Electromagnetic ultrasonic bolt stress detection system based on space-time complementary algorithm

Abstract

The application provides an electromagnetic ultrasonic bolt stress detection system based on a space-time complementary algorithm, which comprises the following steps: collecting a multi-channel ultrasonic signal of a bolt to be measured to obtain a multi-channel echo signal group; performing time-varying attenuation compensation and adaptive entropy weight denoising on the multi-channel echo signal group to obtain a denoised echo signal group; extracting echo time domain features from the denoised echo signal group, performing segmented time-frequency feature extraction on the denoised echo signal group based on a short-time sliding window to obtain echo time-frequency features, and fusing the echo time domain features and the echo time-frequency features into ultrasonic time-frequency features; performing wave field reconstruction on the denoised echo signal group to obtain a spatial ultrasonic energy map, and extracting ultrasonic spatial features of the spatial ultrasonic energy map based on a self-attention map convolution network; performing correlation detection and space-time complementary modeling on the ultrasonic time-frequency features and the ultrasonic spatial features to obtain ultrasonic space-time joint features, and performing stress analysis on the ultrasonic space-time joint features to obtain bolt stress data.

Description

Technical Field

[0001] This invention relates to the field of machine learning technology, and in particular to an electromagnetic ultrasonic bolt stress detection system based on a spatiotemporal complementary algorithm. Background Technology

[0002] Bolts, as critical load-bearing connection components, are widely used in energy equipment, rail transportation, aerospace, and high-end manufacturing. Their stress state directly affects the stability and safety of the overall structure. In recent years, electromagnetic ultrasonic technology has shown significant advantages in the field of metal stress detection due to its non-contact excitation and reception characteristics. However, it still faces several technical bottlenecks in bolt stress detection. On the one hand, electromagnetic ultrasonic signals have low energy and high noise, resulting in insufficient signal-to-noise ratio of the echo signal. On the other hand, single-channel sampling methods cannot simultaneously reflect the non-uniformity of bolt stress distribution in space. In addition, traditional signal analysis is mostly limited to single-dimensional feature extraction in the time or frequency domain, failing to effectively characterize the multi-scale, nonlinear sound field changes caused by bolt stress, thus limiting the accuracy of bolt stress detection. Summary of the Invention

[0003] (a) Technical problems to be solved

[0004] To address the shortcomings of existing technologies, this invention provides an electromagnetic ultrasonic bolt stress detection system based on a spatiotemporal complementary algorithm. This system has the advantages of spatiotemporal complementary analysis, high signal resolution, and high measurement accuracy, and solves the problems of single analysis dimension, high ultrasonic data noise, and low detection accuracy of traditional methods.

[0005] (II) Technical Solution

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] This invention provides an electromagnetic ultrasonic bolt stress detection system based on a spatiotemporal complementary algorithm, comprising an ultrasonic acquisition module, a signal denoising module, a time-frequency feature extraction module, a spatial feature extraction module, and a stress analysis module, wherein:

[0008] The ultrasonic acquisition module is used to acquire multi-channel ultrasonic signals from the bolt under test based on a preset electromagnetic ultrasonic array, and obtain a multi-channel echo signal group.

[0009] The signal denoising module is used to perform time-varying attenuation compensation and adaptive entropy weight denoising on the multi-channel echo signal group to obtain a denoised echo signal group.

[0010] The time-frequency feature extraction module is used to extract echo time-domain features from the denoised echo signal group, perform segmented time-frequency feature extraction on the denoised echo signal group based on a short-time sliding window to obtain echo time-frequency features, and fuse the echo time-domain features and the echo time-frequency features into ultrasonic time-frequency features;

[0011] The spatial feature extraction module is used to reconstruct the wave field of the noise-reduced echo signal group based on the electromagnetic ultrasonic array and the bolt under test to obtain a spatial ultrasonic energy map, and to perform self-attention map convolution on the spatial ultrasonic energy map to obtain ultrasonic spatial features.

[0012] The stress analysis module is used to perform correlation detection and spatiotemporal complementary modeling on the ultrasonic time-frequency features and ultrasonic spatial features to obtain ultrasonic spatiotemporal joint features, and to perform stress analysis on the ultrasonic spatiotemporal joint features to obtain bolt stress data.

[0013] According to a preferred embodiment of the present invention, when the signal denoising module performs time-varying attenuation compensation and adaptive entropy weight denoising on the multi-channel echo signal group to obtain a denoised echo signal group, it includes:

[0014] The DC bias in the multi-channel echo signal group is filtered out, and the multi-channel echo signal group after the DC bias is filtered out is aligned based on a unified timestamp to obtain an aligned echo signal group.

[0015] Bandpass filtering and time-varying attenuation compensation are performed on each aligned echo signal in the aligned echo signal group to obtain a gain echo signal group.

[0016] Adaptive empirical mode decomposition is performed on the gain echo signal group to obtain a set of signal mode components;

[0017] Calculate the modal component energy entropy set corresponding to the signal modal component set, and perform joint scoring and adaptive classification on each signal modal component in the signal modal component set based on the modal component energy entropy set to obtain modal component class groups;

[0018] Based on the modal component class group, component weights are configured for each signal modal component in the signal modal component group set to obtain a component weight reassembly set, and an initial component weight vector group is constructed based on the component weight reassembly set;

[0019] Based on the initial component weight vector set, the signal mode component set is subjected to least squares filtering and signal reconstruction to obtain the noise-reduced echo signal set.

[0020] According to another preferred embodiment of the present invention, when the signal denoising module performs adaptive empirical mode decomposition on the gain echo signal group to obtain a set of signal mode components, it includes:

[0021] Initialize a white noise signal group, select the gain echo signals in the gain echo signal group one by one as the target gain echo signals, add the white noise signal group to the target gain echo signals to obtain a noisy echo signal group;

[0022] Empirical mode decomposition is performed on the noisy echo signal group, and the first eigenmode function obtained from the mode decomposition is used as the noisy mode function to form a noisy mode function group;

[0023] The noise-added mode function group is integrated and averaged to obtain the noise intrinsic mode function. The first-order residual signal between the target gain echo signal and the noise intrinsic mode function is calculated, and the signal energy ratio between the first-order residual signal and the target gain echo signal is calculated.

[0024] Replace the target gain echo signal with the first-order residual signal, and return to the step of adding the white noise signal group to the target gain echo signal to obtain the noisy echo signal group;

[0025] Until the signal energy ratio is less than the preset energy ratio threshold, all first-order residual signals obtained from multi-level iterations are collected as signal mode components into a signal mode component group, and the signal mode component groups of all target gain echo signals are collected into a signal mode component group set.

[0026] According to another preferred embodiment of the present invention, when the signal denoising module performs joint scoring and adaptive classification of each signal mode component in the signal mode component group set based on the mode component energy entropy set to obtain the mode component class group, it includes:

[0027] The modal component energy entropy in the modal component energy entropy set is selected one by one as the target component energy entropy, and the signal modal component corresponding to the target component energy entropy in the signal modal component set is taken as the target modal component;

[0028] Calculate the cross-channel correlation factor of the target modal component;

[0029] The joint scoring factor is obtained by weighted summation of the cross-channel correlation factor, the target component energy entropy, and the energy proportion of the target modal component.

[0030] The joint scoring factors of the energy entropy of all modal components are aggregated into a joint scoring factor set, and each signal modal component in the signal modal component group set is classified based on the quantile of the joint scoring factor set to obtain modal component class groups.

[0031] According to another preferred embodiment of the present invention, when the spatial feature extraction module performs wavefield reconstruction of the noise-reduced echo signal group based on the electromagnetic ultrasonic array and the bolt under test to obtain a spatial ultrasonic energy map, it includes:

[0032] A bolt mesh model is constructed based on the bolt to be tested, and the bolt mesh model is discretized into bolt pixel mesh;

[0033] Based on the positional relationship between the electromagnetic ultrasonic array and the bolt under test, the sound wave propagation time of each pixel in the bolt pixel grid is calculated and phase compensation is performed to obtain the sound wave propagation time set.

[0034] Based on the sound wave propagation time set, the echo signal amplitude set is extracted from the noise-reduced echo signal set, and the echo signal amplitude set is mapped to each pixel of the bolt pixel grid to obtain the signal amplitude grid.

[0035] The echo signal amplitudes of each pixel in the signal amplitude grid on each channel are superimposed to obtain the superimposed amplitude grid.

[0036] The coherence weights of each pixel in the superimposed amplitude grid are calculated based on the phase coherence method, and the superimposed amplitude grid is coherently weighted based on the coherence weights to obtain a weighted amplitude grid.

[0037] The amplitude energy of each grid in the weighted amplitude grid is calculated and the grid is assigned a value to obtain an ultrasonic energy map. The ultrasonic energy map is then subjected to bilinear interpolation smoothing and logarithmic dynamic compression to obtain a spatial ultrasonic energy map.

[0038] According to another preferred embodiment of the present invention, when the spatial feature extraction module performs self-attention map convolution on the spatial ultrasound energy map to obtain ultrasound spatial features, it includes:

[0039] Each pixel of the spatial ultrasonic energy map is used as a network node, and a node feature set is generated based on the amplitude energy and coordinate information of each network node.

[0040] The k-nearest neighbor algorithm is used to establish node edges for each network node in the spatial ultrasonic energy map, and the edge weight of each node edge is calculated based on the Euclidean distance of each network node using the Gaussian distance weighting algorithm.

[0041] An ultrasonic energy graph structure is constructed based on the node feature set, the network nodes, the node edges, and the edge weights.

[0042] Based on the ultrasonic energy map structure, the node feature set is positionally encoded and linearly mapped to obtain an encoded node feature set. Then, based on the neighborhood subgraph of each network node in the ultrasonic energy map structure, self-attention is calculated on the encoded node feature set to obtain a self-attention weight set.

[0043] Based on the self-attention weight set, the encoding node feature set is subjected to self-attention message aggregation and graph convolution update to obtain the self-attention space feature set;

[0044] Cross-layer feature fusion is performed on the self-attention space feature set, and feature merging and feature normalization are performed on the self-attention space features after cross-layer feature fusion to obtain ultrasound space features.

[0045] According to another preferred embodiment of the present invention, when the time-frequency feature extraction module performs segmented time-frequency feature extraction on the denoised echo signal group based on a short-time sliding window to obtain the echo time-frequency features, it includes:

[0046] The noise-reduced echo signal group is segmented based on a short-time sliding window, and a windowing operation is performed on the segmented noise-reduced echo signal group to obtain a windowed signal sequence group.

[0047] Perform continuous wavelet transform on the window signal sequence group to obtain the echo frequency domain sequence group;

[0048] Multi-scale time-frequency feature statistics are performed on the echo frequency domain sequence group to obtain a primary time-frequency feature sequence group. The multi-scale time-frequency feature statistics include statistically analyzing the energy spectrum distribution, wavelet energy spectrum entropy, instantaneous dominant frequency, scale bandwidth, and instantaneous energy of each echo frequency domain in the echo frequency domain sequence group.

[0049] Ridge smoothing is applied to the instantaneous dominant frequency in the primary time-frequency feature sequence group, and short-time smoothing and differential enhancement are applied to the wavelet energy spectral entropy in the ridge-smoothed primary time-frequency feature sequence group to obtain the enhanced time-frequency feature sequence group.

[0050] The enhanced time-frequency feature sequence group is time-aligned and its features are spliced. Then, linear discriminant analysis is performed on the spliced ​​enhanced time-frequency feature sequence group to reduce its dimensionality, thereby obtaining the echo time-frequency features.

[0051] According to another preferred embodiment of the present invention, when the stress analysis module performs correlation detection and spatiotemporal complementary modeling on the ultrasonic time-frequency features and the ultrasonic spatial features to obtain ultrasonic spatiotemporal joint features, it includes:

[0052] The ultrasonic time-frequency features and ultrasonic spatial features are respectively normalized and aligned in dimension to obtain aligned ultrasonic feature groups;

[0053] The feature correlation matrix of the aligned ultrasound feature group is calculated based on the Pearson correlation coefficient, and the aligned ultrasound feature group is weighted based on the feature correlation matrix to obtain the weighted ultrasound feature group.

[0054] The ultrasound attention weight matrix of the weighted ultrasound feature group is calculated based on the cross-modal attention mechanism, and the attention of the weighted ultrasound feature group is enhanced based on the ultrasound attention weight matrix to obtain the enhanced ultrasound feature group.

[0055] Based on a multilayer perceptron, high-dimensional feature fusion is performed on the enhanced ultrasound feature group to obtain high-dimensional spatiotemporal joint features, and then the high-dimensional spatiotemporal joint features are dimensionally compressed to obtain ultrasound spatiotemporal joint features.

[0056] According to another preferred embodiment of the present invention, when the stress analysis module performs stress analysis on the ultrasonic spatiotemporal joint features to obtain bolt stress data, it includes:

[0057] Based on orthogonal basis functions, the feature domain transformation of the ultrasonic spatiotemporal joint features is performed to obtain the transformed spatiotemporal joint features;

[0058] The transformed spatiotemporal joint features are nonlinearly activated, and the ultrasonic spatiotemporal joint features and the nonlinearly activated transformed spatiotemporal joint features are enhanced and fused based on the residual structure to obtain residual spatiotemporal joint features;

[0059] The spatiotemporal joint features of the residuals are modeled nonlinearly using a pre-trained radial basis function network to obtain spatiotemporal stress data.

[0060] The bolt temperature and bolt parameters of the bolt to be tested are obtained, and physical constraint compensation is performed on the spatiotemporal stress data based on the bolt temperature and bolt parameters to obtain bolt stress data.

[0061] This invention provides an electromagnetic ultrasonic bolt stress detection method based on a spatiotemporal complementary algorithm, comprising:

[0062] Based on a preset electromagnetic ultrasonic array, multi-channel ultrasonic signals are acquired from the bolt under test to obtain a multi-channel echo signal group.

[0063] Time-varying attenuation compensation and adaptive entropy weight denoising are performed on the multi-channel echo signal group to obtain a denoised echo signal group.

[0064] Echo time-domain features are extracted from the denoised echo signal group. Segmented time-frequency features are extracted from the denoised echo signal group based on a short-time sliding window to obtain echo time-frequency features. The echo time-domain features and the echo time-frequency features are then fused into ultrasonic time-frequency features.

[0065] Based on the electromagnetic ultrasonic array and the bolt under test, the wave field of the noise-reduced echo signal group is reconstructed to obtain a spatial ultrasonic energy map, and the spatial ultrasonic energy map is convolved with a self-attention map to obtain ultrasonic spatial features.

[0066] Correlation detection and spatiotemporal complementary modeling are performed on the ultrasonic time-frequency features and ultrasonic spatial features to obtain ultrasonic spatiotemporal joint features. Stress analysis is then performed on the ultrasonic spatiotemporal joint features to obtain bolt stress data.

[0067] (III) Beneficial Effects

[0068] Compared with existing technologies, this invention provides an electromagnetic ultrasonic bolt stress detection system based on a spatiotemporal complementary algorithm, which has the following advantages:

[0069] This electromagnetic ultrasonic bolt stress detection system based on a spatiotemporal complementary algorithm effectively suppresses amplitude attenuation caused by changes in propagation distance and temperature within the bolt through time-varying attenuation compensation. Adaptive empirical mode decomposition separates multi-scale oscillation components in both time and frequency domains, preserving the signal's structural characteristics. A joint scoring mechanism based on energy entropy and cross-channel correlation further distinguishes noise modes from useful modes, achieving dynamic noise identification. By introducing adaptive optimal estimation in the weighted reconstruction stage of least-squares filtering, denoising and reconstruction are performed based on the prior knowledge on the right side of the weight vector, improving the smoothness and real-time performance of signal reconstruction. Ultimately, this significantly enhances the signal-to-noise ratio and structural stability of the electromagnetic ultrasonic echo signal, providing high-precision data input for subsequent bolt stress detection calculations.

[0070] This electromagnetic ultrasonic bolt stress detection system based on a spatiotemporal complementary algorithm achieves spatial domain reconstruction of multi-channel ultrasonic echoes through bolt mesh modeling, acoustic wave propagation time compensation, amplitude superposition, and coherence weighting. This effectively enhances the reflection signal of real defects and suppresses various random noises and spatial artifacts. The spatial ultrasonic energy map obtained through amplitude energy mapping accurately presents the stress concentration region inside the bolt, achieving high-fidelity reconstruction of the ultrasonic field distribution. By constructing the spatial ultrasonic energy map as a graph structure with local topological relationships and using self-attention graph convolution to adaptively model the spatial dependence and local energy changes between nodes, it can simultaneously retain local sensitive features and enhance cross-regional correlation expression. This results in extracted spatial ultrasonic features with higher resolution and robustness, thereby significantly improving the accuracy and stability of bolt stress detection.

[0071] This electromagnetic ultrasonic bolt stress detection system based on a spatiotemporal complementary algorithm highlights the synergistic changes in temporal and frequency characteristics and spatial energy distribution under bolt stress through spatiotemporal complementary modeling based on correlation detection and cross-modal attention mechanisms. By utilizing a multilayer perceptron for high-dimensional fusion, it obtains ultrasonic spatiotemporal joint features with stronger characterization capabilities, overcoming the problem of insufficient modeling of single-domain features in traditional methods. By introducing orthogonal wavelet basis transform for stress analysis, it can enhance the localization characteristics of stress-sensitive frequency bands. Combining residual enhancement structures and radial basis function networks, a more stable nonlinear stress mapping model can be constructed. Finally, by combining bolt temperature and material parameters for physical compensation, the robustness and accuracy of stress detection can be significantly improved. Attached Figure Description

[0072] Figure 1 The diagram shown is a structural diagram of an electromagnetic ultrasonic bolt stress detection system based on a spatiotemporal complementary algorithm according to the present invention.

[0073] Figure 2 The flowchart shown is a method for detecting electromagnetic ultrasonic bolt stress based on a spatiotemporal complementary algorithm according to the present invention. Detailed Implementation

[0074] The following description is intended to disclose the present invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious modifications will occur to those skilled in the art. The basic principles of the invention defined in the following description can be applied to other embodiments, modifications, improvements, equivalents, and other technical solutions that do not depart from the spirit and scope of the invention.

[0075] It is understood that the term "a" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element can be one, while in another embodiment, the number of the element can be multiple, and the term "a" should not be understood as a limitation on the number.

[0076] Example 1:

[0077] Please refer to Figure 1 This invention discloses an electromagnetic ultrasonic bolt stress detection system based on a spatiotemporal complementary algorithm. The system mainly comprises an ultrasonic acquisition module, a signal noise reduction module, a time-frequency feature extraction module, a spatial feature extraction module, and a stress analysis module, wherein:

[0078] The ultrasonic acquisition module is used to acquire multi-channel ultrasonic signals from the bolt under test based on a preset electromagnetic ultrasonic array, and obtain a multi-channel echo signal group.

[0079] The electromagnetic ultrasonic array is an array structure composed of multiple electromagnetic ultrasonic transducers (EMATs) arranged in a preset spatial structure. The spatial structure of the electromagnetic ultrasonic array can be a ring array, a linear array, or a racetrack coil array. The multi-channel ultrasonic signal acquisition refers to the emission of transverse and longitudinal dual-wave or longitudinal single-wave ultrasonic waves from each electromagnetic ultrasonic transducer of the electromagnetic ultrasonic array according to a preset timing sequence based on phased array technology. The echo signals can be received by one or all of the electromagnetic ultrasonic transducers in the array, thereby obtaining a multi-channel echo signal group. The multi-channel echo signal group contains the timestamp information corresponding to the acquisition. The electromagnetic ultrasonic array can perform uniform ultrasonic measurement on the bolt under test, realize non-contact stress measurement, overcome the limitations of traditional piezoelectric ultrasonics which require a coupling agent and have high requirements for the surface condition of the bolt, thereby improving the accuracy of stress analysis.

[0080] The signal denoising module is used to perform time-varying attenuation compensation and adaptive entropy weight denoising on the multi-channel echo signal group to obtain a denoised echo signal group.

[0081] When ultrasonic waves propagate inside the bolt under test, they will be scattered due to the unevenness of the internal grains, impurities or microstructure of the bolt, forming material noise. Furthermore, additional noise may be introduced due to changes in the external environment or temperature. Therefore, it is necessary to denoise the received multi-channel echo signal group.

[0082] In this embodiment of the invention, when the signal denoising module performs time-varying attenuation compensation and adaptive entropy weight denoising on the multi-channel echo signal group to obtain a denoised echo signal group, it includes:

[0083] The DC bias in the multi-channel echo signal group is filtered out, and the multi-channel echo signal group after the DC bias is filtered out is aligned based on a unified timestamp to obtain an aligned echo signal group.

[0084] Bandpass filtering and time-varying attenuation compensation are performed on each aligned echo signal in the aligned echo signal group to obtain a gain echo signal group.

[0085] Adaptive empirical mode decomposition is performed on the gain echo signal group to obtain a set of signal mode components;

[0086] Calculate the modal component energy entropy set corresponding to the signal modal component set, and perform joint scoring and adaptive classification on each signal modal component in the signal modal component set based on the modal component energy entropy set to obtain modal component class groups;

[0087] Based on the modal component class group, component weights are configured for each signal modal component in the signal modal component group set to obtain a component weight reassembly set, and an initial component weight vector group is constructed based on the component weight reassembly set;

[0088] Based on the initial component weight vector set, the signal mode component set is subjected to least squares filtering and signal reconstruction to obtain the noise-reduced echo signal set.

[0089] The DC bias refers to the fixed offset component in the multi-channel echo signal group, which can be obtained by calculating the signal mean of each multi-channel echo signal and subtracting the corresponding signal mean from each multi-channel echo signal. Since electromagnetic ultrasonic arrays often use time-division multiplexing to acquire multi-channel ultrasonic signals in order to prevent signal overlap and interference, there is a time difference between the multi-channel echo signals in the multi-channel echo signal group. In order to prevent deviations during subsequent wavefield reconstruction, it is necessary to align the multi-channel echo signal group after filtering out the DC bias based on a unified timestamp to make the multi-channel echo signals logically synchronized.

[0090] In detail, since the ultrasonic signal attenuates due to increased propagation distance and material absorption when it is transmitted inside the bolt under test, time-varying attenuation compensation is required for the aligned echo signal group. The time-varying attenuation compensation includes calculating an attenuation function based on a polynomial fitting method, calculating an attenuation term based on the attenuation function and the time difference between the signal time and the start time of each aligned echo signal, calculating a compensation term based on the exponential function of the attenuation term, and using the product of the compensation term and the aligned echo signal as the gain echo signal.

[0091] Specifically, when the signal denoising module performs adaptive empirical mode decomposition on the gain echo signal group to obtain a set of signal mode components, it includes:

[0092] Initialize a white noise signal group, select the gain echo signals in the gain echo signal group one by one as the target gain echo signals, add the white noise signal group to the target gain echo signals to obtain a noisy echo signal group;

[0093] Empirical mode decomposition is performed on the noisy echo signal group, and the first eigenmode function obtained from the mode decomposition is used as the noisy mode function to form a noisy mode function group;

[0094] The noise-added mode function group is integrated and averaged to obtain the noise intrinsic mode function. The first-order residual signal between the target gain echo signal and the noise intrinsic mode function is calculated, and the signal energy ratio between the first-order residual signal and the target gain echo signal is calculated.

[0095] Replace the target gain echo signal with the first-order residual signal, and return to the step of adding the white noise signal group to the target gain echo signal to obtain the noisy echo signal group;

[0096] Until the signal energy ratio is less than the preset energy ratio threshold, all first-order residual signals obtained from multi-level iterations are collected as signal mode components into a signal mode component group, and the signal mode component groups of all target gain echo signals are collected into a signal mode component group set.

[0097] In this context, the signal mean of all white noise signals in the white noise signal group is 0. Empirical Mode Decomposition (EMD) is an adaptive data-driven analysis method for processing nonlinear and non-stationary signals. EMD decomposes complex signals into several intrinsic mode functions (IMFs) with clear physical meaning. Each IMF represents an oscillation mode at different time scales in the original signal, thus clearly presenting the signal's intrinsic structure. The signal energy ratio can be a short-time energy ratio or an interval energy ratio, obtained by calculating the ratio between the square of the first-order residual signal's amplitude and the square of the target gain echo signal's amplitude. The energy ratio threshold can be between 10^-3 and 10^-4. Since the signal mean of the white noise signal is 0, after ensemble averaging of the noisy mode function group, the noise IMF reflects the modal components of the noise, while the signal mode component group represents the modal components of the denoised target gain echo signal.

[0098] In detail, calculating the modal component energy entropy set corresponding to the signal modal component set includes calculating the short-time energy corresponding to the energy of each signal modal component in each signal modal component set, normalizing the short-time energy in each signal modal component set to obtain the energy distribution and energy ratio, and calculating the modal component energy entropy of each modal component based on the energy ratio. The modal component energy entropy can be calculated using the Shannon entropy algorithm.

[0099] Specifically, when the signal denoising module performs joint scoring and adaptive classification of each signal mode component in the signal mode component group set based on the mode component energy entropy set to obtain mode component class groups, it includes:

[0100] The modal component energy entropy in the modal component energy entropy set is selected one by one as the target component energy entropy, and the signal modal component corresponding to the target component energy entropy in the signal modal component set is taken as the target modal component;

[0101] Calculate the cross-channel correlation factor of the target modal component;

[0102] The joint scoring factor is obtained by weighted summation of the cross-channel correlation factor, the target component energy entropy, and the energy proportion of the target modal component.

[0103] The joint scoring factors of the energy entropy of all modal components are aggregated into a joint scoring factor set, and each signal modal component in the signal modal component group set is classified based on the quantile of the joint scoring factor set to obtain modal component class groups.

[0104] The cross-channel correlation factor can be calculated using the Spearman rank correlation coefficient. When performing a weighted summation based on the cross-channel correlation factor, the target component energy entropy, and the energy proportion of the target modal component, the weighted value of the cross-channel correlation factor, the weighted value of the absolute difference between a preset constant term and the target component energy entropy, and the weighted value of the target modal component are summed. The constant term can be 1. Classifying each signal modal component in the signal modal component group set based on the quantiles of the joint scoring factor set to obtain modal component groups refers to splitting the joint scoring factor set into scoring factors according to the quantiles of the joint scoring factor set. The signal modal component set is classified into modal component groups according to the correspondence between each joint scoring factor and each signal modal component. The step of performing least squares filtering and signal reconstruction on the signal modal component set using the initial component weight vector set to obtain the denoised echo signal set refers to: selecting the initial component weight vector in the initial component weight vector set one by one as the target weight vector, taking the modal component set corresponding to the target weight vector in the signal modal component set as the target modal component set, generating the tissue regression vector of the target modal component set, and performing least squares filtering and signal reconstruction on the tissue regression vector based on the target weight vector to obtain the denoised echo signal.

[0105] In this embodiment of the invention, time-varying attenuation compensation effectively suppresses the amplitude attenuation caused by changes in the propagation distance and temperature within the bolt; adaptive empirical mode decomposition separates multi-scale oscillation components in both time and frequency domains, preserving the structural characteristics of the signal; a joint scoring mechanism based on energy entropy and cross-channel correlation further distinguishes noise modes from useful modes, achieving dynamic noise identification; by introducing adaptive optimal estimation in the weighted reconstruction stage of least squares filtering, denoising and reconstruction can be performed based on the prior knowledge on the right side of the weight vector, improving the smoothness and real-time performance of signal reconstruction, ultimately significantly improving the signal-to-noise ratio and structural stability of the electromagnetic ultrasonic echo signal, providing high-precision data input for subsequent bolt stress detection calculations.

[0106] The time-frequency feature extraction module is used to extract echo time-domain features from the denoised echo signal group, perform segmented time-frequency feature extraction on the denoised echo signal group based on a short-time sliding window to obtain echo time-frequency features, and fuse the echo time-domain features and the echo time-frequency features into ultrasonic time-frequency features.

[0107] The step of extracting echo time-domain features from the noise-reduced echo signal group refers to extracting time-domain features such as the first wave arrival time, peak drift, envelope energy, and phase delay of each noise-reduced echo signal in the noise-reduced echo signal group, and then combining these time-domain features into echo time-domain features.

[0108] Specifically, when the time-frequency feature extraction module performs segmented time-frequency feature extraction on the denoised echo signal group based on a short-time sliding window to obtain the echo time-frequency features, it includes:

[0109] The noise-reduced echo signal group is segmented based on a short-time sliding window, and a windowing operation is performed on the segmented noise-reduced echo signal group to obtain a windowed signal sequence group.

[0110] Perform continuous wavelet transform on the window signal sequence group to obtain the echo frequency domain sequence group;

[0111] Multi-scale time-frequency feature statistics are performed on the echo frequency domain sequence group to obtain a primary time-frequency feature sequence group. The multi-scale time-frequency feature statistics include statistically analyzing the energy spectrum distribution, wavelet energy spectrum entropy, instantaneous dominant frequency, scale bandwidth, and instantaneous energy of each echo frequency domain in the echo frequency domain sequence group.

[0112] Ridge smoothing is applied to the instantaneous dominant frequency in the primary time-frequency feature sequence group, and short-time smoothing and differential enhancement are applied to the wavelet energy spectral entropy in the ridge-smoothed primary time-frequency feature sequence group to obtain the enhanced time-frequency feature sequence group.

[0113] The enhanced time-frequency feature sequence group is time-aligned and its features are spliced. Then, linear discriminant analysis is performed on the spliced ​​enhanced time-frequency feature sequence group to reduce its dimensionality, thereby obtaining the echo time-frequency features.

[0114] The signal segmentation refers to segmenting the signal according to the window length and sliding step of the short-time sliding window, and using the signal corresponding to the window at each sliding time as the window signal. The windowing operation refers to applying a window to each window signal after signal segmentation using a Hanning window or a Hamming window. The ridge smoothing refers to smoothing the ridge region with the most concentrated energy in the instantaneous main frequency. The short-time smoothing and differential enhancement of the wavelet energy spectrum entropy refers to performing a moving average or low-pass filtering on the obtained wavelet energy spectrum entropy sequence in the time dimension, and performing differential calculation of the neighborhood spectral entropy on the short-time smoothed wavelet energy spectrum entropy sequence to further amplify the details of spectral entropy changes, enhance the sensitivity to signal abrupt changes, and thus improve the reliability of detection. The fusion of the echo time-domain features and the echo time-frequency features into ultrasonic time-frequency features refers to performing dimensional splicing or splicing fusion of the echo time-domain features and the echo time-frequency features.

[0115] Specifically, by using short-time sliding windows and continuous wavelet transform for feature extraction, multi-scale time-frequency analysis of non-stationary electromagnetic ultrasonic echo signals can be performed. This ensures that the time-frequency features have good time resolution in the high-frequency band and excellent frequency resolution in the low-frequency band, accurately describing the transient response under bolt stress changes. The energy spectrum distribution, spectral entropy, and main frequency bandwidth extracted through multi-scale statistical analysis reflect the local oscillation characteristics of the signal. Ridge smoothing and differential enhancement improve the sensitivity to abrupt changes, making the stress reflection boundary more obvious. This provides high-resolution time-frequency features for subsequent stress detection, improving the accuracy of stress detection.

[0116] The spatial feature extraction module is used to reconstruct the wave field of the noise-reduced echo signal group based on the electromagnetic ultrasonic array and the bolt under test to obtain a spatial ultrasonic energy map, and to perform self-attention map convolution on the spatial ultrasonic energy map to obtain ultrasonic spatial features.

[0117] Existing bolt stress testing methods are mostly limited to single-dimensional stress testing, often yielding simple bolt stress values ​​without considering the overall stress distribution inside the bolt, resulting in low accuracy of stress testing.

[0118] Specifically, when the spatial feature extraction module performs wavefield reconstruction of the noise-reduced echo signal group based on the electromagnetic ultrasonic array and the bolt under test to obtain a spatial ultrasonic energy map, it includes:

[0119] A bolt mesh model is constructed based on the bolt to be tested, and the bolt mesh model is discretized into bolt pixel mesh;

[0120] Based on the positional relationship between the electromagnetic ultrasonic array and the bolt under test, the sound wave propagation time of each pixel in the bolt pixel grid is calculated and phase compensation is performed to obtain the sound wave propagation time set.

[0121] Based on the sound wave propagation time set, the echo signal amplitude set is extracted from the noise-reduced echo signal set, and the echo signal amplitude set is mapped to each pixel of the bolt pixel grid to obtain the signal amplitude grid.

[0122] The echo signal amplitudes of each pixel in the signal amplitude grid on each channel are superimposed to obtain the superimposed amplitude grid.

[0123] The coherence weights of each pixel in the superimposed amplitude grid are calculated based on the phase coherence method, and the superimposed amplitude grid is coherently weighted based on the coherence weights to obtain a weighted amplitude grid.

[0124] The amplitude energy of each grid in the weighted amplitude grid is calculated and the grid is assigned a value to obtain an ultrasonic energy map. The ultrasonic energy map is then subjected to bilinear interpolation smoothing and logarithmic dynamic compression to obtain a spatial ultrasonic energy map.

[0125] In detail, the bolt mesh model can be a two-dimensional mesh model or a three-dimensional mesh model. The dimension of the bolt mesh model is determined by the electromagnetic ultrasonic array and the stress detection requirements. For example, when the electromagnetic ultrasonic array is a linear array or only needs to detect bolt stress in a plane, the bolt mesh model is a two-dimensional mesh model. The method for calculating the sound wave propagation time is based on the sound velocity model of the material used in the bolt to be tested. The calculation of the coherence weight of each pixel in the superimposed amplitude mesh based on the phase coherence method refers to calculating the phase coherence score based on the weighted algorithm of phase consistency and coherence, and normalizing the phase coherence score based on the sigmoid normalization function to obtain the coherence weight.

[0126] Specifically, the amplitude energy calculation refers to calculating the square of the amplitude of each pixel or the envelope amplitude of the signal corresponding to each pixel. The envelope amplitude can be calculated using the Hilbert transform algorithm. The grid amplitude refers to assigning values ​​or mapping grayscale values ​​to each pixel using the calculated amplitude energy.

[0127] Specifically, when the spatial feature extraction module performs self-attention map convolution on the spatial ultrasound energy map to obtain ultrasound spatial features, it includes:

[0128] Each pixel of the spatial ultrasonic energy map is used as a network node, and a node feature set is generated based on the amplitude energy and coordinate information of each network node.

[0129] The k-nearest neighbor algorithm is used to establish node edges for each network node in the spatial ultrasonic energy map, and the edge weight of each node edge is calculated based on the Euclidean distance of each network node using the Gaussian distance weighting algorithm.

[0130] An ultrasonic energy graph structure is constructed based on the node feature set, the network nodes, the node edges, and the edge weights.

[0131] Based on the ultrasonic energy map structure, the node feature set is positionally encoded and linearly mapped to obtain an encoded node feature set. Then, based on the neighborhood subgraph of each network node in the ultrasonic energy map structure, self-attention is calculated on the encoded node feature set to obtain a self-attention weight set.

[0132] Based on the self-attention weight set, the encoding node feature set is subjected to self-attention message aggregation and graph convolution update to obtain the self-attention space feature set;

[0133] Cross-layer feature fusion is performed on the self-attention space feature set, and feature merging and feature normalization are performed on the self-attention space features after cross-layer feature fusion to obtain ultrasound space features.

[0134] Specifically, establishing node edges for each network node in the spatial ultrasound energy graph based on the k-nearest neighbor algorithm refers to calculating the neighboring network nodes of each network node based on the k-nearest neighbor algorithm and constructing node edges between the network node and its neighboring network nodes. Constructing the ultrasound energy graph structure based on the node feature set, the network node, the node edge, and the edge weight refers to constructing a primary graph structure based on each network node and each node edge, and mapping and assigning values ​​to the primary graph structure using the node feature set and each edge weight to obtain the ultrasound energy graph structure. The neighborhood subgraph is a subgraph composed of the area occupied by each network node and its corresponding neighboring network node in the ultrasound energy graph structure. The cross-layer feature fusion refers to fusing based on the self-attention spatial features corresponding to each neighborhood subgraph.

[0135] Specifically, by performing bolt mesh modeling, acoustic wave propagation time compensation, amplitude superposition, and coherence weighting, spatial domain reconstruction of multi-channel ultrasonic echoes is achieved. This effectively enhances the real defect reflection signal and suppresses various random noises and spatial artifacts. The spatial ultrasonic energy map obtained based on amplitude energy mapping can accurately present the stress concentration area inside the bolt, achieving high-fidelity reconstruction of the ultrasonic field distribution. By constructing the spatial ultrasonic energy map as a graph structure with local topological relationships and using self-attention graph convolution to adaptively model the spatial dependence and local energy changes between nodes, local sensitive features can be preserved and cross-regional correlation expression can be enhanced. This results in the extracted spatial ultrasonic features having higher resolution and robustness, thereby significantly improving the accuracy and stability of bolt stress detection.

[0136] The stress analysis module is used to perform correlation detection and spatiotemporal complementary modeling on the ultrasonic time-frequency features and ultrasonic spatial features to obtain ultrasonic spatiotemporal joint features, and to perform stress analysis on the ultrasonic spatiotemporal joint features to obtain bolt stress data.

[0137] Traditional signal analysis is mostly limited to single-dimensional feature extraction in the time or frequency domain, failing to effectively characterize the multi-scale, nonlinear sound field changes in the spatial domain caused by bolt stress, thus limiting the accuracy of bolt stress detection.

[0138] Specifically, when the stress analysis module performs correlation detection and spatiotemporal complementary modeling on the ultrasonic time-frequency features and the ultrasonic spatial features to obtain the ultrasonic spatiotemporal joint features, it includes:

[0139] The ultrasonic time-frequency features and ultrasonic spatial features are respectively normalized and aligned in dimension to obtain aligned ultrasonic feature groups;

[0140] The feature correlation matrix of the aligned ultrasound feature group is calculated based on the Pearson correlation coefficient, and the aligned ultrasound feature group is weighted based on the feature correlation matrix to obtain the weighted ultrasound feature group.

[0141] The ultrasound attention weight matrix of the weighted ultrasound feature group is calculated based on the cross-modal attention mechanism, and the attention of the weighted ultrasound feature group is enhanced based on the ultrasound attention weight matrix to obtain the enhanced ultrasound feature group.

[0142] Based on a multilayer perceptron, high-dimensional feature fusion is performed on the enhanced ultrasound feature group to obtain high-dimensional spatiotemporal joint features, and then the high-dimensional spatiotemporal joint features are dimensionally compressed to obtain ultrasound spatiotemporal joint features.

[0143] In this process, z-score normalization can be used for feature standardization. Dimension alignment refers to mapping ultrasound time-frequency features and ultrasound spatial features to the same dimensional space using a linear projection layer. Correlation weighting refers to assigning different correlation weights to features of different dimensions based on the magnitude of their correlation in the correlation matrix, thereby achieving feature weighting. The ultrasound attention weight matrix for calculating the relevant ultrasound feature group based on the cross-modal attention mechanism refers to using the dimension-aligned ultrasound time-frequency features as key vectors and the dimension-aligned ultrasound spatial features as query vectors to calculate the ultrasound attention weight matrix. Principal component analysis or an autoencoder can be used for dimensional compression.

[0144] Specifically, when the stress analysis module performs stress analysis on the ultrasonic spatiotemporal joint features to obtain bolt stress data, it includes:

[0145] Based on orthogonal basis functions, the feature domain transformation of the ultrasonic spatiotemporal joint features is performed to obtain the transformed spatiotemporal joint features;

[0146] The transformed spatiotemporal joint features are nonlinearly activated, and the ultrasonic spatiotemporal joint features and the nonlinearly activated transformed spatiotemporal joint features are enhanced and fused based on the residual structure to obtain residual spatiotemporal joint features;

[0147] The spatiotemporal joint features of the residuals are modeled nonlinearly using a pre-trained radial basis function network to obtain spatiotemporal stress data.

[0148] The bolt temperature and bolt parameters of the bolt to be tested are obtained, and physical constraint compensation is performed on the spatiotemporal stress data based on the bolt temperature and bolt parameters to obtain bolt stress data.

[0149] The orthogonal basis functions can be orthogonal wavelet basis functions, which can better extract the localized features of ultrasonic spatiotemporal joint features in different frequency bands. They can be nonlinearly activated using a rectified linear unit (ReLU). The radial basis function network (RBFN) is a neural network trained from multiple residual spatiotemporal features pre-labeled with stress data. The bolt parameters include the bolt's specifications and metal material. By using the bolt temperature and bolt parameters, a stress-sound velocity-temperature compensation model can be constructed, and the compensation model can be used to perform physical constraint compensation on the spatiotemporal stress data.

[0150] In detail, by using spatiotemporal complementary modeling based on correlation detection and cross-modal attention mechanisms, the synergistic variation law of temporal-frequency features and spatial energy distribution under bolt stress can be highlighted. By using multilayer perceptron for high-dimensional fusion, ultrasonic spatiotemporal joint features with stronger characterization capabilities can be obtained, overcoming the problem of insufficient modeling of single-domain features by traditional methods. By introducing orthogonal wavelet basis transformation for stress analysis, the localization characteristics of stress-sensitive frequency bands can be enhanced. Combining residual enhancement structure and radial basis function network can construct a more stable nonlinear stress mapping model. Finally, by combining bolt temperature and material parameters for physical compensation, the robustness and accuracy of stress detection can be significantly improved.

[0151] Example 2:

[0152] Please refer to Figure 2 This invention discloses an electromagnetic ultrasonic bolt stress detection method based on a spatiotemporal complementary algorithm, comprising the following steps:

[0153] Based on a preset electromagnetic ultrasonic array, multi-channel ultrasonic signals are acquired from the bolt under test to obtain a multi-channel echo signal group.

[0154] Time-varying attenuation compensation and adaptive entropy weight denoising are performed on the multi-channel echo signal group to obtain a denoised echo signal group.

[0155] Echo time-domain features are extracted from the denoised echo signal group. Segmented time-frequency features are extracted from the denoised echo signal group based on a short-time sliding window to obtain echo time-frequency features. The echo time-domain features and the echo time-frequency features are then fused into ultrasonic time-frequency features.

[0156] Based on the electromagnetic ultrasonic array and the bolt under test, the wave field of the noise-reduced echo signal group is reconstructed to obtain a spatial ultrasonic energy map, and the spatial ultrasonic energy map is convolved with a self-attention map to obtain ultrasonic spatial features.

[0157] Correlation detection and spatiotemporal complementary modeling are performed on the ultrasonic time-frequency features and ultrasonic spatial features to obtain ultrasonic spatiotemporal joint features. Stress analysis is then performed on the ultrasonic spatiotemporal joint features to obtain bolt stress data.

[0158] The processes described above with reference to the flowcharts in the embodiments disclosed in this invention can be implemented as computer software programs. The embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), it performs the functions defined in the methods of this application. It should be noted that the computer-readable medium described above in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wire segments, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless segments, wire segments, optical fibers, RF, etc., or any suitable combination thereof.

[0159] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0160] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the present invention. The purpose of the present invention has been fully and effectively achieved. The functions and structural principles of the present invention have been shown and explained in the embodiments. Without departing from the stated principles, the implementation of the present invention may have any variations or modifications.

Claims

1. An electromagnetic ultrasonic bolt stress detection system based on a spatiotemporal complementary algorithm, characterized in that, The system includes an ultrasonic acquisition module, a signal noise reduction module, a time-frequency feature extraction module, a spatial feature extraction module, and a stress analysis module, wherein: The ultrasonic acquisition module is used to acquire multi-channel ultrasonic signals from the bolt under test based on a preset electromagnetic ultrasonic array, and obtain a multi-channel echo signal group. The signal denoising module is used to perform time-varying attenuation compensation and adaptive entropy weight denoising on the multi-channel echo signal group to obtain a denoised echo signal group. The time-frequency feature extraction module is used to extract echo time-domain features from the denoised echo signal group, perform segmented time-frequency feature extraction on the denoised echo signal group based on a short-time sliding window to obtain echo time-frequency features, and fuse the echo time-domain features and the echo time-frequency features into ultrasonic time-frequency features; A spatial feature extraction module is used to reconstruct the wavefield of the denoised echo signal group based on the electromagnetic ultrasonic array and the bolt under test to obtain a spatial ultrasonic energy map, and to perform self-attention map convolution on the spatial ultrasonic energy map to obtain ultrasonic spatial features. Specifically, when performing wavefield reconstruction of the denoised echo signal group based on the electromagnetic ultrasonic array and the bolt under test to obtain a spatial ultrasonic energy map, the spatial feature extraction module includes: constructing a bolt mesh model based on the bolt under test, and discretizing the bolt mesh model into a bolt pixel mesh; calculating the sound wave propagation time and performing phase compensation on each pixel of the bolt pixel mesh based on the positional relationship between the electromagnetic ultrasonic array and the bolt under test to obtain a sound wave propagation time set; and based on the sound wave propagation time set... The wave propagation time set extracts the echo signal amplitude set from the noise-reduced echo signal set, and maps the echo signal amplitude set to each pixel of the bolt pixel grid to obtain the signal amplitude grid; the echo signal amplitudes of each pixel in the signal amplitude grid on each channel are superimposed to obtain the superimposed amplitude grid; the coherence weight of each pixel in the superimposed amplitude grid is calculated based on the phase coherence method, and the superimposed amplitude grid is coherently weighted based on the coherence weight to obtain the weighted amplitude grid; the amplitude energy of each grid in the weighted amplitude grid is calculated and the grid is assigned a value to obtain the ultrasonic energy map; the ultrasonic energy map is then subjected to bilinear interpolation smoothing and logarithmic dynamic compression to obtain the spatial ultrasonic energy map; The stress analysis module is used to perform correlation detection and spatiotemporal complementary modeling on the ultrasonic time-frequency features and ultrasonic spatial features to obtain ultrasonic spatiotemporal joint features, and to perform stress analysis on the ultrasonic spatiotemporal joint features to obtain bolt stress data.

2. The electromagnetic ultrasonic bolt stress detection system based on a spatiotemporal complementary algorithm according to claim 1, characterized in that, When the signal denoising module performs time-varying attenuation compensation and adaptive entropy weight denoising on the multi-channel echo signal group to obtain a denoised echo signal group, it includes: The DC bias in the multi-channel echo signal group is filtered out, and the multi-channel echo signal group after the DC bias is filtered out is aligned based on a unified timestamp to obtain an aligned echo signal group. Bandpass filtering and time-varying attenuation compensation are performed on each aligned echo signal in the aligned echo signal group to obtain a gain echo signal group. Adaptive empirical mode decomposition is performed on the gain echo signal group to obtain a set of signal mode components; Calculate the modal component energy entropy set corresponding to the signal modal component set, and perform joint scoring and adaptive classification on each signal modal component in the signal modal component set based on the modal component energy entropy set to obtain modal component class groups; Based on the modal component class group, component weights are configured for each signal modal component in the signal modal component group set to obtain a component weight reassembly set, and an initial component weight vector group is constructed based on the component weight reassembly set; Based on the initial component weight vector set, the signal mode component set is subjected to least squares filtering and signal reconstruction to obtain the noise-reduced echo signal set.

3. The electromagnetic ultrasonic bolt stress detection system based on a spatiotemporal complementary algorithm according to claim 2, characterized in that, When the signal denoising module performs adaptive empirical mode decomposition on the gain echo signal group to obtain a set of signal mode components, it includes: Initialize a white noise signal group, select the gain echo signals in the gain echo signal group one by one as the target gain echo signals, add the white noise signal group to the target gain echo signals to obtain a noisy echo signal group; Empirical mode decomposition is performed on the noisy echo signal group, and the first eigenmode function obtained from the mode decomposition is used as the noisy mode function to form a noisy mode function group; The noise-added mode function group is integrated and averaged to obtain the noise intrinsic mode function. The first-order residual signal between the target gain echo signal and the noise intrinsic mode function is calculated, and the signal energy ratio between the first-order residual signal and the target gain echo signal is calculated. Replace the target gain echo signal with the first-order residual signal, and return to the step of adding the white noise signal group to the target gain echo signal to obtain the noisy echo signal group; Until the signal energy ratio is less than the preset energy ratio threshold, all first-order residual signals obtained from multi-level iterations are collected as signal mode components into a signal mode component group, and the signal mode component groups of all target gain echo signals are collected into a signal mode component group set.

4. The electromagnetic ultrasonic bolt stress detection system based on a spatiotemporal complementary algorithm according to claim 3, characterized in that, When the signal denoising module performs joint scoring and adaptive classification of each signal mode component in the signal mode component group set based on the modal component energy entropy set to obtain the modal component class group, it includes: The modal component energy entropy in the modal component energy entropy set is selected one by one as the target component energy entropy, and the signal modal component corresponding to the target component energy entropy in the signal modal component set is taken as the target modal component; Calculate the cross-channel correlation factor of the target modal component; The joint scoring factor is obtained by weighted summation of the cross-channel correlation factor, the target component energy entropy, and the energy proportion of the target modal component. The joint scoring factors of the energy entropy of all modal components are aggregated into a joint scoring factor set, and each signal modal component in the signal modal component group set is classified based on the quantile of the joint scoring factor set to obtain modal component class groups.

5. The electromagnetic ultrasonic bolt stress detection system based on a spatiotemporal complementary algorithm according to claim 1, characterized in that, The spatial feature extraction module, when performing self-attention map convolution on the spatial ultrasound energy map to obtain ultrasound spatial features, includes: Each pixel of the spatial ultrasonic energy map is used as a network node, and a node feature set is generated based on the amplitude energy and coordinate information of each network node. The k-nearest neighbor algorithm is used to establish node edges for each network node in the spatial ultrasonic energy map, and the edge weight of each node edge is calculated based on the Euclidean distance of each network node using the Gaussian distance weighting algorithm. An ultrasonic energy graph structure is constructed based on the node feature set, the network nodes, the node edges, and the edge weights. Based on the ultrasonic energy map structure, the node feature set is positionally encoded and linearly mapped to obtain an encoded node feature set. Then, based on the neighborhood subgraph of each network node in the ultrasonic energy map structure, self-attention is calculated on the encoded node feature set to obtain a self-attention weight set. Based on the self-attention weight set, the encoding node feature set is subjected to self-attention message aggregation and graph convolution update to obtain the self-attention space feature set; Cross-layer feature fusion is performed on the self-attention space feature set, and feature merging and feature normalization are performed on the self-attention space features after cross-layer feature fusion to obtain ultrasound space features.

6. The electromagnetic ultrasonic bolt stress detection system based on a spatiotemporal complementary algorithm according to claim 1, characterized in that, When the time-frequency feature extraction module performs segmented time-frequency feature extraction on the denoised echo signal group based on a short-time sliding window to obtain the echo time-frequency features, it includes: The noise-reduced echo signal group is segmented based on a short-time sliding window, and a windowing operation is performed on the segmented noise-reduced echo signal group to obtain a windowed signal sequence group. Perform continuous wavelet transform on the window signal sequence group to obtain the echo frequency domain sequence group; Multi-scale time-frequency feature statistics are performed on the echo frequency domain sequence group to obtain a primary time-frequency feature sequence group. The multi-scale time-frequency feature statistics include statistically analyzing the energy spectrum distribution, wavelet energy spectrum entropy, instantaneous dominant frequency, scale bandwidth, and instantaneous energy of each echo frequency domain in the echo frequency domain sequence group. Ridge smoothing is applied to the instantaneous dominant frequency in the primary time-frequency feature sequence group, and short-time smoothing and differential enhancement are applied to the wavelet energy spectral entropy in the ridge-smoothed primary time-frequency feature sequence group to obtain the enhanced time-frequency feature sequence group. The enhanced time-frequency feature sequence group is time-aligned and its features are spliced. Then, linear discriminant analysis is performed on the spliced ​​enhanced time-frequency feature sequence group to reduce its dimensionality, thereby obtaining the echo time-frequency features.

7. The electromagnetic ultrasonic bolt stress detection system based on a spatiotemporal complementary algorithm according to claim 1, characterized in that, When the stress analysis module performs correlation detection and spatiotemporal complementary modeling on the ultrasonic time-frequency features and ultrasonic spatial features to obtain the ultrasonic spatiotemporal joint features, it includes: The ultrasonic time-frequency features and ultrasonic spatial features are respectively normalized and aligned in dimension to obtain aligned ultrasonic feature groups; The feature correlation matrix of the aligned ultrasound feature group is calculated based on the Pearson correlation coefficient, and the aligned ultrasound feature group is weighted based on the feature correlation matrix to obtain the weighted ultrasound feature group. The ultrasound attention weight matrix of the weighted ultrasound feature group is calculated based on the cross-modal attention mechanism, and the attention of the weighted ultrasound feature group is enhanced based on the ultrasound attention weight matrix to obtain the enhanced ultrasound feature group. Based on a multilayer perceptron, high-dimensional feature fusion is performed on the enhanced ultrasound feature group to obtain high-dimensional spatiotemporal joint features, and then the high-dimensional spatiotemporal joint features are dimensionally compressed to obtain ultrasound spatiotemporal joint features.

8. The electromagnetic ultrasonic bolt stress detection system based on a spatiotemporal complementary algorithm according to claim 1, characterized in that, When the stress analysis module performs stress analysis on the ultrasonic spatiotemporal joint features to obtain bolt stress data, it includes: Based on orthogonal basis functions, the feature domain transformation of the ultrasonic spatiotemporal joint features is performed to obtain the transformed spatiotemporal joint features; The transformed spatiotemporal joint features are nonlinearly activated, and the ultrasonic spatiotemporal joint features and the nonlinearly activated transformed spatiotemporal joint features are enhanced and fused based on the residual structure to obtain residual spatiotemporal joint features; The spatiotemporal joint features of the residuals are modeled nonlinearly using a pre-trained radial basis function network to obtain spatiotemporal stress data. The bolt temperature and bolt parameters of the bolt to be tested are obtained, and physical constraint compensation is performed on the spatiotemporal stress data based on the bolt temperature and bolt parameters to obtain bolt stress data.

9. A method for detecting electromagnetic ultrasonic bolt stress based on a spatiotemporal complementary algorithm, characterized in that, The method includes: Based on a preset electromagnetic ultrasonic array, multi-channel ultrasonic signals are acquired from the bolt under test to obtain a multi-channel echo signal group. Time-varying attenuation compensation and adaptive entropy weight denoising are performed on the multi-channel echo signal group to obtain a denoised echo signal group. Echo time-domain features are extracted from the denoised echo signal group. Segmented time-frequency features are extracted from the denoised echo signal group based on a short-time sliding window to obtain echo time-frequency features. The echo time-domain features and the echo time-frequency features are then fused into ultrasonic time-frequency features. Based on the electromagnetic ultrasonic array and the bolt under test, wavefield reconstruction is performed on the denoised echo signal group to obtain a spatial ultrasonic energy map. Then, self-attention map convolution is performed on the spatial ultrasonic energy map to obtain ultrasonic spatial features. Specifically, the wavefield reconstruction based on the electromagnetic ultrasonic array and the bolt under test to obtain the spatial ultrasonic energy map includes: constructing a bolt mesh model based on the bolt under test and discretizing the bolt mesh model into bolt pixel meshes; calculating the sound wave propagation time and performing phase compensation on each pixel of the bolt pixel mesh based on the positional relationship between the electromagnetic ultrasonic array and the bolt under test to obtain a set of sound wave propagation times; and reconstructing the spatial ultrasonic energy map from the denoised echo signal group based on the set of sound wave propagation times. Echo signal amplitude sets are extracted from the echo signal set and mapped onto each pixel of the bolt pixel grid to obtain a signal amplitude grid. The echo signal amplitudes of each pixel in the signal amplitude grid on each channel are superimposed to obtain a superimposed amplitude grid. The coherence weight of each pixel in the superimposed amplitude grid is calculated based on the phase coherence method, and the superimposed amplitude grid is coherently weighted based on the coherence weight to obtain a weighted amplitude grid. The amplitude energy of each grid in the weighted amplitude grid is calculated and the grid is assigned a value to obtain an ultrasonic energy map. The ultrasonic energy map is then subjected to bilinear interpolation smoothing and logarithmic dynamic compression to obtain a spatial ultrasonic energy map. Correlation detection and spatiotemporal complementary modeling are performed on the ultrasonic time-frequency features and ultrasonic spatial features to obtain ultrasonic spatiotemporal joint features. Stress analysis is then performed on the ultrasonic spatiotemporal joint features to obtain bolt stress data.

Images