Insulator non-destructive detection method and system based on voiceprint recognition

Abstract

The present application relates to the field of voiceprint detection, and more particularly to a kind of insulator nondestructive testing method and system based on voiceprint recognition.The method comprises the following steps: the vibration excitation is carried out to insulator, and feature sample set is constructed by collecting vibration response signal;Feature sample set is decomposed by sliding time window, and abnormal feature sample set is extracted;Abnormal feature sample set is analyzed by principal component dimension reduction, and potential defect feature is marked;Acoustic fingerprint signal library is constructed;Potential defect feature is input into acoustic fingerprint signal library for similarity analysis, and defect identification report is output;Defect development analysis is carried out based on defect identification report, and detection report is output.The present application realizes the accurate identification and development trend evaluation of insulator potential defect, and improves the accuracy and efficiency of defect detection.

Description

Technical Field

[0001] This invention relates to the field of voiceprint detection, and more particularly to a non-destructive testing method and system for insulators based on voiceprint recognition. Background Technology

[0002] Insulators are critical equipment in power grid infrastructure, and their quality directly determines the safety and stability of transmission lines. During power grid operation, insulators bear the dual responsibility of electrical insulation and mechanical support, making their reliability paramount. However, during the production, transportation, and storage of insulators, issues such as internal cracks, bubbles, material spalling, or structural defects may arise due to process deviations, material defects, and impact damage. If these hidden dangers are not detected before grid connection, they will worsen under high voltage, environmental stress, and long-term operation, potentially leading to equipment failure or power grid accidents. Therefore, power companies need to establish a strict quality control system when procuring insulators, conducting random inspections and screenings before grid connection to eliminate substandard products and ensure the safety and reliability of power grid operation.

[0003] As power grid equipment ages, the risk of insulator aging and damage also increases. To further enhance equipment monitoring capabilities, the State Grid Equipment Department issued the "Notice on the 2024 Material Technical Supervision and Special Spot Check and Testing of Electrical Equipment Performance," explicitly requiring comprehensive technical supervision and testing of insulators. Therefore, timely and accurate detection of potential defects and problems in insulators is crucial for preventing accidents and malfunctions.

[0004] Traditional insulator quality inspection methods mainly include visual inspection, electrical testing, infrared imaging, ultrasonic testing, and X-ray inspection. While these methods can detect insulator defects to some extent, they still have many shortcomings in actual production quality inspection: visual inspection relies on human experience, resulting in low efficiency and accuracy; electrical testing, while assessing insulation performance, struggles to precisely locate defects and involves cumbersome procedures; infrared imaging and ultrasonic testing, although suitable for detecting certain internal defects, have high requirements for equipment cost and operating environment; X-ray inspection offers high precision but is expensive and poses radiation safety concerns. Therefore, existing inspection technologies face challenges such as low operational efficiency, insufficient applicability, and high cost when meeting the needs of large-scale insulator sampling inspections. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention proposes a non-destructive testing method and system for insulators based on voiceprint recognition, thereby resolving at least one of the aforementioned technical issues.

[0006] To achieve the above objectives, the present invention provides a non-destructive testing method for insulators based on voiceprint recognition, comprising the following steps: Step S1: Vibrate the insulator and collect the vibration response signal to construct a feature sample set; Step S2: Perform sliding window decomposition on the feature sample set to extract the abnormal feature sample set; Step S3: Perform principal component dimensionality reduction analysis on the abnormal feature sample set and label potential defect features; Step S4: Construct an acoustic fingerprint signal database; input the potential defect features into the acoustic fingerprint signal database for similarity analysis, and output a defect identification report; Step S5: Analyze the defect development based on the defect identification report and output a detection report.

[0007] This specification provides a non-destructive testing system for insulators based on voiceprint recognition, used to perform the non-destructive testing method for insulators based on voiceprint recognition as described above, including: The acquisition unit is used to excite the insulator by vibration and collect vibration response signals to construct a feature sample set. The sample unit is used to perform sliding window decomposition on the feature sample set to extract the abnormal feature sample set; The analysis unit is used to perform principal component dimensionality reduction analysis on the abnormal feature sample set and to label potential defect features. The defect identification unit is used to construct an acoustic fingerprint signal database; input the potential defect features into the acoustic fingerprint signal database for similarity analysis, and output a defect identification report; The trend analysis unit is used to analyze defect development based on the defect identification report and output a detection report.

[0008] The beneficial effects of this invention are as follows: By applying controllable vibration excitation to the insulator, its internal structure (such as cracks, pores, delamination, aging interfaces, etc.) generates differentiated acoustic and vibration responses under forced vibration conditions, effectively stimulating the dynamic characteristics of potential defects and avoiding the problem of weak and poorly discernible signals under natural operating conditions. Vibration excitation makes the differences in modal responses, resonant frequencies, and energy distributions corresponding to different structural states more obvious, thereby increasing the proportion of defect-related features in the acquired signal, enhancing the signal-to-noise ratio, and facilitating subsequent feature analysis. Through sliding time window decomposition, long-term sequence signals are divided into multiple local time segments, effectively capturing transient anomalies, non-stationary features, and intermittent defect responses, avoiding the problem of overall analysis masking local anomalies. The sliding time window mechanism can perform overlapping analysis on the signal, allowing short-term anomalies caused by defects to repeatedly appear in multiple time windows, improving the stability and reliability of anomaly feature detection. Principal component analysis (PCA) can map high-dimensional, highly correlated acoustic and vibration anomaly features to a low-dimensional orthogonal feature space, retaining the main energy and key information, reducing redundant features, and improving analysis efficiency. The principal components after dimensionality reduction often correspond to the main patterns of insulator structural state changes, which helps to highlight feature changes caused by potential defects and make the distribution structure of abnormal features clearer in the feature space. The acoustic fingerprint signal library uniformly models and stores the acoustic and vibration features corresponding to different defect types and development stages, forming a structured and scalable defect knowledge base that facilitates long-term accumulation and updates. Through similarity analysis, the abnormal features to be tested are compared with known defect features in the fingerprint library, effectively suppressing the influence of environmental noise and individual differences, achieving more stable and reliable defect type identification. The acoustic fingerprint library can simultaneously cover multiple defect modes (such as microcracks, through cracks, internal voids, material aging, etc.), enabling comprehensive identification of complex defect conditions and improving the system's applicability. Based on defect type identification, the severity, development trend, and potential risks of the defects are analyzed, avoiding only qualitative conclusions and improving the engineering guidance value of the detection results. Through defect development analysis, it is possible to determine whether the insulator is in an early deterioration, stable defect, or rapid deterioration stage, providing maintenance personnel with a basis for prioritizing maintenance and preventive maintenance. The final test report integrates acoustic and vibration analysis results, defect identification conclusions, and development trend assessments to achieve standardized output of test results, facilitating archiving, comparative analysis, and integration with intelligent operation and maintenance systems. Attached Figure Description

[0009] Figure 1 This is a schematic diagram of the steps of a non-destructive testing method for insulators based on voiceprint recognition according to the present invention; Figure 2 This is a detailed flowchart illustrating the implementation steps of step S1. Figure 3 This is a detailed flowchart illustrating the implementation steps of step S2; Figure 4 This is a schematic diagram of an insulator structure. Detailed Implementation

[0010] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0011] This application provides a method and system for non-destructive testing of insulators based on voiceprint recognition. The execution entities of the method and system include, but are not limited to, mechanical equipment, data processing platforms, cloud server nodes, and network upload devices mounted on the system, which can be considered as general computing nodes in this application. The data processing platform includes, but is not limited to, at least one of an audio-visual management system, an information management system, and a cloud-based data management system.

[0012] Please see Figures 1 to 4 This invention provides a non-destructive testing method for insulators based on voiceprint recognition, comprising the following steps: Step S1: Vibrate the insulator and collect the vibration response signal to construct a feature sample set; Step S2: Perform sliding window decomposition on the feature sample set to extract the abnormal feature sample set; Step S3: Perform principal component dimensionality reduction analysis on the abnormal feature sample set and label potential defect features; Step S4: Construct an acoustic fingerprint signal database; input the potential defect features into the acoustic fingerprint signal database for similarity analysis, and output a defect identification report; Step S5: Analyze the defect development based on the defect identification report and output a detection report.

[0013] In the embodiments of the present invention, see Figure 1 The diagram below illustrates the steps of a non-destructive testing method for insulators based on voiceprint recognition according to the present invention. In this example, the steps of the non-destructive testing method for insulators based on voiceprint recognition include: Step S1: Vibrate the insulator and collect the vibration response signal to construct a feature sample set; In this embodiment, a broadband vibration excitation is applied to the insulator to obtain the dynamic response of its overall and local structures, thereby constructing a joint acoustic-vibration feature sample set. The vibration excitation uses a random acoustic excitation signal covering 100 Hz to 25 kHz, which can excite the low-frequency overall mode and high-frequency local defect-sensitive mode of the insulator. Vibration response signals are acquired at multiple points at key structural locations of the insulator, including the edge of the skirt, the middle of the rod, and the connection end, using a high-sensitivity accelerometer to collect vibration data. The frequency response range of the vibration sensor covers at least 30 kHz to ensure accurate recording of high-frequency characteristics. During the acquisition process, continuous signals are sampled for an appropriate time length, such as 20 s to 60 s, to ensure that the vibration reaches a steady-state response. The acquired multi-channel vibration data is processed using feature extraction methods, including power spectral density analysis, modal band energy distribution calculation, and modal coupling strength statistics, to form a preliminary vibration response feature set. Near-field acoustic radiation signals are acquired through a microphone array and subjected to time-domain feature analysis to extract peak value, root mean square value, kurtosis, and autocorrelation features.

[0014] Step S2: Perform sliding window decomposition on the feature sample set to extract the abnormal feature sample set; In this embodiment, after obtaining a complete set of acoustic and vibration joint feature samples, a sliding time window decomposition is performed to capture temporal variation features, thereby identifying anomalous features. The sliding time window processing is achieved by setting a fixed-length time window and a certain overlap ratio. The time window length is typically 50 ms to 200 ms, and the step size is 25% to 50% of the window length, ensuring temporal continuity and sensitivity to anomalous changes. The feature subset corresponding to each time window is mapped to a low-dimensional feature space, forming an acoustic signature trajectory. Subsequently, the rate of change and amplitude of change of the trajectory are calculated, and statistical analysis is performed on the changes between consecutive time windows. Trajectory regions with high rates of change and large amplitudes are identified as potential anomalous segments, and the corresponding feature samples are selected to form an anomalous feature sample set. This processing method can not only eliminate feature interference in stable states but also detect minute structural anomalies at an early stage.

[0015] Step S3: Perform principal component dimensionality reduction analysis on the abnormal feature sample set and label potential defect features; In this embodiment, principal component analysis (PCA) is used to reduce the dimensionality of the extracted abnormal feature sample set, eliminating feature redundancy and highlighting key defect-related features. The abnormal feature sample set is standardized to ensure uniformity in the dimensions of each feature, guaranteeing the effectiveness of PCA. Subsequently, the covariance matrix of the sample set is calculated and eigenvalue decomposition is performed. Principal components with high variance contribution rates are extracted as low-dimensional feature subsets; typically, principal components with a cumulative variance contribution rate of 85%–95% are selected to ensure information integrity. In the low-dimensional feature space, clustering and statistical analysis are performed on the abnormal feature samples to identify feature dimensions highly correlated with the defect state as potential defect features.

[0016] Step S4: Construct an acoustic fingerprint signal database; input the potential defect features into the acoustic fingerprint signal database for similarity analysis, and output a defect identification report; In this embodiment, after obtaining potential defect features, an acoustic fingerprint signal library is constructed for defect matching. The acoustic fingerprint signal library includes standard acoustic print templates for normal and typical defect states. Each template contains low-dimensional acoustic print feature mean, covariance, and statistical distribution information to characterize the central tendency and fluctuation range of the acoustic vibration features under that state. After the potential defect features are input into the acoustic fingerprint signal library, a matching score with each defect template is obtained through feature distance measurement or similarity calculation methods. The matching score is normalized to 0-1; a higher value indicates that the potential defect feature is closer to the corresponding template. Subsequently, a similarity judgment threshold (usually 0.7-0.85) is set. When the score is not lower than the threshold, a corresponding defect is determined to exist; otherwise, no defect is determined. Step S5: Analyze the defect development based on the defect identification report and output a detection report.

[0017] In this embodiment, based on the defect identification report, the insulator defect status is analyzed over a continuous time period to form a defect development trend. The defect identification results for the continuous time period are arranged in chronological order to construct time-series structural state data, including defect type, confidence level, and characteristic deviation value. Trend change calculations are performed on the time-series data to analyze the defect structure change curve and its development direction. When the defect characteristic index shows a continuous upward trend over time, the defect is determined to be in the expansion stage, and an early warning signal is generated; when the defect characteristic index remains stable, it is recorded as stable stage information. The defect type, development stage, early warning information, and confidence level are integrated to form a complete inspection report.

[0018] In this embodiment, see Figure 2 The diagram below illustrates the detailed implementation steps of step S1. In this embodiment, the detailed implementation steps of step S1 include: Design an acoustic excitation source; based on the acoustic excitation source, vibrate the insulator, and use acoustic sensors to collect data synchronously at multiple points to extract random vibration response signals and near-field acoustic radiation signals; the acoustic excitation source covers a wideband random excitation signal in the frequency range of 100Hz to 25kHz.

[0019] Power spectral density analysis was performed on the random vibration response signal to extract the modal frequency band energy distribution, modal coupling strength and frequency stability, and the vibration response feature set was obtained by fitting. Time-domain feature analysis based on near-field acoustic radiation signals is performed to extract multi-dimensional voiceprint representation vectors. A joint modeling of the vibration response feature set and the multidimensional acoustic signature vector is performed to construct a feature sample set.

[0020] In this embodiment, a broadband acoustic excitation source is constructed to achieve random vibration excitation of the insulator. The acoustic excitation source uses an electroacoustic transducer with broadband response characteristics, its effective operating frequency range covering 100 Hz to 25 kHz, used to simultaneously excite the overall structural vibration response and local high-frequency response characteristics of the insulator. The excitation signal adopts broadband random noise, generated by a signal generation module within the range of 100 Hz to 25 kHz, and then amplified to drive the acoustic excitation source output. The excitation sound pressure level is controlled within the range of 90 dB to 110 dB to ensure sufficient and stable excitation energy. A fixed distance is maintained between the acoustic excitation source and the insulator, ranging from 100 mm to 300 mm. By adjusting the directivity of the sound source, the random sound field is uniformly applied to the overall structure of the insulator. Under the acoustic excitation, the random vibration response and near-field acoustic radiation signal generated by the insulator are acquired synchronously at multiple points. Vibration response signals were acquired using multi-point vibration sensors positioned at key structural locations on the insulator, including the skirt area, connection ends, and central structural locations, to comprehensively characterize the insulator's dynamic response characteristics. Near-field acoustic radiation signals were acquired using multi-point acoustic sensors positioned near the insulator surface, with the distance between the acoustic sensors and the insulator surface set to 20 mm–50 mm, to obtain near-field acoustic radiation information caused by structural vibration. Vibration and acoustic signals were acquired synchronously under unified timing conditions, with a sampling frequency set not lower than 51.2 kHz to cover the highest analysis frequency of 25 kHz.

[0021] The vibration signal undergoes preprocessing, including DC component removal, bandpass filtering, and window function weighting, to reduce spectral leakage and interference from non-target frequency bands. Subsequently, power spectral density estimation is used to perform frequency domain analysis on the vibration signal, obtaining energy distribution characteristics across different frequency ranges. Based on the power spectral density results, the modal frequency band energy distribution characteristics of the insulator in different frequency bands are extracted, and modal coupling strength characteristics are extracted by combining the frequency domain correlation between vibration signals from multiple measurement points. Bandpass filtering and amplitude normalization are applied to the acoustic radiation signal to eliminate the influence of background noise and amplitude differences on the analysis results. During time domain analysis, statistical characteristics such as the root mean square value, peak value, peak factor, kurtosis, and skewness of the acoustic signal are calculated to extract information on acoustic radiation intensity and non-stationary characteristics. Simultaneously, short-time energy analysis and autocorrelation analysis methods are combined to extract transient and stochastic characteristics of the acoustic signal. For multi-point near-field acoustic radiation signals, the spatial distribution consistency characteristics of acoustic radiation can also be extracted by analyzing the correlation between signals from different acoustic sensors. Vibration and acoustic signature features are time-aligned and scale-unified to ensure consistent correspondence between the two types of features under the same excitation conditions. Subsequently, a feature-level fusion method is used to combine the vibration response features and acoustic signature features to form a unified acoustic-vibration joint feature vector. To reduce feature redundancy and highlight features sensitive to insulator state changes, correlation analysis and dimensionality reduction are performed on the joint features.

[0022] In the above technical steps, it is necessary to understand a principle of vibration acoustic detection, specifically: when using instruments for testing, the insulator vibrates under the excitation force emitted by the instrument, which can be regarded as forced vibration of the insulator. When a typical multi-degree-of-freedom system is subjected to an external force F and undergoes forced vibration, the equilibrium equation of the system can be obtained according to d'Alembert's principle: In the formula: M, C and K represent the mass matrix, the viscous damping matrix and the stiffness coefficient matrix of the structure, respectively; , These are the displacement, velocity, and acceleration matrices of the system, respectively, and F is the excitation force signal.

[0023] From the solution of the above equation, we can see that if the degree of freedom p is subjected to a dynamic load... By applying the effect of [the law], we obtain the displacement response power spectrum function of degree of freedom k, that is: In the formula: Let be the power spectral density function of the displacement response with degree of freedom k; For a given degree of freedom p, the dynamic load acting on it... The self-power spectral density function, It is the i-th mode shape of the undamped system; r represents the number of degrees of freedom; It is the transfer function of the structure's random vibration response. It is conjugate. Represented as: In the formula, It is the damping ratio of the j-th mode of vibration of the system, where i is the imaginary unit.

[0024] As can be seen from the above equation, if the structure is subjected to stationary random vibration excitation in the p-degree of freedom, then the vibration response of the structure in the k-degree of freedom is also a stationary random signal. Since forced vibration of a multi-degree-of-freedom system can only excite relatively low-order vibrations, it is only necessary to retain a finite number of mode shapes according to the solution accuracy when solving the problem.

[0025] To obtain the structural transfer function shown in the above equation, the structure needs to be excited. Commonly used excitation signals include sinusoidal steady-state excitation and random excitation. The excitation method used in this paper is random excitation generated by a vibration testing instrument. The purpose is to induce vibration in the structure, and to indirectly determine whether the structure has been damaged by analyzing its vibration response signal.

[0026] Since the excitation signal is a random vibration, theoretically it does not have a frequency spectrum. However, its power can be considered uniformly distributed with respect to frequency. Therefore, the power spectrum analysis method is generally used for random vibration response. According to the above equation, when the phase information of the excitation is ignored, we can obtain: If the input excitation is a random vibration signal, or if a flat frequency spectrum is required for the input signal, it can be approximated as a finite-bandwidth white noise signal. In this case, its autocorrelation function only approaches infinity at 0, and remains 0 at all other positions. According to the Wiener-Khinchin theorem, its power density spectrum is the Fourier transform of the autocorrelation function. Therefore, its power spectrum can be considered a constant C. From this, we can derive: The above equation demonstrates that the power spectrum characteristics of the system's vibration response under random vibration input are consistent with the structural spectrum characteristics, which is also the theoretical basis of vibration flaw detectors. The analysis of vibration problems is not only time-dependent but also highly frequency-dependent; therefore, using power spectral density to analyze random vibration problems is more reasonable. However, in practical applications, whether the vibration response signal can reflect the structural spectrum characteristics of the system is also affected by the degree to which the excitation signal conforms to the finite-bandwidth white noise assumption.

[0027] In practical applications of spectral analysis or power spectrum estimation of system transfer functions, it is often necessary to treat the excitation and response signals as truncated signals over infinite time and employ Fourier transforms. This inevitably leads to power spectrum leakage. Furthermore, discretizing sensor data also affects the accuracy of power spectrum estimation due to varying spectral resolution. Consequently, practical vibration testing instruments rarely obtain truly accurate system transfer functions.

[0028] As described in the previous section, insulators are typically approximately cylindrical, with flanges at both ends serving a connecting function. They are generally made of cast iron and have high mechanical strength. For example... Figure 4 As shown, for insulators used to support the busbar, the lower end is usually fixed with bolts, while the upper end supports the busbar. The load is relatively small, so it is usually considered to be in a free state. Therefore, the entire insulator can be equivalent to an upright cantilever beam structure.

[0029] The formulas listed in the previous section are general formulas for the vibration of multi-degree-of-freedom structures. Their solutions show the relationship between the structure's transfer function and its natural frequencies, which differ depending on the object being studied. The vibration patterns of insulators can be described by equations for bending, longitudinal, and torsional vibrations. Theoretically, the natural frequencies of the insulator can be solved from these equations, allowing the construction of the system transfer function for an equivalent model. The vibration equations are as follows: Bending vibration: In the formula: ω is the column bending deflection; x is the column cross-sectional coordinate; t is time; E is the elastic modulus of the insulator material; ρ is the relative density of the insulator part; A is the cross-sectional area of ​​the insulator; q(x, t) is the excitation force.

[0030] Longitudinal vibration: In the formula: u is the longitudinal displacement of the column.

[0031] Torsional vibration: In the formula: θ(x, t) is the torsion angle of the column cross section; GJ is the torsional stiffness of the column material; f(x, t) is the moment of inertia per unit length; f(x, t) is the excitation torque.

[0032] Because the vibration modes of insulators are quite complex and difficult to obtain analytical solutions, finite element analysis is generally used to approximate numerical solutions for the mode shapes and natural frequencies. However, determining insulator damage does not require precise solutions; it is sufficient to excite a specific vibration mode, and the changes in system parameters can be determined from the vibration response signal.

[0033] Insulators subjected to random excitation primarily undergo bending deformation, similar to that of a cantilever beam. The stress limit restricts the degree of deformation before failure. Exceeding this deformation limit will result in damage; this strength limit is reflected in the stress level as the stress limit. Based on the above analysis, if the insulator is equivalent to a vertical columnar cantilever beam, and an external force is applied to induce bending vibration, its natural vibration frequency has the following expression: In the formula: ω is the natural vibration frequency of the columnar structure; k is the solution to the Krylov equation; I is the static moment of inertia of the cross section at the damage location of the columnar structure; The mass of the cross-section of the columnar structure.

[0034] As can be seen from the above formula, the natural frequency is mainly related to the insulator length, the moment of inertia of the cross section, and the mass per unit length. When a certain cross section of the insulator is damaged, its moment of inertia changes, which in turn causes the natural frequencies of each order to change. Therefore, the change in the natural frequency can be used to determine whether the insulator is damaged.

[0035] The power spectrum of the random vibration response signal, derived in the previous section, can reflect the system's transfer function. Since the system's transfer function is closely related to the structure's natural frequency and mode shape, defects affect the structure's stiffness and thus its transfer function, which is reflected in the power spectrum of the random vibration response signal.

[0036] In this embodiment, see Figure 3 The diagram below illustrates the detailed implementation steps of step S2. In this embodiment, the detailed implementation steps of step S2 include: The feature sample set is decomposed by a sliding time window to obtain sample sets for multiple time periods; The sample set is subjected to voiceprint feature space mapping to generate voiceprint trajectories for different time periods; Calculate the rate of change and amplitude of change of the voiceprint trajectory; Based on the rate and magnitude of change, stable trajectories are identified, and abnormal abrupt change locations are identified, marking the abrupt change points of abnormal trajectories. Based on the abnormal trajectory mutation points, the source is located and traced, and an abnormal feature sample set is extracted.

[0037] In this embodiment, the sliding time window decomposition employs a combination of fixed window length and fixed step size to segment the continuous acoustic vibration feature sequence over time. The window length is set based on the temporal stability of the acoustic vibration features, typically within the range of 50ms to 200ms, to balance short-term feature sensitivity and statistical stability. The window step size is set to 25% to 50% of the window length to ensure some overlap between adjacent time periods, thereby improving the ability to capture abnormal changes. Through the sliding time window method, the original acoustic vibration feature sample set is decomposed into multiple interconnected but temporally continuous sub-sample sets, each corresponding to the acoustic vibration feature state of the insulator within a specific time period. For the multiple time-segment sample sets obtained through sliding time window decomposition, acoustic vibration features within each time period are processed by acoustic fingerprint feature space mapping to construct an acoustic fingerprint trajectory that evolves over time. Acoustic fingerprint feature space mapping is achieved through feature transformation and dimensionality reduction methods, mapping the original high-dimensional acoustic vibration joint features to a low-dimensional discriminative space, enabling the features of different time periods to be continuously expressed in the form of trajectories. During the mapping process, the relative distance relationships and structural similarities between acoustic signature features are maintained to ensure that the trajectory changes accurately reflect the evolution trend of the insulator's state. Each time window corresponds to an acoustic signature feature point in the feature space, and they are connected in chronological order to form an acoustic signature trajectory.

[0038] After obtaining the acoustic signature trajectory, the dynamic characteristics of the trajectory in the feature space are quantitatively analyzed, focusing on calculating the rate of change and amplitude of change of the acoustic signature trajectory between adjacent time periods. The rate of change describes the movement speed of the acoustic signature feature points per unit time, reflecting how fast the acoustic signature features change over time; the amplitude of change describes the displacement distance of the acoustic signature feature points in the feature space, used to characterize the strength of the state change. During the calculation process, distance measurements are performed on the acoustic signature feature points corresponding to adjacent time windows to obtain the change parameters between consecutive time periods, and the change parameters are smoothed to reduce the impact of instantaneous noise on the results. By statistically analyzing the distribution of the rate of change and amplitude of change over a longer time range, the threshold interval corresponding to the stable trajectory is determined. When the rate of change and amplitude of change of the acoustic signature trajectory are both within the threshold interval in multiple consecutive time periods, the trajectory segment is determined to be a stable trajectory. Subsequently, when the rate of change or amplitude of change of the acoustic signature trajectory significantly exceeds the stable interval in a certain time period, the position is determined to be an abnormal trajectory abrupt change point. This abrupt change point usually corresponds to the time position where the state of the insulator structure changes significantly.

[0039] Based on the time window location corresponding to the abrupt change in the abnormal trajectory, the acoustic and vibration joint features within that time period are traced back, and feature samples within several time windows before and after the change occur are extracted to construct an anomalous feature sample set. The traceback time range can be set to 100 ms to 500 ms before and after the change point to fully capture the anomalous evolution process. The extracted anomalous feature sample set contains typical acoustic and vibration feature information of insulators under abnormal conditions, which can be significantly distinguished from steady-state features.

[0040] In this embodiment, step S3 includes the following steps: Principal component dimensionality reduction analysis is performed on the abnormal feature sample set to extract a low-dimensional feature subset; Based on the aforementioned low-dimensional feature subset, a correlation analysis of key defects is performed to label potential defect features.

[0041] In this embodiment, the abnormal feature sample set undergoes feature standardization to ensure that all feature dimensions are compared on a uniform scale, avoiding the unbalanced impact of features with different dimensions on the dimensionality reduction results. Based on this, a feature covariance matrix is ​​constructed by calculating the covariance relationship between each feature dimension in the abnormal feature sample set, used to characterize the overall distribution characteristics of the acoustic-vibration joint features under abnormal conditions. Subsequently, eigenvalue decomposition is performed on the covariance matrix, and the principal components are sorted according to their eigenvalues, prioritizing the retention of principal components that contribute significantly to the overall feature variance. The number of principal components retained is determined based on the cumulative contribution rate, typically set to the number of principal components that contribute 85% to 95% of the cumulative variance, thereby significantly reducing the feature dimensionality while ensuring the integrity of the main information. Through principal component mapping, the original high-dimensional abnormal feature samples are projected onto a low-dimensional feature space composed of principal components, forming a low-dimensional feature subset. Difference metrics between low-dimensional features in abnormal and stable states are calculated, including the degree of feature mean shift, the magnitude of variance change, and the change in feature distribution morphology, thereby quantifying the response intensity of each low-dimensional feature to the defect state. Feature dimensions that exhibit significant changes in anomalous samples but remain relatively consistent in stable samples are identified as key features strongly correlated with defect states. The correlation between features and defect types is further validated by analyzing the clustering trends and separation degrees of low-dimensional features among different anomalous samples. The feature dimensions selected through correlation analysis are then labeled to form a set of potential defect features.

[0042] In this embodiment, step S4 includes the following steps: Construct an acoustic fingerprint signal library; the acoustic fingerprint signal library includes standard acoustic fingerprint templates for normal and typical defective states and their feature statistical scores; The potential defect features are input into an acoustic fingerprint signal database for similarity calculation to obtain a similarity score; Set a similarity judgment threshold; determine the existence of similarity scores based on the similarity judgment threshold. If the similarity score is not less than the similarity judgment threshold, it is determined that a matching defect type exists, and a defect existence judgment result is generated. If the similarity score is less than the similarity judgment threshold, it is judged to be without defects; Based on the defect existence determination results, voiceprint deviation analysis and confidence calculation are performed, and a defect identification report is output.

[0043] In this embodiment, the acoustic fingerprint signal library consists of acoustic fingerprint templates in normal state and acoustic fingerprint templates in various typical defect states. Each type of template corresponds to a set of standardized acoustic fingerprint features and their statistical distribution characteristics. During the construction process, the acoustic-vibration joint features obtained under different states are classified and organized. Feature samples in a structurally stable state are categorized as acoustic fingerprint templates in normal state, while samples with typical defect features such as cracks, loosening, and aging are categorized as acoustic fingerprint templates in the corresponding defect states. For each type of acoustic fingerprint template, its feature mean vector, covariance matrix, and feature fluctuation range are extracted using statistical analysis methods to describe the central tendency and discrete characteristics of the acoustic fingerprint features in that state. The acoustic fingerprint signal library uses the same feature dimension, normalization method, and feature space expression form throughout the construction process to ensure comparability between different acoustic fingerprint templates. After the acoustic fingerprint signal library is constructed, the potential defect features marked in the aforementioned steps are input into the acoustic fingerprint signal library, and similarity calculations are performed with various standard acoustic fingerprint templates. Similarity calculation takes a low-dimensional feature subset as input and uses feature distance measurement or distribution matching to quantitatively evaluate the similarity between potential defect features and state templates in the acoustic fingerprint signal database. The calculation focuses on the overall differences between feature vectors and the consistency of feature distribution structure to ensure that the similarity score accurately reflects the matching degree between potential defect features and standard voiceprint templates. For each type of state template in the acoustic fingerprint signal database, a corresponding similarity score is calculated, forming a set of similarity score results. This score value is typically normalized to the range of 0 to 1; a higher value indicates a higher degree of similarity between the potential defect feature and the corresponding voiceprint template.

[0044] After obtaining the similarity scores between potential defect features and various voiceprint templates, the presence of defects is determined by setting a similarity judgment threshold. The similarity judgment threshold is determined based on the statistical characteristics of different state templates in the acoustic fingerprint signal database, comprehensively considering the similarity distribution range between normal and defective voiceprints to ensure good discriminative ability. During threshold setting, the upper similarity limit corresponding to the normal state voiceprint template and the lower similarity limit corresponding to the defective state voiceprint template are analyzed, and a judgment threshold range between the two is selected, typically set to a specific value within the range of 0.7 to 0.85. Subsequently, the similarity scores of potential defect features are compared with the judgment threshold. When the similarity score corresponding to a certain defective state template is not less than the similarity judgment threshold, it is determined that the potential defect feature matches the defect type, and a defect presence judgment result is generated; when the similarity scores corresponding to all defective state templates are less than the similarity judgment threshold, it is determined to be a defect-free state. After obtaining the defect presence judgment result, voiceprint deviation analysis and confidence calculation are further performed on the voiceprint features determined to have defects to improve the reliability and interpretability of the defect identification results. Voiceprint deviation analysis assesses the degree of deviation of the current voiceprint features from the standard template by comparing the feature offset between potential defect features and the corresponding defect state voiceprint template, reflecting the severity of the defect. Confidence calculation comprehensively considers the similarity score, the magnitude of voiceprint deviation, and the score differences between different defect templates to quantitatively evaluate the reliability of the defect identification results. Confidence scores are typically given as a percentage; higher values ​​indicate more reliable defect identification results. The defect type determination results, similarity scores, voiceprint deviation analysis results, and confidence information are integrated to form a complete defect identification report.

[0045] In this embodiment, the specific steps for performing voiceprint deviation analysis and confidence calculation based on the defect existence determination result, and outputting a defect identification report are as follows: Based on the defect existence determination result, the potential defect features are subjected to multiple standard defect voiceprint deviation calculations to obtain the deviation value of each standard defect voiceprint. Confidence levels are calculated based on the deviation values, and confidence levels for various defect types are output. Based on the stated confidence level, defect identification analysis is performed, and a defect identification report is output.

[0046] In this embodiment, after determining the existence of defects, the deviation between the identified potential defect voiceprint features and various standard defect voiceprint templates is calculated to quantify the degree of difference between the current voiceprint features and various defect states. The standard defect voiceprint templates are derived from typical defect state templates established in the acoustic fingerprint signal database. Each template includes the mean low-dimensional voiceprint features, feature distribution range, and statistical fluctuation characteristics under the corresponding defect state. During the deviation calculation, the potential defect features are mapped to a feature space consistent with the standard defect voiceprint templates, and the feature offset between them and various standard defect voiceprint templates is calculated. The offset considers not only the distance relationship of the feature vectors in the overall space but also comprehensively evaluates the degree of difference in each feature dimension to ensure that the deviation calculation accurately reflects subtle changes in voiceprint morphology. In this way, the deviation value of the potential defect features relative to each standard defect voiceprint template can be obtained. The deviation value is usually a continuous value; the smaller the deviation, the closer the potential defect features are to the voiceprint features of the corresponding defect type. After obtaining the deviation values ​​between potential defect features and various standard defect voiceprint templates, the confidence levels for different defect types are calculated based on these deviation values. The core of the confidence level calculation lies in mapping the magnitude of the voiceprint deviation to the degree of confidence in defect matching, thereby quantifying the probability that the current voiceprint feature belongs to a certain defect type. During the calculation process, the deviation values ​​corresponding to various defects are normalized to eliminate the influence of numerical scale differences between different feature dimensions and different defect templates. Subsequently, based on the inverse relationship between the deviation value and the defect correlation, a confidence level mapping relationship is constructed, so that defect types with smaller deviation values ​​correspond to higher confidence levels, while defect types with larger deviation values ​​correspond to lower confidence levels. To enhance the stability of the results, the statistical characteristics of the deviation distribution can also be introduced to smooth the confidence levels, avoiding excessive influence of single feature anomalies on the results.

[0047] During the identification and analysis process, the confidence levels of each defect type are compared to determine the defect type with the highest confidence level as the most likely defect category corresponding to the current voiceprint feature. When the confidence levels of multiple defect types are close, the presence of composite defects or early evolutionary states of defects can be determined by combining the confidence level distribution characteristics. When the confidence levels of all defect types are lower than the preset confidence threshold, the current voiceprint feature is determined to lack obvious defect characteristics. The defect identification results, the confidence level values ​​corresponding to each defect type, and the voiceprint deviation analysis conclusions are comprehensively compiled to form a structured defect identification report output.

[0048] In this embodiment, step S5 includes the following steps: Continuous identification and judgment are performed based on the defect identification report to obtain defect identification results for multiple time periods; Based on the defect identification results, insulator structure state analysis is performed, and time-series structure state data is extracted. Based on the time-series structural state data, trend change calculations are performed to identify the defect structure change curve; The development direction of the defect structure change curve is analyzed. When the defect development direction is detected to be time-series growth, it is determined that the defect is in the expansion stage and an early warning signal is output. When a defect is detected to be developing in a continuous direction, the information is recorded and a detection report is output.

[0049] In this embodiment, potential defect features within a continuous time window are input into the acoustic fingerprint matching and confidence analysis process to obtain the defect type, confidence level, and deviation value corresponding to each time period. By arranging the identification results of continuous time periods in a time sequence, a record of the defect state evolution of the insulator at different time points can be formed. The time window length is typically set to 50 ms to 200 ms, with a step size of 25% to 50% of the window length to balance detection accuracy and dynamic response continuity. During the identification process, abnormally low confidence time periods are filtered out using a confidence threshold to avoid misjudgment, while defect matching results with higher confidence are retained as continuous identification data. The defect type, confidence level, and deviation value of each time period are mapped to structural state indicators and arranged in chronological order to form time-series structural state data. The time-series structural state data may include dimensional information such as defect presence identification, defect type label, corresponding confidence level value, and feature deviation magnitude for each time window to comprehensively reflect the state change trend of the insulator within a continuous time period. By statistically analyzing the changes in structural state indicators for each time period, the stability, potential evolution trend, and stage characteristics of the defect state can be revealed.

[0050] After obtaining the time-series structural state data, trend change analysis is performed to identify the dynamic evolution curve of insulator defects. Trend change calculation quantifies the characteristics of defect state changes over time by comparing the rate and magnitude of change of structural state indicators within continuous time periods. Specific methods include binary sequence analysis of defect presence indicators, smoothing of confidence levels and characteristic deviations, and calculation of the gradient between each time period and the previous time period. Through cumulative analysis of the gradient, a defect structure change curve is formed, reflecting the trajectory of defect characteristics over time. Curve analysis focuses on the magnitude, direction, and continuity of change, effectively distinguishing between the expansion, stable, and latent stages of defects. The time window and step size settings are consistent with the continuous identification steps to ensure the temporal resolution and dynamic traceability of the curve. Analyzing the changing trends of structural state indicators within continuous time periods, when multiple consecutive time windows show a continuous upward trend in defect presence indicators or characteristic deviations, the defect development direction is determined to be time-series growth, indicating that the defect is in the expansion stage. Development direction analysis can combine methods such as curve slope calculation, mean rate of change, and trend persistence assessment to ensure the distinction between long-term cumulative changes and short-term fluctuations. When a defect is determined to be in the expansion stage, a corresponding early warning signal is generated. The early warning signal includes the defect type, defect development rate, and current structural status indicators.

[0051] During the analysis of defect structure change curves, if the structural state indicators over consecutive time periods show that the defect development direction remains unchanged, i.e., the defect characteristic indicators remain stable over time without a significant upward trend, then the defect is judged to be in a stable stage or a temporarily non-developing stage. In this case, the defect identification results, confidence levels, and structural state indicators for each time period are organized and recorded to form a complete detection information sequence. The defect status and related characteristic data of the stable stage are integrated and output as a detection report, which includes information such as defect type, current state description, confidence level interval, and a summary of historical change trends.

[0052] In this embodiment, a non-destructive testing system for insulators based on voiceprint recognition is provided, for performing the non-destructive testing method for insulators based on voiceprint recognition as described above, including: The acquisition unit is used to excite the insulator by vibration and collect vibration response signals to construct a feature sample set. The sample unit is used to perform sliding window decomposition on the feature sample set to extract the abnormal feature sample set; The analysis unit is used to perform principal component dimensionality reduction analysis on the abnormal feature sample set and to label potential defect features. The defect identification unit is used to construct an acoustic fingerprint signal database; input the potential defect features into the acoustic fingerprint signal database for similarity analysis, and output a defect identification report; The trend analysis unit is used to analyze defect development based on the defect identification report and output a detection report.

[0053] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0054] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement it. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein are implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A non-destructive testing method for insulators based on voiceprint recognition, characterized in that, Includes the following steps: Step S1: Vibrate the insulator and collect the vibration response signal to construct a feature sample set; Step S2: Perform sliding window decomposition on the feature sample set to extract the abnormal feature sample set; Step S3: Perform principal component dimensionality reduction analysis on the abnormal feature sample set and label potential defect features; Step S4: Construct an acoustic fingerprint signal database; input the potential defect features into the acoustic fingerprint signal database for similarity analysis, and output a defect identification report; Step S5: Analyze the defect development based on the defect identification report and output a detection report.

2. The non-destructive testing method for insulators based on voiceprint recognition according to claim 1, characterized in that, The specific steps of step S1 are as follows: Design an acoustic excitation source; based on the acoustic excitation source, vibrate the insulator, and use acoustic sensors to collect data synchronously at multiple points to extract random vibration response signals and near-field acoustic radiation signals; Power spectral density analysis was performed on the random vibration response signal to extract the modal frequency band energy distribution, modal coupling strength and frequency stability, and the vibration response feature set was obtained by fitting. Time-domain feature analysis based on near-field acoustic radiation signals is performed to extract multi-dimensional voiceprint representation vectors. A joint modeling of the vibration response feature set and the multidimensional acoustic signature vector is performed to construct a feature sample set.

3. The non-destructive testing method for insulators based on voiceprint recognition according to claim 2, characterized in that, The acoustic excitation source is a wideband random excitation signal covering the frequency band of 100 Hz to 25 kHz.

4. The non-destructive testing method for insulators based on voiceprint recognition according to claim 1, characterized in that, The specific steps of step S2 are as follows: The feature sample set is decomposed by a sliding time window to obtain sample sets for multiple time periods; The sample set is subjected to voiceprint feature space mapping to generate voiceprint trajectories for different time periods; Calculate the rate of change and amplitude of change of the voiceprint trajectory; Based on the rate and magnitude of change, stable trajectories are identified, and abnormal abrupt change locations are identified, marking the abrupt change points of abnormal trajectories. Based on the abnormal trajectory mutation points, the source is located and traced, and an abnormal feature sample set is extracted.

5. The non-destructive testing method for insulators based on voiceprint recognition according to claim 1, characterized in that, Step S3 is as follows: Principal component dimensionality reduction analysis is performed on the abnormal feature sample set to extract a low-dimensional feature subset; Based on the aforementioned low-dimensional feature subset, a correlation analysis of key defects is performed to label potential defect features.

6. The non-destructive testing method for insulators based on voiceprint recognition according to claim 1, characterized in that, The specific steps of step S4 are as follows: Construct an acoustic fingerprint signal library; the acoustic fingerprint signal library includes standard acoustic fingerprint templates for normal and typical defective states and their feature statistical scores; The potential defect features are input into an acoustic fingerprint signal database for similarity calculation to obtain a similarity score; Set a similarity judgment threshold; determine the existence of similarity scores based on the similarity judgment threshold. If the similarity score is not less than the similarity judgment threshold, it is determined that a matching defect type exists, and a defect existence judgment result is generated. If the similarity score is less than the similarity judgment threshold, it is judged to be without defects; Based on the defect existence determination results, voiceprint deviation analysis and confidence calculation are performed, and a defect identification report is output.

7. The non-destructive testing method for insulators based on voiceprint recognition according to claim 1, characterized in that, The specific steps for performing voiceprint deviation analysis and confidence calculation based on the defect existence determination results, and outputting a defect identification report are as follows: Based on the defect existence determination result, the potential defect features are subjected to multiple standard defect voiceprint deviation calculations to obtain the deviation value of each standard defect voiceprint. Confidence levels are calculated based on the deviation values, and confidence levels for various defect types are output. Based on the stated confidence level, defect identification analysis is performed, and a defect identification report is output.

8. The non-destructive testing method for insulators based on voiceprint recognition according to claim 1, characterized in that, The specific steps of step S5 are as follows: Continuous identification and judgment are performed based on the defect identification report to obtain defect identification results for multiple time periods; Based on the defect identification results, insulator structure state analysis is performed, and time-series structure state data is extracted. Based on the time-series structural state data, trend change calculations are performed to identify the defect structure change curve; The development direction of the defect structure change curve is analyzed. When the defect development direction is detected to be time-series growth, it is determined that the defect is in the expansion stage and an early warning signal is output. When a defect is detected to be developing in a continuous direction, the information is recorded and a detection report is output.

9. A non-destructive testing system for insulators based on voiceprint recognition, characterized in that, The method for performing non-destructive testing of insulators based on voiceprint recognition as described in claim 1 includes: The acquisition unit is used to excite the insulator by vibration and collect vibration response signals to construct a feature sample set. The sample unit is used to perform sliding window decomposition on the feature sample set to extract the abnormal feature sample set; The analysis unit is used to perform principal component dimensionality reduction analysis on the abnormal feature sample set and to label potential defect features. The defect identification unit is used to construct an acoustic fingerprint signal database; input the potential defect features into the acoustic fingerprint signal database for similarity analysis, and output a defect identification report; The trend analysis unit is used to analyze defect development based on the defect identification report and output a detection report.

Images