjected to time-frequency analysis using wavelet transform to obtain a time-frequency matrix containing information on the time variation of different frequency components as the initial time-frequency feature matrix;

[0044] S2. Based on a pre-built database of bolt material properties and the material type of the target bolt, characteristic parameters corresponding to ultrasonic echo propagation of the target bolt material are determined. Based on the characteristic parameters, the initial time-frequency characteristic matrix is ​​weighted to highlight the time-frequency characteristics related to the bolt preload, thereby obtaining a weighted time-frequency characteristic matrix.

[0045] S3, extracting time-frequency feature indicators through the weighted time-frequency feature matrix to form a target feature vector;

[0046] S4. Input the target feature vector into the trained dual-branch prediction model, output the preload energy feature value and the preload state change value, input the pre-built neural network model, and output the bolt preload prediction value.

[0047] Furthermore, in S1, the target bolt is ultrasonically excited by a multimodal ultrasonic excitation device. The multimodal ultrasonic excitation device includes a piezoelectric transducer array and an electromagnetic ultrasonic transducer. The piezoelectric transducer array emits ultrasonic signals of different frequencies and phases to generate multi-frequency composite ultrasonic excitation; the electromagnetic ultrasonic transducer induces Lorentz force on the bolt surface to excite non-contact ultrasound. By controlling the excitation timing and parameter combination of the two transducers, ultrasonic echo signals with multimodal information are obtained.

[0048] Furthermore, the excitation timing and parameter combination of the two transducers are controlled by constructing a control model. The frequency of the ultrasonic signal emitted by the piezoelectric transducer array is given by the formula f p (n z ), the phase is The frequency of the ultrasonic signal emitted by the electromagnetic ultrasonic transducer is f e (n z ), the excitation current is I e (n z );

[0049] According to the time-frequency characteristics of the ultrasonic echo signal, n is obtained by the iterative formula z +1 iteration value, the frequency iteration formula of the ultrasonic signal emitted by the piezoelectric transducer array is: where f p is the ultrasonic signal frequency of the piezoelectric transducer array; the phase iteration formula of the piezoelectric transducer array is: The frequency iteration formula of the ultrasonic signal emitted by the electromagnetic ultrasonic transducer is: The frequency iteration formula of the ultrasonic signal emitted by the electromagnetic ultrasonic transducer is: The iterative formula for the excitation current of the electromagnetic ultrasonic transducer is: In the above formula, Δf p 、 Δfe, Δf e is the corresponding adjustment step, n z is the iteration number, F is the energy function of the difference between the time-frequency characteristics of the echo signal and the multimodal characteristics, and the formula is: is the i-th echo signal z time-frequency feature vectors, is the i-th ideal multimodal z feature vectors, For the i z feature component weights, m z is the total number of eigenvectors;

[0050] Set the time interval to T, at time The piezoelectric transducer array is driven by the piezoelectric transducer ultrasonic signal frequency f p (n z ) emits an ultrasonic signal, and after T time, the electromagnetic ultrasonic transducer is at time According to the ultrasonic signal frequency f of the electromagnetic ultrasonic transducer e (n z ) transmits ultrasonic signals and controls the excitation timing and parameter combination of the transducer.

[0051] Furthermore, wavelet transform is used in S1 to perform time-frequency analysis on the ultrasonic echo signal. According to the material, shape, and size characteristics of the target bolt, the amplitude change, phase offset, and echo energy distribution characteristic information within the frequency band of the bolt ultrasonic echo signal are obtained through the wavelet basis function. The ultrasonic echo signal is decomposed into the frequency range affected by the bolt's own structure and material on ultrasonic propagation using multi-scale layering. Each scale corresponds to a different frequency range, and the wavelet coefficients of each scale are rearranged and combined to form a two-dimensional matrix. The rows of the matrix represent different frequency components, and the columns represent different time points, which serves as the initial time-frequency feature matrix.

[0052] Furthermore, the bolt material property database pre-constructed in S2 includes basic information about the bolts, ultrasonic propagation characteristic parameters under different working conditions, and material microstructure information; by obtaining basic information about the bolts, including but not limited to the material and size of the bolts, ultrasonic propagation experiments are conducted, and the propagation characteristics of the ultrasonic echo, including sound speed, attenuation coefficient, and scattering characteristics, the data are classified, organized, and correlated. The processed data is structured and stored according to the material properties and ultrasonic propagation characteristics of the bolts to construct a bolt material property database.

[0053] Furthermore, in S2, the initial time-frequency feature matrix is ​​weighted to obtain a weighted time-frequency feature matrix, specifically:

[0054] The initial time-frequency feature matrix is ​​M i =[m ij ] p×q , where m ij is the time-frequency eigenvalue at the i-th frequency and j-th time point, i=1,2,…,p is the row of the time-frequency matrix, corresponding to the frequency, j=1,2,…,q is the column of the time-frequency matrix, corresponding to the time, and the characteristic parameter vector is F=(f1,f2,…,f r ), where r is the number of characteristic parameters, construct the weight generation function where α k is the weight coefficient of the kth feature parameter, g ij (f k ) is the characteristic parameter fk The influence function of the time-frequency position (i, j) is: β is the adjustment influence coefficient; through the weight matrix W = [W ij ] p×q Weight the initial time-frequency feature matrix to obtain the weighted time-frequency feature matrix M w =[m ij W ij ] p×q .

[0055] Furthermore, the time-frequency characteristic indicators in S3 include frequency band energy distribution, time-frequency peak change rate and phase mutation characteristics; the frequency band energy distribution is divided into frequency bands according to the bolt material and ultrasonic propagation characteristics to obtain the energy proportion in each frequency band; the time-frequency peak change rate, in the weighted time-frequency feature matrix, identifies the time-frequency peak position, calculates the time-frequency peak offset of the ultrasonic echo, and obtains the dynamic offset of the time-frequency peak in time and frequency as the time-frequency peak change rate; the phase mutation characteristics of the characteristic frequency analyze the phase changes of the frequency at different time points, monitor the position, amplitude and frequency of the phase mutation, and obtain the phase mutation characteristics; the time-frequency characteristic indicators of the frequency band energy distribution, time-frequency peak change rate and phase mutation characteristics are combined to form a target feature vector.

[0056] Further, if Figure 2 As shown in FIG, the dual-branch prediction model in S4 includes a frequency-phase joint branch module and a multi-scale time-frequency energy feature branch module; the frequency-phase joint branch module obtains the coupling characteristics of the ultrasonic echo frequency and phase at different time points in the target feature vector through the frequency-phase correlation operator, and obtains the bolt preload state change value by fusing the data information of the ultrasonic echo frequency and phase; the multi-scale time-frequency energy feature branch module performs multi-scale decomposition on the ultrasonic echo signal, obtains the energy distribution characteristics in the time-frequency domain, and outputs the preload energy characteristic value of the ultrasonic energy transferred under the bolt stress state. The preload energy characteristic value and the bolt preload state change value output by the frequency-phase joint branch module are input into the pre-built neural network model to output the preload prediction value.

[0057] Furthermore, the frequency-phase joint branch module obtains the bolt preload state change value through the frequency-phase correlation operator, specifically:

[0058] Get the target feature vector V and extract the frequency of the i-th time point Phase Constructing frequency-phase correlation operators The formula is: in, It is the rate of change of frequency at adjacent time points. By calculating the coupling characteristic values ​​at different time points, the coupling characteristic sequence is obtained. Establish the preload force change prediction function ΔF(C), the formula is in is the weight of the coupled eigenvalue, δ is a constant bias term, and the state change value of the bolt preload is obtained.

[0059] Furthermore, the neural network model is a fusion attention residual network model, which receives the preload energy eigenvalue and the preload state change value respectively through the input layer, enters the adaptive fusion layer, and adaptively weighted fuses the preload energy eigenvalue and the preload state change value through the weight matrix to form a fusion feature vector, which enters the attention residual block. The feature information in the fusion feature vector that is strongly correlated with the bolt preload prediction is enhanced through the attention mechanism. Through the residual connection structure, the input information is directly transmitted across layers to the output end. After deep feature extraction and conversion by multi-layer attention residual blocks, it is input into the fully connected output layer to output the preload prediction value.

[0060] This embodiment details how to overcome the low prediction accuracy and environmental susceptibility of traditional methods using technologies such as multimodal ultrasonic excitation, wavelet transform time-frequency analysis, material property database weighting, a dual-branch prediction model, and a fused attention residual network model. This method achieves high-precision and reliable bolt preload prediction, effectively reducing the number of bolts with preloads exceeding the specified range. It also maintains stable accuracy across diverse environments, ensuring safe and stable equipment operation.

[0061] Based on Example 1, this example describes a pre-built fusion attention residual network model, such as Figure 2 As shown, specifically:

[0062] The preload energy eigenvalue and preload state change value are the input data of the fusion attention residual network model. The preload energy eigenvalue is obtained from the multi-scale time-frequency energy feature branch module. The formula is: To express the preload energy characteristic value E p , where E s is the ultrasonic energy value at the sth scale, S is the total number of scales, ω s It is the weight coefficient of the corresponding scale. By weighted summation of energy values ​​at different scales, the energy characteristic part that has an important influence on the bolt preload is highlighted.

[0063] Preload state change value, by obtaining the target feature vector V, extracting the frequency of the i-th time point Phase Constructing frequency-phase correlation operators The formula is: in, It is the rate of change of frequency at adjacent time points. By calculating the coupling characteristic values ​​at different time points, the coupling characteristic sequence is obtained. Establish the preload force change prediction function ΔF(C), the formula is in is the weight of the coupled eigenvalue, δ is a constant bias term, and the state change value of the bolt preload is obtained;

[0064] In the adaptive fusion layer, the weight matrix is ​​dynamically adjusted according to the preload energy eigenvalue and preload state change value of the input data, and the preload energy eigenvalue and preload state change value are adaptively weighted fused to form a fusion feature vector;

[0065] In the fusion attention residual network model, after the fusion feature vector enters the attention residual block, the attention mechanism analyzes the correlation between the features of each dimension of the fusion feature vector and the bolt preload prediction, and constructs the attention weight matrix W a To focus on key features, the formula is Among them F f (k a ) is the kth fusion feature vector a dimensional elements, L a is the total number of dimensions of the fused feature vector, γ a is the adjustment coefficient, g a It is a nonlinear mapping function that highlights the feature information that is highly correlated with the bolt preload prediction and strengthens the feature representation;

[0066] The residual connection structure allows input information to be directly transmitted across layers to the output. After being processed layer by layer by multiple attention residual blocks, each attention residual block continuously mines the potential information in the fused feature vector, gradually refining and strengthening the features.

[0067] After the deep feature extraction and conversion of the multi-layer attention residual block, the data enters the fully connected output layer for prediction value mapping. The fully connected output layer maps the high-dimensional feature vector output by the attention residual block to the bolt preload prediction value. In the fully connected output layer, the weight matrix W o (d) and the bias vector b o To map, the input feature vector is F f (d), the dimension is D, and the calculation formula of the predicted value P is The accuracy of the mapping is optimized through training with a large amount of sample data. The back propagation algorithm is used to continuously adjust the values ​​of the weight matrix and bias vector according to the error between the predicted value and the actual preload value. The mean square error loss function is used for continuous iterative training, so that the weight matrix and bias vector can better capture the complex relationship between the eigenvector and the preload, accurately map the eigenvector to the bolt preload prediction value, and achieve accurate prediction of the bolt preload.

[0068] This example describes how to adaptively fuse the preload energy eigenvalues ​​and preload state change values ​​output by the dual-branch prediction module using a fused attention residual network model. The attention mechanism focuses on key features and strengthens representation, while residual connections prevent vanishing gradients. Multi-layer blocks deeply extract features. The fully connected output layer accurately maps the predicted values, thereby deeply exploring the complex relationship between ultrasonic echoes and preload, significantly improving the accuracy and stability of bolt preload predictions and ensuring safe equipment operation.

[0069] Based on Example 1, this example describes a comparative verification of a bolt preload prediction method based on ultrasonic echo time-frequency characteristics, specifically:

[0070] Bolts from wind turbine towers were selected for experimental comparison. A batch of wind turbines with a single unit capacity of 5MW and a tower height of 100 meters used a large number of high-strength bolts with a specification of M30 and a material of 42CrMo for connecting the tower sections. The pre-tightening force of the bolts needs to be strictly controlled between 400kN and 450kN. Insufficient pre-tightening force will lead to increased tower shaking and a greatly increased risk of bolt loosening; excessive pre-tightening force will cause plastic deformation or even fracture of the bolts, seriously threatening the safety of the wind turbine.

[0071] Existing detection methods mainly use the torque method and the strain gauge method. In this experiment, 200 bolts were randomly selected for pre-tightening force control using the torque method, and a high-precision torque wrench was used to operate according to the specified torque. After pre-tightening, the actual pre-tightening force was measured using a professional bolt pre-tightening force measuring instrument. The results show that the pre-tightening force of the torque method is very discrete. The minimum actual pre-tightening force is 362.5kN, the maximum value reaches 487.8kN, the average pre-tightening force is 421.3kN, and the standard deviation is about 28.6kN. Among these 200 bolts, the pre-tightening force of 63 bolts exceeded the specified range, accounting for 31.5%.

[0072] As for the strain gauge method, high-precision strain gauges were pasted on another 200 bolts, and the preload force was calculated by measuring the bolt strain. Although this method has high accuracy in theory, it is cumbersome to operate and is greatly affected by environmental factors. The measurement results show that the minimum preload force is 375.6kN, the maximum is 468.9kN, the average preload force is 425.8kN, and the standard deviation is about 23.4kN. However, due to the complex on-site environment, such as temperature changes and electromagnetic interference, 27 strain gauges had data anomalies, resulting in the inability to accurately measure the preload force of the corresponding bolts, accounting for 13.5%. Even in the effective measurement data, the preload force of 45 bolts still exceeded the specified range, accounting for 22.5%.

[0073] The bolt preload prediction method based on ultrasonic echo time-frequency characteristics of this application was also tested on the same batch of bolts. Using a multimodal ultrasonic excitation device, the piezoelectric transducer array transmits ultrasonic signals in a combination of multiple frequencies (1.2MHz, 2.3MHz, 3.1MHz, 4.5MHz, 5.7MHz) and phase differences (15°, 45°, 75°, 105°, 135°), and the electromagnetic ultrasonic transducer works together to excite non-contact ultrasound. Each bolt was subjected to 15 ultrasonic excitations with different parameter combinations, and a total of 3,000 sets of ultrasonic echo signals were collected.

[0074] Based on the material, shape, and size characteristics of the bolts, the ultrasonic echo signal is decomposed into 12 scales using wavelet transform for time-frequency analysis, constructing an initial time-frequency feature matrix. Combined with a pre-established database of bolt material properties, this initial time-frequency feature matrix is ​​weighted to extract key indicators such as frequency band energy distribution, time-frequency peak change rate, and phase mutation characteristics, forming a target feature vector. This vector is then input into a dual-branch prediction model and a fusion attention residual network model to obtain the preload prediction value and confidence interval.

[0075] Measured on 200 bolts, the actual preload force ranged from a minimum of 402.1 kN to a maximum of 447.9 kN, with an average of 423.5 kN and a standard deviation of approximately 8.5 kN. Only 11 bolts, or 5.5%, exceeded the specified preload force. Furthermore, the model's predicted values ​​were within the 95% confidence interval, generally within ±10 kN, demonstrating high reliability.

[0076] To directly compare the effects of the three methods, Figure 4 As shown in the figure, it can be clearly seen that the torque method has the highest proportion of bolts that are out of range, reaching 31.5%; the strain gauge method is second, at 22.5%; and the method of this application performs best, at only 5.5%.

[0077] In addition, stability tests were conducted under different environmental conditions, simulating high temperature (45°C), low temperature (-30°C), high humidity (90% RH) and strong electromagnetic interference environments commonly found in wind farms. In each environment, 20 bolts were tested using three different methods, such as Figure 5 As shown in the figure, the torque method and the strain gauge method are significantly affected by the environment, and the preload deviation fluctuates greatly; the method of the present application is less affected by the environment and can maintain a relatively stable prediction accuracy in different environments, with the standard deviation fluctuation range within ±2kN.

[0078] This embodiment describes in detail the actual application of wind turbine tower bolt preload detection and the comparison of a large amount of experimental data. It fully proves that the bolt preload prediction method based on ultrasonic echo time-frequency characteristics far exceeds the traditional torque method and strain gauge method in terms of accuracy and stability, can effectively ensure the reliability of wind turbine tower bolt connections, and provide strong technical support for the safe and stable operation of wind farms.

[0079] The above are only preferred embodiments of the present invention, which do not limit the scope of protection of the present invention. For those skilled in the art, the present invention can be modified and varied in various ways. Any changes, modifications, replacements, integrations and parameter changes to these embodiments through conventional substitutions or that can achieve the same functions without departing from the principles and spirit of the present invention fall within the scope of protection of the present invention.

Claims

1. Bolt preload prediction method based on ultrasonic echo time-frequency characteristics, characterized by: include: S1. Ultrasonic excitation is performed on the target bolt to obtain the ultrasonic echo signal generated by the bolt; the ultrasonic echo signal is subjected to time-frequency analysis using wavelet transform to obtain a time-frequency matrix containing information on the time variation of different frequency components as the initial time-frequency feature matrix; S2. Based on a pre-built database of bolt material properties and the material type of the target bolt, characteristic parameters corresponding to ultrasonic echo propagation of the target bolt material are determined. Based on the characteristic parameters, the initial time-frequency characteristic matrix is ​​weighted to highlight the time-frequency characteristics related to the bolt preload, thereby obtaining a weighted time-frequency characteristic matrix. S3, extracting time-frequency feature indicators through the weighted time-frequency feature matrix to form a target feature vector; S4. Input the target feature vector into the trained dual-branch prediction model, output the preload energy feature value and the preload state change value, input the pre-built neural network model, and output the bolt preload prediction value.

2. The bolt preload prediction method based on ultrasonic echo time-frequency characteristics according to claim 1 is characterized in that: In S1, the target bolt is ultrasonically excited by a multimodal ultrasonic excitation device, which includes a piezoelectric transducer array and an electromagnetic ultrasonic transducer. The piezoelectric transducer array emits ultrasonic signals of different frequencies and phases to generate multi-frequency composite ultrasonic excitation; the electromagnetic ultrasonic transducer induces a Lorentz force on the surface of the bolt to excite non-contact ultrasound. By controlling the excitation timing and parameter combination of the two transducers, an ultrasonic echo signal with multimodal information is obtained.

3. The bolt preload prediction method based on ultrasonic echo time-frequency characteristics according to claim 2 is characterized in that: The excitation timing and parameter combination of the two transducers are controlled by constructing a control model, and the formula is: Where f is the ultrasonic signal frequency, Δf is the adjustment step, n z is the iteration number, E is the energy function of the difference between the time-frequency characteristics of the echo signal and the multimodal characteristics, and the formula is: is the i-th echo signal z time-frequency feature vectors, is the i-th ideal multimodal z feature vectors, For the i z feature component weights, m z is the total number of eigenvectors; Set the time interval to T, at time The piezoelectric transducer array is driven by the piezoelectric transducer ultrasonic signal frequency f p (n z ) emits an ultrasonic signal, and after T time, the electromagnetic ultrasonic transducer is at time According to the ultrasonic signal frequency f of the electromagnetic ultrasonic transducer e (n z ) transmits ultrasonic signals and controls the excitation timing and parameter combination of the transducer.

4. The bolt preload prediction method based on ultrasonic echo time-frequency characteristics according to claim 2 is characterized in that: In S1, wavelet transform is used to perform time-frequency analysis on the ultrasonic echo signal. According to the material, shape, and size characteristics of the target bolt, the amplitude change, phase offset, and echo energy distribution characteristic information within the frequency band of the bolt ultrasonic echo signal are obtained through wavelet basis functions. Multi-scale layering is used to decompose the ultrasonic echo signal according to the frequency range of the influence of the bolt's own structure and material on ultrasonic propagation. Each scale corresponds to a different frequency range. The wavelet coefficients of each scale are rearranged and combined to form a two-dimensional matrix. The rows of the matrix represent different frequency components, and the columns represent different time points. This matrix serves as the initial time-frequency feature matrix.

5. The bolt preload prediction method based on ultrasonic echo time-frequency characteristics according to claim 1 is characterized in that: The bolt material property database pre-constructed in S2 includes basic information of the bolts, ultrasonic propagation characteristic parameters under different working conditions, and material microstructure information; by obtaining basic information of the bolts, including but not limited to the material and size of the bolts, ultrasonic propagation experiments are performed, and the propagation characteristics of the ultrasonic echo, including sound speed, attenuation coefficient and scattering characteristics, the data are classified, sorted and correlated, and the processed data is structured and stored according to the material properties and ultrasonic propagation characteristics of the bolts to construct a bolt material property database.

6. The bolt preload prediction method based on ultrasonic echo time-frequency characteristics according to claim 1 is characterized in that: In S2, the initial time-frequency feature matrix is ​​weighted to obtain a weighted time-frequency feature matrix, specifically: The initial time-frequency feature matrix is ​​M i =[m ij ] p×q , where m ij is the time-frequency eigenvalue at the i-th frequency and j-th time point, i=1,2,…,p is the row of the time-frequency matrix, corresponding to the frequency, j=1,2,…,q is the column of the time-frequency matrix, corresponding to the time, and the characteristic parameter vector is F=(f1,f2,…,f r ), where r is the number of characteristic parameters, construct the weight generation function where α k is the weight coefficient of the kth feature parameter, g ij (f k ) is the characteristic parameter f k The influence function of the time-frequency position (i, j) is: β is the adjustment influence coefficient; through the weight matrix W = [W ij ] p×q Weight the initial time-frequency feature matrix to obtain the weighted time-frequency feature matrix M w =[m ij W ij ] p×q .

7. The bolt preload prediction method based on ultrasonic echo time-frequency characteristics according to claim 1 is characterized in that: The time-frequency characteristic indicators in S3 include frequency band energy distribution, time-frequency peak change rate and phase mutation characteristics; the frequency band energy distribution is divided into frequency bands according to the bolt material and ultrasonic propagation characteristics, and the energy proportion in each frequency band is obtained; the time-frequency peak change rate, in the weighted time-frequency characteristic matrix, identifies the time-frequency peak position, calculates the time-frequency peak offset of the ultrasonic echo, and obtains the dynamic offset of the time-frequency peak in time and frequency as the time-frequency peak change rate; the phase mutation characteristic of the characteristic frequency analyzes the phase change of the frequency at different time points, monitors the position, amplitude and frequency of the phase mutation, and obtains the phase mutation characteristic; the time-frequency characteristic indicators of the frequency band energy distribution, the time-frequency peak change rate and the phase mutation characteristic are combined to form a target characteristic vector.

8. The bolt preload prediction method based on ultrasonic echo time-frequency characteristics according to claim 1 is characterized in that: The dual-branch prediction model in S4 includes a frequency-phase joint branch module and a multi-scale time-frequency energy feature branch module; The frequency-phase joint branch module obtains the coupling characteristics of the ultrasonic echo frequency and phase at different time points in the target feature vector through the frequency-phase correlation operator, and obtains the bolt preload state change value by fusing the data information of the ultrasonic echo frequency and phase; the multi-scale time-frequency energy feature branch module performs multi-scale decomposition on the ultrasonic echo signal, obtains the energy distribution characteristics in the time-frequency domain, and outputs the preload energy characteristic value of the ultrasonic energy transferred under the bolt stress state. The preload energy characteristic value and the bolt preload state change value output by the frequency-phase joint branch module are input into the pre-built neural network model to output the preload prediction value.

9. The bolt preload prediction method based on ultrasonic echo time-frequency characteristics according to claim 8, characterized in that: The frequency-phase joint branch module obtains the bolt preload state change value through the frequency-phase correlation operator, specifically: Get the target feature vector V and extract the frequency of the i-th time point Phase Constructing frequency-phase correlation operators The formula is: in, It is the rate of change of frequency at adjacent time points. By calculating the coupling characteristic values ​​at different time points, the coupling characteristic sequence is obtained. Establish the preload force change prediction function ΔF(C), the formula is in is the weight of the coupled eigenvalue, δ is a constant bias term, and the state change value of the bolt preload is obtained.

10. The bolt preload prediction method based on ultrasonic echo time-frequency characteristics according to claim 8, characterized in that: The neural network model is a fusion attention residual network model, which receives the preload energy eigenvalue and the preload state change value respectively through the input layer, enters the adaptive fusion layer, and performs adaptive weighted fusion on the preload energy eigenvalue and the preload state change value through the weight matrix to form a fusion feature vector, which enters the attention residual block. The feature information in the fusion feature vector that is strongly correlated with the bolt preload prediction is enhanced through the attention mechanism. Through the residual connection structure, the input information is directly transmitted across layers to the output end. After deep feature extraction and conversion by multiple layers of attention residual blocks, it is input into the fully connected output layer to output the preload prediction value.

Images