Bolt looseness online monitoring method and system

Abstract

The invention discloses a bolt looseness online monitoring method and system, and the method comprises the steps: collecting an original data set containing a vibration signal of bolt connection, temperature gradient data and structural stress distribution through a multi-mode sensor array, and carrying out the time-space alignment and frequency domain decomposition processing of the original data set, obtaining a multi-dimensional feature matrix of the bolt nodes; based on the multi-dimensional feature matrix, a graph neural network model is adopted to carry out bolt looseness probability calculation, and a real-time looseness probability value of each bolt is output; generating a risk level map based on time evolution according to the real-time loosening probability value; and the risk level map is mapped in real time through a three-dimensional visual interface, and when it is detected that the bolt loosening risk level exceeds a preset threshold value, an early warning message is generated and uploaded to an operation and maintenance platform, and closed-loop monitoring response is completed. According to the embodiment of the invention, accurate assessment, dynamic prediction and visual early warning of the loosening risk can be realized, and the intelligent level of structure safety monitoring is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of on-line monitoring, and particularly relates to a method and system for on-line monitoring of bolt loosening. Background Art

[0002] In large steel structure facilities such as power transmission towers and bridges, the reliability of bolt connections directly affects the safety performance of the overall structure. Traditional bolt loosening detection mainly relies on manual inspections or regular torque checks, which have problems such as low efficiency, poor real-time performance, and difficulty in detecting early loosening hazards. Existing loosening detection methods based on vibration signals or strain monitoring usually use a single sensor to collect data, are easily affected by environmental noise, and lack the comprehensive analysis ability for the coupled effects of multiple factors such as temperature changes and structural force transmission paths. In addition, traditional methods are difficult to achieve dynamic prediction and visual warning of loosening risks, resulting in a lag in operation and maintenance response. Summary of the Invention

[0003] The purpose of the present invention is to provide a method and system for on-line monitoring of bolt loosening to solve the deficiencies in the prior art, and to be able to achieve accurate assessment, dynamic prediction, and visual warning of loosening risks, and improve the intelligent level of structural safety monitoring.

[0004] An embodiment of the present application provides a method for on-line monitoring of bolt loosening, the method comprising: Collecting an original data set containing vibration signals, temperature gradient data, and structural stress distribution of bolt connections through a multi-modal sensor array, and performing spatio-temporal alignment and frequency-domain decomposition processing on the original data set to obtain a multi-dimensional feature matrix of bolt nodes, wherein the multi-modal sensor array includes piezoelectric vibration sheets, infrared thermal imaging units, and fiber Bragg grating sensors; Based on the multi-dimensional feature matrix, using a graph neural network model to calculate the bolt loosening probability, by fusing the vibration frequency-domain energy entropy, temperature-stress coupling coefficient, and structural topology features, and dynamically weighting key nodes through an attention mechanism, and outputting the real-time loosening probability value of each bolt; According to the real-time loosening probability value, combining the historical working condition data of the tower and the prediction result of the environmental load time series, using a physically informed LSTM network to dynamically correct the loosening risk level, and generating a risk level map based on time evolution, wherein the risk level map includes the risk evolution trend of each bolt node within a preset future time period; Real-time mapping the risk level map through a three-dimensional visualization interface, and when it is detected that the bolt loosening risk level exceeds a preset threshold, generating a warning message and uploading it to the operation and maintenance platform to complete the closed-loop monitoring response.

[0005] Optionally, the original dataset containing the vibration signals, temperature gradient data, and structural stress distribution of the bolt connection is collected by the multimodal sensor array, and the original dataset is subjected to spatio-temporal alignment and frequency-domain decomposition processing to obtain a multi-dimensional feature matrix of the bolt node. The multimodal sensor array includes a piezoelectric vibration sheet, an infrared thermal imaging unit, and a fiber Bragg grating sensor, and includes: According to the bolt vibration signal collected by the piezoelectric vibration sheet, wavelet packet transform is used for noise reduction processing to obtain a denoised vibration time-domain signal; According to the denoised vibration time-domain signal, the energy distribution in the 0-5 kHz frequency band is calculated by short-time Fourier transform, and the energy ratio and energy entropy of each frequency band are extracted to obtain a vibration frequency-domain feature vector; According to the temperature distribution image collected by the infrared thermal imaging unit, the bicubic interpolation algorithm is used to improve the spatial resolution, and the radial and axial temperature gradients of the bolt connection surface are calculated to obtain a temperature gradient feature matrix; According to the strain data measured by the fiber Bragg grating sensor and combined with the temperature gradient feature matrix, the stress fluctuation coefficient caused by temperature change is calculated to obtain a temperature-stress coupling feature vector; According to the timestamps of the vibration frequency-domain feature vector, the temperature gradient feature matrix, and the temperature-stress coupling feature vector, the dynamic time warping algorithm is used for time series alignment, and finally a multi-dimensional feature matrix of the bolt node is generated.

[0006] Optionally, based on the multi-dimensional feature matrix, a graph neural network model is used to calculate the bolt loosening probability. By fusing the vibration frequency-domain energy entropy, temperature-stress coupling coefficient, and structural topology features, the key nodes are dynamically weighted through the attention mechanism, and the real-time loosening probability value of each bolt is output, including: According to the mechanical model of the iron tower structure, a bolt connection topology relationship graph is constructed, where the nodes of the topology relationship graph represent the bolt positions and the edges represent the structural force transmission paths; According to the multi-dimensional feature matrix, the vibration frequency-domain energy entropy, temperature gradient feature, and stress coupling coefficient of each bolt node are extracted as the initial node features of the graph neural network; According to the node features and topology relationship, the multi-head attention mechanism is used to calculate the feature propagation weights between nodes, highlighting the nodes with abnormal vibration energy and sudden temperature gradient changes; According to the weighted aggregated node features and combined with the theoretical calculation formula of the pre-tightening force, the real-time loosening probability values of each bolt are calculated through a fully connected neural network.

[0007] Optionally, based on the real-time loosening probability value, combining the historical working condition data of the iron tower and the prediction results of the environmental load time series, a physics-informed LSTM network is used to dynamically correct the loosening risk level, and a risk level map based on time evolution is generated. Among them, the risk level map includes the risk evolution trends of each bolt node within a preset future duration, including: According to the historical cases of bolt loosening recorded in the operation and maintenance database, extract the environmental parameters and equipment status characteristics at the time of loosening, and construct a historical working condition feature library; According to the meteorological forecast data, a temporal convolutional network is used to predict the changes in wind speed, temperature, and humidity in the next 24 hours, and an environmental load prediction curve is obtained; Based on the real-time loosening probability value, the historical working condition feature library, and the environmental load prediction curve, input them into a physics-informed LSTM network for time series prediction; According to the output results of the LSTM network, generate the risk level change curves of each bolt node within the next 24 hours, and construct a three-dimensional spatio-temporal risk level map.

[0008] Optionally, the risk level map is mapped in real time through a three-dimensional visualization interface. When it is detected that the bolt loosening risk level exceeds the preset threshold, a warning message is generated and uploaded to the operation and maintenance platform to complete the closed-loop monitoring response, including: Based on the iron tower BIM model and bolt position information, establish a digital twin on the three-dimensional visualization platform; Based on the risk level map data, dynamically render the risk status of each bolt node on the digital twin, and use three colors of red, yellow, and green to represent different risk levels; According to the preset warning threshold, automatically detect high-risk bolt nodes, and generate a warning message containing the position coordinates, risk level, and maintenance suggestions; Push the message to the corresponding operation and maintenance terminal according to the warning level, and track the disposal feedback information to complete the closed-loop process of monitoring-warning-disposal.

[0009] Optionally, based on the strain data measured by the fiber Bragg grating sensor, combined with the temperature gradient feature matrix, calculate the stress fluctuation coefficient caused by temperature changes to obtain the temperature-stress coupling feature vector, including: According to the original wavelength shift data collected by the fiber Bragg grating sensor, calculate the strain values of each measuring point through the Bragg wavelength demodulation algorithm to obtain the initial strain distribution data; According to the initial strain distribution data, use the spatial Kalman filtering algorithm to eliminate the measurement noise and obtain the smoothed strain field matrix; According to the temperature gradient feature matrix provided by the infrared thermal imaging unit, extract the temperature change rate of the corresponding measuring point, and establish a temperature-strain mapping relationship table in combination with the strain field matrix; According to the temperature-strain mapping relation table, the temperature compensation coefficient is fitted by the least square method, the effective stress fluctuation caused by the pure mechanical load is calculated, and finally the temperature-stress coupling eigenvector is generated.

[0010] Optionally, calculating the real-time loosening probability value of each bolt through a fully connected neural network according to the weighted aggregated node features in combination with the theoretical calculation formula of the pre-tightening force includes: Extracting three key indicators of vibration energy entropy, temperature-stress coupling coefficient and topological connection strength from the weighted node feature vector output by the graph neural network; According to the theoretical formula of the pre-tightening force, the stress fluctuation data in the node features is converted into the percentage of pre-tightening force loss to obtain the theoretical loosening degree reference value; Inputting the theoretical loosening degree reference value and the weighted node features into a three-layer fully connected neural network. The first layer performs feature fusion, the second layer adds physical constraint conditions, and the third layer outputs the probability value; Mapping the output value of the neural network to the interval of 0 - 1 through the Sigmoid activation function, calibrating the probability threshold in combination with the historical fault data, and finally generating the real-time loosening probability value of each bolt.

[0011] Another embodiment of the present application provides an on-line bolt loosening monitoring system, and the system includes: A processing module, configured to collect an original data set including vibration signals, temperature gradient data, and structural stress distribution of bolt connections through a multi-modal sensor array, perform spatio-temporal alignment and frequency-domain decomposition processing on the original data set to obtain a multi-dimensional feature matrix of bolt nodes, wherein the multi-modal sensor array includes a piezoelectric vibration piece, an infrared thermal imaging unit, and a fiber Bragg grating sensor; An output module, configured to calculate the bolt loosening probability based on the multi-dimensional feature matrix by using a graph neural network model, dynamically weight key nodes through an attention mechanism by fusing the vibration frequency-domain energy entropy, temperature-stress coupling coefficient, and structural topology features, and output the real-time loosening probability value of each bolt; A correction module, configured to dynamically correct the loosening risk level by using a physically informed LSTM network according to the real-time loosening probability value in combination with the historical working condition data of the iron tower and the prediction result of the environmental load time series, and generate a risk level map based on time evolution, wherein the risk level map includes the risk evolution trend of each bolt node within a preset future time period; A generation module, configured to map the risk level map in real time through a three-dimensional visualization interface, and when it is detected that the bolt loosening risk level exceeds a preset threshold, generate a warning message and upload it to the operation and maintenance platform to complete the closed-loop monitoring response.

[0012] Another embodiment of the present application provides a storage medium, in which a computer program is stored, and the computer program is configured to execute the method described in any one of the above when running.

[0013] Another embodiment of the present application provides an electronic device, including a memory and a processor. A computer program is stored in the memory, and the processor is configured to run the computer program to execute the method described in any one of the above.

[0014] Compared with the prior art, an online bolt loosening monitoring method provided by the present invention collects an original data set including vibration signals, temperature gradient data, and structural stress distribution of bolt connections through a multi-modal sensor array, performs spatio-temporal alignment and frequency-domain decomposition processing on the original data set to obtain a multi-dimensional feature matrix of bolt nodes; based on the multi-dimensional feature matrix, a graph neural network model is used to calculate the bolt loosening probability, and the real-time loosening probability value of each bolt is output; according to the real-time loosening probability value, a risk level map based on time evolution is generated; the risk level map is mapped in real time through a three-dimensional visualization interface. When it is detected that the bolt loosening risk level exceeds a preset threshold, a warning message is generated and uploaded to the operation and maintenance platform to complete the closed-loop monitoring response, so as to realize the accurate assessment, dynamic prediction, and visualization warning of loosening risks, and improve the intelligent level of structural safety monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a hardware structure block diagram of a computer terminal for an online bolt loosening monitoring method provided by an embodiment of the present invention; Figure 2 It is a schematic flow chart of an online bolt loosening monitoring method provided by an embodiment of the present invention; Figure 3 It is a schematic structural diagram of an online bolt loosening monitoring system provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0016] The embodiments described below with reference to the drawings are exemplary and are only used to explain the present invention and should not be construed as limiting the present invention.

[0017] An embodiment of the present invention first provides an online bolt loosening monitoring method, which can be applied to an electronic device, such as a computer terminal, specifically, a general computer, etc.

[0018] The following takes running on a computer terminal as an example to describe it in detail. Figure 1 It is a hardware structure block diagram of a computer terminal for an online bolt loosening monitoring method provided by an embodiment of the present invention. As Figure 1As shown in the figure, the computer device includes a processor, a memory, and a network interface connected through a system bus. Among them, the memory may include a non-volatile storage medium and an internal memory.

[0019] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions. When the program instructions are executed, the processor can be made to execute any one of the online bolt loosening monitoring methods.

[0020] The processor is used to provide computing and control capabilities to support the operation of the entire computer device.

[0021] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can be made to execute any one of the online bolt loosening monitoring methods.

[0022] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art can understand that Figure 1 the structure shown in the figure is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have a different component layout.

[0023] It should be understood that the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0024] See Figure 2 In the embodiments of the present invention, an online bolt loosening monitoring method is provided, which may include the following steps: S201. Collect an original data set including vibration signals, temperature gradient data, and structural stress distribution of bolt connections through a multi-modal sensor array, perform spatio-temporal alignment and frequency-domain decomposition processing on the original data set to obtain a multi-dimensional feature matrix of bolt nodes, where the multi-modal sensor array includes piezoelectric vibration sheets, infrared thermal imaging units, and fiber Bragg grating sensors; Specifically, based on the bolt vibration signals collected by the piezoelectric vibration piece, wavelet packet transform can be used for noise reduction processing to obtain the denoised vibration time-domain signal; The piezoelectric vibration piece collects the vibration signals of the bolt connection at a sampling rate of 100,000 times per second (100 kHz). The original signal contains high-frequency noise (such as wind noise and equipment electromagnetic interference). The noise reduction processing uses wavelet packet transform (Wavelet Packet Transform, WPT), and its core is to extract the effective components of the signal through multi-scale decomposition. The Symlet wavelet basis (Sym8) is selected because its symmetry can reduce phase distortion and is suitable for the non-stationary characteristics of mechanical vibration signals.

[0025] Signal decomposition: The original signal is decomposed by wavelet packet for 3 layers to obtain 8 sub-bands (2^3 = 8), and the frequency range covers 0 - 50 kHz. For example, the first layer of decomposition is 0 - 25 kHz and 25 - 50 kHz, the second layer is further subdivided into 0 - 12.5 kHz, 12.5 - 25 kHz, etc., and finally each sub-band in the third layer has a bandwidth of about 6.25 kHz.

[0026] Threshold denoising: The improved Sureshrink threshold algorithm is adopted to dynamically adjust the threshold according to the noise level of each sub-band. For example, the high-frequency sub-band (such as 25 - 50 kHz) has a high proportion of noise energy, and the threshold is set to 1.5 times the standard deviation of the signal energy in this frequency band; the threshold of the low-frequency sub-band (such as 0 - 6.25 kHz) is set to 0.8 times the standard deviation to retain more effective signals.

[0027] Signal reconstruction: Only retain the sub-bands with significant energy (exceeding the threshold) for inverse transformation. For example, if the characteristic frequencies caused by a certain bolt loosening are concentrated in 8 - 12 kHz, then retain the corresponding sub-band (the local frequency band in the 12.5 - 25 kHz sub-band of the second layer), and the remaining high-frequency noise sub-bands are suppressed.

[0028] The signal-to-noise ratio (SNR) of the denoised signal is increased from 15 dB to 35 dB, and the periodic impact components (such as the micro-slip vibration caused by bolt loosening) in the time-domain waveform are clearly visible.

[0029] According to the denoised vibration time-domain signal, calculate the energy distribution in the 0 - 5 kHz frequency band through short-time Fourier transform, extract the energy proportion and energy entropy of each frequency band, and obtain the vibration frequency-domain feature vector; The denoised vibration signal is subjected to time-frequency analysis through short-time Fourier transform (Short-Time Fourier Transform, STFT), focusing on the 0 - 5 kHz frequency band (the characteristic frequency range of bolt loosening).

[0030] Parameter configuration: Window Type: Hanning Window is adopted. The window length is 256 sampling points (corresponding to a time window of 2.56 milliseconds), and the overlap rate is 75% (sliding 64 points each time). This window design can balance time resolution and frequency resolution and reduce spectral leakage.

[0031] Frequency Range: The signal sampling rate is 100 kHz. The STFT analysis range covers 0 - 50 kHz, but only the data in the frequency band of 0 - 5 kHz is extracted. This frequency band contains the micro-slip vibration characteristics caused by bolt loosening.

[0032] Sub-band Division and Energy Calculation: The 0 - 5 kHz band is divided into 5 equal-width sub-bands (0 - 1 kHz, 1 - 2 kHz, 2 - 3 kHz, 3 - 4 kHz, 4 - 5 kHz), and each sub-band has a width of 1 kHz.

[0033] Calculate the energy proportion of each sub-band: For example, within a certain time window, if the energy of the 3 - 4 kHz sub-band accounts for 40% of the total energy of 0 - 5 kHz, it indicates that there are significant vibration anomalies in this frequency band.

[0034] Energy Entropy Calculation: Based on the principle of Shannon entropy, quantify the degree of dispersion of the energy distribution of each sub-band. The higher the energy entropy value, the more dispersed the energy distribution, which may indicate the superposition of multi-band vibrations caused by loosening. For example, the entropy value of a normal bolt is usually lower than 1.8, while it may exceed 2.5 in the loosened state.

[0035] Feature Vector Generation: Each time window outputs a 6-dimensional vector, including the energy proportions of 5 sub-bands (such as [0.1, 0.2, 0.3, 0.25, 0.15]) and 1 energy entropy value (such as 2.7).

[0036] This vector forms a continuous frequency-domain feature stream through time series splicing for subsequent graph neural network analysis of bolt status.

[0037] According to the temperature distribution image collected by the infrared thermal imaging unit, use the bicubic interpolation algorithm to improve the spatial resolution, calculate the radial and axial temperature gradients of the bolt joint surface, and obtain the temperature gradient feature matrix; The infrared thermal imaging unit collects the temperature distribution image of the bolt connection area at a rate of 1 frame per second, and the original resolution is 640×480 pixels. The resolution is increased to 1280×960 pixels through bicubic interpolation to accurately capture the details of the temperature gradient.

[0038] Interpolation Processing: Interpolation principle: Perform bicubic polynomial fitting on the 16×16 neighborhood (original image) of each low-resolution pixel to calculate the temperature value of the high-resolution pixel. For example, if the temperature distribution in a certain area of the original image is 25°C to 30°C, small differences such as 26.5°C can be identified after interpolation.

[0039] Effect improvement: After interpolation, the edge clarity of the contact surface between the bolt head and the flange in the image is improved, and the transition in the temperature jump area (such as the friction heat generation point) is smoother, reducing the gradient calculation error.

[0040] Temperature gradient calculation: Radial gradient: With the bolt axis as the center, calculate the temperature change rate (unit: °C / mm) every 0.5 mm along the radial direction. For example, the temperature gradient at the edge of a certain bolt reaches 1.2 °C / mm due to friction, which is significantly higher than 0.3 °C / mm in the central area.

[0041] Axial gradient: Measure the temperature change every 1 mm along the length of the bolt. For example, the axial gradient near the flange surface is 0.8 °C / mm, while the gradient in the area far from the flange drops to 0.2 °C / mm, reflecting the attenuation characteristics of heat conduction along the bolt.

[0042] Feature matrix construction: Divide the bolt connection surface into a 10×10 grid (a total of 100 grids), and store the radial and axial gradient values in each grid. For example, the gradient value of a certain grid is [1.2, 0.8], indicating a radial gradient of 1.2 °C / mm and an axial gradient of 0.8 °C / mm.

[0043] Finally, generate a 20×10 temperature gradient feature matrix (each grid contains 2 gradient values), comprehensively characterizing the thermal distribution anomalies on the bolt connection surface.

[0044] According to the strain data measured by the fiber Bragg grating sensor, combined with the temperature gradient feature matrix, calculate the stress fluctuation coefficient caused by temperature changes to obtain the temperature-stress coupling feature vector; Specifically, according to the strain data measured by the fiber Bragg grating sensor, combined with the temperature gradient feature matrix, calculating the stress fluctuation coefficient caused by temperature changes to obtain the temperature-stress coupling feature vector may include: According to the original wavelength shift data collected by the fiber Bragg grating sensor, calculate the strain value of each measurement point through the Bragg wavelength demodulation algorithm to obtain the initial strain distribution data; The core principle of fiber Bragg grating sensors is to reflect the strain changes of the measured object through the shift of the Bragg wavelength. Each fiber Bragg grating sensor is preset with a reference Bragg wavelength (e.g., 1550 nm) at the time of factory. When the bolted connection structure deforms, the change in the period of the fiber Bragg grating will cause the reflection wavelength to shift. Assuming that the calibration parameter of the sensor is that a wavelength shift of 1 picometer (pm) corresponds to 1 microstrain (με), the strain value can be directly converted by measuring the wavelength shift.

[0045] The specific process of the Bragg wavelength demodulation algorithm is as follows: Spectral data acquisition: Use a high-precision spectrometer (resolution 0.1 pm) to scan the reflection spectrum of the fiber Bragg grating and capture the peak wavelength. For example, in a certain measurement, the reference wavelength is 1549.8 nm and the current peak wavelength is 1550.3 nm, then the total shift is 0.5 nm (i.e., 500 pm).

[0046] Strain calculation: According to the calibration coefficient (1 pm = 1με), a 500 pm shift corresponds to a strain of 500 με.

[0047] Preliminary temperature compensation correction: Since the fiber Bragg grating is sensitive to temperature (temperature coefficient is about 10 pm / °C), the ambient temperature change needs to be recorded. For example, if the temperature increases by 2°C, then 20 pm (2°C × 10 pm / °C) needs to be deducted from the total shift, and the remaining 480 pm corresponds to an effective strain of 480 με.

[0048] By processing all measurement points one by one (for example, 50 fiber Bragg grating sensors are arranged at the bolt connection), the initial strain distribution data is generated. For example, the strain at measurement point A is 300 με, and at measurement point B is -150 με (negative value indicates compressive strain).

[0049] According to the initial strain distribution data, the spatial Kalman filtering algorithm is used to eliminate measurement noise and obtain a smoothed strain field matrix; The spatial Kalman filtering algorithm suppresses random noise by fusing the spatial correlation of adjacent measurement points. Assume that the measurement points are distributed in a 5×10 grid, and the state vector of each measurement point includes the strain value and its first derivative (strain change rate). The filtering process is divided into two stages: prediction and update: Prediction stage: According to the strain value and change rate at the previous moment, predict the current strain. For example, the strain at measurement point A at the previous moment is 300 με and the change rate is 10 με / s, then the predicted current strain is 310 με.

[0050] Considering the influence of the spatial neighborhood, if the strain at adjacent measurement point B is -150 με, then the predicted value is adjusted through a weight coefficient (such as 0.2), and the final predicted strain = 310 με × 0.8 + (-150 με) × 0.2 = 218 με.

[0051] Update stage: The predicted value and the actual measured value (e.g., the actually measured value is 320 με) are weighted and averaged. The Kalman gain is set to 0.7, then the updated strain = 218 με × 0.3 + 320 με × 0.7 = 289.4 με.

[0052] Update the covariance matrix to reduce the influence of noise.

[0053] After 10 iterations of filtering, the random noise (such as ±50 με fluctuations) in the original data is suppressed to within ±5 με, generating a smooth strain field matrix. For example, the strain at measurement point A converges from 320 ± 50 με to 295 ± 5 με, and the strain at measurement point B converges from -160 ± 50 με to -155 ± 5 με.

[0054] According to the temperature gradient feature matrix provided by the infrared thermal imaging unit, extract the temperature change rate of the corresponding measurement points, and establish a temperature-strain mapping relationship table in combination with the strain field matrix; The infrared thermal imaging unit acquires the temperature distribution image of the bolted joint surface at a speed of 5 frames per second (resolution 0.1 °C). Align the temperature data with the fiber Bragg grating measurement point positions through image registration technology. For example, measurement point A corresponds to the image coordinates (x = 120, y = 80), and the temperature data at this position rises from 30 °C to 32 °C (change rate 0.4 °C / minute).

[0055] Construction process of the temperature-strain mapping relationship table: Time alignment: Align the temperature data and the strain data according to the millisecond-level time stamps. For example, at a certain moment t, the temperature at measurement point A is 31.5 °C and the strain is 295 με.

[0056] Change rate calculation: Statistically calculate the temperature change rate within every 10-second window. For example, if the temperature rises from 31 °C to 31.5 °C in 30 seconds, the change rate is 1 °C / minute.

[0057] Filling the mapping table: Store the data in five columns according to the measurement point number, time stamp, temperature value, temperature change rate, and strain value, as shown in Table 1 for example: Table 1

[0058] By analyzing the mapping table, the correlation law between temperature and strain can be found. For example, for every 1 °C increase in temperature, the strain at measurement point A increases by 20 με (thermal expansion effect), while for measurement point B, it shows -10 με (compression) due to structural constraints.

[0059] According to the temperature-strain mapping relationship table, fit the temperature compensation coefficient by the least squares method, calculate the effective stress fluctuation caused by the pure mechanical load, and finally generate the temperature-stress coupling eigenvector.

[0060] The least squares method is used to separate the temperature effect and the mechanical load effect. Assume that the strain (ε) is linearly superimposed by the temperature strain (ε t ), and the mechanical strain (ε m ): ε = ε t + ε m = αΔT + ε m , where α is the temperature compensation coefficient and ΔT is the temperature change.

[0061] Fitting process: Data sampling: Select 100 groups of temperature-strain data (ΔT ranges from -5°C to +10°C, with a step size of 0.5°C).

[0062] Linear regression: Fit the relationship between ε and ΔT by the least squares method to obtain the slope α. For example, the fitting result of measuring point A is α = 20 με / °C, R² = 0.95 (high correlation).

[0063] Temperature compensation: Subtract the temperature effect from the total strain. For example, at a certain moment, the temperature of measuring point A rises by 2°C, then calculate the temperature strain ε t = 20 με / °C × 2°C = 40 με. If the measured total strain ε = 300 με, then calculate the mechanical strain as ε m = 300 - 40 = 260 με.

[0064] Calculation of effective stress fluctuation: According to Hooke's law, the stress σ = Ε×ε m , where Ε is the elastic modulus of the material (200 GPa for steel). For example, ε m = 260 με = 260×10 -6 , then σ = 200×10 9 Pa × 260×10 -6 = 52 MPa.

[0065] The finally generated temperature-stress coupling eigenvector contains the following dimensions: temperature compensation coefficient α (unit με / °C); mechanical strain ε m (unit με); effective stress σ (unit MPa); temperature change rate (unit °C / min).

[0066] For example, the eigenvector of a measuring point is [20, 260, 52, 1.0], indicating that the stress caused by the current mechanical load is 52 MPa, and the temperature is rising at a rate of 1°C per minute.

[0067] According to the time stamps of the vibration frequency domain eigenvector, the temperature gradient eigenmatrix, and the temperature-stress coupling eigenvector, the dynamic time warping algorithm is used for time series alignment, and finally a multi-dimensional eigenmatrix of the bolt node is generated.

[0068] Due to the asynchronous timestamps of multimodal data caused by different sensor sampling rates (vibration signal 100 kHz, temperature 1 Hz, strain 1 kHz), dynamic time warping (DTW) is required to align the time series.

[0069] Data segmentation: Taking the 1-second interval of temperature data as the reference segment, the vibration and strain data are downsampled to 1 Hz respectively (taking the mean or peak value within each segment). For example, the vibration frequency domain feature vector is reduced from 1000 Hz to 1 Hz, and the mean entropy value of 1000 samples is taken for each segment.

[0070] DTW alignment: Pairwise alignment is performed on the three 1-Hz sequences of vibration, temperature, and strain to calculate the optimal path. For example, if the starting point of a certain temperature segment is delayed by 0.2 seconds, the strain sequence is stretched by 0.2 seconds to achieve synchronization.

[0071] Feature fusion: The aligned features are concatenated by time point: vibration (6 dimensions), temperature (20×10 = 200 dimensions), stress coupling (5 dimensions), for a total of 211 dimensions.

[0072] The finally generated multi-dimensional feature matrix is N×211, where N is the number of time points. For example, 1-hour data (3600 seconds) forms a 3600×211 matrix for subsequent calculation of the loosening probability.

[0073] By deploying a multimodal sensing network composed of piezoelectric vibration chips, infrared thermal imaging units, and fiber Bragg grating sensors, heterogeneous data such as the vibration spectrum, temperature field distribution, and structural stress of bolt connections are synchronously collected. The wavelet packet transform and short-time Fourier transform are used to jointly analyze the original signal in the time-frequency domain. The dynamic time warping algorithm is used to achieve the spatio-temporal synchronization of multi-source data, and a multi-dimensional feature matrix containing vibration energy entropy, temperature gradient coefficient, and stress coupling characteristics is constructed, breaking through the limitations of traditional single-modal detection, and comprehensively reflecting the bolt connection state through the collaborative perception of multiple physical quantities. Spatio-temporal alignment processing eliminates the acquisition time delay and spatial deviation between sensors, and frequency domain decomposition extracts the characteristic frequency bands strongly related to loosening, providing high-quality input data for subsequent intelligent diagnosis.

[0074] S202, based on the multi-dimensional feature matrix, use a graph neural network model to calculate the bolt loosening probability. By fusing the vibration frequency domain energy entropy, temperature-stress coupling coefficient, and structural topology features, dynamically weight the key nodes through the attention mechanism, and output the real-time loosening probability value of each bolt; Specifically, according to the tower structure mechanics model, a bolt connection topology relationship graph can be constructed, where the nodes of the topology relationship graph represent the bolt positions, and the edges represent the structural force transmission paths; The bolt connection topology diagram is the core model for describing the mechanical properties of the tower structure. First, based on the 3D CAD design drawings or BIM models of the tower, the spatial coordinates and connection relationships of all bolt nodes are extracted. For example, in a high-voltage transmission tower, the main material bolts are usually distributed at key positions such as tower legs and cross arms. Each bolt is an independent node in the topology diagram, and the node attributes include its installation position (such as tower leg node number A1), pre-tightening torque (such as 300 N·m), and material properties (such as 8.8-grade high-strength steel).

[0075] The construction of the force transmission path adopts the finite element analysis method: Static analysis: Apply standard loads (such as wind load, conductor tension) in software such as ANSYS to calculate the stress distribution of each bolt node; Path tracking: Identify the main force transmission paths through the stress nephogram. For example, the tower leg bolts bear vertical pressure, and the cross arm bolts bear bending moment; Edge weight assignment: Set the edge weights according to the stress transfer ratio between nodes. For example, the edge weight from node A to node B = 0.8, indicating that 80% of the load is transmitted through this path.

[0076] For complex structures, the topology relationship is calibrated using experimental data: Strain gauges are arranged on the tower prototype to measure the load distribution under different working conditions, and the edge weights of the finite element model are corrected. For example, if a certain cross arm bolt actually bears an additional 12% load under wind vibration, the corresponding edge weight is corrected from 0.75 to 0.87. The finally generated topology diagram is stored in the format of an adjacency matrix, and the matrix elements represent the force transmission weights between nodes.

[0077] According to the multi-dimensional feature matrix, the vibration frequency domain energy entropy, temperature gradient feature, and stress coupling coefficient of each bolt node are extracted as the initial node features of the graph neural network; The multi-dimensional feature matrix contains three types of core features, which need to be extracted and fused separately: Vibration frequency domain energy entropy: Data source: Time-domain signal collected by a piezoelectric vibration piece (sampling rate 10 kHz); Processing process: Decompose by wavelet packet to 8 layers to obtain 32 frequency bands (0 - 5 kHz); calculate the energy proportion of each frequency band (such as the energy of frequency band 3 accounts for 15%); calculate the energy entropy according to the information entropy formula: If the energy distribution of each frequency band is more uniform, the entropy value is higher (about 1.2 - 1.8 in the normal state, and the entropy value drops to 0.6 - 1.0 when loose because the energy is concentrated in the low-frequency band).

[0078] Temperature gradient feature: Data source: Bolt surface temperature field collected by an infrared thermal imager (resolution 0.1 °C); Processing flow: The temperature field resolution is increased to 0.05 °C by bicubic interpolation; the temperature gradient (ΔT / Δx, unit: °C / mm) is calculated along the bolt axis, which is about 0.2 - 0.5 °C / mm in the normal state; the radial gradient reflects the contact state between the bolt and the connecting plate, and the gradient is abnormal when loose (e.g., locally reaching 1.2 °C / mm).

[0079] Stress coupling coefficient: Data source: Strain data measured by fiber Bragg grating sensors (accuracy 1 με); Processing flow: Kalman filtering is used to eliminate temperature drift (e.g., compensating for 50 με of strain at 25 °C); the mechanical stress σ = E×ε is calculated (E is the elastic modulus, taking 210 GPa for steel); combining the temperature gradient data, the temperature-stress coupling coefficient α = Δσ / ΔT is calculated (the normal value is about 0.05 MPa / °C, and when loose, due to the change in contact thermal resistance, α increases to 0.1 - 0.15 MPa / °C).

[0080] Feature fusion: The feature vector of each node is three-dimensional (energy entropy, axial gradient, α coefficient). For example, the feature vector of node A is [0.73, 0.48 °C / mm, 0.12 MPa / °C], indicating that it is in a potential loose state.

[0081] According to the node features and topological relationships, the multi-head attention mechanism is used to calculate the feature propagation weights between nodes, highlighting the nodes with abnormal vibration energy and sudden changes in temperature gradient; The multi-head attention mechanism (Multi-Head Attention, MHA) is used to capture the dynamic associations between nodes. Taking 4-head attention as an example: Query-Key-Value generation: The feature vector of each node generates Query, Key, and Value matrices through linear transformation, with a dimension of 64×4; for example, the Query vector of node A is [0.32, -0.15,..., 0.07] (64-dimensional).

[0082] Calculation of attention scores: Calculate the attention score of node A to node B: ; if the vibration energy entropy of node B is abnormal (e.g., 0.5) and the temperature gradient changes suddenly (1.1 °C / mm), its Key vector has high values in the corresponding dimensions, resulting in a significant increase in Score.

[0083] Weight assignment and aggregation: The scores are normalized by Softmax to generate attention weights (e.g., the weight of node A to B is 0.8, and to C is 0.2); Weighted aggregation of Value vectors: The new feature of node A = 0.8×Value_B + 0.2×Value_C.

[0084] Abnormal Node Enhancement: Vibration Energy Entropy Threshold: Set the entropy value < 1.0 as abnormal, and the Query vector of the corresponding node is given a higher learning rate during training; Temperature Gradient Mask: When the gradient > 0.8℃ / mm, force the weight of its Key vector to increase by 50%.

[0085] For example, the vibration entropy of a certain cross-arm bolt (node D) is 0.6 and the temperature gradient is 1.0℃ / mm. Its average weight in 4 attention heads reaches 0.7, significantly affecting the feature propagation of adjacent nodes.

[0086] According to the weighted aggregated node features, combined with the theoretical calculation formula of pre-tightening force, calculate the real-time loosening probability value of each bolt through a fully connected neural network.

[0087] Specifically, according to the weighted aggregated node features, combined with the theoretical calculation formula of pre-tightening force, calculating the real-time loosening probability value of each bolt through a fully connected neural network can include: Extract three key indicators: vibration energy entropy, temperature-stress coupling coefficient, and topological connection strength from the weighted node feature vector output by the graph neural network; Three core indicators need to be extracted from the weighted node feature vector output by the graph neural network to quantify the loosening risk of the bolt.

[0088] Vibration Energy Entropy: Data Source: The time-domain vibration signal collected by the piezoelectric vibration piece is decomposed into 8 sub-frequency bands (such as 0 - 625Hz, 625 - 1250Hz, etc.) through wavelet packet transform, and the energy proportion of each frequency band is calculated.

[0089] Entropy Value Calculation: Assume that the energy distribution of a certain bolt in the sub-frequency band is [0.3, 0.2, 0.1, 0.05, 0.05, 0.1, 0.1, 0.1]. Calculate its energy entropy as 1.85 through the Shannon entropy formula (the formula is not expanded). The higher the value, the more dispersed the energy distribution, which may reflect the complication of the vibration mode caused by loosening.

[0090] Abnormal Detection: Historical data shows that the energy entropy range of normal bolts is 1.2 - 1.6. If the current value exceeds 1.8, it is marked as abnormal.

[0091] Temperature-Stress Coupling Coefficient: Data Fusion: After aligning the strain data measured by the fiber Bragg grating sensor with the temperature gradient matrix of the infrared thermal imaging, fit the temperature compensation coefficient by the least squares method. For example, when the temperature of a certain bolt increases by 10℃, the strain increases by 0.02%. After deducting the temperature effect, the pure mechanical strain is 0.15%.

[0092] Coupling coefficient calculation: Define the coefficient = mechanical strain / temperature strain. If the coefficient > 5 (e.g., 0.15% / 0.02% = 7.5), it indicates that the mechanical load is dominant and the stress state of the bolt is stable; if the coefficient < 2, the temperature influence is significant and key monitoring is required.

[0093] Topological connection strength: Structural analysis: Extract the connection degree of bolt nodes based on the iron tower BIM model (e.g., if a bolt connects 3 crossbeams, the connection degree is 3).

[0094] Dynamic weight: Combine the attention weight of the graph neural network (e.g., 0.92). If the attention weight of high-connection-degree bolts (connection degree ≥ 4) is lower than 0.5, it is determined that their topological importance has decreased, which may be due to loosening causing a change in the force transmission path.

[0095] Example: The vibration energy entropy of a certain bolt is 1.9 (abnormal), the temperature-stress coupling coefficient is 6.2 (normal), and the topological connection strength weight is 0.45 (low), comprehensively indicating a relatively high risk of loosening.

[0096] According to the theoretical formula of pre-tightening force, convert the stress fluctuation data in the node characteristics into the percentage of pre-tightening force loss to obtain the theoretical loosening degree reference value; Pre-tightening force loss is a direct manifestation of bolt loosening, and it is necessary to convert the real-time monitored stress fluctuation into the theoretical loosening degree.

[0097] Stress-pre-tightening force mapping: Material parameters: Assume that the bolt material is 8.8-grade steel, the elastic modulus is 210 GPa, and the initial pre-tightening force design value is 50 kN.

[0098] Application of the theoretical formula: According to Hooke's law, stress σ = strain ε × elastic modulus E. For example, if the measured strain reduction is 0.05% (i.e., ε = 0.0005), then the stress loss Δσ = 0.0005 × 210 GPa = 105 MPa.

[0099] Calculation of pre-tightening force loss: The bolt cross-sectional area A = 200 mm², the pre-tightening force loss ΔF = Δσ × A = 105 MPa × 200 mm² = 21 kN, and the loss percentage = 21 kN / 50 kN = 42%.

[0100] Temperature compensation correction: Influence of thermal expansion: When the bolt temperature rises by 10 °C, the thermal expansion strain ε_thermal = α × ΔT (α = 12e-6 / °C), ε_thermal = 12e-6 × 10 = 0.012%.

[0101] Effective mechanical strain: Among the total strain of 0.05%, after deducting the thermal strain of 0.012%, the actual mechanical strain is 0.038%. The corresponding pre-tightening force loss is corrected to ΔF = 0.00038 × 210 GPa × 200 mm² = 15.96 kN, and the loss percentage is 31.9%.

[0102] Generation of looseness reference value: Classification standard: Preset threshold: Loss < 20% is normal (green), 20% - 40% is warning (yellow), > 40% is high risk (red).

[0103] Dynamic adjustment: Combining historical data, if a certain type of bolt frequently fails when the loss is 35%, the threshold of this node will be lowered to 30%.

[0104] Example: The theoretical looseness reference value of a certain bolt is 35%, falling into the yellow warning range, triggering subsequent neural network analysis.

[0105] Input the theoretical looseness reference value and weighted node features into a three-layer fully connected neural network. The first layer performs feature fusion, the second layer adds physical constraint conditions, and the third layer outputs probability values; The design of the fully connected neural network (FCNN) needs to integrate data-driven and physical laws: The first layer (feature fusion layer): Input dimension: 5 features (vibration energy entropy, temperature-stress coefficient, topological connection strength, theoretical looseness, historical failure frequency).

[0106] Processing process: Use 128 neurons and the activation function ReLU. For example, the input of a certain node is [1.9, 6.2, 0.45, 35%, 2 times / year], and after full connection, a high-dimensional feature vector is output.

[0107] Feature enhancement: Perform non-linear transformation (such as square term) on vibration energy entropy and theoretical looseness to capture the coupling effect between them.

[0108] The second layer (physical constraint layer): Injection of constraint conditions: Irreversibility of pre-tightening force: If the theoretical looseness > 100%, force the output probability to be 1.0; Lower limit of energy entropy: If the energy entropy < 0.5 (sensor failure), set the probability to 0.

[0109] Network structure: 64 neurons, activation function ReLU, and weight initialization is constrained (such as the weight of theoretical looseness is not less than 0.3).

[0110] The third layer (probability output layer): Binary classification processing: single neuron output, and the Sigmoid activation function maps the value to 0~1.

[0111] Training data: 5000 groups of historical samples are used (3000 normal and 2000 loose), with the cross-entropy loss function and the Adam optimizer (learning rate 0.001).

[0112] Example inference: After the input features are processed by the first three layers, the output probability is 0.78. If it exceeds the threshold of 0.65, it is determined as a high risk.

[0113] Model verification: Confusion matrix test: accuracy rate is 92%, recall rate is 88%; Real-time performance: the inference time of a single node is <3ms, meeting the requirements of online monitoring.

[0114] The output value of the neural network is mapped to the 0-1 interval through the Sigmoid activation function, and the probability threshold is calibrated in combination with historical fault data, and finally the real-time loosening probability value of each bolt is generated.

[0115] Probability threshold calibration is the key to balancing false alarms and missed alarms: Sigmoid mapping: Function: Compress the original output of the neural network (such as 2.3) to 0~1. For example, 2.3→0.91, -1.2→0.23.

[0116] Probability interpretation: 0.8 indicates an 80% possibility of loosening, and it is necessary to combine the threshold for judgment.

[0117] Threshold dynamic calibration: ROC curve analysis: Draw the curve through historical data, select the optimal threshold (such as 0.65) to make the true positive rate (TPR) reach 85% and the false positive rate (FPR) <10%.

[0118] Working condition adaptability: During extreme weather (such as typhoons), the threshold is temporarily lowered to 0.6 to improve the early warning sensitivity.

[0119] Probability value post-processing: Time smoothing: Take the moving average of the probability values for 10 consecutive seconds (such as [0.7, 0.72, 0.75, 0.8]→0.74) to avoid false alarms caused by instantaneous fluctuations.

[0120] Spatial correlation: If the probabilities of adjacent bolts are all >0.6, trigger a regional-level early warning (such as the risk of loosening of the entire flange).

[0121] Example output: Bolt A: probability 0.73→exceeds the threshold 0.65→generate a high-risk early warning; Bolt B: probability 0.58→below the threshold→marked as under monitoring; Historical calibration: After adjusting the threshold, the missed alarm rate is reduced from 12% to 6%.

[0122] Construct a topological relationship model of the bolt connection system based on graph neural networks, taking features such as vibration energy entropy and temperature-stress coupling coefficient as node attributes, and dynamically capturing the abnormal feature propagation paths of key nodes through the attention mechanism. Integrate the prior knowledge of structural mechanics, calculate the loosening probability of each node by means of weighted aggregation, realize the collaborative diagnosis of the entire connection system rather than isolated judgment, use the graph structure to model the mechanical correlation between bolts, and the attention mechanism can automatically focus on high-risk nodes to avoid missing local anomalies detected by traditional methods. Enhance the mutual verification ability of the states of adjacent bolts through topological feature propagation, and significantly improve the accuracy and robustness of loosening identification.

[0123] S203. According to the real-time loosening probability value, combined with the historical working condition data of the iron tower and the prediction result of the environmental load time series, use the physics-informed LSTM network to dynamically correct the loosening risk level and generate a risk level map based on time evolution, where the risk level map includes the risk evolution trend of each bolt node within a preset future time period. Specifically, according to the historical cases of bolt loosening recorded in the operation and maintenance database, extract the environmental parameters and equipment state characteristics during loosening to construct a historical working condition feature library. The construction of the historical working condition feature library requires integrating multi-source heterogeneous data, including bolt loosening event records, environmental sensor data, and equipment operation logs. First, extract the original records of all loosening cases in the past 5 years from the operation and maintenance database, and each case includes: Environmental parameters: wind speed (0~30m / s), temperature (-20℃~50℃), humidity (10%~100%RH); Equipment state: vibration acceleration (0~10g), preload loss rate (0%~100%), bolt torque value (100~500Nm); Timestamp: time series data before and after the loosening occurs (sampling frequency 1Hz).

[0124] Data preprocessing process: Missing value filling: Use the KNN (K-nearest neighbor, K = 5) algorithm to fill in the missing values according to the mean of similar working condition samples. Outlier removal: Based on the 3σ principle, remove the data points that exceed the mean ± 3 times the standard deviation. Feature standardization: Perform Z-score normalization on parameters with different dimensions such as wind speed and temperature.

[0125] Feature engineering: Time domain features: Calculate the root mean square (RMS) and peak-to-peak value (P2P) of the vibration signal. Frequency domain features: Extract the frequency centroid (FC) of the 0-1kHz frequency band through FFT (Fast Fourier Transform). Coupling feature: Calculate the Pearson correlation coefficient between the temperature gradient and the vibration energy (a threshold > 0.6 indicates a strong correlation).

[0126] The finally constructed historical operating condition feature library contains more than 10,000 samples, and each sample is represented by a 15-dimensional feature vector. For example: [wind speed 6.5 m / s, temperature 25 °C, humidity 60%, RMS = 2.3 g, P2P = 8.7 g, FC = 320 Hz, correlation coefficient 0.72].

[0127] Based on the meteorological forecast data, a temporal convolutional network is used to predict the changing trends of wind speed, temperature, and humidity in the next 24 hours, obtaining the environmental load prediction curve; The temporal convolutional network (TCN) adopts a causal dilated convolutional structure, and its core advantage lies in the ability to capture long-term temporal dependencies. The network configuration is as follows: Input layer: Meteorological data (wind speed, temperature, humidity) in the past 72 hours, with a time resolution of 1 hour; Convolutional layer: 4 layers of dilated convolution, with dilation coefficients of 1, 2, 4, and 8 respectively, 64 filter numbers, and a kernel size of 3; Output layer: Predicted values of the three variables (wind speed, temperature, humidity) in the next 24 hours.

[0128] Training details: Loss function: Weighted sum of MAE (Mean Absolute Error) and RMSE (Root Mean Square Error) (weight 0.7:0.3); Optimizer: AdamW (learning rate 0.001, weight decay 0.01); Regularization: Dropout rate 0.2 to prevent overfitting.

[0129] Example of the prediction process: Input data in the past 72 hours: Wind speed sequence [5.2, 5.8,..., 6.1] m / s, temperature sequence [22, 23,..., 25] °C, humidity sequence [55%, 58%,..., 60%]; TCN outputs the prediction in the next 24 hours: Peak wind speed 8.3 m / s (at the 12th hour), temperature fluctuation range 23 °C to 28 °C, humidity rising to 70%; Post-processing: Smooth the prediction curve through Kalman filtering to reduce sudden noise.

[0130] The prediction results are stored in the form of a time series matrix with a time resolution of 15 minutes for subsequent risk prediction calls.

[0131] Based on the real-time loosening probability value, the historical operating condition feature library, and the environmental load prediction curve, input them into a physically informed LSTM network for temporal prediction; The Physics Information Enhanced LSTM (PI-LSTM) introduces a physical constraint layer on the basis of the traditional LSTM to ensure that the prediction results conform to the mechanical laws. The network structure is as follows: Input layer: Real-time loosening probability value (0 - 1), historical feature similarity (nearest neighbor matching score with K = 10), predicted environmental load values (wind speed, temperature, humidity); LSTM layer: Stacked 2 layers, with 128 units in each layer and a dropout rate of 0.3; Physical constraint layer: Embed the pre-tightening force decay model. When predicting the loosening probability P, simultaneously calculate the theoretical pre-tightening force loss ΔF = αP (α = 0.85 is the material decay coefficient). If ΔF exceeds the threshold (e.g., 30%), forcefully correct the prediction value; Output layer: Loosening risk level (0 - 5 levels) for each hour in the next 24 hours.

[0132] Training strategy: Joint loss function: Prediction error: Cross-entropy loss (weight 0.6); Physical constraint loss: |ΔF measured - ΔF predicted| (weight 0.4); Data augmentation: Generate extreme working condition data (e.g., wind speed 15 m / s + humidity 90%) through GAN to improve the robustness of the model; Transfer learning: Use the historical feature library in the pre-training stage and add real-time data in the fine-tuning stage.

[0133] Prediction example: Input: Real-time loosening probability 0.35, similar working condition matching score 0.82, predicted wind speed 7.2 m / s; LSTM output: The risk level in the next 6 hours rises from level 2 to level 4; The physical constraint layer detects that ΔF = 0.35 × 0.85 = 29.75%, approaching the threshold of 30%, triggering the risk level to be revised upwards to level 5.

[0134] According to the output results of the LSTM network, generate the risk level change curves of each bolt node within the next 24 hours and construct a three-dimensional spatio-temporal risk level map.

[0135] The construction of the three-dimensional spatio-temporal risk level map is divided into two stages: data fusion and visualization rendering: Data fusion: Spatial mapping: According to the GPS coordinates of the bolts on the iron tower (accuracy ±0.1 m), map the risk level to a three-dimensional grid (grid size 0.5 m × 0.5 m × 0.5 m); Temporal interpolation: Perform cubic spline interpolation on the hourly risk levels output by the LSTM to generate a minute-level continuous curve; Heat map generation: Define color gradients based on risk level values (0 - 5): Level 0 (green, RGB(0, 255, 0)); Level 3 (yellow, RGB(255, 255, 0)); Level 5 (red, RGB(255, 0, 0)).

[0136] Visual rendering: Engine selection: Use the Unity3D engine, supporting WebGL export; Dynamic interaction: Clicking on the bolt node can pop up detailed prediction data (such as the risk trend in the next 6 hours); Multi - perspective display: Support top - down view, side view, and structural section view, and can overlay and display the wind speed vector field.

[0137] Example atlas: The predicted risk level of the #32 bolt node of a transmission tower is Level 3 (yellow) at 14:00, rising to Level 5 (red) by 18:00. The 3D model simultaneously shows that the surrounding wind speed increases from 6 m / s to 9 m / s; Operation and maintenance personnel can view the risk diffusion path through VR devices and lock in high - risk areas in advance.

[0138] Combining the historical working condition library and environmental load prediction data, embed the physical evolution law of bolt loosening (such as the vibration - temperature - stress coupling equation) in the LSTM network to dynamically correct the current loosening probability. Generate the future risk evolution curve through time - series prediction, upgrade the static detection to a dynamic risk assessment including the time dimension. The physical information enhancement mechanism makes the prediction results conform to mechanical laws and avoids the over - fitting risk of pure data - driven methods. The spatio - temporal risk atlas can predict the development trend of loosening, support the formulation of preventive maintenance decisions, and greatly improve the foresight of the monitoring system.

[0139] S204, real - time map the risk level atlas through a 3D visualization interface. When it detects that the bolt loosening risk level exceeds the preset threshold, generate a warning message and upload it to the operation and maintenance platform to complete the closed - loop monitoring response.

[0140] Specifically, a digital twin can be established on the 3D visualization platform according to the tower BIM model and bolt position information; BIM model parsing and lightweight processing: Extract the original BIM (Building Information Model) data from the tower design file in the IFC (Industry Foundation Classes) format. Extract the geometric coordinates (accuracy ±1 mm), material properties (such as the tensile strength of M20 bolts is grade 8.8), and connection relationships of bolt nodes through BIM parsing tools (such as IFC Open Shell).

[0141] Lightweight processing of the model: The mesh simplification algorithm (such as Quadric Edge Collapse Decimation) is used to reduce the number of triangular patches from the 100,000 level to the 10,000 level to ensure smooth rendering in the WebGL engine. At the same time, key topological information is retained, such as the bolt hole pitch (standard 50mm) and flange thickness (20mm).

[0142] Digital twin construction and data mapping: Import the lightweight model into the Unity3D or Three.js engine, and bind a unique ID (such as Bolt_001) to each bolt node. The ID is associated with multi-source sensor data, including the vibration signal acquisition timestamp (such as 2023-10-05T14:30:00Z), temperature gradient (such as axial gradient 2.5℃ / m), and stress coupling coefficient (0.85).

[0143] Establish a dynamic data channel: Receive the risk level data (JSON format) from the backend server in real time through the WebSocket protocol, and the data update frequency is 1Hz.

[0144] Spatial coordinate calibration and fault tolerance mechanism: Use the ICP (Iterative Closest Point) algorithm to align the BIM model with the actual point cloud scan data (such as LiDAR acquisition) to eliminate installation errors (such as bolt hole position offset ≤ 2mm).

[0145] Set the tolerance threshold: If the deviation between the sensor data and the model position exceeds 5mm, trigger an abnormal alarm and start the manual review process.

[0146] According to the risk level atlas data, dynamically render the risk status of each bolt node on the digital twin, and use red, yellow, and green to represent different risk levels; Risk level color coding rules: Red (high risk): Loosening probability ≥ 0.8 or temperature gradient ≥ 5℃ / m; Yellow (medium risk): 0.5 ≤ loosening probability < 0.8 and temperature gradient < 5℃ / m; Green (low risk): Loosening probability < 0.5 and temperature gradient < 2℃ / m.

[0147] Implementation of dynamic rendering technology: Shader programming: Write a custom shader on the GPU side, and calculate the color interpolation in real time according to the risk value. For example, use HSL (hue, saturation, brightness) space interpolation: red (H = 0°), yellow (H = 60°), green (H = 120°), and the risk value 0~1 is mapped to the 0°~120° interval of the H value.

[0148] Particle effect enhancement: For high-risk bolts, a particle system (such as a red pulsating halo) is superimposed. The particle emission frequency is positively correlated with the risk value (e.g., when the risk is 0.9, the frequency is 10 Hz).

[0149] Multi-level of detail (LOD) optimization: Near perspective (<10 m): Display the complete bolt model and the floating risk label (such as "Risk: 0.92"); Far perspective (≥10 m): Simplify to a colored cube marker, and the size increases with the risk value (e.g., the high-risk marker is enlarged to 1.2 times).

[0150] Cross-platform compatibility guarantee: For mobile devices (iOS / Android), use the compressed texture format (ASTC 4x4) to control the rendering latency within 30 ms; The desktop version supports VR headsets (such as Oculus Rift), and reduces the sense of dizziness through asynchronous time warping (ATW) technology.

[0151] According to the preset warning threshold, automatically detect high-risk bolt nodes and generate a warning message containing the position coordinates, risk level, and maintenance suggestions; Warning trigger logic and threshold setting The system dynamically triggers different levels of alarms by continuously monitoring the loosening probability value (range 0 - 1) and the temperature gradient change rate (unit: °C / m / min) of the bolts, in combination with the preset multi-level warning thresholds: Level 1 warning (urgent): Triggered when the loosening probability value of a certain bolt is ≥0.9 for 3 consecutive samples (each interval is 1 second), or the temperature gradient suddenly increases by more than 5 °C / m within 1 minute. For example, if the loosening probability values of a certain bolt are 0.91, 0.93, and 0.95 at 14:30:00, 14:30:01, and 14:30:02 respectively, then the level 1 warning is immediately activated.

[0152] Level 2 warning (severe): The single-sample loosening probability ≥0.8, or the temperature gradient change rate ≥2 °C / m / min (e.g., it only takes 1 minute to rise from 2 °C / m to 4 °C / m).

[0153] Level 3 warning (prompt): The loosening probability ≥0.6 and has not decreased for 10 minutes (e.g., continuously higher than 0.6 from 14:30 to 14:40).

[0154] Warning message content generation When the threshold conditions are met, the system automatically generates a structured warning message, including the following core information: Position coordinates: Precise positioning through the three-dimensional coordinates of the bolts (such as X = 125.6 m, Y = 34.8 m, Z = 89.2 m) bound in the BIM model, with an accuracy of up to millimeters.

[0155] Risk level: Labeled as "High risk (0.92)", "Medium risk (0.75)", etc. according to the threshold matching result, and the value is reserved to two decimal places.

[0156] Maintenance suggestions: Match historical work order data from the knowledge graph library. For example: If the temperature gradient > 4°C / m, it is recommended to "check the flange seal and replace the aging gasket".

[0157] Recommend torque values (such as "Immediately tighten to 120 N·m ± 5%") in combination with bolt specifications (such as M20 bolts) to ensure compliance with mechanical standards.

[0158] Additional data: Associate the infrared thermal imaging map at the current moment (such as Bolt001_thermal_image.png) for the operation and maintenance personnel to visually view the temperature abnormal area.

[0159] Push the message to the corresponding operation and maintenance terminal according to the warning level, and track the disposal feedback information to complete the closed-loop process of monitoring - warning - disposal.

[0160] Hierarchical push strategy: Level 1 warning: Triggered simultaneously by SMS, App push (high-priority notification) and audible and visual alarms. If not confirmed within 10 seconds, it will automatically escalate to the operation and maintenance supervisor; Level 2 warning: App push + email notification, with a response timeout threshold of 30 minutes; Level 3 warning: Only recorded in the work order system, and a patrol task will be generated the next day.

[0161] Feedback tracking and work order management: Work order state machine: To be accepted → Assigned → In processing → To be inspected → Closed; Mobile integration: The operation and maintenance personnel can quickly locate by scanning the bolt QR code (encoding format QR Code Model 2) through the App and upload the vibration spectrum comparison map before and after processing (such as the energy entropy drops from 0.9 to 0.3 after processing).

[0162] Closed-loop verification mechanism: Automatic review: After the warning is closed, the system continuously monitors the data of this bolt in the next 1 hour. If the risk value does not drop to the safe range (<0.5), a new work order will be automatically opened; Manual spot check: Randomly select 5% of the closed work orders and require the upload of a torque wrench calibration report (PDF format, including the calibration certificate number).

[0163] Data archiving and knowledge iteration: Warning data is stored in a time series database (such as InfluxDB) with a retention period of 10 years; The risk prediction model is updated quarterly through offline reinforcement learning, such as adding the characteristics of bolt loosening patterns under typhoon weather.

[0164] Build a 3D visualization platform based on digital twin technology, and map the risk level to the corresponding position of the tower BIM model. Intuitively display the risk distribution using color coding. When the detected value exceeds the threshold, automatically trigger a hierarchical early warning mechanism, generate a work order containing precise positioning and maintenance suggestions, and push it to the operation and maintenance terminal to achieve the visual presentation and rapid response of the monitoring results, forming a complete closed loop from data collection to disposal feedback. 3D visualization reduces the threshold for data understanding, and the hierarchical early warning mechanism optimizes the resource allocation efficiency to ensure that high-risk nodes are given priority for disposal.

[0165] It can be seen that by collecting the original data set containing vibration signals, temperature gradient data and structural stress distribution of bolt connections through a multi-modal sensor array, performing spatio-temporal alignment and frequency-domain decomposition processing on the original data set, a multi-dimensional feature matrix of bolt nodes is obtained; based on the multi-dimensional feature matrix, a graph neural network model is used to calculate the bolt loosening probability, and the real-time loosening probability value of each bolt is output; according to the real-time loosening probability value, a risk level map based on time evolution is generated; the risk level map is mapped in real time through a 3D visualization interface. When the detected bolt loosening risk level exceeds the preset threshold, an early warning message is generated and uploaded to the operation and maintenance platform to complete the closed-loop monitoring response, so as to realize the accurate assessment, dynamic prediction and visual early warning of loosening risks, and improve the intelligent level of structural safety monitoring.

[0166] Another embodiment of the present invention provides an on-line bolt loosening monitoring system. Refer to Figure 3 , the system may include: A processing module 301, configured to collect an original data set containing vibration signals, temperature gradient data and structural stress distribution of bolt connections through a multi-modal sensor array, perform spatio-temporal alignment and frequency-domain decomposition processing on the original data set, and obtain a multi-dimensional feature matrix of bolt nodes, wherein the multi-modal sensor array includes piezoelectric vibration chips, infrared thermal imaging units and fiber Bragg grating sensors; An output module 302, configured to calculate the bolt loosening probability based on the multi-dimensional feature matrix, fuse the vibration frequency-domain energy entropy, temperature-stress coupling coefficient and structural topology features, dynamically weight key nodes through an attention mechanism, and output the real-time loosening probability value of each bolt; A correction module 303, configured to dynamically correct the loosening risk level according to the real-time loosening probability value, combine the historical working condition data of the tower and the prediction result of the environmental load time series, and use a physics-informed LSTM network to generate a risk level map based on time evolution, wherein the risk level map includes the risk evolution trend of each bolt node within a preset future time period; A generation module 304 is configured to map the risk level map in real time through a three-dimensional visualization interface. When it detects that the risk level of bolt loosening exceeds a preset threshold, it generates a warning message and uploads it to the operation and maintenance platform to complete the closed-loop monitoring response.

[0167] An embodiment of the present invention further provides a storage medium, in which a computer program is stored. Among them, the computer program is set to execute the steps in any one of the above method embodiments when running.

[0168] Specifically, in this embodiment, the above storage medium can be set to store a computer program for executing the following steps: S201, collect an original data set including vibration signals, temperature gradient data, and structural stress distribution of bolt connections through a multi-modal sensor array, perform spatio-temporal alignment and frequency-domain decomposition processing on the original data set to obtain a multi-dimensional feature matrix of bolt nodes, where the multi-modal sensor array includes piezoelectric vibration plates, infrared thermal imaging units, and fiber Bragg grating sensors; S202, based on the multi-dimensional feature matrix, use a graph neural network model to calculate the bolt loosening probability. By fusing the vibration frequency-domain energy entropy, temperature-stress coupling coefficient, and structural topology features, dynamically weight key nodes through an attention mechanism, and output the real-time loosening probability value of each bolt; S203, according to the real-time loosening probability value, combined with the historical working condition data of the iron tower and the prediction result of the environmental load time series, use a physics-informed LSTM network to dynamically correct the loosening risk level, and generate a risk level map based on time evolution, where the risk level map includes the risk evolution trend of each bolt node within a preset future time period; S204, map the risk level map in real time through a three-dimensional visualization interface. When it detects that the risk level of bolt loosening exceeds a preset threshold, generate a warning message and upload it to the operation and maintenance platform to complete the closed-loop monitoring response.

[0169] An embodiment of the present invention further provides an electronic device, including a memory and a processor. A computer program is stored in the memory, and the processor is set to run the computer program to execute the steps in any one of the above method embodiments.

[0170] Specifically, the above electronic device may further include a transmission device and an input / output device. Among them, the transmission device is connected to the above processor, and the input / output device is connected to the above processor.

[0171] Specifically, in this embodiment, the above processor can be set to execute the following steps through a computer program: S201. Collect an original data set containing vibration signals, temperature gradient data, and structural stress distribution of bolt connections through a multi-modal sensor array, and perform spatio-temporal alignment and frequency-domain decomposition processing on the original data set to obtain a multi-dimensional feature matrix of bolt nodes. Among them, the multi-modal sensor array includes a piezoelectric vibration piece, an infrared thermal imaging unit, and a fiber Bragg grating sensor; S202. Based on the multi-dimensional feature matrix, use a graph neural network model to calculate the bolt loosening probability. By fusing the vibration frequency-domain energy entropy, temperature-stress coupling coefficient, and structural topology features, dynamically weight key nodes through an attention mechanism, and output the real-time loosening probability value of each bolt; S203. According to the real-time loosening probability value, combined with the historical working condition data of the iron tower and the time series prediction result of the environmental load, use a physics-informed LSTM network to dynamically correct the loosening risk level, and generate a risk level map based on time evolution. Among them, the risk level map includes the risk evolution trend of each bolt node within a preset future time period; S204. Real-time map the risk level map through a three-dimensional visualization interface. When it is detected that the bolt loosening risk level exceeds a preset threshold, generate a warning message and upload it to the operation and maintenance platform to complete the closed-loop monitoring response.

[0172] The structure, features, and function effects of the present invention have been described in detail based on the embodiments shown in the drawings. The above is only the preferred embodiment of the present invention, but the present invention is not limited to the scope shown in the drawings. Any changes made according to the concept of the present invention, or equivalent embodiments modified into equivalent changes, should still be within the protection scope of the present invention as long as they do not exceed the spirit covered by the description and the drawings.

Claims

1. An on-line monitoring method for bolt loosening, characterized in that The method includes: Collecting an original data set containing vibration signals, temperature gradient data, and structural stress distribution of bolt connections through a multi-modal sensor array, and performing spatio-temporal alignment and frequency-domain decomposition processing on the original data set to obtain a multi-dimensional feature matrix of bolt nodes. Among them, the multi-modal sensor array includes piezoelectric vibration chips, infrared thermal imaging units, and fiber Bragg grating sensors; Based on the multi-dimensional feature matrix, using a graph neural network model to calculate the bolt loosening probability, by fusing the vibration frequency-domain energy entropy, temperature-stress coupling coefficient, and structural topology features, dynamically weighting key nodes through an attention mechanism, and outputting the real-time loosening probability value of each bolt; According to the real-time loosening probability value, combined with the historical working condition data of the iron tower and the prediction result of the environmental load time series, using a physics-informed LSTM network to dynamically correct the loosening risk level, and generating a risk level map based on time evolution. Among them, the risk level map includes the risk evolution trend of each bolt node within a preset future time period; Real-time mapping the risk level map through a three-dimensional visualization interface. When it is detected that the bolt loosening risk level exceeds the preset threshold, generating a warning message and uploading it to the operation and maintenance platform to complete the closed-loop monitoring response.

2. The method according to claim 1, characterized in that, The collecting an original data set containing vibration signals, temperature gradient data, and structural stress distribution of bolt connections through a multi-modal sensor array, and performing spatio-temporal alignment and frequency-domain decomposition processing on the original data set to obtain a multi-dimensional feature matrix of bolt nodes. Among them, the multi-modal sensor array includes piezoelectric vibration chips, infrared thermal imaging units, and fiber Bragg grating sensors, includes: According to the bolt vibration signal collected by the piezoelectric vibration chip, performing noise reduction processing using wavelet packet transform to obtain a denoised vibration time-domain signal; According to the denoised vibration time-domain signal, calculating the energy distribution in the 0-5 kHz frequency band through short-time Fourier transform, and extracting the energy ratio and energy entropy of each frequency band to obtain a vibration frequency-domain feature vector; According to the temperature distribution image collected by the infrared thermal imaging unit, using bicubic interpolation algorithm to improve the spatial resolution, and calculating the radial and axial temperature gradients of the bolt connection surface to obtain a temperature gradient feature matrix; According to the strain data measured by the fiber Bragg grating sensor, combined with the temperature gradient feature matrix, calculating the stress fluctuation coefficient caused by temperature change to obtain a temperature-stress coupling feature vector; According to the timestamps of the vibration frequency-domain feature vector, temperature gradient feature matrix, and temperature-stress coupling feature vector, using the dynamic time warping algorithm for time series alignment, and finally generating a multi-dimensional feature matrix of bolt nodes.

3. The method according to claim 2, wherein The based on the multi-dimensional feature matrix, using a graph neural network model to calculate the bolt loosening probability, by fusing the vibration frequency-domain energy entropy, temperature-stress coupling coefficient, and structural topology features, dynamically weighting key nodes through an attention mechanism, and outputting the real-time loosening probability value of each bolt, includes: According to the structural mechanics model of the iron tower, constructing a bolt connection topology relationship graph, where the nodes of the topology relationship graph represent bolt positions, and the edges represent structural force transmission paths; Extract the vibration frequency-domain energy entropy, temperature gradient feature, and stress coupling coefficient of each bolt node based on the multi-dimensional feature matrix as the initial node features of the graph neural network; Calculate the feature propagation weights between nodes using the multi-head attention mechanism according to the node features and topological relationships, highlighting the nodes with abnormal vibration energy and sudden changes in temperature gradient; Calculate the real-time loosening probability value of each bolt through a fully connected neural network based on the weighted aggregated node features in combination with the theoretical calculation formula of the pre-tightening force.

4. The method according to claim 3, wherein Based on the real-time loosening probability value, combined with the historical operating conditions data of the iron tower and the prediction results of the environmental load time series, use a physics-informed LSTM network to dynamically correct the loosening risk level and generate a risk level map based on time evolution, where the risk level map includes the risk evolution trends of each bolt node within a preset future time period, including: Extract the environmental parameters and equipment status features at the time of loosening according to the historical cases of bolt loosening recorded in the operation and maintenance database, and construct a historical operating conditions feature library; According to the meteorological forecast data, use a temporal convolutional network to predict the changing trends of wind speed, temperature, and humidity in the next 24 hours to obtain the environmental load prediction curve; Based on the real-time loosening probability value, the historical operating conditions feature library, and the environmental load prediction curve, input them into a physics-informed LSTM network for time series prediction; According to the output results of the LSTM network, generate the risk level change curve of each bolt node within the next 24 hours and construct a three-dimensional spatio-temporal risk level map.

5. The method according to claim 4, wherein The risk level map is mapped in real time through a three-dimensional visualization interface. When it is detected that the bolt loosening risk level exceeds the preset threshold, a warning message is generated and uploaded to the operation and maintenance platform to complete the closed-loop monitoring response, including: Establish a digital twin on the three-dimensional visualization platform according to the iron tower BIM model and bolt position information; Dynamically render the risk status of each bolt node on the digital twin according to the risk level map data, and use three colors of red, yellow, and green to represent different risk levels; Automatically detect high-risk bolt nodes according to the preset warning threshold, and generate a warning message including position coordinates, risk level, and maintenance suggestions; Push the message to the corresponding operation and maintenance terminal according to the warning level, and track the disposal feedback information to complete the closed-loop process of monitoring-warning-disposal.

6. The method according to claim 2, wherein According to the strain data measured by the fiber Bragg grating sensor, combined with the temperature gradient feature matrix, calculate the stress fluctuation coefficient caused by temperature changes to obtain the temperature-stress coupling feature vector, including: According to the original wavelength shift data collected by the fiber Bragg grating sensor, calculate the strain value of each measurement point through the Bragg wavelength demodulation algorithm to obtain the initial strain distribution data; According to the initial strain distribution data, use the spatial Kalman filtering algorithm to eliminate the measurement noise and obtain the smoothed strain field matrix; Extract the temperature change rate of the corresponding measurement point according to the temperature gradient feature matrix provided by the infrared thermal imaging unit, and establish a temperature-strain mapping relationship table in combination with the strain field matrix; According to the temperature-strain mapping relationship table, fit the temperature compensation coefficient by the least squares method, calculate the effective stress fluctuation caused by the pure mechanical load, and finally generate the temperature-stress coupling feature vector.

7. The method according to claim 3, characterized in that, Calculating the real-time loosening probability values of each bolt through a fully connected neural network based on the weighted aggregated node features in combination with the theoretical calculation formula of the pre-tightening force, including: Extracting three key indicators, namely, vibration energy entropy, temperature-stress coupling coefficient, and topological connection strength, from the weighted node feature vector output by the graph neural network; Converting the stress fluctuation data in the node features into the percentage of pre-tightening force loss according to the theoretical formula of the pre-tightening force to obtain the theoretical loosening degree reference value; Inputting the theoretical loosening degree reference value and the weighted node features into a three-layer fully connected neural network. The first layer performs feature fusion, the second layer adds physical constraint conditions, and the third layer outputs probability values; Mapping the output value of the neural network to the interval of 0-1 through the Sigmoid activation function, calibrating the probability threshold in combination with historical fault data, and finally generating the real-time loosening probability values of each bolt.

8. An on-line bolt loosening monitoring system, characterized in that, The system includes: A processing module, configured to collect an original data set including vibration signals, temperature gradient data, and structural stress distribution of bolt connections through a multi-modal sensor array, perform spatio-temporal alignment and frequency-domain decomposition processing on the original data set to obtain a multi-dimensional feature matrix of bolt nodes, wherein the multi-modal sensor array includes a piezoelectric vibration piece, an infrared thermal imaging unit, and a fiber Bragg grating sensor; An output module, configured to calculate the bolt loosening probability based on the multi-dimensional feature matrix by using a graph neural network model, dynamically weight key nodes through an attention mechanism by fusing vibration frequency-domain energy entropy, temperature-stress coupling coefficient, and structural topology features, and output the real-time loosening probability value of each bolt; A correction module, configured to dynamically correct the loosening risk level by using a physically informed LSTM network according to the real-time loosening probability value in combination with the historical working condition data of the iron tower and the prediction result of the environmental load time series, and generate a risk level map based on time evolution, wherein the risk level map includes the risk evolution trend of each bolt node within a preset future time period; A generation module, configured to map the risk level map in real time through a three-dimensional visualization interface, and when it is detected that the bolt loosening risk level exceeds a preset threshold, generate a warning message and upload it to the operation and maintenance platform to complete the closed-loop monitoring response.

9. A storage medium, characterized in that, A computer program is stored in the storage medium, wherein the computer program is set to execute the method according to any one of claims 1-7 when running.

10. An electronic device, comprising a memory and a processor, characterized in that, A computer program is stored in the memory, and the processor is set to run the computer program to execute the method according to any one of claims 1-7.

Images