Self-powered transmission line fitting aeolian vibration damage diagnosis system and method

Abstract

The invention relates to the technical field of vibration monitoring, in particular to a self-powered transmission line fitting aeolian vibration damage diagnosis system and method, and the system comprises a sensing collection module, a signal decoupling module, a damage identification module, a damage association module and a risk assessment module. According to the method, stress wave velocity and acceleration data are synchronously collected, time alignment is implemented, feature coupling precision is enhanced, wave crest offset and energy density are respectively extracted by using moving average filtering and wavelet transform, effective data segments are dynamically screened, and environmental noise interference is suppressed. A stress wave propagation change rate is quantified based on a path attenuation model, a continuous energy abnormal node is matched to realize damage positioning, a breeze response abnormal region is identified by combining vibration direction change and a signal envelope offset degree, multi-dimensional features are coded and subjected to risk judgment through a neural network, a damage positioning and risk assessment closed-loop framework is formed, and the risk assessment accuracy is improved. And the spatial resolution and evaluation precision of aeolian vibration damage identification under complex working conditions are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of vibration monitoring, and in particular to a micro-wind vibration damage diagnosis system and method for self-powered transmission line fittings. Background Technology

[0002] Vibration monitoring technology involves real-time or periodic monitoring of mechanical vibrations generated by various structures or equipment during operation to understand their working status and structural health. Its core aspects include vibration signal acquisition, vibration characteristic analysis and identification, fault early warning, and life assessment. This technology is widely used in industries such as power, transportation, aviation, and construction. Particularly in power transmission systems, vibration monitoring of components such as line fittings and conductors can effectively support operation and maintenance decisions and safety management, improving the reliability and stability of the transmission system.

[0003] The traditional self-powered transmission line fittings micro-wind vibration damage diagnosis system addresses the frequent micro-wind vibration problems of transmission line fittings in low-wind-speed environments. It diagnoses damage by deploying sensing devices on-site and analyzing vibration data. Traditionally, this type of system uses piezoelectric or electromagnetic energy harvesting devices to acquire minute vibration energy for local power supply. Then, it uses accelerometers to collect the minute vibration responses of the fittings and determines the vibration state through characteristic frequency extraction or time-domain fluctuation trends, thus achieving preliminary identification of fatigue damage caused by micro-wind vibration of the fittings.

[0004] Traditional methods rely on a single accelerometer to acquire vibration signals. In weak vibration scenarios, these methods are easily limited by the sensor's frequency response range, leading to the loss of high-frequency components. Characteristic frequency extraction suffers from spectral aliasing. When using time-domain fluctuation trend analysis, there is a lack of a dynamic attenuation compensation mechanism for stress wave propagation paths, making it difficult to distinguish between structural resonance and energy anomalies caused by damage. Piezoelectric energy harvesting devices experience significant output power fluctuations under low-frequency, low-amplitude vibrations, affecting the stability of the sensor's continuous sampling frequency. Damage identification based on empirical thresholds does not establish an attenuation correlation model between vibration energy distribution and damage location, resulting in positioning errors and reducing the reliability of fatigue damage assessment. Summary of the Invention

[0005] This invention addresses the technical problems of traditional methods that rely on a single accelerometer to acquire vibration signals. These problems include susceptibility to sensor frequency response limitations in weak vibration scenarios, resulting in the loss of high-frequency components, spectral aliasing in characteristic frequency extraction, lack of dynamic attenuation compensation mechanism for stress wave propagation paths when using time-domain fluctuation trend analysis, difficulty in distinguishing between structural resonance and energy anomalies caused by damage, significant output power fluctuations of piezoelectric energy harvesting devices under low-frequency micro-amplitude vibrations affecting the stability of continuous sensor sampling frequency, and the lack of an attenuation correlation model between vibration energy distribution and damage location in damage identification based on empirical thresholds, leading to positioning errors and reducing the reliability of fatigue damage assessment. The invention provides a micro-wind vibration damage diagnosis system and method for self-powered transmission line fittings.

[0006] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: On the one hand, a micro-wind vibration damage diagnosis system for self-powered transmission line fittings is provided, the system comprising: The sensor acquisition module is used to acquire stress wave velocity parameters and acceleration time series data, use a synchronizer to complete time alignment, generate time-aligned dual-mode data, and transmit it to the signal decoupling module. The signal decoupling module is used to extract stress wave parameters from the time-aligned dual-mode data, call the moving average filtering algorithm to extract the wave crest time offset, input the vibration signal to the wavelet transform algorithm to extract energy density, filter data segments that do not exceed the offset threshold, generate synchronous feature data, and transmit it to the damage identification module. The damage identification module is used to input the synchronous feature data into the path attenuation model to calculate the propagation change rate, match continuous energy anomaly nodes to determine the damage location, output damage location data, and transmit it to the damage association module. The damage association module is used to receive the damage location data, extract the vibration direction change and signal envelope offset of the damaged section, identify the abnormal area of ​​light wind response, generate light wind response damage data, and transmit it to the risk assessment module. The risk assessment module is used to input the wind response damage data into a neural network to complete feature encoding and risk discrimination, and output the wind vibration damage level assessment result.

[0007] As a further aspect of the present invention, the size of the filtering window of the moving average filtering algorithm is automatically set according to the sensor configuration and the dominant frequency of the stress wave; The wavelet transform algorithm uses the db4 wavelet, with a decomposition level of 3 to 5 levels, depending on the sampling frequency and signal length; The offset threshold is 2 to 3 times the time required for the wave to propagate within the sensor spacing; The time-aligned dual-modal data specifically includes multi-dimensional time labels, cross-modal synchronization accuracy, and a unified sampling structure. The synchronization feature data includes peak time correction values, energy density distribution maps, and anomalous segment time indexes. The damage location data specifically includes damage node numbers, path attenuation ratios, and spatial propagation trajectories. The wind response damage data includes envelope offset curves, vibration direction change values, and wind anomaly identification markers. The wind vibration damage level assessment specifically refers to risk level labels, feature vector encoding, and damage distribution heatmaps.

[0008] As a further aspect of the present invention, the sensing acquisition module includes: The data acquisition submodule acquires time-series data of stress wave velocity and acceleration, sets sensor frequency and channel parameters, configures signal sensitivity and recording trigger threshold, records channel signals and marks their signal source type, extracts all continuous signal data, and generates stress wave velocity data and acceleration time-series data. The record trigger threshold is set within the range of the original amplitude statistical mean ± standard deviation; The synchronization alignment submodule calls the stress wave velocity data and acceleration time series data, calculates the starting offset based on the timestamp, calibrates the position of the starting frame and the time axis through the synchronizer, and filters the data channel group whose delay error is lower than the time alignment reference value to obtain synchronized time alignment data. The time alignment reference value is derived from the delay response measurement experiment of the synchronizer and the statistical deviation of the sampling period between multiple sensors, and is controlled within 1 to 3 sampling periods; The dual-modal generation submodule, based on the synchronous time-aligned data, maps two types of data points according to their sequence positions, analyzes their amplitude difference rate, filters out data outside the fluctuation range, integrates qualified data to establish a unified time series, and generates dual-modal signal data. The fluctuation range is the data segment in which the amplitude difference rate between the stress wave velocity and the acceleration signal at the corresponding sequence position continuously exceeds a preset threshold.

[0009] As a further aspect of the present invention, the signal decoupling module includes: The moving average filtering submodule acquires the stress wave parameters in the dual-mode signal data. Based on the moving average filtering algorithm, it segments the stress wave amplitude according to the time window, calculates the time offset changes of multiple wave peaks, selects the data segments with obvious offset amplitudes and aligns them with the reference window to generate the wave peak time offset. The wavelet transform submodule acquires vibration signal data, performs multi-scale decomposition of frequency components based on the wavelet transform algorithm, calculates the energy density value at the statistical scale, calculates the average energy distribution ratio of each group of signals within the frequency range, and generates energy density features. The data filtering submodule calls the energy density feature and peak time offset to perform a joint comparison of the offset and energy distribution of the signal within the same time window, removes data segments that exceed the offset threshold, and generates synchronization feature data.

[0010] As a further aspect of the present invention, the peak time offset is calculated using the following formula: ; in, This represents the time offset of the k-th wave peak, in seconds. This represents the time point of the i-th peak, in seconds. This represents the amplitude corresponding to the i-th peak, in V. This represents the arithmetic mean of the amplitudes within the k-th window segment, in units of V. This represents the standard deviation of the amplitude within the k-th segment of the window, in units of V. This represents the length of the moving average filter time window. This represents the minimum time interval threshold between adjacent peaks, measured in seconds (s).

[0011] As a further aspect of the present invention, the damage identification module includes: The feature extraction submodule acquires the signal amplitude, signal phase, and distance between nodes from the synchronization feature data, inputs them into the path attenuation model, calculates the propagation change rate, and analyzes the differences in signal propagation between nodes to obtain propagation change rate data. The path attenuation model is a physical model based on the regression relationship between signal energy and distance. It takes the distance between nodes, signal amplitude and spectrum information as inputs and outputs the propagation change rate index. The energy anomaly matching submodule analyzes the energy change trend of the nodes based on the propagation change rate data, compares the signal propagation characteristics with the preset benchmark value, matches consecutive energy anomaly nodes and calculates the anomaly ratio to obtain node anomaly ratio data. The damage localization submodule, based on the node anomaly ratio data and through the set damage identification criteria, determines the number of nodes with anomalies, outputs the corresponding node number, locates the damage location, and obtains damage localization data.

[0012] As a further aspect of the present invention, the propagation rate of change is calculated using the following formula: ; in, Represents a node and nodes The rate of change of propagation between them, in dB / m. Representative node The signal amplitude, in mV. Representative node The signal amplitude, in mV. For nodes With nodes The distance between them, in meters. Representative node With nodes The phase difference between them, in rad. Represents a node With nodes Environmental factor coefficients between This represents the directional coupling factor, measured in rad. This represents a very small positive constant that avoids the denominator being zero. Representative node to Spectral attenuation factor along the path.

[0013] As a further aspect of the present invention, the damage association module includes: The vibration analysis submodule calculates the standard deviation and range of the vibration direction of the damaged section based on the damage location data, extracts the vibration direction change rate of adjacent sections using the sliding window method, compares the change rate with the preset direction change threshold, and generates the vibration direction change rate. The offset quantization submodule calls the vibration direction change rate, extracts the signal envelope peak and valley point sequence, calculates the difference coefficient between adjacent peak and valley points, and constructs the offset index calculation formula by combining the envelope baseline offset to generate the envelope offset index. The anomaly identification submodule integrates the envelope offset index and the vibration direction change rate to establish a three-dimensional feature space mapping relationship. It uses a density clustering algorithm to identify outlier distribution areas in the feature space, marks areas where the outlier density exceeds a preset density threshold as anomaly response areas, and generates wind response damage data.

[0014] As a further aspect of the present invention, the risk assessment module includes: The damage signal preprocessing submodule acquires the original vibration signal from the wind response damage data, performs bandpass filtering to eliminate environmental noise interference, linearly scales the amplitude of the time-domain waveform, and generates preprocessed signal data. The vibration feature encoding submodule calls the preprocessed signal data into the neural network convolutional layer to extract time-frequency features, eliminates dimensional differences through batch normalization, and applies nonlinear activation processing to obtain the vibration feature matrix; The neural network structure uses a 3×3 convolutional kernel with a stride of 2; The risk level discrimination submodule inputs the vibration feature matrix into a fully connected network to calculate the category probability, uses the Softmax function to transform the probability distribution, performs classification decision based on a preset threshold, and generates a light wind vibration damage level assessment result.

[0015] On the other hand, a method for diagnosing wind vibration damage to fittings of self-powered transmission lines, wherein the method is based on the aforementioned system for diagnosing wind vibration damage to fittings of self-powered transmission lines, includes the following steps: S1: Obtain stress wave propagation velocity parameters and triaxial acceleration time series data through a piezoelectric sensor array, and use a synchronizer to perform millisecond-level alignment calculations on the timestamps of the two sets of data to generate time-aligned dual-modal data; S2: Call the time-aligned dual-mode data, use the moving average filtering algorithm to extract the wave peak time offset, input the vibration acceleration signal into the wavelet transform algorithm to calculate the energy density spectrum, filter the data segments with wave peak time offset less than the offset threshold, and generate synchronization feature data. S3: Input the synchronous feature data into the path attenuation model to calculate the rate of change of stress wave propagation velocity, match abnormal nodes in the energy density spectrum whose energy attenuation rate of continuous sampling points exceeds the benchmark ratio, and output damage location data. S4: Based on the damage location data, extract the direction angle change of the vibration acceleration vector in the corresponding section, calculate the root mean square offset of the signal envelope and the baseline envelope through Hilbert transform, and generate wind response damage data. S5: Input the wind response damage data into a pre-trained neural network for feature encoding, calculate the damage probability distribution through the Softmax function, and output the wind vibration damage level assessment result.

[0016] The beneficial effects of this invention are as follows: By synchronously acquiring stress wave velocity parameters and acceleration data in two modes, a time alignment mechanism is established to enhance the accuracy of feature coupling. Moving average filtering and wavelet transform algorithms are used to extract peak offset and energy density features respectively. A dynamic threshold is constructed to screen effective data segments, suppressing the interference of environmental noise on feature extraction. The stress wave propagation rate of change is quantified based on the path attenuation model. Damage spatial localization is achieved by combining continuous energy anomaly node matching. The vibration direction change rate and signal envelope offset are fused to identify anomalous areas of wind response. Multidimensional damage features are encoded and risk association is judged through neural networks, forming a closed-loop analysis framework for damage localization and risk assessment, thereby improving the spatial resolution and assessment dimensions of wind vibration damage identification under complex working conditions. Attached Figure Description

[0017] Figure 1 This is a system flowchart of the present invention; Figure 2This is a system block diagram of the present invention; Figure 3 This is a flowchart of the method in this invention. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0020] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid unnecessary detail that would obscure the description of the invention. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0021] This invention provides a micro-wind vibration damage diagnosis system for self-powered transmission line fittings, such as... Figure 1 The flowchart shown is for a micro-wind vibration damage diagnosis system for self-powered transmission line fittings. The system includes: The sensor acquisition module is used to acquire stress wave velocity parameters and acceleration time series data, use a synchronizer to complete time alignment, generate time-aligned dual-mode data, and transmit it to the signal decoupling module. The signal decoupling module is used to extract stress wave parameters from time-aligned dual-mode data, call the moving average filtering algorithm to extract the wave crest time offset, input the vibration signal to the wavelet transform algorithm to extract energy density, filter data segments that do not exceed the offset threshold, generate synchronous feature data, and transmit it to the damage identification module. The damage identification module is used to input synchronous feature data into the path attenuation model to calculate the propagation change rate, match continuous energy anomaly nodes to determine the damage location, output damage location data, and transmit it to the damage association module. The damage correlation module is used to receive damage location data, extract the vibration direction change and signal envelope offset of the damaged section, identify areas with abnormal wind response, generate wind response damage data, and transmit it to the risk assessment module. The risk assessment module is used to input light wind response damage data into a neural network to complete feature encoding and risk discrimination, and output the light wind vibration damage level assessment result.

[0022] The time-aligned bimodal data specifically includes multi-dimensional time labels, cross-modal synchronization accuracy, and a unified sampling structure. The synchronization feature data includes peak time correction values, energy density distribution maps, and anomalous segment time indexes. The damage location data specifically includes damage node numbers, path attenuation ratios, and spatial propagation trajectories. The wind response damage data includes envelope offset curves, vibration direction change values, and wind anomaly identification markers. The wind vibration damage level assessment specifically refers to risk level labels, feature vector encoding, and damage distribution heatmaps.

[0023] Specifically, such as Figure 2 As shown, the sensor acquisition module includes: The data acquisition submodule acquires time-series data of stress wave velocity and acceleration, sets sensor frequency and channel parameters, configures signal sensitivity and recording trigger threshold, records channel signals and marks their signal source type, extracts all continuous signal data, and generates stress wave velocity data and acceleration time-series data. The record triggers a threshold that is set within the range of the original amplitude statistical mean ± standard deviation. The data acquisition submodule is used to acquire time-series data of stress wave velocity and acceleration. First, the sensor frequency and channel parameters are set. The sensor frequency setting process needs to be determined based on the material response characteristics of the monitored transmission line fittings. Taking aluminum alloy fittings as an example, their response to light wind vibration is most pronounced in the 30–100Hz frequency range; therefore, the sensor operating frequency is set to 60Hz. A single-channel bandwidth of 100Hz is set through a simulated sampling process. A 24-channel acquisition board is selected, and each channel is assigned a number and bound to a specific fitting installation point. For example, the connection point of lead wire 6 on tower A is numbered CH06, and the corresponding sensor is installed at this point and uniquely marked. Signal sensitivity and... The trigger threshold is recorded. Sensitivity needs to be calibrated according to the sensor specification manual. For example, if a piezoelectric accelerometer with a sensitivity of 100mV / g is used, the input gain should be set to 10 times in the data acquisition card to amplify the input signal to the 1V / g level. The trigger threshold is set by statistically analyzing the amplitude mean ± standard deviation. Specifically, 60 seconds of basic noise data are collected under wind-free conditions, recording 600 data points per second, resulting in a total of 36,000 raw data points. The maximum amplitude range of the raw data is [-0.035, 0.042]V, with a calculated mean of 0.0035V and a standard deviation of 0.008V. Therefore, the trigger threshold is set to... The range is [−0.0085, 0.0155]V. If the value exceeds this range, recording is triggered, and the current channel data is saved. Subsequently, the channel signal is recorded and the signal source type is marked. Based on the location indicated by the channel binding point, the signal source is labeled as "suspension clamp," "tension clamp," "spacer joint," etc., and the code is written in the data labeling field. Finally, all continuous signal data is extracted, and the complete data stream within 60 seconds after the saved trigger signal is read to form a complete stress wave velocity data array. and acceleration time series data For example, after point CH06 was triggered at 12:00 on June 1st, the following data pair array was generated: , These arrays will serve as input data for subsequent synchronization and analysis processes.

[0024] Table 1: Sensor Trigger Threshold Setting Table As shown in Table 1, the trigger threshold range of the CH06 sensor is set to [−0.0085, 0.0155] V. This setting is obtained by calculating the mean and standard deviation of real-time monitoring data. Its sensitivity is 1 V / g after conversion, which can effectively record the weak stress wave or vibration response data that occur in the corresponding structure.

[0025] The synchronization alignment submodule calls stress wave velocity data and acceleration time series data, calculates the starting offset based on the timestamp, calibrates the position of the starting frame and time axis through a synchronizer, and filters data channel groups with delay errors lower than the time alignment reference value to obtain synchronized time alignment data. The time alignment reference value is derived from the delay response measurement experiment of the synchronizer and the statistical deviation of the sampling period between multiple sensors, and is controlled within 1 to 3 sampling periods; Synchronous alignment submodule calls stress wave velocity data With acceleration time series data In its execution process, the first step is to compare the timestamps of the two data sequences with the time values ​​of their first valid sampling points, for example... , The calculated initial offset is Seconds, then based on the sampling frequency ,Will Converted to frame offset Frame, system for acceleration sequence Perform a data shift operation to... The elements in the sequence are shifted forward 60 frames to align with the starting frame, completing the initial calibration of the starting frame and the timeline position. Then, a synchronizer detects the real-time synchronization error of each channel pair. This error is determined by the difference in recorded channel acquisition times and the time difference corresponding to the point with the highest correlation coefficient in the data sequence. For example, the synchronizer records the time difference of a certain channel pair... and The maximum cross-correlation peak occurs in frame 72, with a sampling period of 1.667ms, from which the calculated delay error is: If the time alignment reference value is within 3 sampling periods, i.e., within 5.001ms, then the channel group does not meet the alignment requirements and is removed from the candidate group. To filter the channel groups that meet the alignment requirements, the system calculates the frame-level error value between each pair of channels in each group and executes... The result is greater than or equal to Those that are not properly positioned are classified as non-qualified channel groups, and the frame-level offset is controlled within [a certain range]. Those that fall within the specified range are considered to be in the synchronization qualified group, and the data after corresponding synchronization time alignment is retained; the qualified judgment threshold is set from the statistical deviation and response delay test of the sampling period of the synchronizer during the experiment of acquiring data from multiple channels. The statistics show that under standard conditions, the sensor sampling period is... Therefore, the maximum response delay offset should be controlled within Within; when to and After synchronous data processing, the calculated frame offset is 2 frames and the delay error is 3.334ms, which meets the time alignment reference value limit. Therefore, the data of this channel group enters the subsequent dual-modal processing flow, and finally forms a synchronous time-aligned data sequence. and .

[0026] The dual-modal generation submodule, based on synchronized time-aligned data, maps two types of data points according to sequence position, analyzes their amplitude difference rate, filters out data outside the fluctuation range, integrates qualified data to establish a unified time series, and generates dual-modal signal data. The fluctuation range is the data segment in which the amplitude difference rate between the stress wave velocity and the acceleration signal at the corresponding sequence position continuously exceeds a preset threshold. The dual-modal generation submodule is based on the aforementioned synchronized time-aligned data sequence. and First, perform sequence position mapping, that is, let and Establish a one-to-one correspondence, where For the first At each frame time point, the amplitude difference rate is calculated for the two corresponding values ​​in each frame, and the difference rate is defined as follows: During the execution process Calculate sequentially, when consecutive occurrences within a certain time period When the number of frames exceeds a set threshold, such as 50 frames, this range is identified as a fluctuation range and excluded. The corresponding frame points are removed from the original data. For example... Inside Data within the time range consistently between 0.65 and 0.92 was discarded; the retained frame data was reordered to form a unified time series. And construct new bimodal data pairs of sequences. In this example, if the initial data length is 36,000 points, and the remaining frame count after removing fluctuation ranges is 31,200 points, then... The length is 31200; the final output is this dual-mode signal data array for downstream algorithms, in which... and These represent the instantaneous stress wave velocity and acceleration response values, respectively, with amplitude ranges remaining within the ranges of 0.5~6.5m / s and 0.1~2.0m / s². The standard coefficient for determining fluctuation values ​​with a difference rate higher than 0.6 is derived from the statistical analysis of the difference rate of known stable channel data pairs in the sample set. More than 80% are below 0.6, so this is used as the screening criterion. A frame threshold of 50 frames is set, which is derived from the minimum detectable disturbance period within a 5-second continuous sampling period.

[0027] Specifically, such as Figure 2 As shown, the signal decoupling module includes: The moving average filtering submodule obtains stress wave parameters from dual-mode signal data. Based on the moving average filtering algorithm, it segments the stress wave amplitude by time window, calculates the time offset change of the peak of each segment, selects the data segment with obvious offset amplitude and aligns it with the reference window to generate the peak time offset. The size of the filtering window in the moving average filtering algorithm is automatically set based on the sensor configuration and the dominant frequency of the stress wave. The moving average filtering submodule is used to process stress wave parameters in dual-mode signal data. In the scenario of power transmission line fitting inspection, the sliding window setting is first determined based on the configured stress wave sensor parameters. The sensor spacing is set to 1.5m, the stress wave main frequency is 25kHz, and the sampling rate is 100kHz. After conversion, the window duration is 1 / 4 of the main frequency period, that is, the window length is 10µs, corresponding to 1000 data points. After dividing the original stress wave signal according to this sliding window, the signal of each window segment is processed. First, the waveform amplitude sequence of the sampling points in the segment is read, and the sample data segment is set as an array of length 1000. V, calculate the arithmetic mean of the array. For example, if the average value is 1.35V, calculate the amplitude standard deviation. The standard deviation was set to 0.15V, and then the time points corresponding to the local maxima in each segment were extracted to form the peak time series. The peak times are set as follows: ms, total For each wave peak, calculate the absolute value of the time difference between adjacent wave peaks, such as... The result is ms, settings s, The peak time offset is calculated using the following formula: ; in, This represents the time offset of the k-th wave peak, in seconds. This represents the time point of the i-th peak, in seconds. This represents the amplitude corresponding to the i-th peak, in V. This represents the arithmetic mean of the amplitudes within the k-th window segment, in units of V. This represents the standard deviation of the amplitude within the k-th segment of the window, in units of V. This represents the length of the moving average filter time window. This represents the minimum time interval threshold between adjacent peaks, measured in seconds (s).

[0028] ; The calculated peak time offset was 2.12 ms. This value was then compared with a significant offset threshold. The aforementioned sensor spacing was 1.5 m, the wave velocity was 5000 m / s, and the propagation time was... The propagation time is 3 times the time, which is 0.9ms. Obviously, 2.12ms > 0.9ms, indicating that this segment is a significant offset segment. Next, the entire signal segment will be shifted forward along the time axis. The peak time alignment is completed in seconds, and the peak time offset array is finally generated. This indicates the peak delay of multiple corresponding reference segments.

[0029] Table 2: Example Table of Peak Time Offset Calculation Table 2 lists the time, amplitude, squared difference from the mean, adjacent time difference, and weighted result of each peak point in the example data segment, which are ultimately used to calculate the peak offset. The result shows that the current data segment offset value exceeds the allowed threshold range, indicating that it is a valid abnormal data segment.

[0030] The wavelet transform submodule acquires vibration signal data, performs multi-scale decomposition of frequency components based on the wavelet transform algorithm, calculates the energy density value at the statistical scale, calculates the average energy distribution ratio of each group of signals within the frequency range, and generates energy density features. The wavelet transform algorithm uses the db4 wavelet, with a decomposition level of 3 to 5 levels, depending on the sampling frequency and signal length; The wavelet transform submodule acquires vibration signal data collected by the sensor, in time series format. Taking a single-segment signal with 2048 sampling points as an example, with a sampling frequency set to 100kHz, this signal represents the vibration response of transmission line fittings at a certain moment. First, the Daubechiesdb4 wavelet function is selected as the basis wavelet, based on its good time-frequency localization characteristics and experimental results demonstrating high energy focusing capability in structural vibration signal analysis. The signal is then decomposed layer by layer, with a decomposition level of 4. The decomposition process is as follows: In the first layer, the original signal undergoes low-pass and high-pass filtering operations to obtain approximate coefficients. and detail coefficient Then The second layer of filtering is performed again as the input signal to obtain... and This process continues until the fourth layer is completed and approximate coefficients are obtained. and multi-layer detail coefficient Each decomposition layer divides the frequency range into half of the current frequency band. When the sampling frequency is 100kHz, the detail coefficients of the first layer are... The corresponding frequency band is 25–50kHz, the second layer is 12.5–25kHz, the third layer is 6.25–12.5kHz, and the fourth layer is 3.125–6.25kHz. The fourth layer approximation coefficient... The corresponding frequency band is 0–3.125kHz. For each set of detail coefficients, its energy density value is calculated, which is obtained by first squaring the coefficients and then summing them. Then divide by the number of sampling points corresponding to that layer. Obtaining energy density For example, if the length of the second-level detail coefficients in a certain data segment is set to 256 points and the sum of squares is 58.4, then the energy density is... V² / point, then normalize the multi-layer energy density by adding the energy density values ​​as the denominator and using the current layer's energy density as the numerator, to calculate the energy distribution ratio. Assuming the energy densities of the four layers are respectively , , , The total energy density is 0.637, and the proportion of the second layer is calculated as follows: The energy ratios of multiple layers are calculated and combined into eigenvectors in the same manner. This vector represents the energy distribution characteristics of the current signal segment within the frequency range. By performing this process on multiple signal segments, the energy density variation pattern of the entire vibration signal sequence can be obtained as a basis for subsequent feature comparison and anomaly detection. The energy ratio value range is divided as follows: If a certain frequency band Classified as high-energy segment Classified as medium energy level. If the second layer is classified as a low-energy segment, for example, the proportion of the second layer is 0.358, which falls into the medium-energy range. If the second layer frequency segment is a high-energy segment in three consecutive data segments and the amplitude increases continuously, then there is a phenomenon of mid-frequency energy surge caused by local structural loosening or abnormal connection.

[0031] The data filtering submodule calls the energy density feature and the peak time offset to perform a joint comparison of the offset and energy distribution of the signal within the same time window, removes data segments that exceed the offset threshold, and generates synchronization feature data. The offset threshold is 2 to 3 times the time required for the wave to propagate within the sensor spacing; The data filtering submodule compares the peak time offset with the energy distribution feature vector within the same time window. First, it obtains the offset value of each data segment from the moving average submodule. The data is then compared with a preset significant offset threshold, which is set at 0.9ms. If a data segment has an offset greater than 0.9ms, it is considered a candidate anomaly segment. In the example, the three offset segments are set to [0.25, 0.98, 1.25]ms. The latter two segments are greater than the threshold and proceed to the energy feature comparison stage. The energy distribution feature vector is synchronously obtained from the wavelet transform submodule. For example, the corresponding energy distributions are... and Compare it with the feature vector of the set reference window Perform cosine similarity calculation, setting the similarity threshold to 0.95. The calculation formula is as follows: ; in, Representing the The proportion of energy distribution of each frequency band in the current detected signal segment represents the proportion of energy density in that frequency range of the current segment, with a value range of [0, 1]. Representing the The energy distribution ratio of each frequency band in the reference signal band is used as a comparison benchmark, with a value range of [0, 1].

[0032] The similarity of the first set of features is The second group is Both data segments exceeded the set similarity threshold, so these two data segments were retained as feature synchronization segments. Then, a time axis reconstruction operation was performed on the signal segment and the reference segment, that is, the peak position was shifted based on the peak time of the reference segment. After time alignment, a synchronization feature matrix under a unified time axis was formed. Finally, segments with offsets not exceeding the threshold but energy distribution similarity lower than the threshold were removed. For example, a segment with an offset of 0.35ms but a similarity of 0.65 was excluded. The feature matrix in the screening results is used for subsequent damage identification or structural trend analysis.

[0033] Specifically, such as Figure 2 As shown, the damage identification module includes: The feature extraction submodule acquires the signal amplitude, signal phase, and distance between nodes from the synchronization feature data, inputs them into the path attenuation model, calculates the propagation change rate, and analyzes the differences in signal propagation between nodes to obtain propagation change rate data. The path attenuation model is a physical model based on the regression relationship between signal energy and distance. It takes the distance between nodes, signal amplitude and spectrum information as inputs and outputs the propagation change rate index. The feature extraction submodule acquires data such as signal amplitude, signal phase, and distance between nodes by synchronizing feature data. Signal amplitude is measured in mV and obtained through voltage amplitude conversion. For example, if the sensor output voltage is 0.82V, the corresponding amplitude is 820mV. Signal phase is obtained by comparing the phase difference between the sampled signal at the receiving end and the reference transmitter signal at the same frequency. The phase difference is converted to radians. If the sensor sampling point delay is 1.25ms and the frequency is 10Hz, the corresponding phase difference is... Distance information is obtained by combining GIS geographic data with 3D ranging. The propagation change rate is calculated using the path attenuation model, and the formula is as follows: ; in, Represents a node and nodes The rate of change of propagation between them, in dB / m. Representative node The signal amplitude, in mV. Representative node The signal amplitude, in mV. For nodes With nodes The distance between them, in meters. Representative node With nodes The phase difference between them, in rad. Represents a node With nodes Environmental factor coefficients between This represents the directional coupling factor, measured in rad. This represents a very small positive constant that avoids the denominator being zero. Representative node to Spectral attenuation factor along the path.

[0034] in, , All values ​​are in mV. The original dBm values ​​need to be converted using the following formula: Taking the signal amplitude of node 1 as an example of -60.3dBm, the calculation is as follows: The signal at node 2 is -62.1 dBm, corresponding to Distance between nodes Set the spectral attenuation factor Take the directional coupling factor Phase difference Let environmental factor coefficients be set. Avoiding the zero factor Substitute into the calculation: ; Similarly, the signal amplitudes from node 2 to node 3 are -62.1 dBm and -65.0 dBm, respectively, which are converted to... , ,distance Spectral attenuation factor , The phase difference is Environmental coefficient Similarly, substitute the values ​​into the formula to calculate: ; The final propagation change rates for the two segments were 0.352 dB / m and 0.0974 dB / m, respectively. The differences between the two are as follows: ; This difference needs to be compared with the system baseline, which is set at 0.02 dB / m. This setting is based on statistical analysis: the mean of the difference in propagation rate between adjacent paths under normal operating conditions is 0.015 dB / m, and the standard deviation is 0.0025 dB / m. The baseline is calculated as follows: ; Since 0.2546 > 0.02, it indicates that there is an abnormal propagation change difference between the paths from node 1 to node 2 and from node 2 to node 3. Record the segment number and output the propagation change rate data and the difference results.

[0035] Table 3: Calculation Results of Path Propagation Change Rate Table 3 shows the calculation results of the propagation change rate based on real-time sampling data. The numerical results can be used as input for subsequent energy anomaly matching. The results indicate that there are significant differences in propagation characteristics between node paths, and further anomaly location needs to be determined.

[0036] The energy anomaly matching submodule analyzes the energy change trend of nodes based on propagation change rate data, compares the signal propagation characteristics with preset benchmark values, matches consecutive energy anomaly nodes and calculates the anomaly ratio to obtain node anomaly ratio data. The energy anomaly matching submodule receives the propagation change rate data for each path segment output by the path attenuation model. By constructing a correspondence between propagation paths and time series, it archives the energy change trend of each node under different time windows, records the trajectory of the propagation change rate within continuous time periods, and calculates the cumulative difference in propagation change rates between adjacent paths of a node. If the change rate of a node in multiple consecutive path segments corresponds to a larger change rate for surrounding nodes, then that node is marked as a potential anomaly. Specifically, the operation is as follows: First, using the node number as an index, the propagation change rates of the paths in which the node participates are arranged in chronological order. For example, if node 2 forms a path with node 1 at time T1 with a propagation change rate of 0.352, and forms a path with node 3 at time T2 with a propagation change rate of 0.0974, the change rate difference corresponding to node 2 is calculated as follows: The difference is added to the cumulative anomaly index of node 2. Secondly, a preset baseline range is defined as the upper and lower limits of the normal fluctuation range of the propagation change rate set in the system. The mean of the baseline change rate is set to 0.015, and the variance to 0.0025. The upper and lower limits are then respectively... If any calculated change difference exceeds the specified interval, it is considered an anomaly. Further, the number of anomaly markers for a node within n consecutive time periods is accumulated. For example, if node 2 exhibits 3 instances of exceeding the propagation change rate limit within the last 5 windows, its anomaly count is 3. Then, using the anomaly ratio calculation formula: Anomaly Ratio = Number of Node Anomalies / Total Number of Observations, taking node 2 as an example, if it is marked as an anomaly 3 times within 5 periods, its anomaly ratio is... The same operation is performed on the nodes to obtain the node anomaly ratio data. At the same time, in order to ensure the continuity and validity of the anomalous nodes, it is also necessary to introduce the spatial correlation evaluation between adjacent nodes. The nodes with anomaly ratio greater than a set threshold, such as 0.4, and the difference between the anomaly ratio of the adjacent nodes and the node with the anomaly ratio is no greater than 0.2 are marked as a continuous anomalous node group. For example, if the anomaly ratio of node 2 is 0.6, node 3 is 0.45, and node 1 is 0.42, then nodes 2, 3, and 1 can form a continuous anomalous node sequence. The whole process does not involve calling specific algorithms, but is only based on basic operation methods such as parameter sorting, difference calculation, interval judgment, and ratio analysis.

[0037] The damage localization submodule determines the number of abnormal nodes based on the node anomaly ratio data and the set damage identification criteria, outputs the corresponding node number, locates the damage location, and obtains damage localization data. The damage localization submodule takes the node anomaly ratio data output by the previous submodule as input. It then uses a set damage identification standard to quantitatively determine the abnormal state of each node. The standard is based on whether the number of node anomalies within a continuous time window exceeds a set threshold N, combined with whether the anomaly ratio exceeds a judgment threshold R, to locate abnormal nodes. The execution process is as follows: First, the standard parameters N=3 times and R=0.4 are set, meaning that if the number of node anomalies is greater than or equal to 3 within 5 observation windows, and the anomaly ratio is higher than 0.4, it is marked as a faulty node. Taking node 2 as an example, its number of anomalies is 3, the total number of windows is 5, and its anomaly ratio... If the ratio is 0.6, it meets the set standard and is judged as abnormal. Then, the node numbers judged as abnormal are compiled and output as damage node numbers. For example, if nodes 2 and 3 are both marked as abnormal, then [2, 3] will be output. On this basis, further distribution screening is carried out through spatial distribution constraints, requiring that the node numbers must be continuous or no more than 1 number unit apart. Otherwise, sporadic abnormal points are removed. For example, if node 5 is abnormal alone but the abnormal ratio of its neighboring nodes is less than 0.2, then node 5 will not be included in the damage location results. Finally, the abnormal node numbers obtained by screening are output as the damage location results, which constitute the final damage location data. In the entire judgment process, the operations involved are all about accumulating and counting the number of abnormalities, comparing the ratio value with the threshold, judging the integer difference between node numbers and matching the number sequence, etc. No model or algorithm form is used to express them, so as to ensure the directness and repeatability of the operation.

[0038] Specifically, such as Figure 2 As shown, the damage correlation module includes: The vibration analysis submodule calculates the standard deviation and range of vibration direction in the damaged section based on the damage location data, extracts the rate of change of vibration direction in adjacent sections using the sliding window method, compares the rate of change with the preset direction change threshold, and generates the vibration direction change rate. First, calculating the standard deviation and range of the vibration direction in the damaged section is the first step in analyzing the changes in vibration characteristics. Damage location data is acquired by measuring the vibration signals of the line using sensors; these signals reflect the impact of light winds on the line fittings. In application, assuming a transmission line is equipped with sensors, after collecting vibration data, the first step is to calculate the standard deviation of the vibration direction in that section. ) and range ( The formula for calculating the standard deviation is: ; in, This represents the vibration data collected each time. The mean of the data. Let represent the number of data points. Assuming the 10 collected vibration data points are: [3.4, 3.6, 3.8, 4.0, 4.2, 4.1, 4.3, 3.9, 4.0, 3.8], first calculate the mean of these data points. Then calculate the standard deviation using the formula above, and obtain the standard deviation of the vibration direction. .

[0039] The range of vibration direction is obtained by the difference between the maximum and minimum values: ; Regarding the above data, , Therefore, the range .

[0040] Next, the sliding window method is used to extract the rate of change of vibration direction between adjacent segments. The sliding window method detects the changing trend of vibration characteristics by calculating the change within a continuous data window. In this application, a window size of 5 data points is set, and the rate of change of vibration direction within the window is calculated step by step. Assuming the vibration data within the window is: [3.4, 3.6, 3.8, 4.0, 4.2], the rate of change is: ; Then, the rate of change is compared with a preset directional change threshold. If the rate of change is greater than a certain threshold, the vibration of that section is considered to have changed significantly. Assuming the preset directional change threshold is 0.1, when the rate of change is 0.16, which is greater than the threshold, it indicates that the vibration characteristics of that section have changed significantly and there is damage.

[0041] The offset quantization submodule calls the vibration direction change rate, extracts the signal envelope peak and valley point sequence, calculates the difference coefficient of the distance between adjacent peak and valley points, and constructs the offset index calculation formula by combining the envelope baseline offset to generate the envelope offset index. In the offset quantization submodule, the peak-valley sequence of the signal envelope is first obtained by extracting the rate of change of vibration direction. Assuming the peak-valley sequence of the vibration signal envelope is [3.2, 3.5, 3.7, 3.4, 3.8], the peak-valley spacing of the envelope signal is [0.3, 0.2, -0.3, 0.4] (in seconds). To calculate the peak-valley spacing difference coefficient for this segment, it is necessary to further calculate the rate of change of vibration direction and the relevant parameters of the envelope baseline for this segment.

[0042] The formula for calculating the difference coefficient between peak and valley points is: ; in, This represents the coefficient of difference between peak and valley distances. Representing the The distance between signal peaks and valleys, in seconds. Representing the The rate of change of the direction of segment vibration, in rad. Representing the The envelope baseline value for each segment, in units of V. The average value of the baseline envelope of the representative segment is expressed in V. A tiny positive constant, in V, is introduced to avoid the situation where the denominator is zero.

[0043] First, calculate the rate of change of vibration direction ( ), assuming the signal is in the first... The rate of change of the vibration direction of the segment is: These values ​​are extracted using the sliding window method. Next, the envelope baseline value of the segment is calculated. Assuming the signal envelope baseline values ​​are [0.04, 0.05, 0.03, 0.06, 0.05] (in V), calculate its average value: To avoid the case where the denominator is zero, a very small positive constant is added to V. V.

[0044] ; Next, a formula for calculating the offset exponent is constructed based on the envelope baseline offset. Assuming the envelope baseline offset is 0.02V, and the reference value ( The value is 0.05V, and the offset index is calculated as follows: ; This indicates that the overall signal offset is quite significant, providing a basis for further anomaly identification.

[0045] As shown above, by accurately calculating parameters such as the rate of change of vibration direction, the difference coefficient between peak and valley points, and the offset of the envelope baseline, the offset quantization submodule can effectively extract the variation pattern of the signal, providing basic data for subsequent anomaly identification.

[0046] The anomaly identification submodule integrates the envelope offset index and the vibration direction change rate to establish a three-dimensional feature space mapping relationship. It uses a density clustering algorithm to identify outlier distribution areas in the feature space and marks areas where the outlier density exceeds a preset density threshold as an abnormal response area, generating wind response damage data. In the anomaly identification submodule, the envelope offset index and vibration direction change rate generated in the previous stage are first used as core input features. A corresponding three-dimensional feature vector is constructed for each monitoring segment, where the first dimension is the envelope offset index value of the segment, the second dimension is the corresponding vibration direction change rate, and the third dimension is the spacing difference coefficient within the segment. A complete three-dimensional feature point set is constructed by traversing the monitoring segments. Based on this, density estimation operations are performed on multiple points sequentially. First, the distance between points in the feature space is calculated using Euclidean distance. For example, for two feature points... and Its distance is Based on this, by setting the neighborhood radius minimum number of neighbors To count the points at The number of neighbors within the neighborhood, if the number of neighbors is greater than If the point is true, it is considered a core point; otherwise, it is considered a boundary point or noise point. For example, let's define... , In application, distance measurements and neighborhood point counts are performed on the feature points formed in each segment. For example, if a feature point is formed in segment 12... , its in A total of 4 neighboring points were found within the range, satisfying the condition. Therefore, it is set as the core point, and conversely, if it is another feature point... If only one neighbor point is found, it is considered a noise point and does not participate in clustering. This process is repeated until the feature points are clustered. Then, the density value of the core points in each cluster is statistically analyzed. Density is defined as the number of core points per unit volume. The feature space is divided using voxel segmentation or mesh generation, with the volume of each cubic unit set to [value missing]. For example, in a certain region, all eight core points fall within a volume of Within the region, the density is Compare with preset density threshold The system identifies the area as an abnormal response zone and then extracts the monitoring segment index corresponding to the clustered area that exceeds the density threshold as an abnormal segment, thus completing the determination and screening of damage data caused by the light wind response.

[0047] Table 4: Sample Point Data Table for Three-Dimensional Feature Space Table 4 lists the three-dimensional feature vector values ​​of some monitoring sections. Many of these indicators are derived from the calculation results of the previous stage, providing an input basis for cluster recognition.

[0048] In the above process, by quantifying the density distribution of multiple feature points in the feature space, the anomaly identification submodule can effectively separate segments with abnormal response characteristics, providing reliable structured data support for subsequent system processing.

[0049] Specifically, such as Figure 2 As shown, the risk assessment module includes: The damage signal preprocessing submodule acquires the original vibration signal from the wind response damage data, performs bandpass filtering to eliminate environmental noise interference, linearly scales the amplitude of the time-domain waveform, and generates preprocessed signal data. The damage signal preprocessing submodule is used to acquire the raw vibration signal from the light wind response damage data. The vibration signal comes from a high-sensitivity accelerometer deployed at the transmission line fittings. This sensor continuously acquires signals at a sampling frequency of 2000Hz and uploads the data to the data processing terminal in time series format. When performing bandpass filtering, a filter bandwidth range of 5Hz to 150Hz is selected to retain the typical frequency band of light wind vibration and suppress low-frequency wind load disturbances and high-frequency electromagnetic interference. The bandpass filter adopts an IIR filter structure of order 4. In the MATLAB environment, bidirectional filtering is used to eliminate phase shift distortion. Then, the filtered signal is subjected to time-domain amplitude linear scaling, which is achieved by calling the maximum value of the signal. and minimum value Calculate the scaling factor as follows: ; And by mapping each sample point in the original signal Mapped to the range [-1, 1], the mapping method is as follows: ; Suppose the acquired raw signal data is an array: Then there is a maximum value. minimum value Substitute into the calculation to obtain the scaling factor. Taking the third point, 0.05, as an example, its scaled value is: ; This process is executed sequentially on each sampling point to form preprocessed signal data for subsequent neural network input. For "bandpass filtering to eliminate environmental noise interference," the process is as follows: using the set upper and lower cutoff frequencies as boundaries, a filter coefficient matrix is ​​constructed and applied one by one to the Fourier transform spectrum of the signal. Frequency components below 5Hz and above 150Hz are set to zero, and then the signal is inversely transformed back to the time domain to obtain the vibration signal after noise interference removal. For "linear amplitude scaling," the maximum and minimum amplitudes at the midpoint of the signal are first extracted. The scaling factor and offset factor are constructed using the difference between these two values ​​and the average of the center point values. Then, the original waveform is linearly transformed point by point. This process does not involve probability distributions or statistical standard deviations; numerical normalization is only performed based on interval standards to prevent drastic amplitude fluctuations from affecting subsequent feature extraction. For acquiring the "raw vibration signal," the sampling time was set to 10 seconds, resulting in 20,000 data points per acquisition. If minor friction caused by wind-induced micro-motion occurs at the location of the hardware, the resulting periodic small-amplitude vibrations will be mixed into the raw signal with an extremely low signal-to-noise ratio. Therefore, non-target frequency components must be removed through the above steps. To more clearly demonstrate the change in signal amplitude before and after preprocessing, the scaled data of the vibration signal from a specific hardware sample point after the above processing is shown below: Table 5: Raw and Scaled Data of Light Wind Vibration Signals As shown in Table 5, the scaled data is strictly mapped to the range [-1, 1], preserving the signal's vibration rhythm and corresponding amplitude changes. The preprocessed signal data obtained at the end of the paragraph is the signal sequence after filtering and scaling described above.

[0050] The vibration feature encoding submodule calls the preprocessed signal data into the neural network convolutional layer to extract time-frequency features, eliminates dimensional differences through batch normalization, and applies nonlinear activation processing to obtain the vibration feature matrix; The neural network structure uses 3×3 convolutional kernels with a stride of 2; The vibration feature encoding submodule calls the preprocessed signal data that has undergone scaling as the input signal sequence. This sequence is a time series data with a length of 20,000. First, it is sequentially input into a convolutional layer with a stride of 2 and a kernel size of 3×3. This layer performs local feature extraction on the two-dimensional unfolded result of the input signal. The input dimension is unfolded according to [number of samples, number of channels, time length] to [1, 1, 20000]. The convolution operation slides the window once between every two data points, that is, every three consecutive data points form a sliding window and slides it one by one to sample local waveform feature values. Through this process, several filter response channel values ​​can be obtained, forming a preliminary feature tensor matrix. Then, the batch normalization process is called to perform mean and standard deviation statistics on the feature values ​​in each channel. Specifically, the overall average value is calculated for each group of input channel feature vectors. with standard deviation Then perform the following transformation: each feature point is... Normalization is performed without introducing an activation function; only the feature distribution is adjusted to maintain a uniform scale. Then, the ReLU activation function is applied to the normalized data, setting values ​​less than zero to zero and leaving values ​​greater than zero unchanged to suppress negative interference, thus obtaining the final nonlinear vibration feature matrix. The sampling of stage parameters during this process is illustrated in the following example: Assume three consecutive scaled acceleration values ​​are [0.2857, -0.4286, 1], and the initial weights of the convolution kernel are [0.2, 0.3, 0.5]. Then the convolution output is... Then, in batch normalization, if the mean of the feature points in this batch is 0.3 and the standard deviation is 0.2, the normalized result will be: Since the value is greater than 0, the ReLU activation result is still 0.6428. If the value after normalization is -0.6, the activation output is 0. The final vibration feature matrix is ​​composed of each convolution response, the result after normalization and activation, and has a unified numerical scale and nonlinear enhancement characteristics.

[0051] The risk level discrimination submodule inputs the vibration feature matrix into a fully connected network to calculate the category probability, uses the Softmax function to transform the probability distribution, performs classification decision based on a preset threshold, and generates a light wind vibration damage level assessment result. The risk level discrimination submodule obtains the vibration feature matrix output by the aforementioned convolutional network and uses it as the input vector for the fully connected neural network. Each row of this matrix represents a response channel, and each column represents the normalized activation feature value at the corresponding position after convolution extraction. First, the feature value vector of each channel is expanded into a one-dimensional array, and then the channel data is concatenated and combined into a complete input vector. When connecting to a fully connected layer, a weighted summation and bias offset operation are performed on the input vector. The calculation process for each output node is as follows: ; in, For the first element in the input vector 1 eigenvalue, For the first Weight coefficients for each feature value This is a bias term.

[0052] Let the weight vector of a certain output unit be... The input vector is With a bias of 0.05, the output value is... The output value is transformed using the Softmax function into a probability distribution of multiple risk levels. The expression for the Softmax function is: ; in, Indicates the first Normalized probability of class, For the first Class score, For the first The index mapping value of the class score, To output the total number of classes, The index of the item being traversed.

[0053] If the final three outputs are [0.34, 0.51, 0.23], then the corresponding exponents are [1.405, 1.666, 1.258], and the normalized probabilities are respectively... Based on a classification threshold of 0.4, the system determines that the highest probability value of 0.3848 does not exceed the threshold, triggering a secondary confidence strategy. This involves calculating the inter-class distance to determine if there is a central tendency or if the model parameters need retraining. This threshold of 0.4 is the probability cutoff value corresponding to the highest accuracy rate in the original data. The specific setting process involves extracting the predicted probability distribution from 100 samples and calculating the mean and variance of the probability distribution among correctly predicted samples. If the mean is 0.405 and the variance is 0.012, the threshold is conservatively set to 0.4 by rounding down. Finally, the system outputs a wind vibration damage level assessment result, which serves as one of the grading indicators for the diagnostic system and is returned to the front-end interface.

[0054] Please see Figure 3 The micro-wind vibration damage diagnosis system for self-powered transmission line fittings is implemented, including the following steps: S1: Obtain stress wave propagation velocity parameters and triaxial acceleration time series data through a piezoelectric sensor array, and use a synchronizer to perform millisecond-level alignment calculations on the timestamps of the two sets of data to generate time-aligned dual-modal data; S2: Call time-aligned dual-modal data, use the moving average filtering algorithm to extract the wave peak time offset, input the vibration acceleration signal into the wavelet transform algorithm to calculate the energy density spectrum, filter the data segments with wave peak time offset less than the offset threshold, and generate synchronization feature data; S3: Input the synchronous feature data into the path attenuation model to calculate the rate of change of stress wave propagation velocity, match abnormal nodes in the energy density spectrum whose energy attenuation rate of continuous sampling points exceeds the benchmark ratio, and output damage location data. S4: Based on the damage location data, extract the direction angle change of the vibration acceleration vector in the corresponding section, calculate the root mean square offset of the signal envelope and the baseline envelope through Hilbert transform, and generate wind response damage data. S5: Input the wind response damage data into a pre-trained neural network for feature encoding, calculate the damage probability distribution through the Softmax function, and output the wind vibration damage level assessment result.

[0055] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0056] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0057] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0058] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0059] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0060] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0061] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A micro-wind vibration damage diagnosis system for self-powered transmission line fittings, characterized in that, The system includes: The sensor acquisition module is used to acquire stress wave velocity parameters and acceleration time series data, use a synchronizer to complete time alignment, generate time-aligned dual-mode data, and transmit it to the signal decoupling module. The signal decoupling module is used to extract stress wave parameters from the time-aligned dual-mode data, call the moving average filtering algorithm to extract the wave crest time offset, input the vibration signal to the wavelet transform algorithm to extract energy density, filter data segments that do not exceed the offset threshold, generate synchronous feature data, and transmit it to the damage identification module. The damage identification module is used to input the synchronous feature data into the path attenuation model to calculate the propagation change rate, match continuous energy anomaly nodes to determine the damage location, output damage location data, and transmit it to the damage association module. The damage association module is used to receive the damage location data, extract the vibration direction change and signal envelope offset of the damaged section, identify the abnormal area of ​​light wind response, generate light wind response damage data, and transmit it to the risk assessment module. The risk assessment module is used to input the wind response damage data into a neural network to complete feature encoding and risk discrimination, and output the wind vibration damage level assessment result.

2. The micro-wind vibration damage diagnosis system for self-powered transmission line fittings according to claim 1, characterized in that, The filtering window size of the moving average filtering algorithm is automatically set according to the sensor configuration and the dominant frequency of the stress wave; The time-aligned dual-modal data specifically includes multi-dimensional time labels, cross-modal synchronization accuracy, and a unified sampling structure. The synchronization feature data includes peak time correction values, energy density distribution maps, and anomalous segment time indexes. The damage location data specifically includes damage node numbers, path attenuation ratios, and spatial propagation trajectories. The wind response damage data includes envelope offset curves, vibration direction change values, and wind anomaly identification markers. The wind vibration damage level assessment specifically refers to risk level labels, feature vector encoding, and damage distribution heatmaps.

3. The micro-wind vibration damage diagnosis system for self-powered transmission line fittings according to claim 1, characterized in that, The sensing acquisition module includes: The data acquisition submodule acquires time-series data of stress wave velocity and acceleration, sets sensor frequency and channel parameters, configures signal sensitivity and recording trigger threshold, records channel signals and marks their signal source type, extracts all continuous signal data, and generates stress wave velocity data and acceleration time-series data. The synchronization alignment submodule calls the stress wave velocity data and acceleration time series data, calculates the starting offset based on the timestamp, calibrates the position of the starting frame and the time axis through the synchronizer, and filters the data channel group whose delay error is lower than the time alignment reference value to obtain synchronized time alignment data. The dual-modal generation submodule, based on the synchronous time-aligned data, maps two types of data points according to their sequence positions, analyzes their amplitude difference rate, filters out data outside the fluctuation range, integrates qualified data to establish a unified time series, and generates dual-modal signal data. The fluctuation range is the data segment in which the amplitude difference rate between the stress wave velocity and the acceleration signal at the corresponding sequence position continuously exceeds a preset threshold.

4. The micro-wind vibration damage diagnosis system for self-powered transmission line fittings according to claim 3, characterized in that, The signal decoupling module includes: The moving average filtering submodule acquires the stress wave parameters in the dual-mode signal data. Based on the moving average filtering algorithm, it segments the stress wave amplitude according to the time window, calculates the time offset changes of multiple wave peaks, selects the data segments with obvious offset amplitudes and aligns them with the reference window to generate the wave peak time offset. The wavelet transform submodule acquires vibration signal data, performs multi-scale decomposition of frequency components based on the wavelet transform algorithm, calculates the energy density value at the statistical scale, calculates the average energy distribution ratio of each group of signals within the frequency range, and generates energy density features. The data filtering submodule calls the energy density feature and peak time offset to perform a joint comparison of the offset and energy distribution of the signal within the same time window, removes data segments that exceed the offset threshold, and generates synchronization feature data.

5. The micro-wind vibration damage diagnosis system for self-powered transmission line fittings according to claim 4, characterized in that, The peak time offset is calculated using the following formula: ； in, This represents the time offset of the k-th wave peak, in seconds. This represents the time point of the i-th peak, in seconds. This represents the amplitude corresponding to the i-th peak, in V. This represents the arithmetic mean of the amplitudes within the k-th window segment, in units of V. This represents the standard deviation of the amplitude within the k-th segment of the window, in units of V. This represents the length of the moving average filter time window. This represents the minimum time interval threshold between adjacent peaks, measured in seconds (s).

6. The micro-wind vibration damage diagnosis system for self-powered transmission line fittings according to claim 4, characterized in that, The damage identification module includes: The feature extraction submodule acquires the signal amplitude, signal phase, and distance between nodes from the synchronization feature data, inputs them into the path attenuation model, calculates the propagation change rate, and analyzes the differences in signal propagation between nodes to obtain propagation change rate data. The path attenuation model is a physical model based on the regression relationship between signal energy and distance. It takes the distance between nodes, signal amplitude and spectrum information as inputs and outputs the propagation change rate index. The energy anomaly matching submodule analyzes the energy change trend of the nodes based on the propagation change rate data, compares the signal propagation characteristics with the preset benchmark value, matches consecutive energy anomaly nodes and calculates the anomaly ratio to obtain node anomaly ratio data. The damage localization submodule, based on the node anomaly ratio data and through the set damage identification criteria, determines the number of nodes with anomalies, outputs the corresponding node number, locates the damage location, and obtains damage localization data.

7. The micro-wind vibration damage diagnosis system for self-powered transmission line fittings according to claim 6, characterized in that, The propagation rate of change is calculated using the following formula: ； in, Represents a node and nodes The rate of change of propagation between them, in dB / m. Representative node The signal amplitude, in mV. Representative node The signal amplitude, in mV. For nodes With nodes The distance between them, in meters. Representative node With nodes The phase difference between them, in rad. Represents a node With nodes Environmental factor coefficients between This represents the directional coupling factor, measured in rad. This represents a very small positive constant that avoids the denominator being zero. Representative node to Spectral attenuation factor along the path.

8. The micro-wind vibration damage diagnosis system for self-powered transmission line fittings according to claim 6, characterized in that, The damage correlation module includes: The vibration analysis submodule calculates the standard deviation and range of the vibration direction of the damaged section based on the damage location data, extracts the vibration direction change rate of adjacent sections using the sliding window method, compares the change rate with the preset direction change threshold, and generates the vibration direction change rate. The offset quantization submodule calls the vibration direction change rate, extracts the signal envelope peak and valley point sequence, calculates the difference coefficient between adjacent peak and valley points, and constructs the offset index calculation formula by combining the envelope baseline offset to generate the envelope offset index. The anomaly identification submodule integrates the envelope offset index and the vibration direction change rate to establish a three-dimensional feature space mapping relationship. It uses a density clustering algorithm to identify outlier distribution areas in the feature space, marks areas where the outlier density exceeds a preset density threshold as anomaly response areas, and generates wind response damage data.

9. The micro-wind vibration damage diagnosis system for self-powered transmission line fittings according to claim 8, characterized in that, The risk assessment module includes: The damage signal preprocessing submodule acquires the original vibration signal from the wind response damage data, performs bandpass filtering to eliminate environmental noise interference, linearly scales the amplitude of the time-domain waveform, and generates preprocessed signal data. The vibration feature encoding submodule calls the preprocessed signal data into the neural network convolutional layer to extract time-frequency features, eliminates dimensional differences through batch normalization, and applies nonlinear activation processing to obtain the vibration feature matrix; The risk level discrimination submodule inputs the vibration feature matrix into a fully connected network to calculate the category probability, uses the Softmax function to transform the probability distribution, performs classification decision based on a preset threshold, and generates a light wind vibration damage level assessment result.

10. A method for diagnosing wind vibration damage to fittings in self-powered transmission lines, characterized in that, The method is used to implement the micro-wind vibration damage diagnosis system for self-powered transmission line fittings according to any one of claims 1-9, and includes the following steps: S1: Obtain stress wave propagation velocity parameters and triaxial acceleration time series data through a piezoelectric sensor array, and use a synchronizer to perform millisecond-level alignment calculations on the timestamps of the two sets of data to generate time-aligned dual-modal data; S2: Call the time-aligned dual-mode data, use the moving average filtering algorithm to extract the wave peak time offset, input the vibration acceleration signal into the wavelet transform algorithm to calculate the energy density spectrum, filter the data segments with wave peak time offset less than the offset threshold, and generate synchronization feature data. S3: Input the synchronous feature data into the path attenuation model to calculate the rate of change of stress wave propagation velocity, match abnormal nodes in the energy density spectrum whose energy attenuation rate of continuous sampling points exceeds the benchmark ratio, and output damage location data. S4: Based on the damage location data, extract the direction angle change of the vibration acceleration vector in the corresponding section, calculate the root mean square offset of the signal envelope and the baseline envelope through Hilbert transform, and generate wind response damage data. S5: Input the wind response damage data into a pre-trained neural network for feature encoding, calculate the damage probability distribution through the Softmax function, and output the wind vibration damage level assessment result.

Images