Bolt posture detection method

Abstract

According to the bolt posture detection method provided by the invention, the accuracy and integrity of data are ensured by combining the image of the bolt and acquiring the point cloud data of the bolt; images or point cloud data are decomposed into a wavelet domain, a noisy high-frequency sub-band is processed, noise is separated from effective signals of the bolt, the influence of the noise is suppressed, and accurate detection of the posture of the bolt is facilitated; according to the effective signal of the bolt, the optimal fractional order domain with the most concentrated effective signal energy of the bolt is determined, noise is effectively suppressed, the signal-to-noise ratio of the signal is improved, and the anti-interference capability during posture detection of the telegraph pole bolt is improved; the interference of the effective signal of the bolt is suppressed through filtering, and the filtered bolt signal is converted into the time domain signal, so that the weak signal after bolt enhancement is easier to detect, and the accuracy of the bolt attitude detection result is improved.

Description

Technical Field

[0001] The present invention relates to the technical fields of signal processing and electromagnetic compatibility, and more specifically to a method for detecting the posture of bolts. Background Art

[0002] With the increasing demand for industrial and residential electricity, the requirements for power transmission are also gradually increasing. Therefore, the correct installation and posture accuracy of bolts on utility poles pose extremely high requirements for the safe operation of power equipment. However, in order to reduce the demand for manual labor in a strong electricity environment to ensure personal safety, the use of traditional methods for detecting bolt postures still has problems such as low accuracy, weak anti-interference ability, and weak adaptability to complex environments, especially in an environment with extremely strong electromagnetic interference.

[0003] Although existing methods based on digital filtering and data fusion algorithms have removed some high-frequency noise and interference signals to a certain extent, the problem of ignoring weak signals in a strong electromagnetic field is still difficult to solve. Therefore, it is urgent to solve the detection defects under electromagnetic interference, which is of great significance for improving the safety of power transmission and the accuracy and stability of bolt posture detection. Summary of the Invention

[0004] To solve the problems of weak detection ability and weak anti-interference ability of weak signals of bolt postures in a complex electromagnetic field environment, the present invention proposes a method for detecting bolt postures.

[0005] In order to achieve the above technical effects, the technical solution of the present invention is as follows:

[0006] A method for detecting bolt postures, comprising the following steps:

[0007] Collect images of bolts and obtain point cloud data of bolts, where the point cloud data includes geometric shapes and spatial position information of bolts;

[0008] After decomposing the image or point cloud data into the wavelet domain, process the image or point cloud data containing high-frequency noise, separate the high-frequency noise from the effective image or point cloud data of the bolt, and obtain the effective signal of the bolt;

[0009] According to the effective signal of the bolt from which the noise has been separated, determine the optimal fractional order domain where the energy of the effective signal of the bolt is most concentrated;

[0010] Filter the effective signal of the bolt in the optimal fractional order domain to obtain a fractional order domain signal;

[0011] Convert the filtered fractional order domain signal back to a time domain signal;

[0012] Analyze the bolt position according to the time-domain signal to obtain the detection result of the bolt position.

[0013] Compared with the prior art, the beneficial effects of the technical solution of the present invention are as follows:

[0014] The present invention combines dual data acquisition of three-dimensional point cloud spatial positioning and two-dimensional image features to ensure the accuracy and integrity of the data; decomposes the image or point cloud data into the wavelet domain, processes the noisy high-frequency subbands, separates the noise from the effective signal of the bolt, suppresses the influence of the noise, and is conducive to accurately detecting the bolt position; determines the optimal fractional-order domain where the effective signal energy of the bolt is most concentrated according to the effective signal of the bolt; suppresses interference by filtering the effective signal of the bolt, converts the filtered bolt signal into a time-domain signal, so that the enhanced weak signal of the bolt is more easily detected, and improves the accuracy of the bolt position detection result. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a flowchart of the bolt position detection method shown in the embodiment of the present invention.

[0016] Figure 2 It is different wavelet bases in the wavelet domain shown in the embodiment of the present invention.

[0017] Figure 3 It is a reconstructed noise comparison diagram shown in the embodiment of the present invention.

[0018] Figure 4 It is a comparison diagram of detail coefficients and approximation coefficients at different levels after the signal shown in the embodiment of the present invention undergoes wavelet transform.

[0019] Figure 5 It is a signal sampling example diagram of the fractional-order domain shown in the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] Here, the exemplary embodiments will be described in detail, and the examples are shown in the drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0021] The terms used in this invention are for the purpose of describing specific embodiments only and are not intended to limit the invention. The singular forms "a", "said", and "the" used in this invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0022] It should be understood that although the terms first, second, third, etc. may be used in this invention to describe various information, such information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of this invention, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "when" or "while" or "in response to a determination".

[0023] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0024] Embodiment 1

[0025] As Figure 1 shown, it is a flowchart of the bolt pose detection method of this embodiment.

[0026] In a bolt pose detection method proposed in this embodiment, the following steps are included:

[0027] Collect the image of the bolt and obtain the point cloud data of the bolt, where the point cloud data includes the geometric shape and spatial position information of the bolt;

[0028] After decomposing the image or point cloud data into the wavelet domain, process the image or point cloud data with high-frequency noise, separate the high-frequency noise from the effective image or point cloud data of the bolt, and obtain the effective signal of the bolt;

[0029] According to the effective signal of the bolt from which the noise has been separated, determine the optimal fractional-order domain where the energy of the effective signal of the bolt is most concentrated;

[0030] Perform filtering processing on the effective signal of the bolt in the optimal fractional-order domain to obtain the fractional-order domain signal;

[0031] Convert the filtered fractional-order domain signal back to the time-domain signal;

[0032] Analyze the bolt pose based on the time-domain signal to obtain the detection result of the bolt pose.

[0033] In this embodiment, by combining the image of the bolt and the acquisition of the point cloud data of the bolt, the accuracy and integrity of the data are ensured; the image or point cloud data is decomposed into the wavelet domain, the noisy high-frequency subbands are processed, the noise is separated from the effective signal of the bolt, the influence of the noise is suppressed, which is conducive to accurately detecting the posture of the bolt; according to the effective signal of the bolt, the optimal fractional-order domain where the effective signal energy of the bolt is most concentrated is determined, effectively suppressing the noise, improving the signal-to-noise ratio of the signal, and enhancing the anti-interference ability when detecting the posture of the pole bolt; the effective signal of the bolt is filtered to suppress interference, and the filtered bolt signal is converted into a time-domain signal, so that the enhanced weak signal of the bolt is more easily detected, improving the accuracy of the bolt posture detection result.

[0034] Embodiment 2

[0035] Specifically, on the basis of Embodiment 1, specific implementation examples are combined to illustrate the solution, further reflecting the technical effects of the solution. Specifically:

[0036] A high-precision vision sensor and a laser displacement sensor are used to comprehensively collect the images of the pole bolts. To ensure the accuracy and integrity of the data, the vision sensor needs to take pictures from multiple angles, and the laser sensor performs three-dimensional scanning to obtain the geometric shape and spatial position information of the bolts.

[0037] In an alternative embodiment, the decomposing the image or point cloud data into the wavelet domain includes the following steps:

[0038] Let the original bolt signal be f(m)), the original bolt signal is convolved with a low-pass filter and a high-pass filter to obtain low-frequency coefficients and high-frequency coefficients; the expression is:

[0039]

[0040] Among them, f(m) represents the original bolt signal, m represents the position index of the original signal, n represents the position index of the wavelet coefficient, cA1(n) represents the low-frequency coefficient, cD1(n) represents the high-frequency coefficient, h represents the low-pass filter, and g represents the high-pass filter.

[0041] In an alternative embodiment, the processing of the image or point cloud data with high-frequency noise to separate the high-frequency noise from the effective image or point cloud data of the bolt includes the following steps:

[0042] The low-frequency coefficients and high-frequency coefficients are respectively upsampled to obtain the upsampled low-frequency coefficients and upsampled high-frequency coefficients, and the upsampled coefficients are respectively convolved with the conjugate filters of the low-pass filter and the high-pass filter and then added to obtain the reconstructed signal of the bolt; the expression is:

[0043]

[0044] Among them, f(n) represents the reconstructed signal of the bolt, m represents the position index of the wavelet coefficient, and n represents the position index of the reconstructed signal. represents the conjugate filter of the low-pass filter. represents the conjugate filter of the high-pass filter. represents the sampled low-frequency coefficient. represents the sampled high-frequency coefficient.

[0045] In an alternative embodiment, the mean square error is calculated based on the reconstructed signal of the bolt and the original signal of the bolt; its expression is:

[0046]

[0047] Among them, MSE represents the mean square error, N represents the number of signal samples, x(i) represents the i-th data point of the original bolt signal, represents the i-th data point of the bolt reconstructed signal, and i represents the i-th data point.

[0048] In an alternative embodiment, the mean square errors are sorted from small to large to obtain the minimum mean square error, the optimal decomposition level is obtained based on the minimum mean square error, and the effective signal of the bolt is output according to the optimal decomposition level.

[0049] Furthermore, the bolt data obtained by the sensor in this embodiment mainly includes rich details such as the bolt contour and texture. The db4 wavelet basis of the Daubechies wavelet family is selected in this embodiment. Different wavelet bases in the wavelet domain are as Figure 2 shown. It can effectively capture the mutation part of the signal, is beneficial to retaining the bolt edge and fine texture, and is convenient for subsequent pose recognition. According to the sampling frequency and noise characteristics of the sensor data, the mean square error between the reconstructed data and the original data at different decomposition levels is calculated, the accuracy of bolt feature extraction after processing at different levels is compared, the optimal decomposition level is determined, and the corresponding effective bolt signal is output according to the optimal decomposition level. In this embodiment, the Mallat algorithm is used to decompose the image or point cloud data collected by the sensor. First, the original signal is assumed. The original signal is convolved with the low-pass filter and the high-pass filter to obtain the low-frequency coefficient and the high-frequency coefficient. The low-frequency coefficient of the first layer is used as the input, and the above convolution and downsampling operations are repeated to obtain the low-frequency coefficient and the high-frequency coefficient of the second layer; the low-frequency coefficient of the second layer is used as the input, and the above convolution and downsampling operations are repeated to obtain the low-frequency coefficient and the high-frequency coefficient of the third layer; the low-frequency coefficient of the n-th layer is used as the input, and the above convolution and downsampling operations are repeated to obtain the low-frequency coefficient and the high-frequency coefficient of the (n + 1)-th layer, so as to obtain the decomposition results of multiple layers.

[0050] In this embodiment, the Mallat algorithm mainly functions in the wavelet domain. In the transform domain, noise usually concentrates in certain high-frequency sub-bands. After decomposing the data into the wavelet domain using the Mallat algorithm, a noise comparison graph is reconstructed by reconstructing the signal as shown in Figure 3 . The noisy high-frequency sub-bands are processed to separate the noise from the effective signal of the bolt and suppress the influence of noise. For example, in a complex electromagnetic environment, the bolt images collected may be accompanied by a large amount of noise. This step can remove the high-frequency noise, improve the image quality, and facilitate the accurate detection of the bolt posture. For a complex environment (non-linear noise), in the wavelet domain, the signal can be analyzed at different scales. For the signal of the pole bolt, the original bolt posture can still be retained on the premise of reducing noise, which is convenient for the secondary processing of the signal.

[0051] In this embodiment, in the wavelet domain, first, the original signal collected by the sensor is decomposed at multiple scales, and the signal is decomposed into sub-bands of different frequencies. By selecting an appropriate wavelet basis function, the noise and useful information in the signal can be effectively separated. Among them, noise usually appears as high-frequency components, while the characteristic signals of the bolt are concentrated in the low-frequency and middle-frequency parts. By performing adaptive threshold processing on the high-frequency sub-bands, the noise can be significantly suppressed while retaining the key features of the signal. After reconstructing the signal through the inverse wavelet transform, most of the high-frequency noise is filtered out, and the signal-to-noise ratio of the signal is significantly improved, laying a foundation for subsequent processing. The comparison graph of the detail coefficients and approximation coefficients at different levels after the signal undergoes wavelet transform is shown in Figure 4 .

[0052] In an alternative embodiment, obtaining the fractional order according to the effective signal of the bolt from which the noise has been separated includes the following steps:

[0053] When the fluctuation of the chirp signal exceeds the preset fluctuation threshold, the fractional order is obtained according to the chirp slope and sampling frequency of the signal; its expression is:

[0054]

[0055] where p represents the fractional order of energy focusing, μ represents the chirp slope, and f s represents the sampling frequency;

[0056] When the fluctuation of the chirp signal does not exceed the preset fluctuation threshold, the fractional order is obtained according to the time width and bandwidth of the signal; its expression is:

[0057]

[0058] where p represents the fractional order of energy focusing, B represents the signal bandwidth, and T represents the signal time width.

[0059] In an alternative embodiment, the fractional order is obtained by denoising the signal-to-noise ratio to obtain the fractional order when there is no noise; its expression is:

[0060] p = p0 + k × SNR

[0061] Where p represents the initially determined fractional order when there is no noise, p0 represents the initially determined fractional order, k represents an adjustment coefficient determined according to the specific application scenario, and SNR represents the signal-to-noise ratio.

[0062] Exemplarily, in practical applications, the presence of noise may affect the selection of the fractional order. Therefore, the fractional order can be dynamically adjusted by the signal-to-noise ratio (SNR).

[0063] In this embodiment, the fractional order is dynamically adjusted by the signal-to-noise ratio (SNR), enabling the denoising parameters to be optimized according to the real-time noise, and enhancing the robustness of the algorithm in complex noise environments. Compared with the fixed parameter method, this dynamic mechanism can maintain stable signal processing performance in different signal-to-noise ratio scenarios, thereby improving the efficiency and accuracy of signal processing.

[0064] In an alternative embodiment, determining the optimal fractional order domain where the effective signal energy of the bolt is most concentrated includes the following steps:

[0065] Converting the effective signal in the time domain to the fractional Fourier domain to obtain the optimal fractional order domain; wherein, the fractional Fourier domain is obtained according to the fractional Fourier function; the expression of the fractional Fourier function is:

[0066]

[0067] Where X p (u) represents the fractional Fourier function, u represents the transformed independent variable, K p (t, u) represents the kernel function, t represents time, and x(t) represents the signal.

[0068] Furthermore, the selection of the fractional order is crucial for the analysis effect of the bolt effective signal. The optimal fractional order can be automatically selected according to the characteristics of the bolt effective signal or the analysis purpose. Taking the energy concentration degree of the bolt effective signal in the fractional Fourier domain as the optimization target, the order p value is dynamically adjusted in real time to make the transformation result reach the best.

[0069] Even further, the form of the kernel function in the fractional Fourier function is related to the order, and its expression is:

[0070] K p (t, u) = A p exp[jπ(at 2 - 2btu + cu 2 )]

[0071] wherein, A p represents a normalization factor and p≠0, ±2, ±4, …, j represents the imaginary part in a complex number, p represents the fractional order, and a, b, and c represent quadratic coefficients.

[0072] Exemplarily, when programming, the NumPy library of Python can be used for matrix operations to construct a discrete fractional Fourier transform (discrete fractional Fourier transform) matrix, which is multiplied by an input signal vector to complete signal transformation.

[0073] In this embodiment, it can enhance signal features and improve the signal-to-noise ratio for detecting the posture of pole bolts and can focus the signal on a specific fractional domain region. An example diagram of signal sampling in the fractional domain is as Figure 5 shown. At the same time, the signal and noise are separated in the fractional domain. By selecting a suitable fractional domain for filtering and other processing, noise is effectively suppressed, and the signal-to-noise ratio of the signal is improved, so that the detection result is more accurate and reliable. Since the fractional domain is concentrated in a certain region while noise is usually distributed in a wider region, most noise interference can be removed by setting a suitable threshold or filter in the fractional domain, and the anti-interference ability during detection is improved.

[0074] In an alternative embodiment, filtering the effective bolt signal in the optimal fractional domain to obtain a fractional domain signal includes the following steps:

[0075] Calculating a gain matrix for adjusting parameter estimates; its expression is:

[0076]

[0077] wherein, K(n) represents the gain matrix, P(n - 1) represents the covariance matrix at the previous moment, x(n) represents the input vector, λ represents the forgetting factor, T represents the transpose, and n represents the nth step;

[0078] Multiplying the error between the observed output and the predicted output by the gain matrix and then adding it to the parameter estimate at the previous moment to obtain a new parameter estimate; its expression is:

[0079]

[0080] where represents the parameter estimate at the nth step, represents the parameter estimate at the previous moment, y(n) represents the observed output, and λ represents the forgetting factor (usually 0 < λ ≤ 1 to control the weight of historical data);

[0081] Calculating a covariance matrix to cope with the change of the signal in a strong electromagnetic field; its expression is:

[0082]

[0083] Among them, P(n) represents the covariance matrix, and P(n - 1) represents the covariance matrix at the previous moment.

[0084] Furthermore, according to the specific application scenario and requirements for filter performance: If the requirement for computational complexity is low and the requirement for real-time performance is high, the LMS algorithm can be selected; if the requirement for filtering accuracy is high and the signal changes are relatively complex, the RLS algorithm can be selected.

[0085] Exemplarily, in this embodiment, the RLS algorithm is used to solve the continuous change and influence of signals in a complex environment by using its real-time performance and forgetting factor. Although its computational complexity is high, its convergence speed is faster than that of traditional stochastic gradient algorithms such as LMS. The RLS algorithm can automatically adjust the parameters of the filter according to the statistical characteristics of the input signal. When there is periodic interference, such as power frequency interference in the power system, the adaptive filter can adjust the weights to form a notch in the frequency response of the filter at the interference frequency, effectively suppressing the periodic interference. The RLS with a forgetting factor used in the present invention can enhance the tracking ability of the time-varying system, continuously adjust the filtering parameters, make the output signal closer to the true signal, and has a very important role in the posture detection of the pole bolt under the condition of electromagnetic field interference.

[0086] In an alternative embodiment, the conversion of the filtered fractional-domain signal back to the time-domain signal includes the following steps: The inverse discrete fractional Fourier transform matrix is obtained by transposing the matrix and taking the complex conjugate of each element; its expression is:

[0087]

[0088] where F -α represents the inverse discrete fractional Fourier transform matrix, F α represents the original matrix, and H represents the conjugate transpose operation.

[0089] Exemplarily, after denoising the effective signal of the bolt, the inverse transform corresponding to the forward transform is used to convert the filtered fractional-domain signal into the time-domain signal. The inverse transform matrix of the discrete fractional Fourier transform is multiplied by the filtered bolt signal vector, and the inverse transform is completed to convert the filtered fractional-domain signal back to the time-domain signal, which is convenient for outputting the bolt detection result.

[0090] In this embodiment, the fractional Fourier transform can perform fine time-frequency analysis on signals in the fractional domain, reveal the characteristics of weak signals at different time and frequency scales, and capture key characteristics such as the position and shape contained in the weak bolt signals. For example, in the case of weak reflection signals caused by bolt surface wear and corrosion, the fractional Fourier transform can extract weak characteristic signals related to bolt wear and corrosion from a complex background. In addition, during long-term use, the bolt may rotate by a small angle, resulting in extremely weak signal changes. The fractional Fourier transform can utilize its sensitivity to the rotation angle to effectively extract such weak rotational signal characteristics in the transform domain and accurately detect whether the bolt has rotated by a small angle, making weak signals easier to detect.

[0091] In the fractional transform domain, the energy distribution characteristics of signals can better adapt to the noise characteristics in complex environments. By initially estimating the order of the fractional Fourier transform and using an adaptive search algorithm to optimize the order, the fractional domain with the most concentrated signal energy is found. In the optimal fractional domain, the recursive least squares method based on adaptive filtering is used to further filter the signal. The RLS algorithm has the characteristics of fast convergence and low computational complexity, can dynamically adjust the filter parameters, and effectively suppress chirp-like noise and electromagnetic interference. At the same time, the weak signal is enhanced in the fractional domain. By analyzing the energy distribution characteristics of the signal, the gain of the low-energy region is adjusted to improve the visibility and detectability of the weak signal.

[0092] In this embodiment, noise suppression and signal enhancement are respectively performed in the wavelet domain and the fractional domain, making full use of the following characteristics of the two transform domains: First, the wavelet domain is good at processing non-stationary signals and multi-scale features and can effectively separate noise and useful signals; Second, the fractional domain can adapt to the frequency change characteristics in complex environments, further filter out specific types of noise, and enhance weak signals.

[0093] Through this staged transform domain processing, this embodiment can significantly improve the signal quality in complex environments and provide reliable data support for subsequent bolt pose detection.

[0094] Embodiment 3

[0095] In this embodiment, a bolt pose detection method is verified based on Embodiment 1 and Embodiment 2.

[0096] The test and verification work of this method is carried out. The test and verification work is carried out by constructing various types of signal models, including sine wave signals, linear frequency modulation signals, or more non-linear signals, and adding different intensities and types of noise interferences, including but not limited to Gaussian white noise, impulse noise, etc., and constructing a three-dimensional model for test and inspection. This method is compared and tested with traditional time-domain adaptive filtering algorithms, data fusion algorithms, and deep learning algorithms. Through multiple model simulation experiments, the superiority and reliability of this method are verified based on the test indicators.

[0097] In this embodiment, the method is tested and verified by constructing various signal models and a three-dimensional model, and this method is compared and tested with traditional time-domain adaptive filtering algorithms, data fusion algorithms, and deep learning algorithms. It is verified that this method realizes noise suppression and signal enhancement in the transform domain, and through wavelet domain noise reduction, fractional order domain filtering, and weak signal enhancement, it realizes more accurate detection of the bolt posture in a complex environment, providing reliable technical support for power inspection and infrastructure maintenance.

[0098] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, rather than limiting the implementation manners of the present invention. For those of ordinary skill in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to enumerate all the implementation manners here. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included in the protection scope of the claims of the present invention.

Claims

1. A method for detecting the posture of a bolt, characterized in that, It includes the following steps: Collect the image of the bolt and obtain the point cloud data of the bolt. The point cloud data includes the geometric shape and spatial position information of the bolt; After decomposing the image or point cloud data into the wavelet domain, process the image or point cloud data with high-frequency noise, separate the high-frequency noise from the effective image or point cloud data of the bolt, and obtain the effective signal of the bolt; According to the effective signal of the bolt from which the noise has been separated, determine the optimal fractional-order domain where the energy of the effective signal of the bolt is most concentrated; Perform filtering processing on the effective signal of the bolt in the optimal fractional-order domain to obtain the fractional-order domain signal; Convert the filtered fractional-order domain signal back to the time-domain signal; Analyze the posture of the bolt according to the time-domain signal to obtain the detection result of the bolt posture.

2. The method for detecting the posture of a bolt according to claim 1, wherein The decomposing the image or point cloud data into the wavelet domain includes the following steps: Let the original signal of the bolt be f(m). The original signal of the bolt is convolved with a low-pass filter and a high-pass filter to obtain low-frequency coefficients and high-frequency coefficients. The expression is: where f(m) represents the original signal of the bolt, m represents the position index of the original signal, n represents the position index of the wavelet coefficient, cA1(n) represents the low-frequency coefficient, cD1(n) represents the high-frequency coefficient, h represents the low-pass filter, and g represents the high-pass filter.

3. A bolt posture detection method according to claim 2, characterized in that, The processing the image or point cloud data with high-frequency noise and separating the high-frequency noise from the effective image or point cloud data of the bolt includes the following steps: Upsample the low-frequency coefficients and high-frequency coefficients respectively to obtain the upsampled low-frequency coefficients and upsampled high-frequency coefficients. Add the coefficients after upsampling convolved with the conjugate filters of the low-pass filter and the high-pass filter respectively to obtain the reconstructed signal of the bolt. The expression is: Among them, f(n) represents the reconstructed signal of the bolt, m represents the position index of the wavelet coefficient, and n represents the position index of the reconstructed signal. represents the conjugate filter of the low-pass filter. represents the conjugate filter of the high-pass filter. represents the sampled low-frequency coefficient. represents the sampled high-frequency coefficient.

4. A method for detecting the posture of a bolt according to claim 3, characterized in that, Calculate the mean square error according to the reconstructed signal of the bolt and the original signal of the bolt. The expression is: Among them, MSE represents the mean square error, N represents the number of sample points of the signal, x(i) represents the i-th data point of the original bolt signal, represents the i-th data point of the reconstructed bolt signal, and i represents the i-th data point.

5. A bolt posture detection method according to claim 4, characterized in that, Sort the mean square errors from small to large to obtain the minimum mean square error. Obtain the optimal decomposition level according to the minimum mean square error and output the effective signal of the bolt according to the optimal decomposition level.

6. A bolt posture detection method according to claim 1, characterized in that The obtaining the fractional order according to the effective signal of the bolt from which the noise has been separated includes the following steps: When the fluctuation of the chirp signal exceeds the preset fluctuation threshold, obtain the fractional order according to the chirp slope and sampling frequency of the signal. The expression is: Among them, p represents the fractional order of energy focusing, μ represents the frequency modulation slope, and f s represents the sampling frequency; When the fluctuation of the chirp signal does not exceed the preset fluctuation threshold, obtain the fractional order according to the time width and bandwidth of the signal. The expression is: where p represents the fractional order of energy focusing, B represents the signal bandwidth, and T represents the signal time width.

7. A bolt posture detection method according to claim 6, characterized in that, The fractional order is denoised according to the signal-to-noise ratio to obtain the fractional order without noise; The expression is: p = p0 + k × SNR where p represents the initially determined fractional order without noise, p0 represents the initially determined fractional order, k represents the adjustment coefficient determined according to the specific application scenario, and SNR represents the signal-to-noise ratio.

8. A method for detecting the posture of a bolt according to claim 7, characterized in that, The determining the optimal fractional-order domain where the energy of the effective signal of the bolt is most concentrated includes the following steps: Convert the effective signal in the time domain to the fractional Fourier domain to obtain the optimal fractional-order domain. Among them, the fractional Fourier domain is obtained according to the fractional Fourier function. The expression of the fractional Fourier function is: Among them, X p (u) represents the fractional Fourier function, u represents the transformed independent variable, K p (t, u) represents the kernel function, t represents time, and x(t) represents the signal.

9. A bolt posture detection method according to any one of claims 1 to 8, characterized in that, Filtering the effective bolt signal in the optimal fractional-order domain to obtain a fractional-order domain signal, which includes the following steps: Calculating a gain matrix for adjusting the parameter estimation value; its expression is: Where K(n) represents the gain matrix, P(n - 1) represents the covariance matrix at the previous moment, x(n) represents the input vector, λ represents the forgetting factor, T represents the transpose, and n represents the nth step; Multiplying the error between the observed output and the predicted output by the gain matrix, and then adding it to the parameter estimation value at the previous moment to obtain a new parameter estimation value; its expression is: Among them, represents the parameter estimation at the nth step, represents the parameter estimation at the previous moment, y(n) represents the observed output, and λ represents the forgetting factor (0 < λ ≤ 1); Calculating a covariance matrix for dealing with the change of the signal in a strong electromagnetic field; its expression is: Where P(n) represents the covariance matrix, and P(n - 1) represents the covariance matrix at the previous moment.

10. A bolt posture detection method according to claim 9, characterized in that, Converting the filtered fractional-order domain signal back to a time-domain signal, which includes the following steps: The inverse discrete fractional Fourier transform matrix is obtained by transposing the matrix and taking the complex conjugate of each element; its expression is: Among them, F -α represents the discrete inverse fractional Fourier transform matrix, F α represents the original matrix, and G represents the conjugate transpose operation.

Images