Method and system for detecting internal defects of hardware fitting based on ultrasonic waves

Abstract

The invention relates to the technical field of defect detection, in particular to a hardware fitting internal defect detection method and system based on ultrasonic waves. The method comprises the following steps: acquiring ultrasonic waveform data of a to-be-analyzed hardware fitting; after an initial interface return band in the ultrasonic waveform data is extracted, a contact state feature set of the to-be-analyzed hardware fitting is determined; based on the contact state feature set and historical detection data, determining dynamic health reference waveform data matched with the features of the to-be-analyzed hardware fitting; performing data processing on the dynamic health reference waveform data and the ultrasonic waveform data to determine residual signal data; and based on the residual signal data and a preset defect judgment rule, identifying a detection result containing the type and severity of the real defect. The objective of the invention is to solve the problem in the prior art that the identification accuracy of the real defects of the hardware fitting is low, so that the false alarm rate of the defects of the hardware fitting is increased, and the judgment of the health condition of the power transmission network hardware fitting is affected.

Description

Technical Field

[0001] This invention relates to the field of defect detection technology, and specifically to a method and system for detecting internal defects in hardware based on ultrasound. Background Technology

[0002] As critical load-bearing components in power transmission and distribution networks, fittings may develop fatigue cracks or corrosion defects due to long-term exposure to multiple stresses. To ensure stable line operation, maintenance departments regularly use ultrasonic non-destructive testing (NDT) technology to perform internal flaw detection on fittings. However, in actual field testing, the acoustic contact between the ultrasonic probe and the fitting surface has a decisive impact on data quality. Due to complex field environments, such as rough, uneven, and corroded fitting surfaces, or limited operating conditions, such as high-altitude operations and extreme weather, the contact between the ultrasonic probe and the fitting is often unstable or insufficient. Poor contact can lead to severe attenuation and distortion of the ultrasonic signal when it enters the fitting, and even generate a large amount of abnormal noise signals. These abnormal signals may manifest as additional peaks or baseline drift in the waveform, showing some similarity to the echo characteristics of actual defects.

[0003] Existing cloud platform data analysis methods struggle to distinguish between signal changes stemming from genuine defects within the fittings and acoustic artifacts caused by poor contact when receiving abnormal waveforms resulting from poor contact. This leads to a decrease in the accuracy of identifying genuine defects and a significant increase in the false alarm rate for healthy fittings. Furthermore, differences in surface characteristics and acoustic impedance among different fitting models can cause signal distortion patterns caused by contact problems to overlap with genuine defect characteristics, exacerbating the risk of misjudgment and consequently affecting the assessment of the health status of power transmission network fittings. Summary of the Invention

[0004] The purpose of this invention is to provide an ultrasonic-based method and system for detecting internal defects in hardware fittings, which addresses the problem that existing technologies have low accuracy in identifying actual defects in hardware fittings, leading to an increased false alarm rate and affecting the assessment of the health status of hardware fittings in power transmission networks.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a method for detecting internal defects in hardware based on ultrasound, comprising: Obtain the ultrasonic waveform data of the hardware to be analyzed; After extracting the initial interface echo segment from the ultrasonic waveform data, the set of contact state features of the hardware to be analyzed is determined. Based on the contact state feature set and historical detection data, dynamic health reference waveform data matching the characteristics of the hardware to be analyzed is determined; The dynamic health reference waveform data and the ultrasonic waveform data are processed to determine the residual signal data. Based on the residual signal data and the preset defect judgment rules, the detection results are identified, which contain the type and severity of the actual defects.

[0006] Furthermore, the step of determining the contact state feature set of the hardware to be analyzed after extracting the initial interface echo segment from the ultrasonic waveform data includes: Based on the ultrasonic waveform data, the model and service life of the hardware to be analyzed are obtained; Extract the initial interface echo band of the standard health reference waveform of the hardware model in a brand new state; Based on the service duration and a preset acoustic evolution trajectory model, a set of health characteristics is determined. The initial interface echo segment of the standard health reference waveform is modified and reconstructed using the set of health features to obtain the set of contact state features of the hardware to be analyzed.

[0007] Further, the step of processing the dynamic health reference waveform data and the ultrasonic waveform data to determine the residual signal data includes: After aligning the dynamic health reference waveform data with the ultrasonic waveform data in time, the time-aligned dynamic health reference waveform data and ultrasonic waveform data are subtracted point by point to obtain the initial residual waveform. The initial residual waveform is decomposed into residual signals of different frequency sub-bands by multi-scale wavelet decomposition. Adaptive threshold filtering is performed on the low-frequency subband residual signal in the residual signals of the different frequency subbands, and nonlinear gain adjustment is performed on the high-frequency subband residual signal in the residual signals of the different frequency subbands to obtain residual signal data.

[0008] Further, the steps of performing adaptive threshold filtering on the low-frequency subband residual signals in the residual signals of the different frequency subbands, and performing nonlinear gain adjustment on the high-frequency subband residual signals in the residual signals of the different frequency subbands to obtain residual signal data include: Based on each frequency sub-band, after determining the local energy distribution, instantaneous frequency change rate, and signal sparsity index of each frequency sub-band, the local clustering region of suspected weak defect signals in each frequency sub-band, as well as the distribution pattern of residual noise or non-defect artifacts, are identified. The residual signal within the local clustering region is amplified by local gain, and the residual noise or non-defect artifacts outside the local clustering region are suppressed by adaptive threshold, resulting in the suppressed frequency sub-band residual signal. Based on the distribution pattern, the residual signal of each frequency sub-band is nonlinearly gain adjusted to obtain the adjusted frequency sub-band residual signal. The suppressed frequency subband residual signal and the adjusted frequency subband residual signal are reconstructed to obtain the residual signal data.

[0009] Furthermore, based on the residual signal data and preset defect judgment rules, the step of identifying the detection results containing the type and severity of the actual defects includes: Extract potential defect echoes from residual signal data; The amplitude, duration, energy, and arrival time of the potential defect echoes are analyzed to identify the defect signals of the hardware to be analyzed. Based on the defect signals of the hardware to be analyzed and the preset defect judgment rules, the detection results that include the type and severity of the actual defects are identified.

[0010] Furthermore, the step of analyzing the amplitude, duration, energy, and arrival time of the potential defect echo to identify the defect signal of the hardware to be analyzed includes: The potential defect echo is subjected to resolution time-frequency decomposition to obtain the time-frequency distribution data of the potential defect echo; Separate each independent time-frequency component of the overlapping echo in the time-frequency distribution data; Feature extraction is performed on each of the independent time-frequency components to determine the center frequency, bandwidth, time delay, and energy concentration of each independent time-frequency component; Defects are identified by using the center frequency, bandwidth, time delay, and energy concentration of each independent time-frequency component, thus obtaining the defect signal of the hardware to be analyzed.

[0011] Furthermore, the step of separating each independent time-frequency component of the overlapping echo in the time-frequency distribution data includes: After performing local enhancement processing on the time-frequency distribution data, the enhanced time-frequency distribution data is segmented to obtain each preliminary separated component; After determining the time-frequency center, energy moment, and sparsity index of each of the preliminary separation components, it is determined that there is a fusion region between two adjacent preliminary separation components where the component frequency is within the frequency threshold and the energy distribution is continuous. Each fusion region is subjected to iterative signal decomposition and separation processing to obtain each independent time-frequency component.

[0012] Further, based on the distribution pattern, the step of performing nonlinear gain adjustment on the residual signal of each frequency sub-band to obtain the adjusted frequency sub-band residual signal includes: By utilizing the characteristics of the distribution pattern, after determining the instantaneous bandwidth, energy distribution continuity, and signal symmetry indices of each frequency sub-band, the characteristics of each broadband defect signal across the frequency sub-band are identified. The low-frequency subband and high-frequency subband of each broadband defect signal feature are fused to obtain each broadband processed subband. Adaptive threshold filtering of energy gradient and local signal-to-noise ratio and nonlinear gain adjustment are performed on each of the broadband processing subbands to obtain the adjusted frequency subband residual signal.

[0013] Further, the step of performing adaptive threshold filtering and nonlinear gain adjustment on each of the broadband processing sub-bands based on energy gradient and local signal-to-noise ratio to obtain the adjusted frequency sub-band residual signal includes: Multi-scale edge detection is performed on each of the broadband processing sub-bands to obtain edge intensity data for each broadband processing sub-band; Potential defect signal boundary regions and noise artifact boundary regions are marked in each edge intensity data; Based on the boundary regions of defective signals and noise artifacts, the overall energy distribution and instantaneous frequency variation trend of each broadband processing sub-band are determined. By utilizing the overall energy distribution and instantaneous frequency variation trend of each broadband processing sub-band, nonlinear gain adjustment is performed on each broadband processing sub-band to obtain the adjusted frequency sub-band residual signal.

[0014] The present invention also provides an ultrasonic-based system for detecting internal defects in hardware, the system comprising: The acquisition module is used to acquire the ultrasonic waveform data of the hardware to be analyzed; The set determination module is used to extract the initial interface echo segment from the ultrasonic waveform data and then determine the contact state of the hardware to be analyzed. The waveform determination module is used to determine dynamic health reference waveform data that matches the characteristics of the hardware to be analyzed, based on the contact state feature set and historical detection data. The signal determination module is used to process the dynamic health reference waveform data and the ultrasonic waveform data to determine the residual signal data. The identification module is used to identify the detection results containing the type and severity of real defects based on the residual signal data and preset defect judgment rules.

[0015] Compared with the prior art, the ultrasonic-based method and system for detecting internal defects in hardware of the present invention has the following advantages: This invention discloses an ultrasonic-based method and system for detecting internal defects in hardware fittings. By acquiring ultrasonic waveform data of the hardware to be analyzed and extracting the initial interface echo segment, the system determines the set of contact state characteristics of the hardware, effectively evaluating the acoustic contact quality between the probe and the hardware surface. Based on the contact state characteristic set and historical detection data, dynamic health reference waveform data matching the characteristics of the hardware to be analyzed is determined. This dynamic reference waveform fully considers the actual service condition and acoustic evolution of the hardware, thus more accurately reflecting its health baseline. Subsequently, the dynamic health reference waveform data and ultrasonic waveform data are processed to determine residual signal data. The residual signal effectively filters out signal distortion and noise caused by poor contact conditions, highlighting the echo characteristics of true defects. Finally, based on the residual signal data and preset defect judgment rules, the detection results containing the type and severity of true defects are identified. By introducing dynamic health reference waveforms and refined residual signal processing, it is possible to effectively distinguish false defect signals caused by poor contact from real defect signals inside the fittings. At the same time, it can adapt to the differences in surface characteristics and acoustic impedance of different fitting models, overcome the risk of misjudgment caused by the differences in fitting models, and thus avoid misjudging low-quality data as new defects, significantly improving the ability to identify real defects and reducing the false alarm rate. Attached Figure Description

[0016] To more clearly illustrate the specific embodiments of the present invention, the accompanying drawings used in the specific embodiments will be briefly described below. In all the drawings, the elements or parts are not necessarily drawn to scale.

[0017] Figure 1 This is an exploded view of an ultrasonic-based method for detecting internal defects in hardware according to the present invention.

[0018] Figure 2 This is a structural block diagram of an ultrasonic-based internal defect detection system for hardware according to the present invention.

[0019] In the diagram: 210, Acquisition module; 220, Set determination module; 230, Waveform determination module; 240, Signal determination module; 250, Identification module.

[0020] The implementation and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] The following drawings disclose several embodiments of the present invention. For clarity, many practical details will be described in the following description. However, it should be understood that these practical details are not intended to limit the invention. That is, in some embodiments of the invention, these practical details are not essential. Furthermore, for the sake of simplicity, some conventional structures and components will be shown in the drawings in a simple schematic manner.

[0022] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0023] Furthermore, in this invention, the use of terms such as "first" and "second" is for descriptive purposes only and does not specifically refer to any order or sequence, nor is it intended to limit the invention. They are merely used to distinguish components or operations described using the same technical terms, and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of various embodiments can be combined with each other, but only if they are feasible for those skilled in the art. If a combination of technical solutions is contradictory or impossible to implement, such a combination should be considered nonexistent and not within the scope of protection claimed by this invention.

[0024] Existing methods for detecting internal defects in power transmission fittings often suffer from poor acoustic contact between the ultrasonic probe and the fitting surface in complex field environments. This leads to severe attenuation and distortion of the ultrasonic signal as it enters the fitting, generating a large amount of noise and creating abnormal signals. These abnormal signals may manifest as additional peaks or baseline drift in the waveform, exhibiting similarities to the echo characteristics of genuine defects. Current technologies struggle to distinguish whether signal changes originate from actual internal defects or are acoustic artifacts caused by poor probe-fitting surfaces, leading to false alarms. This negatively impacts the comprehensive assessment of the overall health status of power transmission network fittings and subsequent maintenance.

[0025] To further understand the content, features, and effects of this invention, the following embodiments are provided, and detailed descriptions are given below in conjunction with the accompanying drawings: Please see Figure 1 This invention provides a method for detecting internal defects in hardware based on ultrasound, comprising the following steps: S100. Acquire the ultrasonic waveform data of the fitting to be analyzed. The ultrasonic waveform data is a record of the amplitude of the ultrasonic signal received during the inspection of the fitting using an ultrasonic probe, showing how this amplitude changes over time. It contains information about the internal structure and defects of the fitting. Specifically, this can be achieved by placing the ultrasonic probe on the surface of the fitting and emitting ultrasonic pulses into the fitting. After receiving the ultrasonic signal reflected from inside the fitting, the probe converts it into an electrical signal and performs digital processing to obtain the ultrasonic waveform data. For example, an existing ultrasonic flaw detector can be used, with its built-in sensors and data acquisition unit converting the acquired analog ultrasonic signals into digital waveform data, which can then be stored in local memory or transmitted to a remote server.

[0026] S200. After extracting the initial interface echo segment from the ultrasonic waveform data, the contact state feature set of the hardware to be analyzed is determined. The initial interface echo segment is the first reflected wave generated when the ultrasonic wave enters the surface of the hardware from the probe, and its characteristics are closely related to the contact state between the probe and the hardware. The contact state feature set is a set of parameters describing the acoustic coupling quality between the probe and the hardware surface, such as contact pressure and the uniformity of coupling agent distribution. Specifically, a time window can be set, and the first significant energy peak in the ultrasonic waveform data within this time window can be identified as the initial interface echo segment. Then, parameters such as amplitude attenuation, waveform distortion, or energy distribution of this echo segment can be analyzed to quantify the contact state. For example, when the contact is poor, the amplitude of the initial interface echo may be significantly reduced, and the waveform may broaden or split. These parameters constitute the contact state feature set of the hardware to be analyzed.

[0027] S300. Based on the contact state feature set and historical detection data, determine dynamic health reference waveform data that matches the characteristics of the hardware to be analyzed. The dynamic health reference waveform data is obtained through model prediction or historical data matching based on factors such as the model, service duration, and environmental conditions of the hardware to be analyzed, representing the ultrasonic waveform that should exist under ideal health conditions. Specifically, by introducing dynamic health reference waveform data, for example, a database can be established to store historical health ultrasonic waveform data of different models of hardware under different service durations and contact states. After obtaining the contact state feature set of the hardware to be analyzed, the historical health waveform data that most closely matches the current hardware model, service duration, and contact state can be retrieved from this database and used as the dynamic health reference waveform data. Simultaneously, a machine learning model can also be used to predict the ultrasonic waveform that the hardware to be analyzed should have under healthy conditions under the current contact state by training on historical data.

[0028] S400. The dynamic health reference waveform data and the ultrasonic waveform data are processed to determine the residual signal data. The residual signal data is the difference signal obtained by comparing the actual ultrasonic waveform data of the hardware to be analyzed with the dynamic health reference waveform data. This difference signal often contains defect information. In this embodiment, the dynamic health reference waveform data and the actual ultrasonic waveform data can be time-aligned and then subtracted point by point to obtain the initial residual waveform. If there are no defects inside the hardware and the contact state is accurately compensated, the residual waveform should be close to zero. Any significant non-zero residual signal may indicate the presence of defects inside the hardware. This eliminates signal changes caused by differences in contact state, thereby highlighting signal anomalies caused by actual defects inside the hardware.

[0029] S500. Based on the residual signal data and preset defect judgment rules, identify the detection results containing the type and severity of real defects. The preset defect judgment rules are logic or models established based on a large number of defect samples and expert experience, used to identify defect features in the residual signal. Specifically, after obtaining the residual signal data, it needs to be further analyzed to identify whether defect signals exist and to determine the type (e.g., cracks, corrosion, and voids) and severity (e.g., defect size and depth) of the defects. For example, defect judgment rules can be preset, based on parameters such as the amplitude, duration, energy, frequency characteristics, and relative position to the time axis of the residual signal. When certain characteristics of the residual signal meet the preset rules, a defect can be determined to exist. For example, a high-amplitude, short-duration residual pulse may indicate a tiny crack, while a low-amplitude, long-duration residual signal may indicate a corroded area. By matching the residual signal data with these rules, the type and severity of the real defects can be identified, and the corresponding detection results can be output.

[0030] In this embodiment, a dynamic reference system adaptable to different contact states and hardware characteristics is constructed to effectively distinguish signal changes caused by external interference (such as poor contact) from signal anomalies caused by actual defects within the hardware. Specifically, ultrasonic waveform data of the hardware to be analyzed is acquired to provide raw input for subsequent analysis. Then, by extracting the initial interface echo band and determining the contact state feature set, the acoustic coupling quality between the probe and the hardware surface is quantified. This solves the problem of misjudgment caused by poor contact state. Subsequently, based on the determined contact state feature set and a large amount of historical detection data, a dynamic health reference waveform data highly matched to the characteristics of the hardware currently being analyzed can be generated. The dynamic reference waveform is no longer static but is adjusted according to actual working conditions, thus more accurately reflecting the ultrasonic response of healthy hardware under current conditions. Then, the actually acquired ultrasonic waveform data and the dynamic health reference waveform data undergo fine data processing, such as time alignment and point-by-point subtraction, to obtain residual signal data. This residual signal data eliminates interference caused by differences in contact state and individual hardware differences to the greatest extent, making the abnormal information contained therein more likely to originate from actual defects within the hardware. Finally, by comparing the residual signal data with the preset defect judgment rules, the detection results containing the type and severity of actual defects can be identified. This improves the accuracy and reliability of the detection, ensuring accurate judgment of internal defects in hardware.

[0031] Specifically, by determining the set of contact state characteristics of the hardware to be analyzed, the acoustic coupling quality between the probe and the hardware surface can be effectively quantified. Based on this, dynamic health reference waveform data matching the characteristics of the hardware to be analyzed is determined. The health reference waveform can adaptively adjust according to the actual working conditions of the hardware to be analyzed (including contact state, hardware model, and service life). By processing the actual ultrasonic waveform data with this highly matched dynamic health reference waveform data, the resulting residual signal data can maximally eliminate signal interference caused by non-defect factors (such as poor contact), thereby more accurately reflecting the true defect information inside the hardware. Finally, based on the residual signal data and preset defect judgment rules, the detection results containing the type and severity of true defects are identified, significantly improving the accuracy and reliability of defect identification. Therefore, this application not only solves the problem of misjudgment caused by contact issues, but also provides a more accurate solution for the detection of internal defects in hardware through dynamic reference and residual analysis, thereby greatly improving detection efficiency and diagnostic accuracy.

[0032] In some embodiments of this application described above, the step of determining the contact state feature set of the hardware to be analyzed after extracting the initial interface echo segment from the ultrasonic waveform data includes: Based on the ultrasonic waveform data, the model number and service life of the hardware to be analyzed are obtained. This is achieved by consulting hardware records, nameplate information, or through automatic sensor identification, to determine the specific model of the hardware currently under test and the length of time it has been in use. The hardware model number is crucial for determining its initial acoustic characteristics, while the service life reflects the cumulative effects of material aging, corrosion, or stress changes that may occur during actual operation.

[0033] Extract the initial interface echo segment of the standard health reference waveform for the hardware model in its brand-new state. Specifically, extracting the initial interface echo segment of the standard health reference waveform for the hardware model in its brand-new state means retrieving the ultrasonic test waveform of the hardware model from a pre-established database or through experimental measurement, when the hardware model is undamaged and in an ideal health state. The initial interface echo segment of the standard health reference waveform represents the ideal reflection characteristics of ultrasound at the interface between the hardware and the probe, serving as a benchmark for subsequent evaluation of the contact state.

[0034] Based on the service duration and a preset acoustic evolution trajectory model, a set of health characteristics is determined. Specifically, determining the set of health characteristics based on the service duration and a preset acoustic evolution trajectory model refers to using a pre-established mathematical or empirical model that can describe the changes in the acoustic properties of the fitting over service time. This model considers the influence of factors such as material fatigue and microstructural changes on the ultrasonic wave propagation characteristics. By inputting the service duration of the fitting to be analyzed into this model, the expected changes in its acoustic properties relative to a completely new state under the current service condition can be predicted, thereby obtaining characteristic parameters reflecting the current health state, i.e., the set of health characteristics.

[0035] The initial interface echo segment of the standard health reference waveform is modified and reconstructed using the aforementioned set of health features to obtain the contact state feature set of the hardware to be analyzed. Specifically, this modification and reconstruction of the initial interface echo segment of the standard health reference waveform using the set of health features involves applying the aforementioned determined set of health features to the initial interface echo segment of the standard health reference waveform. The expected acoustic response of the hardware under the current service duration is simulated. Through modification and reconstruction, an initial interface echo segment that more closely reflects the actual health state of the hardware to be analyzed is obtained. This echo segment is considered the contact state feature set of the hardware to be analyzed and is used for subsequent comparison with the actual detection waveform.

[0036] This invention, by acquiring the fitting model and service duration, and combining a standard health reference waveform and an acoustic evolution trajectory model, can more accurately predict the ideal initial interface echo characteristics of the fitting under its current service state. Specifically, the fitting model provides the acoustic baseline of its original design, while the service duration, through the acoustic evolution trajectory model, quantifies the impact of potential material performance degradation on the ultrasonic signal during long-term operation. Therefore, by adaptively modifying and reconstructing the standard health reference waveform under the new state, a set of contact state characteristics that fully considers the fitting's own characteristics and service history can be obtained. This avoids the errors that may arise from directly using a single static health reference waveform, allowing subsequent analysis of actual ultrasonic waveform data to more accurately focus on anomalies caused by defects, rather than normal fluctuations caused by changes in the fitting's own state. This achieves a refined determination of the fitting contact state characteristic set. Compared to simply extracting the initial interface echo band, this solution, by introducing the fitting model, service duration, and acoustic evolution trajectory model, can more accurately reflect the true acoustic characteristics of the fitting under analysis under current service conditions. It helps to eliminate non-defect signal fluctuations caused by individual differences in hardware and service aging, thereby improving the accuracy and reliability of subsequent defect detection, reducing false alarm rate and false negative rate, and making defect identification results more targeted and reliable.

[0037] In some embodiments of this application described above, the step of processing the dynamic health reference waveform data and the ultrasonic waveform data to determine the residual signal data includes: After aligning the dynamic health reference waveform data and the ultrasonic waveform data in time, the time-aligned dynamic health reference waveform data and the ultrasonic waveform data are subtracted point-by-point to obtain the initial residual waveform. Specifically, time alignment refers to adjusting the correspondence between the dynamic health reference waveform data and the ultrasonic waveform data on the time axis using signal processing techniques, such as cross-correlation algorithms or dynamic time warping (DTW) algorithms, to eliminate time deviations caused by minor changes in the propagation path or inconsistent acquisition starting points. Its purpose is to ensure that subsequent point-by-point subtraction operations are compared at the same physical time points of the signals, thereby improving the accuracy of the residual signal. The point-by-point subtraction operation refers to subtracting the value of each sampling point in the dynamic health reference waveform data at the corresponding time point from the value of each sampling point in the ultrasonic waveform data after time alignment, thus obtaining the initial residual waveform. The initial residual waveform reflects the difference between the analyzed hardware and its health status, and may contain defect information and noise.

[0038] The initial residual waveform is decomposed using multi-scale wavelet decomposition to obtain residual signals in different frequency sub-bands. Multi-scale wavelet decomposition refers to using wavelet transform to decompose the initial residual waveform into signal components with multiple different frequency ranges (i.e., frequency sub-bands). For example, Daubechies wavelet or Symlet wavelet can be used for three-level or four-level decomposition to obtain low-frequency sub-band residual signals and multiple high-frequency sub-band residual signals. The purpose is to separate the defect features from noise in the residual signal, because defect signals usually exhibit significant features within a specific frequency range, while noise may be distributed over a wider frequency range.

[0039] Adaptive threshold filtering is applied to the low-frequency subband residual signals in the residual signals of different frequency subbands, and nonlinear gain adjustment is performed on the high-frequency subband residual signals in the residual signals of different frequency subbands to obtain residual signal data. Adaptive threshold filtering refers to dynamically adjusting the filtering threshold for the low-frequency subband residual signals based on their local statistical characteristics (such as mean, variance, or energy) to effectively suppress low-frequency noise components while preserving potential low-frequency defect characteristics. For example, an adaptive threshold algorithm based on local signal-to-noise ratio or local energy can be used. Nonlinear gain adjustment refers to nonlinear amplification or attenuation processing of the high-frequency subband residual signals based on their signal strength, sparsity, or local energy distribution. Its purpose is to enhance weak high-frequency defect signals while suppressing high-frequency noise or non-defect artifacts, thereby improving the detectability of high-frequency defect signals.

[0040] Specifically, the ultrasonic waveform data of the hardware to be analyzed is acquired, and dynamic health reference waveform data matching the hardware model and service duration is obtained from historical testing data. The acquired ultrasonic waveform data and the dynamic health reference waveform data are then time-aligned. For example, the optimal time offset can be determined using a method based on the peak value of the cross-correlation function, and then one waveform is time-shifted to align it with the other waveform. After time alignment, the two waveforms are subtracted point by point to obtain the initial residual waveform. For example, the formula for calculating the initial residual waveform is: ; in, Let be the ultrasonic waveform data at point t. The dynamic health reference waveform data at point t. The initial residual waveform at point t is given.

[0041] Subsequently, the initial residual waveform undergoes multi-scale wavelet decomposition. For example, the db4 wavelet basis can be used for three-level decomposition, resulting in a low-frequency approximation component and three high-frequency detail components (D1, D2, D3), which constitute residual signals of different frequency sub-bands. Adaptive threshold filtering is then applied to the low-frequency approximation component. Specifically, the standard deviation of the low-frequency approximation component within a local window is determined, and a threshold is dynamically set based on this standard deviation, setting signal components below this threshold to zero to suppress low-frequency noise. Nonlinear gain adjustment is then applied to the three high-frequency detail components. For example, a sigmoid function or a logarithmic function can be used to adjust the gain of the high-frequency components, amplifying signals with smaller amplitudes while maintaining or slightly suppressing signals with larger amplitudes to enhance the visibility of weak defect signals. Finally, the low-frequency approximation component after adaptive threshold filtering and the high-frequency detail components after nonlinear gain adjustment are reconstructed to obtain the final residual signal data, which will be used for subsequent defect assessment.

[0042] This invention effectively solves the problems of noise interference and non-defect information confusion in traditional direct data processing by introducing steps such as time alignment, multi-scale wavelet decomposition, adaptive threshold filtering, and nonlinear gain adjustment. First, time alignment ensures the accuracy of comparisons and avoids spurious residuals caused by time deviations. Second, multi-scale wavelet decomposition decomposes the residual signal into different frequency sub-bands, enabling effective separation of defect signals and noise. Specifically, low-frequency sub-bands typically contain structural differences or large defect information, while high-frequency sub-bands may contain minor defects or high-frequency noise. Adaptive threshold filtering of the low-frequency sub-band residual signal effectively removes low-frequency noise while retaining true low-frequency defect characteristics. Simultaneously, nonlinear gain adjustment of the high-frequency sub-band residual signal enhances weak high-frequency defect signals, making them more prominent against a noisy background, thereby improving the detection capability of minor defects. Therefore, the solution proposed in this application can more accurately extract residual signal data related to internal defects in hardware, providing a cleaner and more reliable input for subsequent defect identification. It can generate cleaner and more representative residual signal data, thus providing a more reliable basis for subsequent identification of defect types and severity, and reducing the risk of false alarms and false negatives.

[0043] In some embodiments of this application described above, the steps of performing adaptive threshold filtering on the low-frequency subband residual signal in the residual signals of different frequency subbands, and performing nonlinear gain adjustment on the high-frequency subband residual signal in the residual signals of different frequency subbands to obtain residual signal data include: Based on each of the different frequency sub-bands, after determining the local energy distribution, instantaneous frequency change rate, and signal sparsity index of each frequency sub-band, local clustering regions of suspected weak defect signals and the distribution patterns of residual noise or non-defect artifacts are identified in each frequency sub-band. Specifically, when determining the local energy distribution of each frequency sub-band, methods such as sliding window or wavelet packet analysis can be used to determine the energy intensity of the signal within the local time-frequency region, revealing the degree of energy concentration in different regions. The instantaneous frequency change rate can be obtained using techniques such as Hilbert transform or short-time Fourier transform, used to characterize the signal frequency variation characteristics over time, which is crucial for identifying defect signals with specific frequency modulation characteristics. The signal sparsity index can be evaluated using L1 norm or basis pursuit algorithms to quantify the compactness of the signal in a specific transform domain; signals with high sparsity are often more likely to be transient defect echoes. This allows for more accurate identification of local clustering regions of suspected weak defect signals, which typically exhibit high local energy, specific frequency change patterns, and strong sparsity. Meanwhile, the distribution pattern of residual noise or non-defect artifacts may exhibit characteristics such as uniform energy distribution, random frequency variation, or low sparsity.

[0044] The residual signal within the local clustering region is locally amplified by gain, while residual noise or non-defect artifacts outside the local clustering region are adaptively suppressed by thresholding, resulting in a suppressed frequency sub-band residual signal. Local gain amplification of the residual signal within the local clustering region enhances the weak signal initially identified as a potential defect, making it easier to detect and analyze in subsequent processing. For example, a greater gain can be applied to regions with higher energy based on the intensity of the local energy distribution. Simultaneously, adaptive thresholding of residual noise or non-defect artifacts outside the local clustering region effectively removes background noise and interference, preventing them from confusing defect identification. The adaptive threshold can be dynamically determined based on the statistical characteristics of the noise (such as mean and variance) or through machine learning methods.

[0045] Based on the distribution pattern, nonlinear gain adjustment is performed on the residual signal of each frequency sub-band to obtain the adjusted frequency sub-band residual signal. Specifically, nonlinear gain adjustment of the residual signal of each frequency sub-band based on the distribution pattern can further optimize signal quality. For example, for identified noise or artifact distribution patterns, nonlinear functions (such as the Sigmoid function, Tanh function, or a custom nonlinear mapping function) can be used to compress or attenuate the signal in these regions, while suspected defect signal regions are preserved or moderately enhanced, thereby suppressing noise to the maximum extent without losing defect information.

[0046] The suppressed and adjusted frequency sub-band residual signals are reconstructed to obtain the residual signal data. Specifically, the suppressed and adjusted frequency sub-band residual signals are reconstructed to obtain the final residual signal data. The reconstruction process typically involves recombining the processed frequency sub-band signals through inverse transforms (such as inverse wavelet transforms) to form a residual signal with a higher signal-to-noise ratio and clearer defect characteristics in the time domain.

[0047] Specifically, after ultrasonic testing of the hardware, residual signals from multiple frequency sub-bands were obtained through multi-scale wavelet decomposition. For one frequency sub-band, the local energy distribution of the signal was determined, for example, by obtaining the root mean square value through a sliding window with a width of 50 sampling points. Simultaneously, the instantaneous frequency was obtained using Hilbert transform, and its rate of change was analyzed. Then, the sparsity of the signal was calculated using the L1 norm. Based on these indicators, a certain region on the time axis (e.g., from 10 microseconds to 15 microseconds) was identified as a local clustering region of suspected weak defect signals, exhibiting high local energy, stable instantaneous frequency, and high sparsity. Other regions, however, showed lower energy, random frequency variations, and lower sparsity, and were identified as residual noise distribution patterns. The residual signal within this local clustering region was locally amplified, for example, by increasing its amplitude by a factor of 1.5, to enhance its detectability. Simultaneously, for residual noise outside the local clustering area, an adaptive threshold is set based on its statistical characteristics (such as the root mean square error of the noise). Signal components below this threshold are directly set to zero, thereby achieving noise suppression. Subsequently, based on the identified noise distribution pattern, nonlinear gain adjustment is performed on the residual signal of the entire frequency sub-band. For example, for regions identified as noise, a sigmoid function is used to compress them, making their amplitude approach zero, while for suspected defect regions, their amplitude remains unchanged or is slightly enhanced. Finally, the signal after local gain amplification and adaptive threshold suppression is reconstructed with the signal after nonlinear gain adjustment to obtain residual signal data for that frequency sub-band with a higher signal-to-noise ratio and clearer defect characteristics. By repeating this process on all frequency sub-bands, all processed sub-band signals are finally reconstructed to obtain residual signal data for defect identification.

[0048] In this embodiment, by performing refined feature analysis on the residual signal of each frequency sub-band, including determining local energy distribution, instantaneous frequency change rate, and signal sparsity indices, it is possible to more accurately identify local clustering regions of suspected weak defect signals and distinguish the distribution patterns of residual noise or non-defect artifacts. This allows for local gain amplification of the residual signal within the local clustering region to enhance the potential defect signal, while adaptive threshold suppression of noise outside the region effectively improves the signal-to-noise ratio. Based on the identified distribution pattern, nonlinear gain adjustment is applied to the residual signal to optimize signal quality, ensuring the integrity of the defect signal and suppressing interference to the greatest extent. This step-by-step and refined processing effectively solves the problem of accurately extracting weak defect signals in complex noise environments.

[0049] In some embodiments of this application described above, the step of identifying detection results containing the type and severity of real defects based on the residual signal data and preset defect judgment rules includes: Extracting potential defect echoes from residual signal data. Specifically, after obtaining the residual signal data, this involves using specific signal processing techniques, such as threshold segmentation, peak detection, or pattern recognition algorithms, to identify and separate echo signals that may be caused by internal defects. Echo signals typically exhibit characteristics different from normal signals, such as anomalous amplitude, shape, or time delay. The aim is to initially screen out defect-related signal components from the complex residual signal, laying the foundation for subsequent detailed analysis.

[0050] The amplitude, duration, energy, and arrival time of the potential defect echoes are analyzed to identify the defect signals of the hardware under analysis. This analysis can be understood as quantifying and evaluating multiple dimensions of the extracted potential defect echoes. Amplitude reflects the intensity of the defect echo and may be related to the size or nature of the defect; duration indicates the length of the echo and may be related to the geometry or acoustic characteristics of the defect; energy is the total intensity of the echo signal, providing overall energy information about the defect; and arrival time indicates the depth or location of the defect within the hardware. Through comprehensive analysis of these characteristics, it is possible to more accurately determine whether the potential defect echoes truly originate from actual defects within the hardware and to preliminarily distinguish between different types of defects.

[0051] Based on the defect signals of the hardware to be analyzed and preset defect judgment rules, the detection results containing the types and severity of actual defects are identified. Specifically, this involves comparing the analyzed and identified defect signals of the hardware to be analyzed with the preset defect judgment rules. The preset defect judgment rules can be a series of standards based on experience, physical models, or machine learning training. For example, when the amplitude of a defect signal exceeds a certain threshold, the duration is within a certain range, the energy reaches a certain level, and the arrival time indicates a specific depth, it may be judged as a certain type of defect (such as cracks, voids, and inclusions). The severity (such as minor, moderate, and severe) is further assessed based on the quantified values ​​of these characteristics. The goal is to transform abstract signal characteristics into concrete and actionable defect information, providing a basis for decision-making in the maintenance and repair of hardware.

[0052] In this embodiment, by performing multi-dimensional feature analysis on potential defect echoes in residual signal data and combining it with preset defect judgment rules, real defects can be systematically identified from complex ultrasonic signals. By extracting potential defect echoes, attention is focused on signal regions where defects may exist, effectively reducing the computational burden and noise interference of subsequent analysis. Furthermore, quantitative analysis of key parameters such as echo amplitude, duration, energy, and arrival time allows for the precise capture of the physical characteristics of the defects. Finally, by comparing these quantitative features with preset judgment rules, they are transformed into specific defect types and severity, thereby achieving an accurate and objective assessment of internal defects in hardware.

[0053] In some embodiments of this application described above, the step of analyzing the amplitude, duration, energy, and arrival time of the potential defect echo to identify the defect signal of the hardware to be analyzed includes: The potential defect echo is subjected to resolution time-frequency decomposition to obtain its time-frequency distribution data. This resolution time-frequency decomposition transforms the original one-dimensional time-domain signal into a two-dimensional time-frequency domain representation, using methods such as short-time Fourier transform, wavelet transform, Wigner-Ville distribution, or Hilbert-Huang transform. The goal is to reveal the energy distribution of the signal at different times and frequencies, allowing simultaneous observation of the instantaneous frequency characteristics and temporal evolution of the defect echo, which is crucial for distinguishing different types of defect echoes.

[0054] The process involves separating each independent time-frequency component of the overlapping echoes in the time-frequency distribution data. Specifically, this separation occurs in the time-frequency domain when multiple defect echoes are close to each other in time or frequency, causing their time-frequency representations to overlap. This step utilizes signal processing techniques, such as algorithms based on image segmentation, cluster analysis, or blind source separation, to decompose the overlapping energy regions into their respective independent time-frequency components. This ensures that each identified defect echo can be analyzed independently, avoiding feature confusion caused by overlap.

[0055] Feature extraction is performed on each independent time-frequency component to determine its center frequency, bandwidth, time delay, and energy concentration. Specifically, feature extraction for each independent time-frequency component, determining its center frequency, bandwidth, time delay, and energy concentration, yields key parameters characterizing its physical properties. The center frequency reflects the influence of the defect's size or type on the ultrasonic frequency response; the bandwidth indicates the defect's complexity or scattering characteristics; the time delay is directly related to the defect's depth or location; and the energy concentration quantifies the intensity of the defect echo. The aim is to provide a quantified feature vector for subsequent defect assessment.

[0056] Defects are identified by utilizing the center frequency, bandwidth, delay, and energy concentration of each independent time-frequency component, resulting in the defect signal of the hardware under analysis. Specifically, this defect identification process uses the extracted features as input, combined with pre-defined defect identification rules or machine learning models, to classify or perform regression analysis on each independent time-frequency component. For example, the presence of defects can be determined based on the degree of matching between these features and known defect patterns, and the type (e.g., cracks, voids, and inclusions) and severity of defects can be further identified. The goal is to achieve accurate identification and quantitative assessment of internal defects in hardware.

[0057] In this embodiment, by introducing high-resolution time-frequency decomposition technology, traditional time-domain signal analysis is extended to the time-frequency domain, enabling the simultaneous capture of the time and frequency characteristics of potential defect echoes. This allows even weak or overlapping defect signals to be more clearly distinguished in the time-frequency plot. Subsequently, by separating the independent time-frequency components of overlapping echoes, the problem of mutual interference between multiple defect signals is effectively solved, ensuring the independence and analyzability of each defect echo. Simultaneously, refined features such as center frequency, bandwidth, time delay, and energy concentration are extracted from each independent component. These features can more comprehensively and accurately describe the physical properties of the defect, such as size, shape, location, and material properties. Due to the multi-dimensional and high-resolution features, subsequent defect judgment becomes more accurate and reliable, overcoming the limitations that may arise when relying solely on macroscopic feature analysis.

[0058] In some embodiments of this application described above, the step of separating each independent time-frequency component of the overlapping echo in the time-frequency distribution data includes: After performing local enhancement processing on the time-frequency distribution data, the enhanced time-frequency distribution data is segmented to obtain each preliminary separated component. Specifically, local enhancement processing of the time-frequency distribution data involves applying specific image processing or signal processing algorithms, such as contrast enhancement, edge sharpening, or local energy amplification, to improve the visibility of weak or blurred echo components in the time-frequency distribution data, making them easier to identify and distinguish visually or computationally. This provides clearer input for subsequent segmentation processing, thereby improving the accuracy of the preliminary separated components. Segmentation processing of the enhanced time-frequency distribution data can be understood as dividing the enhanced time-frequency distribution data into multiple independent regions, each region corresponding to a potential preliminary separated component. This can be achieved through methods such as thresholding, connected component analysis, or clustering algorithms, grouping signal energies with similar characteristics in the time-frequency domain together to form discrete preliminary components.

[0059] After determining the time-frequency center, energy moment, and sparsity index of each preliminary separation component, a fusion region is identified between two adjacent preliminary separation components where the component frequency is within a frequency threshold and the energy distribution is continuous. In practical applications, determining the time-frequency center, energy moment, and sparsity index of each preliminary separation component is a quantitative analysis of each component. The time-frequency center characterizes the position of the component on the time-frequency plane; the energy moment reflects the energy distribution characteristics and intensity of the component; and the sparsity index measures the compactness or non-Gaussianity of the component in the time-frequency domain. This provides a quantitative basis for subsequent identification of overlapping regions and fine separation. Simultaneously, identifying a fusion region between two adjacent preliminary separation components where the component frequency is within a frequency threshold and the energy distribution is continuous is achieved by comparing the time-frequency characteristics of adjacent preliminary separation components, identifying regions that are close in frequency and show continuity or overlap in energy distribution. This accurately pinpoints complex overlapping regions requiring deep separation processing, avoiding unnecessary complex calculations for non-overlapping regions.

[0060] Each fused region undergoes iterative signal decomposition to obtain each independent time-frequency component. This iterative decomposition of each fused region can be understood as employing advanced signal processing techniques, such as blind source separation, sparse component analysis, or dictionary-based decomposition methods, within the identified fused region. Through an iterative optimization process, the multiple fused signal components are gradually decoupled until each independent time-frequency component is clearly separated. The goal is to accurately extract each independent defect echo component from complex overlapping signals, providing a clean signal source for subsequent defect feature extraction.

[0061] Specifically, during ultrasonic testing of hardware, two spatially close micro-cracks may exist within the hardware, or the echoes from a crack and a structural boundary may exhibit highly overlapping signals on the time-frequency map. After performing time-frequency decomposition on the potential defect echoes, the resulting time-frequency distribution data will show significant overlap between the two echo components in time and frequency. First, local enhancement processing is applied to this time-frequency distribution data, for example, using two-dimensional Gaussian filtering combined with local contrast stretching, to highlight the energy peaks of each component within the overlapping region. Subsequently, connected component segmentation is performed on the enhanced data, which may initially yield a large, preliminarily separated component containing the two overlapping components. Next, the time-frequency center, energy moment, and sparsity indices of the preliminarily separated component are determined. For example, it may be found that the component has two distinct energy centers, and its sparsity index indicates that it is not a single, compact signal. Based on these indices, a fusion region is identified within this preliminarily separated component, containing two potentially independent components with similar frequencies and continuous energy distribution. Finally, an iterative separation process of signal decomposition is performed on this fusion region. For example, iterative algorithms based on nonnegative matrix factorization or independent component analysis can be used. In each iteration, the algorithm attempts to extract the most significant independent component from the fused signal and subtract it from the original signal. This process is repeated for the remaining signal until all independent components are separated. In this way, even if two defect echoes highly overlap in the time-frequency domain, they can be accurately separated into two independent time-frequency components, providing an accurate basis for subsequent defect feature extraction and judgment. This not only avoids missed or false defects caused by echo overlap but also provides a cleaner and more accurate input for subsequent defect feature extraction and type / severity identification, thus significantly improving the overall performance and reliability of internal defect detection in hardware.

[0062] In some embodiments of this application described above, the step of performing nonlinear gain adjustment on the residual signal of each frequency sub-band based on the distribution pattern to obtain the adjusted frequency sub-band residual signal includes: By utilizing the characteristics of the distribution pattern, after determining the instantaneous bandwidth, energy distribution continuity, and signal symmetry indices of each frequency sub-band, the characteristics of each broadband defect signal spanning the frequency sub-bands can be identified. Specifically, when identifying broadband defect signal characteristics spanning frequency sub-bands using the characteristics of the distribution pattern, the instantaneous bandwidth, energy distribution continuity, and signal symmetry indices of each frequency sub-band can be comprehensively considered. Instantaneous bandwidth reflects the instantaneous expansion of the signal in the frequency domain, energy distribution continuity indicates the smooth transition of signal energy between different frequency sub-bands, and signal symmetry indices help distinguish defect signals from random noise. By comprehensively analyzing these indices, broadband defect signals exhibiting correlation, energy continuity, or specific symmetry patterns across multiple frequency sub-bands can be identified more accurately.

[0063] The low-frequency and high-frequency sub-bands of each broadband defect signal feature are fused to obtain each broadband processed sub-band. This fusion process integrates defect information originally scattered across different frequency sub-bands, forming a more complete representation of the defect signal. Various techniques can be employed for this fusion process, such as simple superposition, weighted averaging, or multi-channel fusion algorithms based on signal correlation. Through fusion, the overall energy of the broadband defect signal can be enhanced, and local noise that lacks correlation in individual sub-bands can be suppressed.

[0064] Adaptive threshold filtering of energy gradient and local signal-to-noise ratio (SNR) and nonlinear gain adjustment are performed on each broadband processing sub-band to obtain the adjusted frequency sub-band residual signal. The purpose of performing adaptive threshold filtering and nonlinear gain adjustment on each broadband processing sub-band is to effectively suppress residual noise that may be introduced during the fusion process while enhancing the defect signal. The energy gradient reflects the intensity of local changes in the signal, helping to highlight defect boundaries; the local SNR assesses the strength of the signal in the current region relative to noise. This enables refined processing of the signal within the broadband processing sub-band, selectively enhancing defect signal regions and effectively suppressing noise regions, thereby obtaining the adjusted frequency sub-band residual signal.

[0065] In this embodiment, by analyzing instantaneous bandwidth, the continuity of energy distribution, and signal symmetry indicators, the overall performance of the defect signal in the frequency domain can be captured more comprehensively, rather than focusing solely on local features. Fusing correlated low-frequency and high-frequency sub-bands concentrates the previously dispersed defect energy, significantly improving the signal-to-noise ratio of the broadband defect signal. Adaptive threshold filtering of energy gradient and local signal-to-noise ratio, along with nonlinear gain adjustment, is applied to the fused broadband sub-band. This allows for refined processing based on the local characteristics and noise level of the defect signal, further enhancing its saliency and effectively suppressing background noise, thereby providing clearer and more reliable residual signal data for subsequent defect identification.

[0066] In some embodiments of this application described above, the step of performing adaptive threshold filtering and nonlinear gain adjustment on each broadband processing sub-band based on energy gradient and local signal-to-noise ratio to obtain the adjusted frequency sub-band residual signal includes: Multi-scale edge detection is performed on each of the broadband processing sub-bands to obtain edge intensity data for each sub-band. Specifically, multi-scale edge detection for each broadband processing sub-band involves applying edge detection operators of different scales (such as the Canny operator, Sobel operator, or LoG operator) to identify abrupt changes or structural boundaries in the signal within the broadband processing sub-band. The purpose of multi-scale detection is to capture defect features of different sizes and intensities, as defect signals may exhibit different edge characteristics at different frequencies and time scales. This results in edge intensity data for each broadband processing sub-band, reflecting the degree of drastic signal changes.

[0067] Potential defect signal boundary regions and noise artifact boundary regions are marked in each edge intensity data set. Marking these regions in each edge intensity data set can be understood as distinguishing between signal boundaries caused by real defects and boundaries caused by noise or non-defect artifacts, based on the characteristics of the edge intensity data and in conjunction with a preset threshold or pattern recognition algorithm. For example, defect signal boundaries typically have high edge intensity and specific spatial continuity, while noise artifact boundaries may appear as randomly distributed low-intensity edges.

[0068] Based on the boundary regions of defective signals and noise artifacts, the overall energy distribution and instantaneous frequency variation trend of each broadband processing sub-band are determined. Specifically, determining the overall energy distribution and instantaneous frequency variation trend of each broadband processing sub-band involves analyzing these marked regions to evaluate the distribution of energy at different frequencies and times within the broadband processing sub-band, as well as the pattern of signal frequency variation over time. For example, defective regions may exhibit concentrated energy and large instantaneous frequency fluctuations, while noise regions may show dispersed energy and irregular instantaneous frequency changes.

[0069] By utilizing the overall energy distribution and instantaneous frequency variation trend of each broadband processing sub-band, nonlinear gain adjustment is performed on each sub-band to obtain the adjusted frequency sub-band residual signal. Specifically, the adjusted frequency sub-band residual signal is obtained by performing nonlinear gain adjustment on each broadband processing sub-band based on the identified defect signal boundaries and noise artifact boundaries, and the energy distribution and frequency variation trend determined accordingly. Differentiated gain adjustments are made to the signals in the broadband processing sub-bands. For example, a larger gain can be applied to the defect signal boundary region to enhance the defect characteristics; for the noise artifact boundary region, a suppressive gain can be applied or filtering can be performed to reduce the noise impact. This allows for more accurate highlighting of the defect signal while effectively suppressing background noise.

[0070] Specifically, when processing a broadband subband, the Canny edge detection operator is used to perform edge detection on the subband at multiple scales, generating edge intensity data. By analyzing this edge intensity data, several high-intensity and continuous edge regions are identified as potential defect signal boundary regions, while some low-intensity and discrete edge regions are identified as noise artifact boundary regions. The energy concentration and instantaneous frequency fluctuation range within the defect signal boundary regions, and the energy dispersion and frequency stability within the noise artifact boundary regions, are then determined. For example, if a defect signal boundary region exhibits high energy concentration and severe frequency fluctuations, a higher gain is applied to that region; if a noise artifact boundary region exhibits low energy dispersion and frequency stability, a lower gain is applied or smoothing is performed. This results in a finely adjusted frequency subband residual signal, where the defect signal is effectively enhanced while noise is significantly suppressed. This application enables more accurate differentiation and processing of defect signals and noise in broadband subbands. Compared to adaptive threshold filtering and nonlinear gain adjustment that only perform energy gradient and local signal-to-noise ratio adjustments, this scheme, through multi-scale edge detection and boundary region identification, can more accurately capture the subtle features of defect signals and perform differentiated processing based on their differences from noise artifacts. This significantly improves the detection sensitivity and signal-to-noise ratio of weak defect signals, reduces false alarm and false negative rates, and thus enhances the accuracy and reliability of internal defect detection in hardware.

[0071] Please see Figure 2 The present invention also provides an ultrasonic-based internal defect detection system for hardware, which includes an acquisition module 210, a set determination module 220, a waveform determination module 230, a signal determination module 240, and an identification module 250.

[0072] The acquisition module 210 is used to acquire the ultrasonic waveform data of the hardware to be analyzed.

[0073] The set determination module 220 is used to extract the initial interface echo segment from the ultrasonic waveform data and then determine the contact state of the hardware to be analyzed.

[0074] The waveform determination module 230 is used to determine dynamic health reference waveform data that matches the characteristics of the hardware to be analyzed based on the contact state feature set and historical detection data.

[0075] The signal determination module 240 is used to process the dynamic health reference waveform data and the ultrasonic waveform data to determine the residual signal data.

[0076] The identification module 250 is used to identify the detection results containing the type and severity of real defects based on the residual signal data and preset defect judgment rules.

[0077] In this embodiment, the acquisition module 210 is responsible for collecting raw ultrasonic data; the set determination module 220 quantifies the contact state between the probe and the fitting by analyzing the initial interface echo band; the waveform determination module 230 generates dynamic health reference waveform data that matches the characteristics of the fitting to be analyzed based on the contact state and historical data; the signal determination module 240 compares the actual data with the reference waveform and extracts the residual signal data; finally, the identification module 250 accurately identifies the actual defects inside the fitting based on the residual signal data and preset rules. This effectively improves the accuracy and reliability of defect detection and provides more precise guidance for the operation and maintenance of power fittings.

[0078] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the present invention specification.

Claims

1. A method for detecting internal defects in hardware based on ultrasound, characterized in that, include: Obtain the ultrasonic waveform data of the hardware to be analyzed; After extracting the initial interface echo segment from the ultrasonic waveform data, the set of contact state features of the hardware to be analyzed is determined. Based on the contact state feature set and historical detection data, dynamic health reference waveform data matching the characteristics of the hardware to be analyzed is determined; The dynamic health reference waveform data and the ultrasonic waveform data are processed to determine the residual signal data. Based on the residual signal data and the preset defect judgment rules, the detection results are identified, which contain the type and severity of the actual defects.

2. The method for detecting internal defects in hardware based on ultrasound according to claim 1, characterized in that, The step of determining the contact state feature set of the hardware to be analyzed after extracting the initial interface echo segment from the ultrasonic waveform data includes: Based on the ultrasonic waveform data, the model and service life of the hardware to be analyzed are obtained; Extract the initial interface echo band of the standard health reference waveform of the hardware model in a brand new state; Based on the service duration and a preset acoustic evolution trajectory model, a set of health characteristics is determined. The initial interface echo segment of the standard health reference waveform is modified and reconstructed using the set of health features to obtain the set of contact state features of the hardware to be analyzed.

3. The method for detecting internal defects in hardware based on ultrasound according to claim 1, characterized in that, The steps of processing the dynamic health reference waveform data and the ultrasonic waveform data to determine the residual signal data include: After aligning the dynamic health reference waveform data with the ultrasonic waveform data in time, the time-aligned dynamic health reference waveform data and ultrasonic waveform data are subtracted point by point to obtain the initial residual waveform. The initial residual waveform is decomposed into residual signals of different frequency sub-bands by multi-scale wavelet decomposition. Adaptive threshold filtering is performed on the low-frequency subband residual signal in the residual signals of the different frequency subbands, and nonlinear gain adjustment is performed on the high-frequency subband residual signal in the residual signals of the different frequency subbands to obtain residual signal data.

4. The method for detecting internal defects in hardware based on ultrasound according to claim 3, characterized in that, The steps of performing adaptive threshold filtering on the low-frequency subband residual signals in the residual signals of the different frequency subbands, and performing nonlinear gain adjustment on the high-frequency subband residual signals in the residual signals of the different frequency subbands to obtain residual signal data include: Based on each frequency sub-band, after determining the local energy distribution, instantaneous frequency change rate, and signal sparsity index of each frequency sub-band, the local clustering region of suspected weak defect signals in each frequency sub-band, as well as the distribution pattern of residual noise or non-defect artifacts, are identified. The residual signal within the local clustering region is amplified by local gain, and the residual noise or non-defect artifacts outside the local clustering region are suppressed by adaptive threshold, resulting in the suppressed frequency sub-band residual signal. Based on the distribution pattern, the residual signal of each frequency sub-band is nonlinearly gain adjusted to obtain the adjusted frequency sub-band residual signal. The suppressed frequency subband residual signal and the adjusted frequency subband residual signal are reconstructed to obtain the residual signal data.

5. The method for detecting internal defects in hardware based on ultrasound according to claim 1, characterized in that, Based on the residual signal data and preset defect judgment rules, the steps for identifying detection results containing the type and severity of real defects include: Extract potential defect echoes from residual signal data; The amplitude, duration, energy, and arrival time of the potential defect echoes are analyzed to identify the defect signals of the hardware to be analyzed. Based on the defect signals of the hardware to be analyzed and the preset defect judgment rules, the detection results that include the type and severity of the actual defects are identified.

6. The method for detecting internal defects in hardware based on ultrasound according to claim 5, characterized in that, The steps for analyzing the amplitude, duration, energy, and arrival time of the potential defect echoes to identify the defect signals of the hardware to be analyzed include: The potential defect echo is subjected to resolution time-frequency decomposition to obtain the time-frequency distribution data of the potential defect echo; Separate each independent time-frequency component of the overlapping echo in the time-frequency distribution data; Feature extraction is performed on each of the independent time-frequency components to determine the center frequency, bandwidth, time delay, and energy concentration of each independent time-frequency component; Defects are identified by using the center frequency, bandwidth, time delay, and energy concentration of each independent time-frequency component, thus obtaining the defect signal of the hardware to be analyzed.

7. The method for detecting internal defects in hardware based on ultrasound according to claim 6, characterized in that, The step of separating each independent time-frequency component of the overlapping echo in the time-frequency distribution data includes: After performing local enhancement processing on the time-frequency distribution data, the enhanced time-frequency distribution data is segmented to obtain each preliminary separated component; After determining the time-frequency center, energy moment, and sparsity index of each of the preliminary separation components, it is determined that there is a fusion region between two adjacent preliminary separation components where the component frequency is within the frequency threshold and the energy distribution is continuous. Each fusion region is subjected to iterative signal decomposition and separation processing to obtain each independent time-frequency component.

8. The method for detecting internal defects in hardware based on ultrasound according to claim 4, characterized in that, Based on the distribution pattern, the step of performing nonlinear gain adjustment on the residual signal of each frequency sub-band to obtain the adjusted frequency sub-band residual signal includes: By utilizing the characteristics of the distribution pattern, after determining the instantaneous bandwidth, energy distribution continuity, and signal symmetry indices of each frequency sub-band, the characteristics of each broadband defect signal across the frequency sub-band are identified. The low-frequency subband and high-frequency subband of each broadband defect signal feature are fused to obtain each broadband processed subband. Adaptive threshold filtering of energy gradient and local signal-to-noise ratio and nonlinear gain adjustment are performed on each of the broadband processing subbands to obtain the adjusted frequency subband residual signal.

9. The method for detecting internal defects in hardware based on ultrasound according to claim 8, characterized in that, The steps of performing adaptive threshold filtering and nonlinear gain adjustment on the energy gradient and local signal-to-noise ratio for each broadband sub-band to obtain the adjusted frequency sub-band residual signal include: Multi-scale edge detection is performed on each of the broadband processing sub-bands to obtain edge intensity data for each broadband processing sub-band; Potential defect signal boundary regions and noise artifact boundary regions are marked in each edge intensity data; Based on the boundary regions of defective signals and noise artifacts, the overall energy distribution and instantaneous frequency variation trend of each broadband processing sub-band are determined. By utilizing the overall energy distribution and instantaneous frequency variation trend of each broadband processing sub-band, nonlinear gain adjustment is performed on each broadband processing sub-band to obtain the adjusted frequency sub-band residual signal.

10. An ultrasonic-based system for detecting internal defects in hardware, characterized in that, The system includes: The acquisition module is used to acquire the ultrasonic waveform data of the hardware to be analyzed; The set determination module is used to extract the initial interface echo segment from the ultrasonic waveform data and then determine the contact state of the hardware to be analyzed. The waveform determination module is used to determine dynamic health reference waveform data that matches the characteristics of the hardware to be analyzed, based on the contact state feature set and historical detection data. The signal determination module is used to process the dynamic health reference waveform data and the ultrasonic waveform data to determine the residual signal data. The identification module is used to identify the detection results containing the type and severity of real defects based on the residual signal data and preset defect judgment rules.

Images