Power transmission line multi-component defect detection method and system giving consideration to lightweight and precision

Abstract

The invention provides a power transmission line multi-component defect detection method and system giving consideration to light weight and precision, and relates to the technical field of power transmission line detection.The method comprises the steps that power transmission line component signals are collected for preliminary defect recognition, evolution analysis, feature fusion and evolution detection are conducted on preliminary defects, and a final evaluation result is generated; according to the invention, the memory occupation is greatly reduced, the system is adaptive to an unmanned aerial vehicle and embedded equipment, the patrol endurance time is prolonged, and the lightweight and real-time performance are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of power transmission line testing technology, specifically to a method and system for detecting defects in multiple components of power transmission lines that balances lightweight design and high precision. Background Technology

[0002] The types of defects that need to be detected in transmission lines include conductor strand breakage, insulator breakdown, and tower corrosion. Traditional detection methods have significant shortcomings in terms of lightweight design, real-time performance, and adaptability to various environments, making it difficult to meet the demands of modern power grids for rapid, accurate, and intelligent inspections. This leads to delayed fault location, high false alarm rates, and increased maintenance costs. Existing transmission line defect detection technologies mainly include manual inspection, UAV visual inspection, and sensor-based signal analysis and impedance measurement methods. Manual inspection uses visual inspection or simple instruments to detect surface defects; UAVs equipped with high-definition cameras or infrared devices use image processing algorithms to identify cracks, corrosion, etc.; sensor systems collect voltage, current, and vibration signals and detect anomalies through Fourier transform or time-domain analysis; impedance measurement technology analyzes changes in line impedance based on relay principles. These technologies have improved detection coverage to some extent, but are still limited by equipment complexity and environmental interference.

[0003] The specific limitations of existing technologies include the following aspects:

[0004] Insufficient lightweight and real-time performance: Existing UAV visual inspection and multi-sensor systems rely on high-performance processors or complex algorithms (such as deep learning), which consume a lot of memory (often exceeding 500MB) and consume a lot of power, resulting in short flight time or long response time (usually exceeding 100ms) for UAVs, making it difficult to meet the needs of rapid localization for long-distance line inspection or sudden failures.

[0005] Environmental noise interference and low detection accuracy: In complex environments such as strong winds in mountainous areas, electromagnetic interference in cities, or salt spray in coastal areas, existing technologies are easily affected by noise, resulting in a high false alarm rate. It is difficult to accurately detect weak defects (such as early cracks) or cross-component chain defects (such as the propagation from conductor to tower), and the positioning error is relatively large (usually more than 0.5km), with insufficient robustness.

[0006] Poor dynamic adaptability and multi-component detection capability: Existing methods mostly use fixed thresholds (such as phase angle shift or spectral energy threshold), which are difficult to dynamically adjust according to different voltage levels (such as 35kV, 110kV, 220kV), line length or environmental conditions, resulting in missed detection of latent defects (such as internal stress accumulation), and limited ability to analyze the propagation path of multi-component chain defects, affecting the comprehensiveness of detection.

[0007] Therefore, a rapid detection method for defects in multiple components of power transmission lines that balances lightweight design and precision is needed to solve the above problems. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention provides a method for detecting defects in multiple components of transmission lines that balances lightweight design with high precision, solving the problems of insufficient lightweight design and real-time performance, as well as poor dynamic adaptability and multi-component detection capabilities in existing technologies.

[0009] Another objective of this invention is to provide a multi-component defect detection system for power transmission lines that balances lightweight design with high precision.

[0010] To achieve the above objectives, the present invention provides the following technical solution:

[0011] A method for detecting defects in multiple components of transmission lines that balances lightweight design and high accuracy includes the following steps:

[0012] The instantaneous phase angles of voltage and current signals from multiple components of a transmission line are collected, and the difference in phase angle offset threshold is calculated.

[0013] The difference between the calculated phase angle offset threshold and the threshold is compared. The phase difference is used to distinguish between normal and defective states, and the phase angle offset is projected onto the specific component position.

[0014] Vibration signals from multiple components of the transmission line are collected, and multi-resolution decomposition is performed to calculate spectral energy and differential spectral energy threshold.

[0015] By comparing the difference between the spectral energy and the spectral energy threshold, energy abrupt regions exceeding the threshold are mapped to multi-component locations;

[0016] The phase difference anomaly data and the spectral energy difference anomaly data are transformed and processed to extract the time-frequency characteristic information of different components, and the overlap coefficient is calculated to analyze the defect propagation path.

[0017] Impedance measurements are performed on the defect propagation path, the root mean square value and duty cycle are calculated, and the impedance anomaly is projected onto the location of multiple components.

[0018] The impedance anomaly data, phase difference anomaly data, and spectral energy difference anomaly data are subjected to principal component analysis, data feature compression and fusion for dimensionality reduction, and imbalanced data are processed to output low-dimensional feature set data.

[0019] Based on the aforementioned low-dimensional feature set data and real-time acquired phasor data, a two-stage process analysis is performed, and compared with the aforementioned defect propagation path results to form a defect evolution link.

[0020] Based on the aforementioned defect evolution chain and low-dimensional feature set data, the standard deviation, peak-to-peak value, and mean are fused to construct a parameter matrix and generate a defect severity assessment.

[0021] Preferably, in the differential step of calculating the phase angle offset threshold, the discrete Fourier transform is applied to calculate the differential of the phase angle offset threshold; the threshold is set based on the statistical mean of historical normal phase angle data plus or minus 5 degrees.

[0022] Preferably, in the step of projecting the phase angle offset to the specific component position, the projection is performed by integrating multi-component topology mapping.

[0023] Preferably, in the steps of calculating spectral energy and differential spectral energy threshold, the Dobby wavelet is used for multi-resolution decomposition to generate approximation coefficients and detail coefficients. The spectral energy is calculated based on the sum of squares of the detail coefficients, and the differential spectral energy threshold is set to twice the standard deviation of the normal spectral energy.

[0024] Preferably, the energy mutation region is mapped to multiple component locations through a component location index table.

[0025] Preferably, in the step of extracting the time-frequency feature information of different components, the signal is processed by maximum overlap wavelet transform and non-subsampled wavelet decomposition is performed.

[0026] Preferably, in the steps of calculating the root mean square value and duty cycle, the root mean square value is calculated based on a 256-point signal window, and the duty cycle includes positive and negative duty cycles, which are determined by the proportion of positive and negative cycle time; the Min-Max scaling normalized impedance signal is used, and the impedance anomaly is mapped to the location of multiple components using a line geometry model.

[0027] Preferably, in the low-dimensional feature set data processing step, principal component analysis extracts several principal component data with a cumulative variance contribution rate exceeding 85% through covariance analysis, compressing the high-dimensional features of the original data into a low-dimensional representation; a synthetic minority oversampling algorithm is used to generate new synthetic samples for minority class defect samples, the interpolation between the original minority class defect samples and their nearest neighbors is calculated, the data distribution is expanded, and the generated new samples are supplemented into the original dataset with a default expansion rate of twice, forming balanced defect feature metadata; the low-dimensional feature set data includes defect category labels and location information.

[0028] Preferably, in the two-stage process analysis steps, the first stage detects anomalies by threshold screening. When the voltage phase angle difference is greater than a set value or the current phase angle difference is greater than a set value, it is marked as a potential defect. The second stage analyzes the abnormal signals based on this, calculates the time evolution law of the phasor difference value, tracks the propagation chain of phasor changes through time series analysis, and identifies chain defects between multiple components. The defect evolution chain is a time-stamped defect evolution diagram, which includes defect type, propagation direction and isolation suggestions.

[0029] Preferably, in the defect severity assessment step, three parameters—standard deviation, peak-to-peak value, and mean—are calculated within a 256-point signal window and integrated into a parameter fusion matrix. By default, the weights of the three types of parameters are all one-third. The comprehensive index output by the parameter fusion matrix is ​​processed by a ReLU-type threshold function. If the comprehensive index is lower than the threshold, the output is zero; if it exceeds the threshold, the output is the original value. The threshold is determined by the mean of the comprehensive index under normal conditions.

[0030] A multi-component defect detection system for transmission lines that balances lightweight design and high precision includes:

[0031] The phase acquisition module is used to acquire the instantaneous phase angle of voltage and current signals from multiple components of the transmission line and calculate the difference of the phase angle offset threshold.

[0032] The global fault detector construction module is connected to the phase acquisition module. It is used to compare the difference between the calculated phase angle offset threshold and the threshold, distinguish between normal and defective states through the phase difference, and project the phase angle offset onto the specific component position.

[0033] The vibration signal acquisition module is used to acquire vibration signals from multiple components of the transmission line and perform multi-resolution decomposition to calculate spectral energy and differential spectral energy threshold.

[0034] The differential comparison module, connected to the vibration signal acquisition module, is used to compare the difference between the spectral energy and the spectral energy threshold. Energy abrupt regions exceeding the threshold are mapped to the positions of multiple components.

[0035] The maximum overlap wavelet transform processing module is used to transform the phase difference anomaly data and the spectral energy difference anomaly data, extract the time-frequency feature information of different components, and calculate the overlap coefficient to analyze the defect propagation path.

[0036] The Morse impedance measurement module is used to measure the impedance of the defect propagation path, calculate the root mean square value and duty cycle, and project the impedance anomaly onto the location of multiple components.

[0037] The principal component analysis feature compression module is used to perform principal component analysis on the impedance anomaly data, phase difference anomaly data, and spectral energy difference anomaly data, compress and fuse the data features to reduce dimensionality, process imbalanced data, and output low-dimensional feature set data.

[0038] The phasor measurement unit synchronous detection module is used to collect phasor data in real time and perform two-stage process analysis based on the low-dimensional feature set data, and compare it with the defect propagation path results to form a defect evolution link.

[0039] The statistical parameter analysis module, based on the aforementioned defect evolution chain and low-dimensional feature set data, fuses the standard deviation, peak-to-peak value, and mean to construct a parameter matrix and generate a defect severity assessment.

[0040] This invention provides the following beneficial effects:

[0041] This solution significantly improves lightweight design and real-time performance through optimized hardware and algorithm design. The system utilizes a low-power STM32 series processor, combined with an RTOS real-time operating system and DSP instruction set acceleration, resulting in a substantial reduction in memory usage. This makes it compatible with drones and embedded devices, extending inspection endurance. Modules such as phase acquisition, vibration signal analysis, and impedance measurement achieve significantly shorter overall response times through efficient data flow management and queue buffer optimization, meeting the needs of long-distance line inspection and rapid fault location. Compared to existing technologies, this solution achieves lightweight deployment on edge devices, significantly improving inspection efficiency and real-time performance, making it particularly suitable for rapid response scenarios involving long-distance lines in mountainous areas and high-voltage lines in urban areas.

[0042] This solution significantly improves noise immunity and detection accuracy through multi-level signal preprocessing and the fusion of advanced algorithms. The system employs preprocessing methods such as low-pass filtering, band-pass filtering, and signal gain amplification to effectively suppress various environmental noises. It combines Dobbech wavelet multi-resolution decomposition and maximum overlap wavelet transform to capture the time-frequency characteristics of weak defects, ensuring high robustness. Principal component analysis and synthetic minority oversampling techniques fuse multi-sensor data to generate balanced feature metadata, significantly improving the detection accuracy of weak defects such as cracks and corrosion, as well as cross-component cascading defects, and greatly reducing positioning errors. The solution can operate stably in complex scenarios such as mountainous areas, urban areas, and coastal areas, reducing false alarm rates and enhancing the reliability of power grid maintenance.

[0043] This solution significantly improves adaptability and detection comprehensiveness through dynamic threshold adjustment and multi-module collaborative design. The system dynamically adjusts thresholds such as phase angle offset, spectral energy, and overlap coefficient based on voltage level, noise level, and line complexity. Combined with an evolution tracker and phasor measurement unit, it analyzes defect propagation paths using maximum overlap wavelet transform and a two-stage process to accurately capture cascading defects from conductors to towers. Parameter fusion matrices and ReLU-type threshold functions optimize latent defect identification, and a few-oversampling techniques are used to balance rare defect data, improving classification accuracy. The solution can adaptively operate in diverse scenarios (such as mountain inspections, urban fault response, and long-term coastal monitoring), comprehensively detecting defects in multiple components, significantly reducing the missed detection rate, and improving the stability and maintenance efficiency of the power grid. Attached Figure Description

[0044] Figure 1 This is a flowchart of the present invention;

[0045] Figure 2This is a system framework diagram of the present invention;

[0046] Figure 3 This is a simulation diagram of the phase difference detection of the present invention;

[0047] Figure 4 This is the spectral energy distribution diagram of the present invention. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Specific Implementation Example 1:

[0050] like Figures 1 to 4 As shown, a rapid defect detection method for multiple components of transmission lines that balances lightweight design and accuracy includes the following steps:

[0051] The phase sensor employs a high-precision measurement device based on GPS synchronization, with a sampling rate of at least 1000Hz, to acquire instantaneous phase angles of voltage and current signals from multiple components of the transmission line (such as conductors and insulators). A Discrete Fourier Transform (DFT) is used to process the signal within a 256-point window to calculate the phase angle offset threshold, which is set based on the statistical mean of historical normal phase angle data plus or minus 5 degrees. The global fault detector module distinguishes between normal and defective states (such as high-impedance faults) by comparing the phase angle difference with the threshold. The projection process uses a pre-built component topology mapping table (containing component numbers and geographic coordinates) to map the phase angle offset to specific component locations, such as the specific locations of insulator breakdown and conductor strand breakage. The system preprocesses the signal with a low-pass filter with a cutoff frequency of 50Hz to eliminate high-frequency noise and ensure robustness against current transformer saturation and measurement errors. The detection response time is controlled within 20 milliseconds to meet the low-power requirements of edge devices.

[0052] The vibration sensor uses a high-sensitivity accelerometer with a sampling rate of 2000Hz to collect vibration signals from transmission line components. The Dobbech mother wavelet uses the db8 type, supporting 8-level decomposition to generate approximation and detail coefficients. Spectral energy is calculated based on the sum of squares of the detail coefficients, and the differential spectral energy threshold is set to twice the standard deviation of the normal spectral energy. The differential comparator only processes energy abrupt changes exceeding the threshold, mapping to multiple components through a component location index table (storing component type and location), such as insulator cracks or abnormal conductor vibration. Performance indicators include detection accuracy and recall, and the reliability of the results is verified through a self-checking mechanism. Signal preprocessing uses a 10-point wide moving average smoothing to enhance robustness in noisy environments, and the detection cycle is controlled within 20 milliseconds, suitable for multi-component inspection in the field.

[0053] The maximum overlap wavelet transform employs a non-downsampling method, with at least four decomposition levels, capturing the time-frequency characteristics across components, including energy distribution spectra. The evolution tracker analyzes defect propagation paths based on overlap coefficients, with a threshold of 0.8; values ​​exceeding this threshold are considered defect propagation. Defect propagation paths are generated by tracking changes in time-frequency characteristics across components (e.g., chain reactions from conductors to the tower), achieving a positioning accuracy of ±0.1 km. Classification is based on a predefined defect pattern library (e.g., cracks, corrosion). The system adapts to different noise interference through adaptive threshold adjustment, with response latency controlled within 1 / 4 cycle (approximately 5 milliseconds), ensuring real-time performance and robustness.

[0054] Morse code relay impedance measurement involves injecting a 1A, 50Hz test current and scanning a long-distance line segment by segment (every 0.5km) to measure the apparent impedance. The root mean square (RMS) value is calculated based on a 256-point signal window, with the positive and negative duty cycles determined by the proportion of positive and negative cycle times. The gap mapping algorithm uses a Min-Max scaling normalized impedance signal, with an anomaly threshold set at 10% deviation from the normal impedance. The projection process uses a line geometry model (including distance and component coordinates) to map impedance anomalies to multiple component locations, providing directional protection. The system reduces false alarms through dual threshold verification (initial threshold + secondary confirmation), with a response time of less than 10 milliseconds, making it suitable for real-time inspection of long-distance lines.

[0055] Impedance anomaly data, phase difference anomaly data, and spectral energy difference anomaly data are used to form a high-dimensional feature matrix. Principal component analysis selects the top k principal components with a cumulative variance contribution rate exceeding 85% to form a low-dimensional feature representation. A synthetic minority oversampling technique balances the data distribution by generating minority class samples (e.g., rare defect types), with the oversampling rate set to 200% by default, generating balanced defect feature metadata. This feature metadata is used by a simple classifier (e.g., the k-nearest neighbor algorithm) for defect classification and localization, including cracks, corrosion, and breakdown, with a localization accuracy of ±0.2 km. The system optimizes computational efficiency through batch processing (1000 samples per batch) to adapt to edge devices.

[0056] Phasor measurement employs GPS synchronization technology with a sampling rate of at least 30 frames / second, acquiring voltage and current phasor data. The two-stage process includes: the first stage filters anomalies based on phase angle difference (voltage phase angle difference > 2 degrees or current phase angle difference > 3 degrees); the second stage precisely locates defects using phasor differences. The evolution tracking module, based on time series analysis, tracks the propagation chain of phasor changes, capturing multi-component cascading defects (such as conductor anomalies caused by insulator breakdown). Fault isolation is achieved by severing propagation chains exceeding a threshold, which is dynamically adjusted based on historical data. The system uses a 10-frame sliding window to analyze dynamic evolution, optimizing robustness to fault initiation angle and impedance changes.

[0057] Signal statistical parameters include standard deviation, peak-to-peak value, and mean, calculated based on a 256-point signal window. A parameter fusion matrix integrates these three types of parameters to form a comprehensive index, with weights assigned equally by default (1 / 3). A ReLU-type threshold function uses the mean of the normal comprehensive index as the threshold, handles non-linear patterns, and generates a defect severity score from 0 to 1; scores exceeding the threshold are considered severe defects. The system optimizes the identification of latent defects (such as internal stress accumulation) by iteratively adjusting the weights (updated every 100 detections), adapting to long-term monitoring scenarios.

[0058] The following is a detailed explanation of the algorithm section above:

[0059] In this method, Discrete Fourier Transform (DFT) is primarily used to process the instantaneous phase angle data of voltage and current signals acquired by phase sensors. The input data is the raw phase angle sequence obtained through a GPS-synchronized high-precision phase measurement device at a sampling rate of at least 1000Hz. The algorithm first divides the signal into 256-point analysis windows. Within each window, the time series is converted into frequency domain features using DFT, extracting the phase angle information of the main frequency components. A phase angle offset threshold is established based on the historical average phase angle under normal conditions, typically set to the average plus or minus 5 degrees. The algorithm compares the current phase angle with the threshold; if it exceeds the threshold, it is considered abnormal. A pre-constructed component topology mapping table is used to project the offset value to specific component locations, such as insulators or conductors, thereby enabling the identification of high-impedance faults and multi-point defects. The final output is a defect detection signal labeled with the specific component location, capable of completing the determination within 20 milliseconds, ensuring both speed and robustness.

[0060] The Dobwich wavelet multiresolution decomposition algorithm takes acceleration signals acquired by a vibration sensor as input, which obtains vibration information of components such as conductors and insulators at a sampling rate of 2000Hz. The algorithm uses a db8 ​​wavelet basis and performs eight-level decomposition on the signal, decomposing the original signal into approximation coefficients and detail coefficients at different scales. Then, the spectral energy of the detail coefficients is calculated, i.e., the energy magnitude within a frequency band is measured by the sum of squares. By comparing the spectral energy with the statistical distribution of normal spectral energy, a threshold of twice the standard deviation is set for energy abrupt changes. When the spectral energy difference in a frequency band exceeds the threshold, the algorithm marks it as an anomalous region and maps this energy anomaly to a specific component using a component location index table, such as an insulator crack or abnormal conductor vibration. The algorithm outputs the energy abrupt change detection results with clearly marked anomalous components and verifies the detection results through a self-checking mechanism to ensure identification within 20 milliseconds and guarantee high reliability.

[0061] The maximum overlap wavelet transform algorithm performs non-downsampling processing on the input vibration and current signals, preserving the full signal length, and generates a time-frequency feature distribution map based on a four-level decomposition. The algorithm calculates the energy distribution changes between different frequency bands, extracts time-frequency feature information across components, and inputs this information into an evolution tracker. The evolution tracker analyzes the defect propagation path by calculating the overlap coefficient between features of different components; overlap coefficients exceeding 0.8 are considered to indicate a defect propagation relationship. Through continuous tracking of time-frequency feature changes across components, the algorithm can reconstruct the dynamic path of defect propagation from one component to another, such as the transmission from a conductor crack to a tower structure. The output is the defect propagation path and classification label, capable of tracking within 5 milliseconds, ensuring a rapid response to real-time defect evolution.

[0062] The Morse code relay impedance measurement algorithm takes current and voltage signals obtained from scanning every 0.5 km as input, sampled by injecting a 1 ampere, 50 Hz test current. The algorithm first calculates the root mean square (RMS) value of the signal within a 256-point window to measure the impedance amplitude; secondly, it calculates the positive and negative duty cycles, i.e., the proportion of signal duration within the positive and negative half-cycles, to analyze impedance asymmetry. Subsequently, the impedance values ​​are standardized using a Min-Max scaling method to eliminate numerical scale differences. Deviations are then identified by comparing with normal impedance data, with a threshold set at 10% of the normal impedance deviation range. If an impedance anomaly is detected, it is projected onto the specific component location using a line geometry model, and a dual-threshold confirmation mechanism is used to reduce false alarms. The final output is the impedance anomaly location result for components in long-distance lines, with a response time of less than 10 milliseconds, suitable for real-time inspection.

[0063] Principal Component Analysis (PCA) and Synthetic Minority Oversampling (SMO) algorithms take high-dimensional feature matrices collected from multiple sensors as input, including vibration, current, and voltage signals. PCA first extracts the top principal components with a cumulative variance contribution exceeding 85% through covariance analysis, compressing the original high-dimensional features into a low-dimensional representation to reduce redundancy. Then, SMO generates new synthetic samples for minority class defect samples (such as rare corrosion or breakdown). By calculating the interpolation between the original minority class samples and their nearest neighbors, the data distribution is expanded. The generated new samples are added to the original dataset at a default expansion rate of twice, thus forming balanced defect feature metadata. The output is a balanced low-dimensional feature set, which effectively improves the accuracy and location precision of subsequent classifiers for identifying defect types such as cracks, corrosion, and breakdown.

[0064] The two-stage process analysis algorithm processes synchronous voltage and current phasor data collected by the phasor measurement unit, with input consisting of 30 frames of synchronous sampling data per second. The first stage rapidly detects anomalies using a threshold filtering method; when the voltage phase angle difference exceeds two degrees or the current phase angle difference exceeds three degrees, it is marked as a potential defect. The second stage performs a detailed analysis of the anomalous signals, calculating the temporal evolution of the phasor differences and inputting this data into the evolution tracking module. The evolution tracking module uses time series analysis to trace the propagation chain of phasor changes, identifying cascading defects between multiple components, such as insulator breakdown leading to abnormal conductor current. The final output is a clear defect propagation chain and fault isolation recommendations. The system can achieve fault isolation by cutting off propagation chains exceeding the threshold, demonstrating robustness and real-time performance.

[0065] The input to the parameter fusion matrix and ReLU threshold function algorithm consists of preprocessed statistical parameters collected by sensors, including standard deviation, peak-to-peak value, and mean. The algorithm first calculates these three parameters within a 256-point signal window, then integrates them into the parameter fusion matrix. By default, the weights of the three types of parameters are each one-third. The comprehensive index output by the matrix is ​​processed using a ReLU threshold function. If the comprehensive index is below the threshold, the output is zero; if it exceeds the threshold, the original value is output. The threshold is determined by the mean of the comprehensive index under normal conditions and is used to distinguish between normal and abnormal modes. The output is a defect severity score, ranging from zero to one. Defects exceeding the threshold are marked as severe defects. The algorithm continuously optimizes its ability to identify latent defects through periodic weight adjustments, making it suitable for long-term monitoring and health assessment.

[0066] The following needs to be explained: Figure 3 (Phase difference detection):

[0067] Image content: The horizontal axis represents time (0–1s), and the vertical axis represents the amplitude of the voltage and current signals. The blue curve represents voltage, and the orange curve represents current.

[0068] In the phase difference acquisition step, the voltage and current signals of the transmission line are acquired by a phase sensor, and then the phase difference between the two is estimated by Discrete Fourier Transform (DFT).

[0069] As shown in the figure, the voltage and current waveforms are roughly at the same frequency but have a phase difference, which is a key characteristic for detecting high-impedance faults and multi-point defects. When a defect occurs, the phase angle shifts, and the delay or lead of the current waveform relative to the voltage waveform in the figure is a direct manifestation of this shift. Because it only relies on phase difference and does not require complex feature extraction, the detection method has the advantage of being lightweight, and at the same time, it has a fast response speed, which can meet the real-time detection requirements of less than 20ms.

[0070] Appendix Figure 4 (Spectral energy distribution):

[0071] Image content: The horizontal axis represents time (0–1s), the vertical axis represents frequency (0–1000Hz), and the color represents the strength of the power spectral density. Yellow represents areas of concentrated energy, and blue represents areas of lower energy.

[0072] The vibration signal acquisition and energy spectrum calculation steps reflect the spectral energy distribution of the signal as it changes over time.

[0073] The graph shows a relatively uniform spectral energy distribution, but with slightly higher energy levels in certain time periods and frequency bands (yellow areas), indicating potential transient energy spikes. The differential comparator specifically targets these spike regions to identify fault points or defects in multiple components.

[0074] Since wavelet decomposition is a time-frequency domain analysis method, it can effectively capture local energy changes in noisy environments, ensuring the robustness and reliability of detection. Specific Implementation Example 2

[0076] like Figures 1 to 4 As shown, in this technical solution, the modules do not operate independently, but are tightly coupled via data flow to form a complete defect detection chain. The system first acquires instantaneous phase angle data of voltage and current signals using a phase acquisition module. This data, acquired at a kilohertz sampling rate, is preprocessed by a low-pass filter and then passed as input to the global fault detector construction module. The global fault detector analyzes the phase angle characteristics within the signal window using discrete Fourier transform and compares them with a preset threshold, outputting a feature data packet containing phase angle offset information. This data packet not only contains the defect status determination result but also carries the location index from the component topology mapping table, thus providing subsequent modules with defect candidate signals with location information.

[0077] Subsequently, the vibration signal acquisition module acquires high-sampling-rate acceleration data in parallel and sends it to the differential comparison module for wavelet multi-resolution decomposition. This module generates energy spectrum features and determines energy abrupt change regions through differential operations. The output is a vibration anomaly marker with energy labels and component coordinates. This output is fused with the results from the phase detector module on the data bus, ensuring that information from different signal sources can be correlated under a unified indexing system, thereby avoiding misjudgments caused by a single sensor.

[0078] Based on this, the maximum overlap wavelet transform processing module receives anomaly marker data from the phase and vibration modules and further performs non-subsampled wavelet decomposition to generate a time-frequency feature map. The evolution tracker calculates the overlap coefficient across components based on these inputs and outputs defect propagation path information. This path data is in an ordered linked list structure, containing the starting component, the ending component, and propagation delay parameters, for use in subsequent impedance measurements and evolution analysis.

[0079] The Morse code impedance measurement module receives the aforementioned path data as a reference. During each 0.5 km gap scan, it focuses on measuring the impedance of the line segments involved in the propagation path, calculating the root mean square value and duty cycle. The module's output is an impedance anomaly matrix, with each row recording the component number, impedance amplitude deviation, and directivity result. This matrix is ​​then passed to the principal component analysis feature compression module, where it is merged with high-dimensional features from the wavelet and phase modules.

[0080] The principal component analysis feature compression module reduces the dimensionality of all input features and generates balanced feature metadata using a synthetic minority oversampling algorithm. The module outputs a compressed and balanced low-dimensional feature set containing defect category labels and location information. This significantly reduces the data volume while maintaining feature integrity, facilitating subsequent real-time processing. This feature set is then fed into the phasor measurement unit's synchronous detection module.

[0081] The phasor measurement unit's synchronous detection module receives low-dimensional feature sets and real-time acquired phasor data, performs a two-stage process analysis, and compares it with the path results of the evolution tracker to form a more refined defect evolution chain. The output is a timestamped defect evolution map, including defect type, propagation direction, and isolation recommendations.

[0082] Finally, the statistical parameter analysis module receives statistical parameters from the evolution path and feature set, fuses the standard deviation, peak-to-peak value, and mean to construct a parameter matrix, and generates a defect severity score using a ReLU threshold function. The output detection report not only includes the defect type and location but also provides a quantified severity result for priority scheduling on the operations and maintenance side.

[0083] In terms of the operating environment, the system achieves inter-module data transmission via an RS-485 bus and a communication rate of over one megabit per second. The data stream is managed in memory in the form of a circular buffer queue, ensuring high throughput and low latency. The hardware platform uses an STM32 series processor in conjunction with an embedded GPU acceleration unit. Wavelet decomposition and principal component analysis operations are accelerated through DSP instruction sets or CUDA kernel calls, while data acquisition and transmission rely on the RTOS real-time operating system for scheduling, guaranteeing millisecond-level response. The development platform is based on a hybrid programming approach using C and Python. The algorithm components are encapsulated as API interfaces. For example, the wavelet decomposition API provides multi-level decomposition and energy spectrum calculation functions, and the PCA API provides covariance matrix decomposition and eigenvector selection functions. When calling these functions, only a data buffer pointer needs to be passed in to receive the processing results.

[0084] System details: The system uses an RS-485 interface for inter-module data transmission, with a transmission rate of at least 1 Mbps, ensuring an overall response time of less than 50 milliseconds. Data flows from the phase acquisition module and vibration signal acquisition module, sequentially to the global fault detector construction module, differential comparison module, maximum overlap wavelet transform processing module, Morse impedance measurement module, principal component analysis feature compression module, and phasor measurement unit synchronous detection module. Finally, a defect report is generated by the statistical parameter analysis module. The system is compatible with edge devices (such as STM32 series processors), with memory usage kept below 200MB, supporting deployment on UAVs or embedded inspection equipment. Data flows between modules are managed through a queue buffer (capacity 1000 data points) to prevent data congestion and improve real-time performance. Specific Implementation Example 3

[0086] like Figures 1 to 4 As shown, the following is the complete working method of a multi-component defect detection method for transmission lines that balances lightweight design and accuracy:

[0087] 1. Actual preparation phase: System setup and data initialization

[0088] Before practical use, a hardware platform was first built. An STM32 series processor was selected as the core controller, integrating a phase sensor (a high-precision device based on GPS synchronization, with a sampling rate of 1000Hz, used to acquire voltage / current phase angles), a vibration sensor (a high-sensitivity accelerometer, with a sampling rate of 2000Hz, used for vibration signals), a phasor measurement unit (GPS synchronized, 30 frames / second, used for phasor synchronization), and a Morse code relay module (used for impedance measurement). These sensors are connected to the processor via an RS-485 interface, forming a modular system that supports deployment on UAVs or as fixed embedded inspection equipment. Memory was kept below 200MB to avoid high power consumption.

[0089] On the software side, API interfaces are developed using a hybrid C and Python programming language. For example, a wavelet decomposition API is used for Dobbech mother wavelet processing, and a principal component analysis API is used for feature compression. Initialization data includes: a pre-built component topology mapping table (storing the component numbers, types, and geographical coordinates, such as the latitude and longitude of conductor segments), a normal signal history database (used for threshold settings, such as phase angle mean plus or minus 5 degrees, spectral energy twice the standard deviation, and impedance 10% deviation), a defect pattern library (predefined time-frequency feature templates for cracks, corrosion, etc.), and a location index table (associating component types with coordinates). This data is imported from the power company's historical inspection records and scheduled through an RTOS real-time operating system to ensure immediate availability after system startup.

[0090] Before inspections, the power maintenance team inputs line parameters (such as length, voltage level, and geographical environment) via an app or console, and the system automatically loads the corresponding model (such as setting a 0.5km gap scan for long-distance lines). Preparation time is usually less than 30 minutes, suitable for routine inspections or emergency fault response.

[0091] 2. Deployment Phase: Equipment Installation and Scene Setup

[0092] In practical deployments, the system is installed on drones or inspection robots to support outdoor environments (such as mountainous areas and high-voltage lines). Drones are equipped with sensor arrays, flying at altitudes controlled between 10-50 meters, scanning along the power transmission line path (at speeds of 20-50 km / h). For fixed deployments, such as installing embedded devices on transmission towers, long-term monitoring is achieved.

[0093] Scenario settings include: selecting the inspection mode (real-time or timed mode). Real-time mode is suitable for fault response, while timed mode is used for routine maintenance. The system synchronizes all sensors via GPS positioning to ensure data consistency (e.g., 30 frames / second synchronization for the phasor measurement unit). Communication uses an RS-485 bus with a transmission rate of over 1Mbps. Data flow between modules is managed through a queue buffer (capacity 1000 data points) to prevent congestion. Low-power design of edge devices (e.g., STM32 processor) ensures a single inspection cycle lasts for over 2 hours.

[0094] Maintenance personnel start the system via a remote console, input the inspection range (such as a specific line section), and the system automatically calibrates the threshold (such as adjusting the phase angle offset threshold based on historical data). Once ready, data collection begins.

[0095] 3. Operational Phase: Data Acquisition, Processing, and Analysis

[0096] During system operation, the modules are tightly coupled via data flow to form a complete detection link. In practice, assuming the inspection of a 10km high-voltage transmission line, the process is as follows:

[0097] Data acquisition begins: The phase acquisition module and vibration signal acquisition module operate in parallel. The phase sensor acquires the instantaneous phase angle of voltage / current (256 points in the window), and the vibration sensor acquires the acceleration signal. The Morse impedance measurement module injects a 1A / 50Hz test current for gap scanning, and the phasor measurement unit acquires synchronous phasor data (30 frames / second). All data undergoes low-pass filtering or moving average smoothing preprocessing to eliminate noise.

[0098] Preliminary Defect Identification (Sp1 and Sp2): Phase data is input to the global fault detector construction module, where a discrete Fourier transform is applied to calculate the phase offset threshold. Defects are distinguished by phase difference, and the offset is projected onto the component location (e.g., insulator breakdown). The module outputs candidate defect signals with location information, with a response time of less than 20ms. Simultaneously, vibration data is input to the differential comparison module, which uses Dobbech wavelet multi-resolution decomposition to calculate spectral energy and differential spectral energy thresholds. Only abrupt change regions are processed, mapped to component locations, and integrated with performance index self-checking to ensure a detection cycle of less than 20ms. These outputs are fused onto the data bus to form preliminary anomaly markers.

[0099] Evolutionary analysis and scanning (Sp3 and Sp4): Initial labels are input to the maximum overlap wavelet transform processing module. The signal is processed using the maximum overlap wavelet transform to capture time-frequency features. An evolutionary tracker is designed to calculate the defect propagation path using the overlap coefficient, enabling multi-point defect localization and classification with a response delay of 1 / 4 cycle. The Morse impedance measurement module receives path information and extends the Morse relay impedance measurement principle. It calculates the root mean square value, positive duty cycle, and negative duty cycle through a 0.5km gap scan. A gap mapping algorithm using Min-Max scaling normalization is designed to project impedance anomalies onto component locations, providing directional protection and reducing false alarms.

[0100] Feature fusion and evolution detection (Sp5 and Sp6): Multi-sensor data is input into the principal component analysis feature compression module, where principal component analysis is applied to compress and fuse features into a low-dimensional representation. Unbalanced data is then integrated and synthesized using minority oversampling techniques to balance the distribution, generating balanced defect feature metadata for defect classification and localization. The feature set is then input into the phasor measurement unit synchronous detection module, which collects synchronous phasor data. Defect evolution is analyzed through a two-stage process, and an evolution tracking module is designed to capture cascading defects, achieving fault isolation and optimizing robustness to fault initiation angle and impedance changes.

[0101] Final Analysis and Evaluation (Sp7): All results are fed into the statistical parameter analysis module, which integrates standard deviation, peak-to-peak value, and mean for multi-parameter analysis. A parameter fusion matrix is ​​designed, and a ReLU-type threshold function is used to handle nonlinear patterns. Latent defects are identified, and severity assessment (0-1 score) is provided. The system optimizes identification through iterative weight adjustment, and the entire link response time is less than 50ms.

[0102] During operation, the data stream is buffered in a circular manner to ensure real-time performance; if a defect is detected, the system automatically triggers an alarm and logs the information.

[0103] 4. Output and Follow-up Stages: Results Presentation and Maintenance Decisions

[0104] The system outputs a defect report, displayed via the console or app, including the defect type (e.g., crack, breakdown), location coordinates, propagation path, severity score, and isolation recommendations. The report format includes visual charts (e.g., evolution diagrams) and text lists to facilitate decision-making by operations and maintenance personnel. For example, if a broken strand is detected in a conductor, the system recommends prioritizing the isolation of that section.

[0105] Subsequently, the team conducts on-site verification or repairs based on the report. The system supports data playback and threshold updates (such as adjusting phase angle thresholds based on new historical data). In long-term use, it is suitable for preventative maintenance, reducing the failure rate. Through the API interface, the system can be integrated into the power grid management system to achieve automated inspection and scheduling. Specific Implementation Example 4

[0107] like Figures 1 to 4 As shown, the following are experimental use cases of this solution:

[0108] Trial Application Case 1: Long-Distance Line Inspection in Mountainous Areas

[0109] Scenario Description: A power company needs to inspect a power transmission line in a mountainous area. The line covers complex terrain and includes components such as conductors, insulators, and towers. The inspection needs to detect cracks, abnormal vibrations, and high-impedance faults. Due to terrain limitations, the inspection will utilize a drone-borne system, emphasizing lightweight design to extend endurance and high precision to cope with wind noise interference.

[0110] Preparation Phase: The operations and maintenance team selected a system based on an embedded processor, integrating phase sensors, vibration sensors, phasor measurement units, and Morse impedance measurement modules. These are connected via high-speed interfaces, ensuring low memory usage and lightweight design. The software pre-loads APIs written in C and Python, including Discrete Fourier Transform APIs and Dobbech wavelet APIs. Initialization data includes a component topology mapping table (recording line component numbers and geographical coordinates), a normal signal history database (for threshold setting), a defect pattern library (crack and corrosion templates), and a location index table, quickly imported from the power company's database. The team inputs line parameters (such as voltage level and mountainous environment) through the console, and the system automatically loads gap scan settings, resulting in short preparation time.

[0111] Deployment Phase: The system is installed on a drone at a suitable flight altitude, with an inspection speed adapted to the mountainous environment. The drone is equipped with a GPS module to ensure sensor time synchronization. High-speed bus transmission of data and queue buffer management of the data stream prevent congestion. System calibration thresholds are set, and flight time is extended to cover the entire line. Maintenance personnel can initiate real-time inspection mode via a remote APP, setting the inspection range to the entire line.

[0112] Operational Phase: The UAV flies along the line, the phase acquisition module acquires the instantaneous phase angles of voltage and current in real time, the vibration signal acquisition module acquires acceleration data, the Morse impedance measurement module injects test current for gap scanning, and the phasor measurement unit synchronously acquires phasor data. The data stream undergoes low-pass filtering or moving average smoothing preprocessing to eliminate wind noise. The global fault detector construction module applies discrete Fourier transform to calculate phase angle shift, detects high-impedance faults, and projects them onto the insulator location. The vibration signal is decomposed by Dobbech wavelet multi-resolution decomposition, and a differential comparator processes spectral energy abrupt changes to map conductor vibration anomalies. The maximum overlap wavelet transform processing module captures time-frequency features, and the evolution tracker calculates the overlap coefficient to locate crack propagation (e.g., from conductor to tower). The Morse impedance measurement module calculates the root mean square value and duty cycle through gap scanning, and projects the impedance anomaly after standardization. The principal component analysis feature compression module fuses multi-sensor data, synthesizes a minority oversampled balanced distribution, and generates balanced feature metadata for classifying crack types. The phasor measurement unit synchronous detection module captures cascading defects through a two-stage process and generates isolation commands. The statistical parameter analysis module integrates standard deviation, peak-to-peak value, and mean, and uses a ReLU threshold function to generate severity scores. The entire process has a short response time, and the data flow is managed through a circular buffer, improving efficiency.

[0113] Output Phase: The system outputs defect reports via the app, such as the location of wire cracks, high severity, and recommendations for isolation and repair. The report includes location coordinates, a propagation path diagram, and classification tags, allowing the maintenance team to plan repairs accordingly. The data playback function supports threshold optimization, improving inspection efficiency.

[0114] Case Study 2: Response to Sudden Faults in Urban High-Voltage Lines

[0115] Scenario Description: A sudden power outage occurs on a city power transmission line, suspected to be due to insulator breakdown or broken conductor strands, requiring rapid fault location. The system is deployed on fixed embedded equipment, emphasizing high-precision positioning and real-time response to minimize power outage time.

[0116] Preparation Phase: Embedded devices are selected, integrating phase sensors, vibration sensors, phasor measurement units, and Morse impedance measurement modules, featuring high-speed interface connections and low memory usage. The software loads Discrete Fourier Transform API, Principal Component Analysis API, etc., and initializes data including a line component topology mapping table, a historical signal database (for threshold setting), a defect pattern library, and a location index table, which are quickly imported from the urban power grid system. The console inputs line parameters (such as voltage level and urban environment), loads real-time mode and gap scan settings, and has a short preparation time.

[0117] Deployment Phase: The equipment is fixed on the starting tower of the line, GPS synchronizes sensor data, high-speed bus transmission is used, and a buffer manages the data stream. The system calibrates thresholds, responds to sudden fault requirements, and operates in real-time detection mode, covering the entire line area.

[0118] Operational Phase: The phase acquisition module acquires phase angles, the vibration signal acquisition module obtains acceleration data, the Morse code impedance measurement module scans impedance, and the phasor measurement unit synchronizes phasors. Data is preprocessed with low-pass filtering to eliminate urban electromagnetic interference. The global fault detector construction module calculates phase angle shifts using discrete Fourier transform, detects breakdown, and projects it onto the insulator location. The differential comparison module uses Dobbech wavelet decomposition to handle spectral energy abrupt changes and locate conductor strand breaks. The maximum overlap wavelet transform processing module generates time-frequency features, and the evolution tracker calculates the overlap coefficient to confirm the fault propagation path. The Morse code impedance measurement module calculates the root mean square value and duty cycle, performs Min-Max scaling normalization, and projects impedance anomalies. The principal component analysis feature compression module fuses data, synthesizes minority oversampling to generate balanced metadata, and classifies breakdown types. The phasor measurement unit synchronous detection module captures cascading defects through a two-stage process and outputs isolation commands. The statistical parameter analysis module uses a ReLU-type threshold function to generate severity scores. The system features short response time and efficient data flow management.

[0119] Output phase: The app displays a report, such as indicating insulator breakdown at a certain location, high severity, and recommending emergency isolation. The report provides precise coordinates and isolation suggestions, enabling the team to quickly repair the damage and significantly reducing power outage time.

[0120] Case Study 3: Long-Term Monitoring in Coastal Areas

[0121] Scenario Description: Coastal power transmission lines are susceptible to salt spray corrosion, requiring long-term monitoring of latent defects (such as internal stress accumulation and corrosion). The system is deployed on fixed equipment, emphasizing high-precision latent defect assessment and low-power long-term operation.

[0122] Preparation Phase: An embedded processor is selected, integrating a complete set of sensor modules, with high-speed interface connections and low memory usage. The API interface is loaded, and initialization data, including a line component topology mapping table, historical signal database, defect pattern library, and location index table, is quickly imported. Parameters (such as voltage level and coastal environment) are input to the console, and timing modes are loaded, resulting in a short preparation time.

[0123] Deployment phase: Equipment is fixed on towers along the line, using GPS-synchronized sensors, high-speed transmission, and buffer zone management. The system is calibrated with thresholds to adapt to salt spray interference, operates in a timed mode, undergoes daily inspections, and has a long battery life.

[0124] Operational Phase: The phase acquisition module, vibration signal acquisition module, Morse impedance measurement module, and phasor measurement unit acquire data in parallel, and preprocessing eliminates salt spray noise. The global fault detector construction module detects phase angle anomalies and projects corrosion locations using discrete Fourier transform. The differential comparison module uses Dobbech wavelet decomposition to locate vibration anomalies. The maximum overlap wavelet transform processing module captures time-frequency features, and the evolution tracker analyzes the propagation path. The Morse impedance measurement module scans impedance, calculates the root mean square value and duty cycle, and projects anomalies. The principal component analysis feature compression module fuses data, synthesizes a minority oversampled equilibrium distribution, and classifies corrosion types. The phasor measurement unit synchronous detection module analyzes cascading defects and generates isolation commands. The statistical parameter analysis module integrates standard deviation, peak-to-peak value, and mean value, uses a ReLU threshold function to assess the severity of latent defects, and iterative weight optimization for identification. The system features short response time and stable data flow.

[0125] Output phase: Daily reports indicate corrosion in a specific wire, indicating high severity and recommending regular inspection, including coordinates and evolution diagrams. Data archiving supports trend analysis, enabling the operations team to plan preventative maintenance, resulting in a reduction in the failure rate. Specific Implementation Example 5

[0127] like Figures 1 to 4 As shown, the following is a supplement to the content of the above embodiments:

[0128] 1. Dynamic adjustment mechanism for threshold setting:

[0129] To adapt to different line conditions (such as voltage level and environmental noise) and ensure robustness and high accuracy of detection, this system designs a dynamic threshold adjustment mechanism, covering phase angle offset threshold, spectral energy difference threshold, overlap coefficient threshold, impedance anomaly threshold, and statistical parameter threshold. The adjustment logic is based on historical data statistical analysis, combined with current line environmental characteristics (such as voltage level and environmental noise level) to dynamically update the threshold. The system achieves automated adjustment through the threshold management module of the embedded processor, with an adjustment cycle triggered every 100 detections to balance real-time performance and stability. Specifically, the system extracts statistical features (such as phase angle mean and spectral energy standard deviation) from the normal signal historical database and calculates the baseline threshold according to the line voltage level (such as low voltage, medium voltage, or high voltage). Then, based on real-time environmental data (such as wind noise or salt spray interference intensity collected by noise sensors), the threshold is dynamically corrected using a weighted standard deviation method, with the weight determined by the environmental noise level (for example, increasing the standard deviation weight when wind noise is strong and decreasing the mean offset weight in salt spray environments). For the phase angle offset threshold, the system dynamically adjusts the deviation range based on the historical average to ensure adaptability to different load conditions. For the spectral energy difference threshold, the system adaptively adjusts the standard deviation multiple according to the environmental noise level. For the overlap coefficient threshold, the system dynamically optimizes the threshold range by combining line complexity and component topology. For the impedance anomaly threshold and statistical parameter threshold, the system updates the reference values ​​in real time through a mean drift algorithm. This dynamic adjustment mechanism ensures that the system maintains high-precision detection in scenarios such as strong winds in mountainous areas, electromagnetic interference in cities, or salt spray in coastal areas, while reducing the computational burden through periodic updates to meet the lightweight requirements of edge devices.

[0130] The following details the initial range, dynamic adjustment criteria, and specific implementation logic for each threshold:

[0131] Phase offset threshold (Sp1):

[0132] Initial range: based on the statistical mean of historical normal phase angle data plus or minus 5 degrees, with a typical range of [-5°, +5°].

[0133] Adjustment Logic: The system calculates the historical average phase angle based on line voltage levels (e.g., 35kV, 110kV, 220kV) and dynamically adjusts the deviation range in conjunction with real-time load fluctuations (e.g., load changes monitored by current sensors). For example, when the load fluctuation of low-voltage lines (35kV) is small, the threshold is narrowed to [-4°, +4°]; when the load fluctuation of high-voltage lines (220kV) is large, the threshold is widened to [-6°, +6°]. Environmental noise levels (e.g., wind noise intensity measured by noise sensors, in dB) affect threshold adjustment. If the noise exceeds 60dB, the threshold is increased by 0.5° to avoid false alarms. The adjustment algorithm uses a mean-shifting method, updating the mean and standard deviation based on a sliding window of the most recent 1000 frames of historical data, recalculating every 100 detections.

[0134] Implementation details: The threshold management module extracts phase angle statistical features (mean and standard deviation) from the historical database of normal signals, and calculates dynamic thresholds using a weighted formula (weights of 0.7 for voltage level and 0.3 for noise level) to ensure adaptability to scenarios such as strong winds in mountainous areas or electromagnetic interference in cities.

[0135] Spectral energy difference threshold (Sp2):

[0136] Initial range: set to twice the standard deviation of the normal spectral energy, with a typical range of [1.5σ, 2.5σ], where σ is the standard deviation of the historical spectral energy (based on the sum of squared detail coefficients of the Dobby wavelet decomposition).

[0137] Adjustment logic: The threshold is dynamically adjusted based on environmental noise levels (such as wind noise and salt spray interference) and line complexity (number of components and topology density). For example, in mountainous strong wind scenarios (noise intensity > 50 dB), the threshold is increased to 2.7σ to reduce false noise triggers; in coastal salt spray environments (humidity > 80%), the threshold is decreased to 1.3σ to improve sensitivity to minor vibration anomalies. The adjustment is based on a weighted standard deviation method, with weights allocated as 0.6 for noise level and 0.4 for line complexity. The standard deviation is recalculated every 100 detections based on the vibration data from the most recent 500 frames.

[0138] Implementation details: The threshold management module dynamically updates the threshold by combining the real-time noise spectrum collected by the vibration sensor with the topological information in the component location index table. The calculation process is integrated into the DSP instruction set to maintain a lightweight design.

[0139] Overlap coefficient threshold (Sp3):

[0140] Initial range: set to 0.8, with a range of [0.75, 0.85], based on the time-frequency feature correlation of the maximum overlap wavelet transform.

[0141] Adjustment logic: Dynamically optimized based on the topological complexity of line components and environmental interference (such as electromagnetic noise and salt spray). For example, in complex dual-loop lines (number of components > 200), the threshold is lowered to 0.73 to capture more potential propagation paths; in simple single-loop lines, it is raised to 0.87 to reduce false alarms. When the environmental noise level is high (e.g., electromagnetic interference > 40dB), the threshold is raised by 0.02 to enhance robustness. The adjustment algorithm is based on the statistical distribution of historical defect propagation data, updated every 100 detections, and recalculated by combining the time-frequency characteristics of the most recent 100 frames.

[0142] Implementation details: The evolution tracker extracts historical propagation features from the defect pattern library, combines them with real-time noise sensor data, and adjusts the threshold through a weighted average algorithm (weights are topological complexity 0.5 and noise level 0.5) to ensure adaptability to different scenarios.

[0143] Impedance anomaly threshold (Sp4):

[0144] Initial range: set to 10% deviation from normal impedance, with a typical range of [8%, 12%], based on the root mean square value of Morse relay impedance measurement.

[0145] Adjustment logic: The threshold is dynamically adjusted based on line length and load conditions. For example, for long-distance lines (>50km), the threshold is widened to 14% to accommodate impedance fluctuations; for short-distance lines (<10km), it is narrowed to 7% to improve accuracy. When environmental noise (such as impedance attenuation caused by salt spray) is high, the threshold is lowered to 6% to capture subtle anomalies. The adjustment is based on a mean-shift algorithm, combined with statistical analysis of the impedance data from the most recent 1000 frames, and is updated every 100 detections.

[0146] Implementation details: The Morse impedance measurement module dynamically updates the threshold using a weighted formula (with a weight of 0.6 for line length and 0.4 for noise level) based on the line geometry model and real-time load data (acquired through a current sensor), resulting in low computational overhead.

[0147] Statistical parameter threshold (Sp7):

[0148] Initial range: Based on the mean of normal comprehensive indicators, the typical range is [μ-0.1μ,μ+0.1μ], where μ is the mean of the comprehensive indicator fusion matrix of standard deviation, peak-to-peak value and mean.

[0149] Adjustment logic: The threshold is dynamically adjusted based on line operating time and environmental conditions. For example, for lines operating for a long time (>1 year), the threshold is widened to [μ-0.15μ, μ+0.15μ] to capture latent defects; for new lines, it is narrowed to [μ-0.08μ, μ+0.08μ] to improve accuracy. When environmental noise is high (e.g., wind noise >50dB), the threshold is increased by 0.02μ to reduce false alarms. The adjustment is based on a weighted standard deviation method, with weights of 0.7 for operating time and 0.3 for noise level, and is updated every 100 detections based on statistical data from the most recent 500 frames.

[0150] Implementation details: The statistical parameter analysis module extracts comprehensive index statistical features from the historical database and dynamically adjusts the threshold through a ReLU-type threshold function and a weighted formula to ensure high accuracy in the detection of hidden defects.

[0151] Overall implementation:

[0152] The threshold management module, scheduled by RTOS, triggers an adjustment every 100 detections. It calculates a weighted threshold based on voltage level (35kV, 110kV, 220kV), noise level (wind noise, electromagnetic interference, salt spray), and line complexity (number of components, topology). The adjustment algorithm runs on the STM32 processor's DSP instruction set, keeping memory usage below 10MB and adjustment time less than 5ms per cycle, ensuring lightweight design. The system generates adaptive thresholds by fusing historical and real-time environmental data (weighted at 0.6 for historical data and 0.4 for real-time data), covering scenarios such as mountainous areas, cities, and coastal areas, improving detection robustness and accuracy.

[0153] 2. Specific measures to address noise pollution:

[0154] To address the noise characteristics of different scenarios (such as urban, coastal, and mountainous areas), this system designs targeted preprocessing methods, expanding the applicability of low-pass filtering (50Hz cutoff frequency) and 10-point moving average smoothing to ensure high robustness against specific noises such as high-frequency electromagnetic interference, signal attenuation caused by salt spray, and wind noise. For high-frequency electromagnetic interference in urban environments, the system adds a bandpass filter to retain effective signal frequency bands, while dynamically adjusting the input signal amplitude through an adaptive gain controller to prevent electromagnetic noise from masking defect characteristics. For signal attenuation caused by salt spray in coastal areas, the system uses a signal gain amplifier, dynamically adjusting the gain coefficient based on salt spray concentration (indirectly measured by a humidity sensor) to ensure stable signal dynamic range. For vibration noise caused by strong winds in mountainous areas, the system combines 10-point moving average smoothing with a high-order low-pass filter to further suppress high-frequency vibration interference, while eliminating wind-dominated frequency bands through frequency domain analysis (e.g., eliminating high-frequency detail coefficients based on Dobbec wavelet decomposition). The selection of preprocessing parameters is based on environmental characteristics; for example, the frequency range of the bandpass filter is dynamically adjusted according to the real-time noise spectrum, and the coefficients of the gain amplifier are calibrated using historical data. All preprocessing modules are integrated into the STM32 processor's DSP instruction set and executed in real time via RTOS scheduling. This ensures that noise suppression does not significantly increase computational overhead, adapts to lightweight designs, and guarantees the integrity of signal characteristics and detection accuracy in complex environments.

[0155] 3. Specific implementation of data fusion between modules:

[0156] To achieve efficient data fusion between modules, this system designs a unified data bus fusion mechanism to ensure that the feature data output by the phase acquisition module, vibration signal acquisition module, and maximum overlap wavelet transform processing module are aligned in time and space, avoiding data conflicts from multiple sensors. The fusion process is implemented on the data bus through unified timestamp alignment, with the following logic: the phase acquisition module outputs a phase offset feature vector (containing timestamp, component ID, and phase angle value); the vibration signal acquisition module outputs a spectral energy feature vector (containing timestamp, component ID, and energy value); and the maximum overlap wavelet transform processing module outputs a time-frequency feature vector (containing timestamp, component ID, and overlap coefficient). These vectors are transmitted to the fusion module via an RS-485 bus. The fusion module uses a weighted average algorithm to integrate the features, with weights allocated based on sensor reliability (default weights: phase feature 0.6, vibration feature 0.4, and time-frequency feature 0.2, calibrated based on historical data). The fusion result generates a unified feature matrix in the format [timestamp, component ID, phase angle value, energy value, overlap coefficient], stored in a 1000-point circular buffer to prevent data congestion. The fusion module ensures data synchronization across multiple sensors through timestamp alignment. If a timestamp discrepancy is detected (e.g., due to communication delay), the system uses linear interpolation to fill in missing data points. To avoid data conflicts, the fusion module prioritizes high-reliability sensor data (e.g., phase data has higher priority than vibration data) and removes outlier data points using an anomaly detection algorithm (based on eigenvalue standard deviation analysis). The fused feature matrix is ​​directly transmitted to the principal component analysis feature compression module and the statistical parameter analysis module to ensure the efficiency and consistency of subsequent classification and localization. This mechanism is optimized through RTOS scheduling, resulting in low computational overhead and adaptability to edge devices.

[0157] 4. Error handling and abnormal situation response mechanisms:

[0158] To ensure system robustness under abnormal conditions, a comprehensive error handling mechanism is designed, covering scenarios such as sensor failure, data loss, and drone communication interruption. For sensor failure (e.g., phase sensor or vibration sensor malfunction), the system monitors sensor output in real time through a built-in diagnostic module. If a signal interruption or abnormal fluctuation (e.g., continuous zero values ​​or values ​​outside the normal range) is detected, it automatically switches to a backup sensor (each module is pre-configured with a backup sensor channel) or uses historical data interpolation to fill in missing signals. The interpolation uses a linear interpolation algorithm, predicting the current value based on the previous 10 frames of data. For data loss (e.g., data packet loss during transmission), the system stores a 1GB buffer locally, recording the original acquired data and intermediate feature data. Lost data is recovered from the buffer using timestamp matching. If recovery is impossible, it is marked as abnormal and the current detection cycle is skipped, while simultaneously triggering a warning to maintenance personnel. For drone communication interruption, the system automatically switches to local processing mode, storing the detection results in the buffer. Once communication is restored, the results are uploaded in batches via the RS-485 bus, with communication recovery detection performed once per second. The anomaly detection module runs via RTOS scheduling, periodically checking sensor status and data integrity (every 100ms) and recording anomaly logs (including timestamps, anomaly types, and handling measures) to ensure stable system operation even in harsh environments (such as storms and electromagnetic interference). The error handling mechanism is optimized using a DSP instruction set, resulting in low computational overhead, suitability for lightweight requirements, and guaranteed integrity and reliability of detection results.

[0159] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising a reference structure" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0160] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for detecting defects in multiple components of transmission lines that balances lightweight design and precision, characterized in that, include: The instantaneous phase angles of voltage and current signals from multiple components of a transmission line are collected, and the difference in phase angle offset threshold is calculated. The difference between the calculated phase angle offset threshold and the threshold is compared. The phase difference is used to distinguish between normal and defective states, and the phase angle offset is projected onto the specific component position. Vibration signals from multiple components of the transmission line are collected, and multi-resolution decomposition is performed to calculate spectral energy and differential spectral energy threshold. By comparing the difference between the spectral energy and the spectral energy threshold, energy abrupt regions exceeding the threshold are mapped to multi-component locations; The phase difference anomaly data and the spectral energy difference anomaly data are transformed and processed to extract the time-frequency characteristic information of different components, and the overlap coefficient is calculated to analyze the defect propagation path. Impedance measurements are performed on the defect propagation path, the root mean square value and duty cycle are calculated, and the impedance anomaly is projected onto the location of multiple components. The impedance anomaly data, phase difference anomaly data, and spectral energy difference anomaly data are subjected to principal component analysis, data feature compression and fusion for dimensionality reduction, and imbalanced data are processed to output low-dimensional feature set data. Based on the aforementioned low-dimensional feature set data and real-time acquired phasor data, a two-stage process analysis is performed, and compared with the aforementioned defect propagation path results to form a defect evolution link. Based on the aforementioned defect evolution chain and low-dimensional feature set data, the standard deviation, peak-to-peak value, and mean are fused to construct a parameter matrix and generate a defect severity assessment.

2. The method for detecting defects in multiple components of transmission lines that balances lightweight design and precision, as described in claim 1, is characterized in that... In the step of calculating the phase angle offset threshold difference, the discrete Fourier transform is applied to calculate the difference of the phase angle offset threshold; the threshold is set based on the statistical mean of historical normal phase angle data plus or minus 5 degrees.

3. The method for detecting defects in multiple components of transmission lines that balances lightweight design and precision, as described in claim 1, is characterized in that... In the step of projecting the phase angle offset to the specific component position, the projection is performed by integrating the topology mapping of multiple components.

4. The method for detecting defects in multiple components of transmission lines that balances lightweight design and precision, as described in claim 1, is characterized in that... In the steps of calculating spectral energy and differential spectral energy threshold, the Dobby wavelet is used for multi-resolution decomposition to generate approximation coefficients and detail coefficients. The spectral energy is calculated based on the sum of squares of the detail coefficients, and the differential spectral energy threshold is set to twice the standard deviation of the normal spectral energy.

5. The method for detecting defects in multiple components of transmission lines that balances lightweight design and precision, as described in claim 1, is characterized in that... The energy mutation region is mapped to multiple component locations through a component location index table.

6. The method for detecting defects in multiple components of transmission lines that balances lightweight design and precision according to claim 1, characterized in that, In the step of extracting time-frequency feature information of different components, the signal is processed by maximum overlap wavelet transform and non-subsampled wavelet decomposition is performed.

7. The method for detecting defects in multiple components of transmission lines that balances lightweight design and precision according to claim 1, characterized in that, In the steps of calculating the root mean square value and duty cycle, the root mean square value is calculated based on a 256-point signal window, and the duty cycle includes positive and negative duty cycles, which are determined by the proportion of positive and negative cycle time. The impedance signal is scaled to Min-Max, and impedance anomalies are mapped to multiple component locations using a line geometry model.

8. The method for detecting defects in multiple components of transmission lines that balances lightweight design and precision according to claim 1, characterized in that, In the low-dimensional feature set data processing step, principal component analysis extracts several principal component data with a cumulative variance contribution rate exceeding 85% through covariance analysis, compressing the high-dimensional features of the original data into a low-dimensional representation; a synthetic minority oversampling algorithm is used to generate new synthetic samples for minority class defect samples, calculates the interpolation between the original minority class defect samples and their nearest neighbors, expands the data distribution, and the generated new samples are supplemented into the original dataset with a default expansion rate of twice, forming balanced defect feature metadata; the low-dimensional feature set data includes defect category labels and location information.

9. A method for detecting defects in multiple components of transmission lines that balances lightweight design and precision, as described in claim 1, is characterized in that... In the two-stage process analysis steps, the first stage detects anomalies by threshold screening. When the voltage phase angle difference is greater than a set value or the current phase angle difference is greater than a set value, it is marked as a potential defect. The second stage analyzes the abnormal signals based on this, calculates the time evolution law of the phasor difference value, and tracks the propagation chain of phasor changes through time series analysis to identify chain defects between multiple components. The defect evolution chain is a time-stamped defect evolution diagram, which includes defect type, propagation direction and isolation suggestions.

10. A method for detecting defects in multiple components of transmission lines that balances lightweight design and precision, as described in claim 1, characterized in that... In the defect severity assessment step, three parameters—standard deviation, peak-to-peak value, and mean—are calculated within a 256-point signal window and integrated into a parameter fusion matrix. By default, the weights of the three types of parameters are all one-third. The comprehensive index output by the parameter fusion matrix is ​​processed by a ReLU-type threshold function. If the comprehensive index is lower than the threshold, the output is zero; if it exceeds the threshold, the output is the original value. The threshold is determined by the mean of the comprehensive index under normal conditions.

11. A detection system for a multi-component defect detection method for transmission lines that balances lightweight design and precision, characterized in that, include: The phase acquisition module is used to acquire the instantaneous phase angle of voltage and current signals from multiple components of the transmission line and calculate the difference of the phase angle offset threshold. The global fault detector construction module is connected to the phase acquisition module. It is used to compare the difference between the calculated phase angle offset threshold and the threshold, distinguish between normal and defective states through the phase difference, and project the phase angle offset onto the specific component position. The vibration signal acquisition module is used to acquire vibration signals from multiple components of the transmission line and perform multi-resolution decomposition to calculate spectral energy and differential spectral energy threshold. The differential comparison module, connected to the vibration signal acquisition module, is used to compare the difference between the spectral energy and the spectral energy threshold. Energy abrupt regions exceeding the threshold are mapped to the positions of multiple components. The maximum overlap wavelet transform processing module is used to transform the phase difference anomaly data and the spectral energy difference anomaly data, extract the time-frequency feature information of different components, and calculate the overlap coefficient to analyze the defect propagation path. The Morse impedance measurement module is used to measure the impedance of the defect propagation path, calculate the root mean square value and duty cycle, and project the impedance anomaly onto the location of multiple components. The principal component analysis feature compression module is used to perform principal component analysis on the impedance anomaly data, phase difference anomaly data, and spectral energy difference anomaly data, compress and fuse the data features to reduce dimensionality, process imbalanced data, and output low-dimensional feature set data. The phasor measurement unit synchronous detection module is used to collect phasor data in real time and perform two-stage process analysis based on the low-dimensional feature set data, and compare it with the defect propagation path results to form a defect evolution link. The statistical parameter analysis module, based on the aforementioned defect evolution chain and low-dimensional feature set data, fuses the standard deviation, peak-to-peak value, and mean to construct a parameter matrix and generate a defect severity assessment.

12. A multi-component defect detection system for transmission lines that balances lightweight design and precision, as described in claim 11, is characterized in that... The phase acquisition module uses discrete Fourier transform to calculate the difference in phase angle offset threshold; the threshold is set based on the statistical mean of historical normal phase angle data plus or minus 5 degrees.

13. A multi-component defect detection system for transmission lines that balances lightweight design and precision, as described in claim 11, is characterized in that... The global fault detector construction module integrates multi-component topology mapping for projection.

14. A multi-component defect detection system for transmission lines that balances lightweight design and precision, as described in claim 11, is characterized in that... The vibration signal acquisition module uses Dobby wavelet for multi-resolution decomposition to generate approximation coefficients and detail coefficients. The spectral energy is calculated based on the sum of squares of the detail coefficients, and the differential spectral energy threshold is set to twice the standard deviation of the normal spectral energy.

15. A multi-component defect detection system for transmission lines that balances lightweight design and precision, as described in claim 11, is characterized in that... The differential comparison module maps energy mutation regions to multiple component locations through a component location index table.

16. A multi-component defect detection system for transmission lines that balances lightweight design and precision, as described in claim 11, is characterized in that... The maximum overlap wavelet transform processing module uses maximum overlap wavelet transform to process the signal and performs non-subsampled wavelet decomposition.

17. A multi-component defect detection system for transmission lines that balances lightweight design and precision, as described in claim 11, is characterized in that... The aforementioned Morse impedance measurement module calculates the root mean square value based on a 256-point signal window, and the duty cycle includes positive and negative duty cycles, determined by the proportion of positive and negative cycle time. It uses Min-Max scaling to normalize the impedance signal and uses a line geometry model to map impedance anomalies to the locations of multiple components.

18. A multi-component defect detection system for transmission lines that balances lightweight design and precision, as described in claim 11, is characterized in that... The principal component analysis feature compression module extracts principal component data with a cumulative variance contribution rate exceeding 85% through covariance analysis, compressing the high-dimensional features of the original data into a low-dimensional representation. A synthetic minority oversampling algorithm is used to generate new synthetic samples for minority class defect samples. The interpolation between the original minority class defect samples and their nearest neighbors is calculated to expand the data distribution. The generated new samples are added to the original dataset at a default expansion rate of twice, forming balanced defect feature metadata. The low-dimensional feature set data includes defect category labels and location information.

19. A multi-component defect detection system for transmission lines that balances lightweight design and precision, as described in claim 11, is characterized in that... The phasor measurement unit synchronous detection module has a two-stage process analysis. In the first stage, anomalies are detected by threshold screening. When the voltage phase angle difference or the current phase angle difference is greater than a set value, it is marked as a potential defect. In the second stage, the abnormal signals are analyzed to calculate the time evolution law of the phasor difference value. The propagation chain of phasor changes is tracked by time series analysis to identify chain defects between multiple components. The defect evolution chain is a time-stamped defect evolution diagram, which includes defect type, propagation direction and isolation suggestions.

20. A multi-component defect detection system for transmission lines that balances lightweight design and precision, as described in claim 11, is characterized in that... The statistical parameter analysis module calculates three parameters—standard deviation, peak-to-peak value, and mean—within a 256-point signal window and integrates them into a parameter fusion matrix. By default, the weights of the three types of parameters are all one-third. The comprehensive index output by the parameter fusion matrix is ​​processed by a ReLU-type threshold function. If the comprehensive index is below the threshold, the output is zero; if it exceeds the threshold, the output is the original value. The threshold is determined by the mean of the comprehensive index under normal conditions.

Images