A method and system for vibration mode identification of power transmission towers based on three-dimensional features
By constructing a three-dimensional feature space and utilizing triaxial acceleration, acoustic signature, and environmental data, the unified characterization and differentiation of vibration modes of power transmission towers were solved, enabling accurate identification and continuous online monitoring of multiple states.
Patent Information
- Application Number
- CN202611541306.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-30
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies cannot incorporate the various states of transmission towers, such as vibration, acoustic signature, and environmental excitation, into the same model for unified characterization and differentiation, and they also ignore the coupling effect between conductors and towers.
A three-dimensional feature recognition method is adopted. By collecting triaxial acceleration, acoustic text and environmental data, a three-dimensional feature space is constructed, which maps different vibration modes, and the current mode is determined by Mahalanobis distance.
It achieves unified characterization and differentiation of multiple vibration modes of power transmission towers, overcomes the problems of fragmented perception dimensions and lack of a unified state characterization framework, and can continuously monitor online and improve discrimination accuracy.
Smart Images

Figure CN123111455A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of power transmission tower condition monitoring technology, and in particular to a method and system for identifying vibration modes of power transmission towers based on three-dimensional features. Background Technology
[0002] Transmission towers are subjected to various load excitations during actual operation, and swaying or vibration phenomena are common. Related research shows that their vibration response is the result of the combined effects of external environmental conditions and the structure's own state. According to the different vibration sources, tower vibration can be divided into: ① Wind-induced vibration. Under wind loads, the tower body and conductors produce continuous vibration. Normal wind-induced vibration amplitude is within a safe range; however, when the wind speed is too high (such as typhoons or hurricanes) or the wind direction is at a specific angle to the line direction, it may cause abnormal large-scale swaying or even twisting of the tower. ② Conductor galloping. Under uneven icing conditions, the conductors generate low-frequency (0.1Hz~3Hz), large-amplitude self-excited vibrations under wind excitation, which are transmitted to the tower crossarms through the suspension strings and tension strings, causing additional tower vibration. Galloping is one of the most serious disasters of transmission lines, which can lead to deformation of the tower crossarms, loosening of bolts, and even tower collapse. ③ Temperature-induced vibration. High summer temperatures increase conductor sag and decrease the material's elastic modulus, causing tower vibration frequency drift and amplitude increase. Low winter temperatures, accompanied by icing loads, place additional weight on the tower, altering its vibration characteristics. ④ Bolt loosening and structural damage. Under long-term alternating loads, bolts at tower connection points may loosen, leading to a decrease in local stiffness and changes in acoustic signature; in severe cases, this can develop into structural damage, with a significant decrease in natural frequency. ⑤ Geological hazard effects. Transmission tower foundations are affected by geological hazards such as landslides, collapses, debris flows, and earthquakes, causing abnormal upward fluctuations that lead to tower swaying, and in severe cases, tower tilting or even collapse.
[0003] Currently, methods for monitoring the vibration status of transmission towers can be mainly categorized as follows: (1) Vibration monitoring methods based on a single type of sensor. Patent CN107340053A discloses "A multi-sensor data fusion system and method for monitoring the stability of transmission line towers", which uses three vibration sensors (tower top, tower waist, and tower base) and one tilt sensor (crossarm). Patent CN121279814B discloses "An online monitoring and early warning processing system for abnormal transmission towers". This type of method is essentially a redundant arrangement of the same type of sensor, focusing only on the vibration response itself, and failing to incorporate external excitation conditions and local structural integrity into a unified analysis framework.
[0004] (2) Bolt loosening detection method based on acoustic signature. Patents CN119935517A, CN118212938B, and CN121499041A all disclose methods for detecting bolt loosening based on acoustic signature. These methods mostly focus on the single fault mode of bolt loosening, require active excitation, and do not conduct joint analysis with the overall vibration mode of the tower or external environmental excitation.
[0005] (3) Identification methods based on finite element simulation and modal analysis. Patent CN115795325B discloses "an automatic verification method and system for identifying modal parameters of transmission towers". This type of method requires active excitation or relies on an accurate finite element model, making it difficult to achieve continuous online monitoring.
[0006] (4) Methods for monitoring conductor galloping. Existing methods mostly identify galloping by installing accelerometers on the conductors or by video monitoring. However, it is difficult to power the sensors installed on the conductors and transmit data, and the coupling effect of conductor galloping on the towers is not included in the assessment.
[0007] In summary, existing technologies have the following main shortcomings: ① Fragmented perception dimensions. Vibration, acoustic signatures, environmental excitations, and conductor states are analyzed independently; ② Lack of a unified state representation framework. It is impossible to incorporate multiple states such as wind-induced vibration, torsion, temperature effects, icing, galloping, and bolt loosening into the same model for unified representation and differentiation; ③ The conductor-tower coupling effect is ignored. Existing methods treat the conductor and tower as independent systems, failing to identify the characteristic patterns of conductor galloping transmitted to the tower. Therefore, this application proposes a method and system for identifying the vibration modes of transmission towers based on three-dimensional features. Summary of the Invention
[0008] Therefore, it is necessary to provide a method, system, and computer equipment for identifying vibration modes of power transmission towers based on three-dimensional features, in order to address the aforementioned technical problems.
[0009] Firstly, this application provides a method for identifying vibration modes of power transmission towers based on three-dimensional features. The method includes: Collect multi-source sensing data from power transmission towers, including triaxial acceleration data, acoustic data, and environmental data; The multi-source sensing data is preprocessed to obtain preprocessed multi-source sensing data. Feature extraction is performed on the preprocessed multi-source sensing data to obtain driving features extracted based on the environmental data, voiceprint features extracted based on the voiceprint data, and vibration features extracted based on the triaxial acceleration data. A three-dimensional feature space is constructed using the driving feature, the acoustic signature feature, and the vibration feature as three coordinate axes, and the various vibration modes of the power transmission tower are mapped to state regions in the three-dimensional feature space. The driving characteristics, acoustic characteristics, and vibration characteristics at the current moment are mapped to three-dimensional feature points in the three-dimensional feature space. Based on the distance between the three-dimensional feature points and each of the state regions, the current vibration mode of the transmission tower is determined.
[0010] In one embodiment, the multi-source sensing data is collected through a multi-source sensing sensor network deployed on the transmission tower; the collected multi-source sensing data from the transmission tower includes: The triaxial acceleration data is collected by installing a triaxial accelerometer on the power transmission tower. By installing sound sensors on the power transmission tower, multi-channel acoustic data is collected. By installing wind speed sensors, temperature sensors, and humidity sensors on the transmission tower, wind speed data, temperature data, and humidity data at the location of the transmission tower are collected respectively, thus forming the environmental data; The triaxial acceleration data, the voiceprint data, and the environmental data are timestamped and aligned using a set time reference to construct the multi-source sensing data. In one embodiment, the multi-source sensing data is preprocessed to obtain the preprocessed multi-source sensing data, which includes: The triaxial acceleration data and the acoustic print data are respectively bandpass filtered, the filtered frequency band of the triaxial acceleration data covers the structural vibration frequency band of the power transmission tower, and the filtered frequency band of the acoustic print data covers the audible sound frequency band; A notch filter is used to suppress power frequency interference in the triaxial acceleration data and the acoustic signature data. Outliers in each channel's data are identified and removed based on the 3σ criterion, and then replaced with linear interpolations of the data before and after the outlier. The preprocessed multi-source sensing data is divided into time windows of a set length, ranging from 1 minute to 60 minutes.
[0011] In one embodiment, the extraction of the driving features includes: extracting wind speed statistical features from the wind speed sequence, the wind speed statistical features including average wind speed, maximum wind speed, wind speed standard deviation, and wind speed change rate; extracting temperature features from the temperature sequence, the temperature features including current temperature value and temperature change rate; determining an icing sensitivity indicator factor based on temperature and humidity, wherein when the temperature is within the icing sensitivity range and the humidity exceeds a set threshold, the icing sensitivity indicator factor takes a first value, otherwise it takes a second value; combining the wind speed statistical features, the temperature features, and the icing sensitivity indicator factor into a driving feature vector, and performing standardization and dimensionality reduction processing on the driving feature vector to obtain the driving feature axis coordinates of the three-dimensional feature space.
[0012] In one embodiment, the extraction of the voiceprint features includes: The acoustic data is processed by frame segmentation and windowing to extract Mel frequency cepstral coefficients and their first and second differences, forming acoustic temporal features. Short-time energy and short-time zero-crossing rate are extracted from the acoustic data of each channel. High-frequency band energy analysis is performed on the acoustic data of each channel to calculate the intensity and temporal kurtosis of the high-frequency impact component, where the intensity of the high-frequency impact component is the ratio of high-frequency band energy to total energy. The multi-channel acoustic energy distribution vector is calculated, along with the ratio of the energy of each channel in the multi-channel acoustic energy distribution vector to the baseline energy of the corresponding channel under pre-calibrated tower health conditions. The energy proportion of the acoustic data in the 200Hz to 800Hz frequency band is extracted as a segment breakage discrimination feature. The acoustic temporal features, short-time energy, short-time zero-crossing rate, high-frequency impact component intensity, temporal kurtosis, multi-channel acoustic energy distribution vector, and segment breakage discrimination feature are combined into an acoustic feature vector. The acoustic feature vector is then standardized and dimensionality-reduced to obtain the acoustic feature axis coordinates in the three-dimensional feature space.
[0013] In one embodiment, the vibration feature extraction includes: performing spectral analysis on the triaxial acceleration data to extract the dominant frequency, vibration amplitude, and low-frequency energy proportion of the X-axis and Y-axis, respectively; extracting torsional components, which include the phase difference between the X-axis and Y-axis accelerations at the dominant frequency, the torsional amplitude, and the torsional dominant frequency, and determining a torsional mode when the phase difference is within a preset torsional judgment interval; extracting conductor-tower coupled vibration features, which include the correlation coefficient between the X-axis and Y-axis accelerations, the Lissajous figure trajectory shape, and the low-frequency amplitude component ratio; combining the X-axis and Y-axis spectral features, the torsional components, and the conductor-tower coupled vibration features into a vibration feature vector, and standardizing and reducing the dimensionality of the vibration feature vector to obtain the vibration feature axis coordinates in the three-dimensional feature space.
[0014] In one embodiment, mapping the various vibration modes of the transmission tower to state regions in the three-dimensional feature space includes: Under the health status of the transmission tower, operational data for a set period of time is collected and a benchmark sample set is extracted. The benchmark status region is determined based on the mean vector and covariance matrix of the benchmark sample set. Based on the tower structure dynamics model and historical fault data calibration, the center position and regional range of multiple vibration modes in the three-dimensional feature space are determined. The multiple vibration modes include wind-induced vibration anomaly, wind-induced torsional anomaly, high-temperature vibration anomaly, icing vibration anomaly, conductor galloping, galloping causing bolt loosening, bolt loosening anomaly, structural damage anomaly, conductor strand breakage anomaly, and icing falling impact. The step of determining the current vibration mode of the transmission tower based on the distance between the three-dimensional feature points and each of the state regions includes: calculating the Mahalanobis distance from the three-dimensional feature points to the center of each of the state regions, determining the vibration mode corresponding to the state region with the smallest Mahalanobis distance as the current vibration mode, and determining the discrimination confidence level based on the smallest Mahalanobis distance and the Mahalanobis distance in the reference state.
[0015] In one embodiment, the method further includes: Calculate the voiceprint energy of each channel in the voiceprint data, determine the abnormal voiceprint channel based on the ratio of the voiceprint energy of each channel to the corresponding channel reference energy in the pre-calibrated healthy state, and determine the installation position of the sound sensor corresponding to the abnormal voiceprint channel as the abnormal part. The three-dimensional feature points at multiple consecutive moments are connected in chronological order to form a state trajectory. When the state trajectory points to the same state region for a set number of consecutive moments and the trajectory velocity increases, a state deterioration warning is issued. When the three-dimensional feature points deviate from the reference state region and do not enter any other state region, an early warning is issued. The vibration mode at future moments is predicted by extrapolation based on the displacement direction and velocity of the state trajectory. After accumulating newly added state samples that have been confirmed to be without misjudgment, an incremental learning method is used to update the center position and covariance matrix of each state region; When the Mahalanobis distance from the three-dimensional feature point to the center of all known state regions exceeds a set threshold and continues for a set duration, a new state discovery alarm is triggered.
[0016] Secondly, this application also provides a vibration mode identification system for power transmission towers based on three-dimensional features. The system includes: The data acquisition module is used to collect multi-source sensing data from the power transmission tower, including triaxial acceleration data, acoustic data, and environmental data. The preprocessing module is used to preprocess the multi-source sensing data to obtain the preprocessed multi-source sensing data. The feature extraction module is used to extract features from the preprocessed multi-source sensing data to obtain driving features extracted based on the environmental data, voiceprint features extracted based on the voiceprint data, and vibration features extracted based on the triaxial acceleration data. The feature space construction module is used to construct a three-dimensional feature space with the driving feature, the acoustic feature and the vibration feature as three coordinate axes respectively, and to map the various vibration modes of the power transmission tower as state regions in the three-dimensional feature space respectively; The modal recognition module is used to map the driving features, acoustic features, and vibration features at the current moment into three-dimensional feature points in the three-dimensional feature space, and determine the current vibration mode of the transmission tower based on the distance between the three-dimensional feature points and each of the state regions.
[0017] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method described in any of the above embodiments.
[0018] The above-mentioned method, system, and computer equipment for identifying vibration modes of transmission towers based on three-dimensional features have at least the following beneficial effects: 1. By synchronously collecting three types of multi-source data that reflect external excitation conditions, local structural integrity, and overall dynamic response, driving information, acoustic fingerprint information, and vibration information are incorporated into the same sensing system, overcoming the shortcomings of existing technologies where vibration, acoustic fingerprint, and environmental excitation are analyzed independently and sensing dimensions are fragmented. 2. By uniformly filtering, removing outliers, and dividing time windows on multi-source data, the effects of power frequency interference, environmental noise, and instantaneous outliers are suppressed, thereby improving the noise resistance and stability of subsequent feature extraction and modality discrimination. 3. By extracting driving features based on environmental data, extracting voiceprint features based on voiceprint data, and extracting vibration features based on triaxial acceleration data, three mutually independent feature dimensions representing "external excitation - local integrity - overall response" are formed, realizing the quantitative representation of the complete causal chain of "excitation → state → response". Among them, the vibration features naturally include the response component transmitted from the conductor galloping to the tower, so that the conductor-tower coupling effect ignored in the existing technology can be included in the identification. 4. By constructing a physically orthogonal three-dimensional feature space and mapping multiple vibration modes such as wind-induced vibration, torsion, temperature effect, icing, conductor galloping, and bolt loosening into a unified state region in the space, the problem of existing technologies lacking a unified state representation framework and being unable to represent and distinguish multiple modes under the same model is solved. 5. Objective quantitative determination of modality attribution is achieved by judging the distance between three-dimensional feature points and state regions. It can be continuously identified online 24 / 7 without active excitation. The accuracy of the discrimination results continues to improve with the accumulation of state region calibration samples, overcoming the shortcomings of existing methods that rely on active excitation or precise finite element models and are difficult to monitor continuously online. Attached Figure Description
[0019] Figure 1 This is an application environment diagram of a transmission tower vibration mode identification method based on three-dimensional features in one embodiment; Figure 2 This is a flowchart illustrating a method for identifying vibration modes of transmission towers based on three-dimensional features in one embodiment. Figure 3 This is a schematic diagram of the arrangement of a multi-source sensing sensor network in one embodiment; Figure 4 This is a flowchart illustrating the data preprocessing steps in one embodiment; Figure 5 This is a schematic diagram of the distribution of three-dimensional feature space and state regions in one embodiment; Figure 6 This is a flowchart illustrating the state determination and result output steps in one embodiment; Figure 7 This is a structural block diagram of a power transmission tower vibration mode identification system based on three-dimensional features in one embodiment; Figure 8 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation
[0020] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0021] The vibration mode identification method for transmission towers based on three-dimensional features provided in this application can be applied to, for example... Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or located in the cloud or on other network servers. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, etc. Portable wearable devices can include smartwatches, smart bracelets, head-mounted devices, etc. Server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers.
[0022] In one embodiment, such as Figure 2 As shown, a method for vibration mode identification of power transmission towers based on three-dimensional features is provided, and this method is applied to... Figure 1 Taking server 104 as an example, the following steps are included: Step 202: Collect multi-source sensing data of the power transmission tower, including triaxial acceleration data, acoustic data and environmental data.
[0023] Multi-source sensing data refers to heterogeneous data collected by a multi-source sensing sensor network deployed on transmission towers and time-stamped. Environmental data serves as the "input" information for tower vibration, characterizing the excitation conditions of the external environment on the tower; acoustic fingerprint data characterizes the local integrity state of the tower structure; and triaxial acceleration data characterizes the overall dynamic response state of the tower. Specifically, the multi-source sensing sensor network collects these three types of data in real time and uploads them to server 104 via wired or wireless means.
[0024] Step 204: Preprocess the multi-source sensing data to obtain preprocessed multi-source sensing data.
[0025] The preprocessing step eliminates the impact of environmental noise, power frequency electromagnetic interference, and outliers on subsequent feature extraction. It also divides the continuously collected data into several time windows to improve feature stability. Specifically, server 104 performs bandpass filtering and power frequency interference suppression on the triaxial acceleration data and acoustic signature data, respectively, and divides the data into time windows of a set length after removing outliers.
[0026] Step 206: Perform feature extraction on the preprocessed multi-source sensing data to obtain driving features extracted based on the environmental data, voiceprint features extracted based on the voiceprint data, and vibration features extracted based on the triaxial acceleration data.
[0027] Among them, driving features, acoustic features, and vibration features are used to quantify the intensity and type of external excitation, the local integrity state of the structure, and the overall dynamic response state of the structure, respectively. Specifically, server 104 calculates the above three types of features in each time window, and obtains the one-dimensional coordinate values of each feature axis after standardization and dimensionality reduction.
[0028] Step 208: Construct a three-dimensional feature space using the driving feature, the acoustic signature feature, and the vibration feature as three coordinate axes, and map the various vibration modes of the power transmission tower as state regions in the three-dimensional feature space.
[0029] The three-dimensional feature space refers to the three-dimensional space spanned by the driving feature axis, the acoustic signature feature axis, and the vibration feature axis. The three coordinate axes are physically independent and orthogonal, forming a complete description space of the tower's vibration state. Specifically, server 104 first calibrates the baseline state region of the tower under healthy conditions, and then determines the regional center and range of various typical abnormal states through theoretical analysis and experimental calibration, so that each vibration mode corresponds to a region in the three-dimensional feature space.
[0030] Step 210: Map the driving features, acoustic features, and vibration features at the current moment to three-dimensional feature points in the three-dimensional feature space, and determine the current vibration mode of the transmission tower based on the distance between the three-dimensional feature points and each of the state regions.
[0031] The position of the three-dimensional feature point in the three-dimensional feature space at the current moment reflects the current excitation conditions, local integrity, and overall response state of the tower. Specifically, server 104 calculates the Mahalanobis distance from the three-dimensional feature point to the center of each state region, determines the vibration mode corresponding to the state region with the smallest Mahalanobis distance as the current vibration mode, and outputs the confidence level and suggested maintenance measures.
[0032] The above-mentioned method for identifying vibration modes of transmission towers based on three-dimensional features maps external environmental excitation, local structural integrity state, and overall structural dynamic response into three physically orthogonal feature dimensions and constructs a unified three-dimensional feature space. This incorporates multi-source heterogeneous information into the same state representation framework, transforming the essential differences between different vibration modes into differences in position and distance in three-dimensional space. This achieves unified representation and accurate differentiation of multiple states, solving the technical problems of fragmented perception dimensions and lack of a unified state representation framework in the background technology.
[0033] In one embodiment, such as Figure 3 As shown, step 202 includes: Step 302: Collect triaxial acceleration data of the power transmission tower by installing a triaxial accelerometer on the power transmission tower.
[0034] The triaxial accelerometer is installed at the top main structural member node of the tower head. This location represents the area of maximum amplitude in the tower's vibration response, allowing for the most comprehensive reflection of the tower's overall vibration modal characteristics. Optionally, the triaxial accelerometer is a MEMS capacitive triaxial accelerometer with a range of at least ±2g, a resolution of at least 0.1mg, and a sampling frequency of at least 100Hz, covering the main frequency band of the tower structure's vibration (typically 0.1Hz to 50Hz). The X-axis is defined as along the track direction, the Y-axis as perpendicular to the track direction (crossarm direction), and the Z-axis as vertically upward.
[0035] Step 304: Collect multi-channel acoustic data by installing sound sensors on the power transmission tower.
[0036] Specifically, sound sensors are installed at several key locations, including the ends of the crossarms of each phase conductor, both ends of the ground crossarm, and at least three of the main tower leg components. In this embodiment, sound sensors are deployed at six key locations on the tower, specifically: channel 1 is at the end of the A-phase conductor crossarm, channel 2 is at the end of the B-phase conductor crossarm, channel 3 is at the end of the C-phase conductor crossarm, channel 4 is at the left end of the ground crossarm, channel 5 is at the right end of the ground crossarm, and channel 6 is at the main tower leg component. The determination of these locations is based on the following: the ends of the crossarms are where the displacement amplitude is largest during tower vibration, and they are also where the conductor-tower connection nodes are located, with dense bolt connections, making them the area with the richest acoustic signature data; the tower legs are the starting point of the vibration transmission path, and acoustic signature data is transmitted from the tower body to here, which can be used to determine the longitudinal location of abnormal sound sources. Optionally, the sound sensor uses an electret condenser microphone or a MEMS digital microphone with a frequency response range of not less than 20Hz to 20kHz, a sensitivity of not less than -26dBFS, and a sampling frequency of not less than 44.1kHz. The number of sound sensors can be reduced to 4 or increased to more than 8, depending on the size of the tower and the required monitoring accuracy.
[0037] Step 306: By installing wind speed sensors, temperature sensors, and humidity sensors on the power transmission tower, wind speed data, temperature data, and humidity data at the location of the power transmission tower are collected respectively to form the environmental data.
[0038] Specifically, wind speed sensors, temperature sensors, and humidity sensors are installed on the tower head or body of the power transmission tower.
[0039] Optionally, the wind speed sensor uses a three-cup anemometer or an ultrasonic anemometer with a measurement range of 0–60 m / s and an accuracy of not less than ±0.5 m / s; the temperature sensor uses a platinum resistance or semiconductor temperature sensor with a measurement range of -40℃ to +85℃ and an accuracy of not less than ±0.5℃; the humidity sensor uses a capacitive or resistive humidity sensor with a measurement range of 0–100%RH and an accuracy of not less than ±3%RH, used to assist in determining icing / galloping weather conditions.
[0040] Step 308: Timestamp align the triaxial acceleration data, the voiceprint data, and the environmental data using a set time reference to construct the multi-source sensing data.
[0041] Specifically, time synchronization is based on GPS or Network Time Protocol (NTP), and the timestamps of each sensor's data are aligned to ensure the temporal consistency of multi-source data. The aforementioned acceleration data, acoustic data, and environmental data are synchronized and integrated to form a multi-source sensing data vector at time t. Data collected continuously for T time points constitutes multi-source sensing data. In this embodiment, by arranging acceleration, sound, and environmental sensors in the maximum amplitude area of the tower head, the dense node area at the end of the crossarm, and the starting point of the transmission path of the tower leg, the multi-source sensing data can completely cover the causal chain of "excitation - local state - overall response" under the same time reference, providing a data foundation for the construction of the three-dimensional feature space.
[0042] In one embodiment, such as Figure 4 As shown, step 204 includes: Step 402: Bandpass filtering is performed on the triaxial acceleration data and the acoustic data respectively. The filtered frequency band of the triaxial acceleration data covers the structural vibration frequency band of the power transmission tower, and the filtered frequency band of the acoustic data covers the audible frequency band.
[0043] Optionally, a fourth-order Butterworth bandpass filter is used to filter the triaxial acceleration signal, with a filtering frequency band of 0.1Hz to 50Hz, to retain the main frequency components of the tower structure vibration and filter out high-frequency environmental noise and sensor self-noise; its amplitude-frequency response has a maximum attenuation of no more than 3dB in the passband and a stopband attenuation of no less than 40dB. A fourth-order Butterworth bandpass filter is used to filter the 6-channel acoustic data, with a filtering frequency band of 20Hz to 20kHz, to retain complete acoustic information and filter out interference from infrasound and ultrasonic frequencies.
[0044] Step 404: Use a notch filter to suppress power frequency interference on the triaxial acceleration data and the acoustic data.
[0045] Specifically, to filter out power frequency electromagnetic interference at 50Hz and its harmonics (100Hz, 150Hz, 200Hz, etc.), the transfer function of the notch filter includes the normalized notch frequency. ω 0=2 π ×50 / fs , fs Sampling frequency, r This is the notch width control parameter (usually taken as 0.95 to 0.99).
[0046] Step 406: Identify and remove outliers in each channel's data based on the 3σ criterion, and replace them with linear interpolations of the data before and after the outlier.
[0047] Specifically, for time series data of a certain channel { x 1, x 2,..., x L (L is the length of the time window), calculate its mean. μ and standard deviation σ :, when | xi - μ |>3 σ At that time, the judgment x i For outliers, use linear interpolation of the data before and after that point for replacement: x i =( x i -1+ x i +1) / 2.
[0048] Step 408: Divide the preprocessed multi-source sensing data into time windows of a set length, the length of which is 1 minute to 60 minutes.
[0049] Specifically, the time window length L is set according to the analysis requirements; in this embodiment, a 30-minute time window is preferred. When the sampling frequency is 100Hz, a 30-minute window contains 180,000 data points, and each time window contains enough data points for statistical feature extraction. A shorter window results in higher real-time performance but slightly lower feature stability; a longer window results in higher feature stability but slightly lower real-time performance.
[0050] In this embodiment, a combination of preprocessing techniques, including filtering, notch filtering, outlier removal, and time window division, is used to retain the dominant frequency components of the tower structure's vibration and complete acoustic information while suppressing power frequency interference and transient noise, thus ensuring consistency of subsequent feature extraction results across time windows.
[0051] In one embodiment, the extraction of driving features in step 206 includes: Step 502: Extract wind speed statistical features from the wind speed sequence. The wind speed statistical features include average wind speed, maximum wind speed, wind speed standard deviation, and wind speed change rate.
[0052] Specifically, the average wind speed reflects the overall wind force level within the time period; the maximum wind speed reflects the peak wind speed; the standard deviation of wind speed reflects the intensity of wind speed fluctuations, i.e., the volatility of wind load; and the rate of change of wind speed reflects the degree of abrupt changes in wind speed, used to identify gusts or sudden wind speed changes. Furthermore, for the wind speed sequence within the nth time window, a wind speed driving intensity index is obtained by weighting the above wind speed characteristics: in, The average wind speed, At maximum wind speed, For wind speed standard deviation, α 1. α 2. α 3 represents the weighting coefficients, determined through principal component analysis, and satisfies... α 1+ α 2+ α 3 = 1.
[0053] Step 504: Extract temperature features from the temperature sequence, the temperature features including the current temperature value and the rate of temperature change.
[0054] Specifically, the current temperature value (the temperature value at the end of the window) and the rate of temperature change are extracted from the temperature sequence within the nth time window. The temperature state type is determined based on the temperature value: a temperature greater than 35℃ is a high temperature state; a temperature less than -5℃ is a low temperature state; a temperature between -5℃ and 35℃ is determined to be a normal temperature state; when the temperature is between -5℃ and 5℃, if the humidity is greater than 80%, it is determined to be an icing-sensitive temperature. Both the endpoint values of -5℃ and 5℃ are included in this temperature range, and the determination of the icing-sensitive temperature takes precedence over the normal temperature state; otherwise, it is classified as a normal temperature state.
[0055] Step 506: Determine the icing sensitivity indicator factor based on temperature and humidity. When the temperature is within the icing sensitivity range and the humidity exceeds a set threshold, the icing sensitivity indicator factor takes a first value; otherwise, it takes a second value.
[0056] Specifically, the icing sensitivity indicator factor is determined based on temperature and humidity: when the temperature is between -5℃ and 5℃ and the humidity is greater than 80%, the icing sensitivity indicator factor is set to 1, indicating that the current meteorological conditions are conducive to conductor icing and galloping; otherwise, it is set to 0.
[0057] Step 508: Combine the wind speed statistical features, the temperature features, and the icing sensitivity indicator into a driving feature vector, and perform standardization and dimensionality reduction on the driving feature vector to obtain the driving feature axis coordinates of the three-dimensional feature space.
[0058] Specifically, the driving feature vector, composed of driving features, is Z-score standardized to eliminate differences in dimensions and orders of magnitude among different features. Principal component analysis is then used to reduce the dimensionality of the standardized driving feature vector, extracting principal components with a cumulative variance contribution rate of 85%. The scores of these principal components are used as the coordinate values of the driving feature axes in the three-dimensional feature space. If the variance contribution rate of the first principal component reaches 85%, its score is used as the coordinate value of the driving feature axis; otherwise, the scores of the first k principal components are taken and weighted by the proportion of each principal component's eigenvalue to obtain the coordinate value of the driving feature axis. k is the minimum number of principal components required to achieve a cumulative variance contribution rate of 85%. In another embodiment, the dimensionality reduction method can also employ linear discriminant analysis, t-distributed random neighborhood embedding, or an autoencoder.
[0059] In this embodiment, the driving feature vector fully characterizes the wind load intensity, pulsation characteristics, abrupt change degree, and temperature and icing meteorological conditions, so that the external excitation conditions can be quantified into a single coordinate value on the driving feature axis.
[0060] In one embodiment, voiceprint feature extraction in step 206 includes: Step 602: Perform frame-segmentation and windowing processing on the voiceprint data.
[0061] Specifically, the voiceprint data within the nth time window is divided into frames and windowed, with a frame length of 20ms to 40ms (preferably 25ms in this embodiment), and the frame shift is 50% of the frame length (i.e., 12.5ms) to ensure smooth transition between frames; a Hamming window is applied to each frame of voiceprint data to reduce spectral leakage.
[0062] Step 604: Extract the Mel frequency cepstral coefficients and their first and second differences to form the audioprint temporal features.
[0063] Specifically, a fast Fourier transform is performed on the windowed voiceprint data of each frame to obtain the power spectrum. The power spectrum is then passed through a Mel-Frequency Cepstral Coefficients (MFCCs) filter bank (26 triangular filters are used in this embodiment, evenly distributed in the Mel-Frequency domain). After taking the logarithm of the output of each filter, a Discrete Cosine Transform is performed to extract the first 13-dimensional Mel-Frequency Cepstral Coefficients. The first-order difference, which represents the rate of change between frames, and the second-order difference, which represents the trend of change between frames, are further extracted. The 13-dimensional Mel-Frequency Cepstral Coefficients (MFCCs) and their first-order and second-order differences are concatenated to form a 39-dimensional feature vector, which serves as the voiceprint temporal feature.
[0064] Step 606: Extract the short-time energy and short-time zero-crossing rate of the voiceprint data.
[0065] Specifically, short-time energy reflects the intensity of the voiceprint data. When the bolts loosen and cause abnormal vibration, the short-time energy of the corresponding channel will increase significantly. The short-time zero-crossing rate reflects the frequency structure of the voiceprint data. There is a difference between the zero-crossing rate in the normal state and the abnormal state.
[0066] Step 608: Perform high-frequency band energy analysis on the multi-channel acoustic data, and calculate the intensity and temporal kurtosis of the high-frequency impact component. The intensity of the high-frequency impact component is the ratio of the high-frequency band energy to the total energy.
[0067] Specifically, high-frequency band (greater than 1kHz) energy analysis is performed on the voiceprint data, and the intensity of the high-frequency impact component is characterized by the ratio of high-frequency band energy to the total energy of the voiceprint data. in, The total energy of the voiceprint signal. It carries high-frequency energy. When the bolt loosens, the intensity of the high-frequency impact component increases significantly. HFC ch Exceeding the normal range.
[0068] Further time-domain analysis of the voiceprint data was performed to detect the presence of periodic spike pulses and to calculate the kurtosis of the time-domain signal. in, For the first preprocessed ch The first channel of voiceprint data k The sound pressure value (unit: Pa) at the sampling point, where K is the sound pressure value at the sampling point within the time window. ch The total number of sampling points for channel voiceprint data. for The mean value. Under normal conditions, the kurtosis of the voiceprint data is close to 3. The impact signal generated by loose bolts causes the time-domain kurtosis to increase significantly (greater than 5), which is used to determine whether there is an impact voiceprint anomaly in the corresponding channel.
[0069] Step 610: Calculate the multi-channel acoustic energy distribution vector and the ratio of the energy of each channel in the multi-channel acoustic energy distribution vector to the reference energy of the corresponding channel under the pre-calibrated tower health condition.
[0070] Specifically, the vector of energy distribution reflects the spatial location of the abnormal sound source, and the channel with the closest distance to the sound source has a greater energy; when the ratio of the energy of a certain channel to the reference energy under healthy tower conditions is greater than 1.5, it indicates that the energy of that channel is significantly abnormal.
[0071] Step 612: Extract conductor strand breakage detection features. When a conductor breaks, the broken conductor will produce vibration sound patterns at a specific frequency under wind load. Its typical feature is a significant increase in energy in the range of 200Hz to 800Hz, and the amplitude changes periodically with wind speed fluctuations. The energy proportion of this specific frequency band is extracted as the strand breakage discrimination feature.
[0072] Step 614: Combine the audioprint temporal features, short-time energy, short-time zero-crossing rate, high-frequency impact component intensity, temporal kurtosis, multi-channel energy distribution, and conductor strand breakage detection features into an audioprint feature vector. Standardize and reduce the dimensionality of the audioprint feature vector to obtain the audioprint feature axis coordinates in the three-dimensional feature space.
[0073] Specifically, referring to the standardization and dimensionality reduction method in step 508, the principal component scores with a cumulative variance contribution rate of 85% are used as the voiceprint feature axis coordinates in the three-dimensional feature space.
[0074] In this embodiment, the voiceprint features take into account the spectral envelope, temporal variation, energy distribution, impact characteristics and spatial distribution information, and correspond local anomalies with sensor installation locations through multi-channel energy ratios, providing a basis for the spatial localization of voiceprints in abnormal locations.
[0075] In one embodiment, vibration feature extraction in step 206 includes: Step 702: Perform spectral analysis on the triaxial acceleration data.
[0076] Specifically, no. n Within a time window, the triaxial acceleration sequence { a x.t}、{ a y.t}、{ a z.t Perform Fast Fourier Transform (FFT) separately to obtain the power spectral density along the X and Y axes: P x ( f ), P y ( f ), P z ( f ): Similarly, we can obtain Py ( f )and Pz ( f The number of FFT points is taken as the smallest power of 2 greater than or equal to the time window length, and the frequency resolution is the ratio of the sampling frequency to the number of FFT points.
[0077] Step 704: Extract the X-axis oscillation component, including the X-axis dominant frequency, X-axis vibration amplitude, and the energy proportion of the X-axis low-frequency band (0.2Hz~5Hz); extract the Y-axis dominant frequency, Y-axis vibration amplitude, and Y-axis low-frequency band energy proportion in the same way.
[0078] Specifically, taking the X-axis as an example: X-axis main frequency f x =argmax P x ( f This reflects the main frequency components of the transmission tower's vibration in the X direction (line direction); X-axis vibration amplitude It reflects the intensity of vibration in the X direction, with units of (m / s²). 2 ); X-axis low-frequency energy percentage: It is used to determine whether the vibration is mainly wind-induced low-frequency oscillation. Under normal circumstances, this value is greater than 0.6.
[0079] Step 706: Extract the torsional component. The torsional component includes the phase difference between the X-axis acceleration and the Y-axis acceleration at the main frequency, the torsional amplitude, and the torsional main frequency. When the phase difference is within the preset torsional determination range, it is determined to be a torsional mode.
[0080] Specifically, calculate the cross-power spectrum of the X-axis acceleration and the Y-axis acceleration: P xy ( f )=F{ ax.t}·F{ ay . t} ∗ Where (·)* represents complex conjugation, taking the phase difference at the dominant frequency: ϕ xy =∠ P xy ( f main ), where ∠ represents the complex argument; when the phase difference is between 60° and 120°, the tower is determined to be in torsional mode; when the phase difference is between -30° and 30° or between 150° and 210°, it is determined to be in unidirectional oscillation mode; simultaneously, the torsional amplitude is calculated: With the twisting main frequency f t =f main When the torsional amplitude increases significantly and the phase difference approaches 90°, it is determined to be an abnormal torsion.
[0081] Step 708: Extract conductor-tower coupled vibration characteristics, which include the correlation coefficient between X-axis acceleration and Y-axis acceleration, the trajectory shape of the Lissajous figure (a two-dimensional trajectory plotted with X-axis acceleration as the abscissa and Y-axis acceleration as the ordinate), and the low-frequency amplitude component ratio.
[0082] Specifically, the Pearson correlation coefficient between the X-axis acceleration and the Y-axis acceleration is calculated: in a ˉ x and a ˉ y These are the mean values of the acceleration along the X-axis and Y-axis, respectively.
[0083] Specifically, a Pearson correlation coefficient greater than 0.7 indicates a strong correlation between the X-axis and Y-axis vibrations, which is a typical characteristic of conductor galloping transmitted to the tower. The Pearson correlation coefficient is close to 0 during normal wind-induced vibration and close to 0 during torsion.
[0084] A Lissajous figure is plotted with the X-axis acceleration as the abscissa and the Y-axis acceleration as the ordinate. Principal component analysis is performed on the data points, and the trajectory shape of the Lissajous figure is identified by combining the polar angle distribution. When the trajectory shape of the Lissajous figure is randomly scattered, it corresponds to normal wind-induced vibration; when it is elliptical or figure-eight shaped, it corresponds to conductor galloping transmission; when it is an inclined straight line, it corresponds to unidirectional oscillation; and when it is circular, it corresponds to torsional vibration.
[0085] Calculate the low-frequency amplitude component ratio: When the low-frequency amplitude component ratio is greater than 0.5, it indicates that the vibration energy is concentrated in the ultra-low frequency band (0.1Hz~3Hz), which is a typical characteristic of conductor galloping transmission.
[0086] Step 710: Construct a dance discrimination index based on the above dance characteristics. When the trajectory shape of the Lissajous figure is an ellipse or a figure-eight shape, the shape term in the dance discrimination index is set to 1; otherwise, it is set to 0. When the dance discrimination index is greater than 0.6, it is determined that conductor dance exists. Alternatively, dance discrimination can be performed using only one or both of the correlation coefficient, the trajectory shape of the Lissajous figure, and the low-frequency amplitude component ratio.
[0087] Step 712: Standardize and reduce the dimensionality of the vibration feature vector to obtain the vibration feature axis coordinates in the three-dimensional feature space.
[0088] Specifically, referring to the standardization and dimensionality reduction methods in step 508, standardization processing and principal component analysis dimensionality reduction are performed, and the principal component scores with a cumulative variance contribution rate of 85% are used as the coordinate values of the vibration characteristic axes in the three-dimensional feature space.
[0089] In this embodiment, by combining phase difference, correlation coefficient, Lissajous figure trajectory shape and low-frequency amplitude component ratio, the coupling response of conductor galloping transmitted to the tower is distinguished from the tower's own vibration response, filling the gap in the prior art where the conductor-tower coupling effect is ignored.
[0090] In one embodiment, such as Figure 5 and Figure 6 As shown, step 208 includes: Step 802: Construct a three-dimensional feature space using the driving feature, the acoustic feature, and the vibration feature as three coordinate axes.
[0091] Specifically, the three-dimensional feature space is defined as the space spanned by the driving feature axis, the acoustic signature feature axis, and the vibration feature axis: Ω={( x , y , z )∣ x ∈R, y ∈R, z ∈R}, where the driving characteristic axis characterizes the intensity and type of external environmental excitation, determined by a combination of wind speed and temperature characteristics; the acoustic signature characteristic axis characterizes the local integrity state of the structure, determined by a combination of multi-channel acoustic signature characteristics; and the vibration characteristic axis characterizes the overall dynamic response state of the structure, determined by a combination of oscillation, torsion, and galloping transmission characteristics. The vibration mode of the tower at any given time corresponds to a characteristic point P=( x , y , z ).
[0092] Step 804: Collect operating data for a set period of time under the health status of the transmission tower and extract a benchmark sample set. Determine the benchmark state region based on the mean vector and covariance matrix of the benchmark sample set.
[0093] Specifically, under the condition that the tower is in good health (confirmed by climbing the tower to ensure all bolts are tight and the structure is intact), at least 720 hours (30 days) of normal operation data are collected, and the three-dimensional feature points of each time window are extracted according to step 206 to form a benchmark sample set. ( N (The total number of reference samples); calculate the mean vector of the reference sample set. μ base With covariance matrix : The baseline state corresponds to the mean vector in three-dimensional space. μ base The ellipsoidal region centered on the chi-square distribution with 3 degrees of freedom, whose boundary is determined by the 95th quantile (taken as 7.815). .
[0094] Step 806: Based on the tower structure dynamics model and historical fault data calibration, determine the center position and region range of multiple vibration modes in the three-dimensional feature space. The multiple vibration modes include wind-induced vibration anomaly, wind-induced torsional anomaly, high-temperature vibration anomaly, icing vibration anomaly, conductor galloping, galloping causing bolt loosening, bolt loosening anomaly, structural damage anomaly, conductor strand breakage anomaly, and icing falling impact.
[0095] Specifically, the center locations and regional ranges of multiple vibration modes in the three-dimensional feature space are determined, including: (1) The wind-induced vibration is abnormal, the driving characteristic axis is significantly offset in the positive direction, the sound pattern is normal, the vibration amplitude exceeds 3 times the standard deviation of the reference value, and the phase difference is close to 0° or 180°. (2) The wind-induced torsion is abnormal, the driving characteristic shaft is moderately positively offset, the sound pattern is normal, the phase difference is between 60° and 120°, and the torsion amplitude is significantly increased. (3) Abnormal high-temperature vibration, temperature greater than 35℃, normal acoustic signature, main frequency drops more than 5% from the reference value and energy shifts to low frequency; (4) Abnormal icing vibration, the temperature is in the icing sensitive temperature range of -5℃ to 5℃, the humidity is greater than 80% and the icing sensitive indicator factor is 1, the sound text is normal, the vibration amplitude increases and the main frequency decreases. (5) When the conductor gallops, the icing sensitivity indicator factor is 1, the acoustic signature is normal, the main frequency is between 0.1Hz and 3Hz, the correlation coefficient is greater than 0.7, the low frequency amplitude component ratio is greater than 0.5, the trajectory shape of the Lissajous figure is coded as an ellipse or a figure "8", and the galloping discrimination index is greater than 0.6. Among them, the driving characteristics of the icing vibration anomaly and the conductor galloping both point to the icing sensitive temperature range. The two are distinguished by the values of the acoustic signature axis and the vibration signature axis. The icing vibration anomaly is characterized by an increase in vibration amplitude, a decrease in main frequency and no low frequency amplitude self-excitation component between 0.1Hz and 3Hz, while the conductor galloping is characterized by a low frequency amplitude self-excitation component and a correlation coefficient greater than 0.7. (6) The bolts loosened due to the dancing motion, and the driving and vibration characteristics were the same as those of the conductor dancing. In addition, the acoustic signature of a specific channel was abnormal and high-frequency impact components appeared. (7) The bolts are loose, the driving characteristics and temperature are normal, the acoustic energy ratio of a specific channel exceeds the limit and the kurtosis is greater than 5, and the vibration amplitude in a specific direction deviates from the reference value by 10% to 30%; (8) Abnormal structural damage, normal driving characteristics, abnormal multi-channel persistent acoustic signature, and a natural frequency drop of more than 5%; (9) Abnormal conductor strand breakage, the strand breakage discrimination feature of a specific channel at the end of the crossarm is significantly enhanced, that is, the energy proportion of the 200Hz to 800Hz frequency band exceeds the set threshold; (10) Ice shedding impact, temperature fluctuates around 0℃, multi-channel time-domain synchronous transient acoustic wave impact and the vibration response shows broadband transient impact components. The above state and the reference state together constitute no less than 11 vibration modes.
[0096] In this embodiment, by simultaneously incorporating excitation conditions, local integrity, and overall response into the calibration, the vibration modes are separated in the three-dimensional feature space, providing a spatial benchmark for unified multi-state discrimination. Alternatively, the construction of the three-dimensional feature space and state calibration can also be obtained by statistically analyzing normal and fault sample data confirmed by on-site operation and maintenance inspections.
[0097] In one embodiment, such as Figure 6 As shown, step 210 includes: Step 902: Calculate the Mahalanobis distance from each of the three-dimensional feature points to the center of each state region.
[0098] Specifically, for the three-dimensional feature point P at the current moment current =( xc , yc , zc ), calculate its Mahalanobis distance to the center of each state region: Where, Σ k For the first k The covariance matrix of the class state (calculated based on the calibrated samples of the corresponding state), and Σ−1 k is its inverse matrix.
[0099] Step 904: Determine the vibration mode corresponding to the state region with the minimum Mahalanobis distance as the current vibration mode.
[0100] Specifically, the three-dimensional feature points are categorized into the state region where the Mahalanobis distance is minimized: Status = argmin k∈{0,1,...,10} dk ( P current The vibration mode corresponding to this state region is the current vibration mode.
[0101] Step 906: Determine the discrimination confidence level based on the minimum Mahalanobis distance and the Mahalanobis distance under the reference state.
[0102] Specifically, the confidence level is calculated based on the minimum and maximum values of the Mahalanobis distance under the baseline condition: in, d min For the minimum Mahalanobis distance, dbase min and dbase max These are the minimum and maximum values of the Mahalanobis distance under the baseline condition, respectively. The confidence level ranges from 0 to 1, with a larger value indicating a more reliable judgment result.
[0103] Step 908: Calculate the voiceprint energy of each channel in the multi-channel voiceprint data, based on the ratio of the voiceprint energy of each channel to the pre-calibrated reference energy of the corresponding channel under healthy conditions: Identify the abnormal voiceprint channel and determine the installation location (R) of the corresponding sound sensor. ch The sensor installation location corresponding to the largest channel is identified as the abnormal location, and the specific name of the location is output (e.g., the end of the C-phase crossarm, the left end of the ground wire crossarm, the main material of the tower leg, etc.). In another embodiment, acoustic fingerprint localization can also be performed using the time difference of arrival localization method or the beamforming spatial spectrum estimation method.
[0104] Step 910: Connect the three-dimensional feature points at multiple consecutive moments in chronological order to form a state trajectory. When the state trajectory points to the same state region for a set number of consecutive moments and the trajectory velocity increases, a state deterioration warning is issued. When the three-dimensional feature points deviate from the reference state region and do not enter any other state region, an early warning is issued, and the vibration mode at future moments is predicted by extrapolation based on the displacement direction and velocity of the state trajectory.
[0105] Specifically, the three-dimensional feature points at multiple consecutive moments are connected in chronological order to form a state trajectory: Γ( t )={P( t 1), P( t 2),...,P( t n )}, in t 1< t 2<...< t n This refers to the sequence number of multiple consecutive moments; Calculate the state trajectory: Displacement vector ΔP( t ): ΔP( t )=P( t )−P( t −Δ t ) velocity v( t ):v( t )=Δ t ΔP( t ) acceleration a( t ): a( t )=Δ t v( t )−v( t −Δ t ); Early warning analysis: When the displacement direction of the state trajectory points to the same abnormal area for more than 3 consecutive time windows and the velocity increases, a state deterioration warning is triggered; when the three-dimensional feature point deviates from the boundary of the reference state area (the Mahalanobis distance is greater than the 90th quantile of the chi-square distribution with 3 degrees of freedom, which is taken as 6.251) but has not yet entered any abnormal area, an early warning is triggered; based on the current trajectory velocity and acceleration, the state position after a set number of time windows is extrapolated and predicted. The abnormal area into which the predicted position falls is the predicted vibration mode.
[0106] Step 912: After accumulating newly added state samples confirmed to be without misjudgment, the center position and covariance matrix of each state region are updated using an incremental learning method. When the Mahalanobis distance from the three-dimensional feature point to the center of all known state regions exceeds a set threshold and remains so for a set duration, a new state discovery alarm is triggered.
[0107] Specifically, the system outputs a tower vibration mode identification report, which includes the tower number and evaluation time, three-dimensional feature space coordinates, deviation of each axis, vibration mode type and confidence level, severity of anomaly, location of anomaly, status trend and early warning information, and recommended maintenance measures. When the accumulated new data exceeds 10% of the current baseline sample size, the baseline parameters are updated. After the new status sample is confirmed to be correct by on-site maintenance inspection, the center position and covariance matrix of each status region are corrected using incremental learning. When the Mahalanobis distance from the three-dimensional feature point to the center of all known status regions exceeds the set threshold (the 99th quantile of the chi-square distribution with 3 degrees of freedom, i.e., 11.345) and this occurs continuously for more than 5 time windows, a new status discovery alarm is triggered.
[0108] The system automatically generates suggested maintenance measures based on the judgment results. For example: during normal wind-induced vibration, conduct regular inspections; during abnormal wind-induced vibration, strengthen inspections and pay attention to weather warnings; during abnormal wind-induced torsion, conduct special torsion tests on the tower and check the weld seams of the main material nodes; during abnormal high-temperature vibration, increase inspections during the summer high-temperature period and check conductor sag and tower deformation; during abnormal icing vibration, check the tower foundation and strengthen inspections before the icing period; during conductor galloping, strengthen galloping monitoring and check the connection status of suspension strings and tension strings; when galloping causes bolts to loosen, immediately arrange to climb the tower to check the crossarm bolts; when bolts loosen, arrange to climb the tower to check the bolt torque of designated locations; when structural damage occurs, immediately organize a structural safety assessment and formulate reinforcement or replacement plans; when conductor strands break, arrange to check conductor strand breakage and replace the conductor if necessary; when icing falls off and impacts occur, check whether there is impact damage to the tower crossarm and foundation.
[0109] In this embodiment, through Mahalanobis distance discrimination and state trajectory analysis, not only can the current vibration mode and its confidence level be given, but also early warning and prediction of future states can be provided before the state enters the abnormal region, providing accurate spatial guidance and time windows for operation and maintenance. Actual measurements show that the three-dimensional feature extraction and state discrimination calculation time for a single tower and single time window is less than 1 second, which can meet the requirements of real-time monitoring.
[0110] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.
[0111] In one embodiment, such as Figure 7 As shown, a vibration mode recognition system for power transmission towers based on three-dimensional features is provided, including: a data acquisition module, a preprocessing module, a feature extraction module, a feature space construction module, and a mode recognition module, wherein: The data acquisition module is used to collect multi-source sensing data from the power transmission tower, including triaxial acceleration data, acoustic data, and environmental data. The preprocessing module is used to preprocess the multi-source sensing data to obtain the preprocessed multi-source sensing data. The feature extraction module is used to extract features from the preprocessed multi-source sensing data to obtain driving features extracted based on the environmental data, voiceprint features extracted based on the voiceprint data, and vibration features extracted based on the triaxial acceleration data. The feature space construction module is used to construct a three-dimensional feature space with the driving feature, the acoustic feature and the vibration feature as three coordinate axes respectively, and to map the various vibration modes of the power transmission tower as state regions in the three-dimensional feature space respectively; The modal recognition module is used to map the driving features, acoustic features, and vibration features at the current moment into three-dimensional feature points in the three-dimensional feature space, and determine the current vibration mode of the transmission tower based on the distance between the three-dimensional feature points and each of the state regions.
[0112] The modules in the aforementioned three-dimensional feature-based transmission tower vibration mode identification system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of each module.
[0113] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 8 As shown, the computer device includes a processor, memory, and network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores data. The network interface communicates with external terminals via a network connection. When executed by the processor, the computer program implements a method for identifying vibration modes of power transmission towers based on three-dimensional features.
[0114] Those skilled in the art will understand that Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.
[0115] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.
[0116] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0117] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.
Claims
1. A method for identifying vibration modes of transmission towers based on three-dimensional features, characterized in that, The method includes: Collect multi-source sensing data from power transmission towers, including triaxial acceleration data, acoustic data, and environmental data; The multi-source sensing data is preprocessed to obtain preprocessed multi-source sensing data. Feature extraction is performed on the preprocessed multi-source sensing data to obtain driving features extracted based on the environmental data, voiceprint features extracted based on the voiceprint data, and vibration features extracted based on the triaxial acceleration data. A three-dimensional feature space is constructed using the driving feature, the acoustic signature feature, and the vibration feature as three coordinate axes, and the various vibration modes of the power transmission tower are mapped to state regions in the three-dimensional feature space. The driving characteristics, acoustic characteristics, and vibration characteristics at the current moment are mapped to three-dimensional feature points in the three-dimensional feature space. Based on the distance between the three-dimensional feature points and each of the state regions, the current vibration mode of the transmission tower is determined.
2. The method according to claim 1, characterized in that, The multi-source sensing data is collected through a multi-source sensing sensor network deployed on the transmission tower; the collected multi-source sensing data from the transmission tower includes: The triaxial acceleration data is collected by installing a triaxial accelerometer on the power transmission tower. By installing sound sensors on the power transmission tower, multi-channel acoustic data is collected. By installing wind speed sensors, temperature sensors, and humidity sensors on the transmission tower, wind speed data, temperature data, and humidity data at the location of the transmission tower are collected respectively, thus forming the environmental data; The triaxial acceleration data, the voiceprint data, and the environmental data are timestamped and aligned using a set time reference to construct the multi-source sensing data.
3. The method according to claim 1, characterized in that, The multi-source sensing data is preprocessed to obtain the preprocessed multi-source sensing data, which includes: The triaxial acceleration data and the acoustic print data are respectively bandpass filtered, the filtered frequency band of the triaxial acceleration data covers the structural vibration frequency band of the power transmission tower, and the filtered frequency band of the acoustic print data covers the audible sound frequency band; A notch filter is used to suppress power frequency interference in the triaxial acceleration data and the acoustic signature data. Outliers in each channel's data are identified and removed based on the 3σ criterion, and then replaced with linear interpolations of the data before and after the outlier. The preprocessed multi-source sensing data is divided into time windows of a set length, ranging from 1 minute to 60 minutes.
4. The method according to claim 1, characterized in that, The extraction of the driving features includes: extracting wind speed statistical features from the wind speed sequence, wherein the wind speed statistical features include average wind speed, maximum wind speed, wind speed standard deviation, and wind speed change rate; Extract temperature features from the temperature sequence, the temperature features including the current temperature value and the rate of temperature change; The icing sensitivity indicator factor is determined based on temperature and humidity. When the temperature is within the icing sensitivity range and the humidity exceeds a set threshold, the icing sensitivity indicator factor takes a first value; otherwise, it takes a second value. The wind speed statistical features, the temperature features, and the icing sensitivity indicator are combined into a driving feature vector. The driving feature vector is then standardized and dimensionality reduced to obtain the driving feature axis coordinates in the three-dimensional feature space.
5. The method according to claim 1, characterized in that, The extraction of the voiceprint features includes: The voiceprint data is subjected to frame-by-frame windowing processing to extract the Mel frequency cepstral coefficients and their first and second differences, which constitute the voiceprint temporal features. Extract the short-time energy and short-time zero-crossing rate of the acoustic data from each channel; perform high-frequency band energy analysis on the acoustic data from each channel, and calculate the intensity and temporal kurtosis of the high-frequency impact component, wherein the intensity of the high-frequency impact component is the ratio of the high-frequency band energy to the total energy; Calculate the multi-channel acoustic energy distribution vector, and the ratio of the energy of each channel in the multi-channel acoustic energy distribution vector to the reference energy of the corresponding channel under the pre-calibrated tower health condition; The energy percentage of the voiceprint data in the 200Hz to 800Hz frequency band is extracted as a segmentation discrimination feature; The audioprint temporal features, short-time energy, short-time zero-crossing rate, high-frequency impact component intensity, temporal kurtosis, multi-channel audioprint energy distribution vector, and strand-breaking discrimination features are combined into an audioprint feature vector. The audioprint feature vector is then standardized and dimensionality-reduced to obtain the audioprint feature axis coordinates in the three-dimensional feature space.
6. The method according to claim 1, characterized in that, The vibration feature extraction includes: Spectral analysis was performed on the triaxial acceleration data to extract the dominant frequency, vibration amplitude, and low-frequency energy proportion for the X and Y axes, respectively. Extract the torsional component, which includes the phase difference between the X-axis acceleration and the Y-axis acceleration at the dominant frequency, the torsional amplitude, and the torsional dominant frequency. When the phase difference is within a preset torsional determination range, it is determined to be a torsional mode. Extract conductor-tower coupled vibration characteristics, which include the correlation coefficient between X-axis acceleration and Y-axis acceleration, Lissajous figure trajectory shape, and low-frequency amplitude component ratio; The X-axis and Y-axis spectral features, the torsional component, and the conductor-tower coupled vibration features are combined into a vibration feature vector. The vibration feature vector is then standardized and dimensionality reduced to obtain the vibration feature axis coordinates in the three-dimensional feature space.
7. The method according to claim 1, characterized in that, Mapping the various vibration modes of the transmission tower into state regions in the three-dimensional feature space includes: Under the health status of the transmission tower, operational data for a set period of time is collected and a benchmark sample set is extracted. The benchmark status region is determined based on the mean vector and covariance matrix of the benchmark sample set. Based on the tower structure dynamics model and historical fault data calibration, the center position and regional range of multiple vibration modes in the three-dimensional feature space are determined. The multiple vibration modes include wind-induced vibration anomaly, wind-induced torsional anomaly, high-temperature vibration anomaly, icing vibration anomaly, conductor galloping, galloping causing bolt loosening, bolt loosening anomaly, structural damage anomaly, conductor strand breakage anomaly, and icing falling impact. The step of determining the current vibration mode of the transmission tower based on the distance between the three-dimensional feature points and each of the state regions includes: calculating the Mahalanobis distance from the three-dimensional feature points to the center of each of the state regions, determining the vibration mode corresponding to the state region with the smallest Mahalanobis distance as the current vibration mode, and determining the discrimination confidence level based on the smallest Mahalanobis distance and the Mahalanobis distance in the reference state.
8. The method according to claim 7, characterized in that, The method further includes: Calculate the voiceprint energy of each channel in the voiceprint data, determine the abnormal voiceprint channel based on the ratio of the voiceprint energy of each channel to the corresponding channel reference energy in the pre-calibrated healthy state, and determine the installation position of the sound sensor corresponding to the abnormal voiceprint channel as the abnormal part. The three-dimensional feature points at multiple consecutive moments are connected in chronological order to form a state trajectory. When the state trajectory points to the same state region for a set number of consecutive moments and the trajectory velocity increases, a state deterioration warning is issued. When the three-dimensional feature points deviate from the reference state region and do not enter any other state region, an early warning is issued. The vibration mode at future moments is predicted by extrapolation based on the displacement direction and velocity of the state trajectory. After accumulating newly added state samples that have been confirmed to be without misjudgment, an incremental learning method is used to update the center position and covariance matrix of each state region; When the Mahalanobis distance from the three-dimensional feature point to the center of all known state regions exceeds a set threshold and continues for a set duration, a new state discovery alarm is triggered.
9. A vibration mode identification system for power transmission towers based on three-dimensional features, characterized in that, The system includes: The data acquisition module is used to collect multi-source sensing data from the power transmission tower, including triaxial acceleration data, acoustic data, and environmental data. The preprocessing module is used to preprocess the multi-source sensing data to obtain the preprocessed multi-source sensing data. The feature extraction module is used to extract features from the preprocessed multi-source sensing data to obtain driving features extracted based on the environmental data, voiceprint features extracted based on the voiceprint data, and vibration features extracted based on the triaxial acceleration data. The feature space construction module is used to construct a three-dimensional feature space with the driving feature, the acoustic feature and the vibration feature as three coordinate axes respectively, and to map the various vibration modes of the power transmission tower as state regions in the three-dimensional feature space respectively; The modal recognition module is used to map the driving features, acoustic features, and vibration features at the current moment into three-dimensional feature points in the three-dimensional feature space, and determine the current vibration mode of the transmission tower based on the distance between the three-dimensional feature points and each of the state regions.
10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 8.
Citation Information
Patent Citations
Multi-sensor data fusion type power transmission line tower stability monitoring system and method
CN107340053A
A method and system for automatically checking modal parameter identification of a power transmission tower
CN115795325B
Tower bolt health detection method and system based on artificial intelligence and voiceprint
CN118212938B
Communication tower bolt loosening positioning system and method based on voiceprint feature extraction
CN119935517A
A power transmission tower abnormality online monitoring and early warning processing system
CN121279814B
