Fine wire defect identification method and system based on eddy current detection

Abstract

The invention discloses a fine wire rod defect identification method and system based on eddy current detection, and relates to the technical field of defect identification, a plurality of fine wire rod detection data are determined according to synchronous detection of an absolute coil channel and a differential coil channel, the plurality of fine wire rod detection data are aligned, and a plurality of fine wire rod defect identification results are obtained. Transient pulse features are extracted in a time domain, defect frequency response features are extracted in a frequency domain, multi-factor fusion is carried out according to the transient pulse features and the defect frequency response features to determine an eddy current detection event, and the accuracy of the eddy current detection event is improved. Determining a plurality of defect features according to comparison between the feature spectrum of each sub eddy current detection item and a preset standard feature spectrum; and determining the defect level of each defect feature based on the feature position of each defect feature, the corresponding feature form and the structural form of the fine wire, and determining the dynamic early warning level along the superposition of the defect levels of the defect features, thereby realizing the dynamic early warning of the fine wire detector on the fine wire.

Description

Technical Field

[0001] This invention relates to the technical field of defect identification, and in particular to a method and system for identifying defects in fine wire based on eddy current detection. Background Technology

[0002] As a core component in critical fields such as electronic equipment, the surface and internal quality of fine wire directly affects the safety of electronic equipment. Therefore, using eddy current testing technology to perform non-destructive testing on fine wire has become an essential step in industrial production.

[0003] In existing technologies, traditional wire detectors typically use a single type of detection coil for detection. If only an absolute coil is used, although it can detect long-gradual defects (such as diameter wear and material abrupt changes), it has extremely low sensitivity to short-term defects such as surface micro-cracks. Therefore, existing technologies lack the ability to combine absolute coil channels with differential coil channels, which makes it impossible to integrate detection information from different dimensions when facing defects with complex shapes, thus reducing the accuracy of eddy current detection events. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art. This invention provides a method and system for identifying defects in fine wire based on eddy current detection.

[0005] This invention provides a method for identifying defects in precision wire based on eddy current detection, comprising: The dual-channel detection architecture is determined based on the detection of the fine wire detector, and the absolute coil channel and the differential coil channel are combined with multiple factors. The absolute coil channel and the differential coil channel perform wide-area detection in different dimensions. Multiple fine wire detection data are determined based on the synchronous detection of the absolute coil channel and the differential coil channel. The multiple fine wire detection data are aligned to extract transient pulse features in the time domain and defect frequency response features in the frequency domain. Eddy current detection events are determined by multi-factor fusion based on transient pulse features and defect frequency response features. In this eddy current detection event, multiple sub-eddy current detection items are identified based on the identification of the eddy current detection event, and the feature maps of each sub-eddy current detection item are marked. Multiple defect features are determined by comparing the feature maps of each sub-eddy current detection item with the preset standard feature maps. Based on the characteristic location, corresponding characteristic shape, and structural shape of each defect feature, the defect level of each defect feature is determined, and the dynamic warning level is determined by superimposing the defect levels of each defect feature, so as to trigger the dynamic warning of the fine wire detector on the fine wire.

[0006] This invention provides a precision wire defect identification system based on eddy current detection, which is applied to the aforementioned precision wire defect identification method based on eddy current detection; the precision wire defect identification system based on eddy current detection includes: The multi-factor combination module is used to determine the dual-channel detection architecture based on the detection of the fine wire detector, and to combine the absolute coil channel and the differential coil channel in a multi-factor combination, so that the absolute coil channel and the differential coil channel can perform wide-area detection in different dimensions; The eddy current detection module is used to determine multiple fine wire detection data based on the synchronous detection of the absolute coil channel and the differential coil channel. The multiple fine wire detection data are aligned to extract transient pulse features in the time domain and defect frequency response features in the frequency domain. The eddy current detection event is determined by multi-factor fusion based on the transient pulse features and defect frequency response features. The defect module is used to identify multiple sub-eddy current detection items based on the identification of the eddy current detection event in the eddy current detection event, mark the feature map of each sub-eddy current detection item, and determine multiple defect features by comparing the feature map of each sub-eddy current detection item with the preset standard feature map. The dynamic early warning module is used to determine the defect level of each defect feature based on its characteristic location, corresponding characteristic shape, and structural shape of the wire. It then determines the dynamic early warning level by superimposing the defect levels of each defect feature, thereby triggering the wire detector to issue a dynamic early warning for the wire.

[0007] Compared with the prior art, the beneficial effects of the present invention are: (1) Based on the detection of the fine wire detector, a dual-channel detection architecture is determined, and the absolute coil channel and the differential coil channel are combined with multiple factors. The absolute coil channel and the differential coil channel perform wide-area detection in different dimensions. Multiple fine wire detection data are determined based on the synchronous detection of the absolute coil channel and the differential coil channel. The multiple fine wire detection data are aligned to extract transient pulse features in the time domain and defect frequency response features in the frequency domain. The eddy current detection event is determined by multi-factor fusion based on transient pulse features and defect frequency response features. The absolute coil channel and the differential coil channel are introduced, and multiple fine wire detection data are controlled. The multi-factor fusion of transient pulse features and defect frequency response features is realized, which improves the accuracy of eddy current detection events.

[0008] (2) In this eddy current detection event, multiple sub-eddy current detection items are identified based on the identification of the eddy current detection event, and the feature maps of each sub-eddy current detection item are marked. Multiple defect features are identified by comparing the feature maps of each sub-eddy current detection item with the preset standard feature maps. The defect level of each defect feature is determined based on the feature position, corresponding feature shape and structural shape of the wire material. The dynamic warning level is determined by superimposing the defect levels of each defect feature, so as to trigger the dynamic warning of the wire material detector on the wire material, control multiple defect features, and realize the superposition of the defect levels of each defect feature, improve the stage dynamic detection of the wire material, improve the accuracy of the dynamic warning level, and realize the dynamic warning of the wire material detector on the wire material.

[0009] (3) Through multi-factor fusion analysis of time-domain and frequency-domain features, and combined with intelligent comparison of feature maps using convolutional neural networks, mechanical vibration and electromagnetic noise interference are effectively suppressed, surface stains are accurately distinguished from real structural defects, and the robustness and reliability of detection are greatly improved. Furthermore, based on electromagnetic inversion technology, the geometric dimensions (length, depth, and orientation) of defects are quantitatively inverted, and multi-dimensional mechanical analysis is performed by integrating the structural morphology of fine wire, thus completing an objective assessment of the safety risk level of defects and achieving a technological leap from qualitative detection to quantitative risk assessment. On this basis, the system automatically matches the defect maintenance knowledge base through a dynamic early warning level triggering mechanism, generates standardized handling suggestions, significantly reduces the reliance on human experience, improves the automation, standardization, and response efficiency of maintenance decisions, and forms a complete closed loop of "detection-assessment-early warning-decision". Attached Figure Description

[0010] Figure 1 This is a flowchart illustrating the method for identifying defects in fine wire based on eddy current detection in an embodiment of the present invention. Figure 2 This is a flowchart illustrating step S11 in the eddy current detection-based precision wire defect identification method in this embodiment of the invention. Figure 3 This is a flowchart illustrating step S12 in the eddy current detection-based precision wire defect identification method in this embodiment of the invention. Figure 4 This is a flowchart illustrating step S13 in the eddy current detection-based precision wire defect identification method in this embodiment of the invention. Figure 5 This is a flowchart illustrating step S14 in the eddy current detection-based precision wire defect identification method in this embodiment of the invention. Figure 6 This is a schematic diagram of the structural composition of the precision wire defect identification system based on eddy current detection in an embodiment of the present invention. Detailed Implementation

[0011] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0012] Please see Figures 1 to 6 A method for identifying defects in fine wire based on eddy current detection is applied to defect identification scenarios. The method includes: Step S11: Determine the dual-channel detection architecture based on the detection of the fine wire detector, and combine the absolute coil channel and the differential coil channel with multiple factors. The absolute coil channel and the differential coil channel perform wide-area detection in different dimensions. Step S12: Based on the synchronous detection of the absolute coil channel and the differential coil channel, multiple fine wire detection data are determined. The multiple fine wire detection data are aligned to extract transient pulse features in the time domain and defect frequency response features in the frequency domain. Based on the transient pulse features and defect frequency response features, multi-factor fusion is performed to determine the eddy current detection event. Step S13: In this eddy current detection event, multiple sub-eddy current detection items are identified based on the identification of the eddy current detection event, and the feature maps of each sub-eddy current detection item are marked. Multiple defect features are determined by comparing the feature maps of each sub-eddy current detection item with the preset standard feature maps. Step S14: Determine the defect level of each defect feature based on its characteristic location, corresponding characteristic shape, and structural shape of the wire, and determine the dynamic warning level by superimposing the defect levels of each defect feature, so as to trigger the dynamic warning of the wire detector on the wire.

[0013] refer to Figure 2 In step S11, the specific steps are as follows: S111: In the fine wire detector, mark multiple detection channels of the fine wire detector, and construct a corresponding dual-channel detection architecture based on the multiple detection channels and the fine wire to be detected. In the dual-channel detection architecture, determine the absolute coil channel and the differential coil channel, and trigger the cooperative detection of the same fine wire by the absolute coil channel and the differential coil channel. S112: Mark the first layer of detection factors for the absolute coil channel and the second layer of detection factors for the differential coil channel. Combine the first layer of detection factors and the second layer of detection factors into multiple factors and balance the detection performance of the absolute coil channel and the differential coil channel at the same position of the fine wire. At this time, the absolute coil channel is responsible for wide-area monitoring of long gradient defects and material abrupt changes; the differential coil channel is responsible for wide-area monitoring of surface defects.

[0014] In the embodiments of this application, multiple detection channels of the fine wire detector are marked, and a corresponding dual-channel detection architecture is constructed based on the multiple detection channels and the fine wire to be detected. In the dual-channel detection architecture, the absolute coil channel and the differential coil channel are determined, and the absolute coil channel and the differential coil channel are triggered to conduct cooperative detection of the same fine wire. This approach takes into account the overall consideration of multiple detection channels and the fine wire to be detected, and ensures the accuracy of the corresponding dual-channel detection architecture.

[0015] At this point, the front-end data acquisition (DAQ) system of the eddy current detector typically has multiple analog-to-digital converter (ADC) channels. Based on the hardware topology, the system logically labels specific physical ports as "Channel 1" and "Channel 2," and uniquely binds them to the physically installed sensors—absolute coils and differential coils. Simultaneously, the "dual-channel detection architecture" refers to establishing a parallel data pipeline. Under this architecture, the system is configured with two independent signal conditioning circuits (including preamplifiers, filters, phase-sensitive detectors, etc.), corresponding to the two labeled channels respectively. The core of the architecture is to ensure that these two channels have a consistent sampling rate and a synchronized sampling clock, thereby eliminating underlying clock deviations for subsequent data fusion.

[0016] The system clearly defines the "absolute coil channel" as responsible for providing low-frequency, slowly varying background signals (such as changes in diameter and conductivity), while the "differential coil channel" is responsible for providing high-frequency, abrupt local signals (such as cracks and pits). "Trigger-based collaborative detection" is a prerequisite for achieving multi-source heterogeneous data fusion, which is usually achieved through hardware triggering or software interruption. When the sensor detects that the wire to be tested has entered the effective detection area (such as through photoelectric switch signals or signal amplitude threshold triggering), the system simultaneously starts data acquisition from both channels.

[0017] More importantly, the cooperative detection requires the system to dynamically adjust the sampling delay (TimeDelayCompensation) at the signal processing level based on the running speed (v) of the wire and the physical installation distance (Δd) between the two coils. Because the two coils are spatially separated, the time points (t1, t2) at which they detect the same physical cross section (PositionP) of the wire are different. The cooperative triggering mechanism must record and compensate for this time difference to ensure that the data streams of the two channels are logically targeted at the same cross section.

[0018] Specifically, on the production line of fine wire, the detector starts a self-test program; the system identifies that port A on the front-end analog board is connected to a through-type absolute coil for full cross-section coverage, and port B is connected to a differential detection coil that is sensitive to local abrupt changes; at the software level, the system marks port A as the "absolute channel" and configures its passband to low frequency to enhance penetration; it marks port B as the "differential channel" and configures its passband to high frequency to improve sensitivity to minute defects; at this point, the dual-channel detection architecture for fine wire is logically established.

[0019] Assume the wire passes through the detection probe at a constant speed of 5 m / s; the absolute coil is installed upstream, and the differential coil is installed downstream, with their centers 20 cm apart; when the head of the wire triggers the photoelectric sensor, the system triggers a cooperative detection command, and both channels simultaneously begin recording data at a frequency of 10 kHz; due to the installation gap, the absolute coil will detect a defect feature on the wire at time t1, while the differential coil will detect the same defect t2 = t1 + (0.2 m / 5 m / s) = t1 + 40 ms later.

[0020] Under the collaborative detection mechanism of S111, the system automatically establishes a "virtual alignment window": when the differential channel collects data at time t2, the system will call the historical data at time t1 from the cache of the absolute channel for packaging; the fine wire is jointly scanned at the same time and at the same location, thereby ensuring the spatiotemporal consistency of feature extraction in subsequent steps.

[0021] Furthermore, the first layer of detection factors for the absolute coil channel and the second layer of detection factors for the differential coil channel are marked. A multi-factor combination is performed on the first and second layer detection factors, and the detection performance of the absolute coil channel and the differential coil channel at the same position of the fine wire is balanced. At this time, the absolute coil channel is responsible for wide-area monitoring of long gradient defects and material abrupt changes; the differential coil channel is responsible for wide-area monitoring of surface defects. The differential coil channel is introduced to be responsible for wide-area monitoring of surface defects.

[0022] At this point, the output signal of the dual-channel detection architecture is defined by attributes and abstracted into two different-dimensional detection factors, introducing the first layer of detection factors and the second layer of detection factors; the first layer of detection factors (absolute coil channel) is labeled: this factor corresponds to the output characteristics of the absolute coil; the absolute coil is sensitive to the changes in the properties of all magnetic or conductive materials within the coverage of the entire detection coil; its signal characteristics are manifested as baseline drift or DC offset; the first layer factor is defined as a global trend factor, used to characterize the macroscopic physical properties of the object under test, such as the average changes in overall diameter, conductivity, and permeability.

[0023] The second layer of detection factors (differential coil channel) is marked: This factor corresponds to the output characteristics of the differential coil; the differential coil consists of two coils wound in opposite directions, which are not sensitive to uniform fields, but are extremely sensitive to field non-uniformity (i.e., gradient); its signal characteristics are transient spikes or AC components; the second layer factor is defined as "local gradient factor", which is used to characterize the microscopic discontinuities of the measured object, such as cracks, pores, pits, etc.

[0024] The global trend signal of the first layer and the local gradient signal of the second layer are jointly processed in the feature space. Since the absolute coil has a low signal-to-noise ratio for small defects and the differential coil is not sensitive to slow changes, directly using a single channel will lead to missed detections or false judgments. The system needs to dynamically adjust the weights of the two factors according to the characteristics of the target to be detected. For example, by using adaptive filtering, the first-layer factor is used to construct a background model and subtract it from the original signal, thereby highlighting the characteristics of the second-layer factor; or the signal amplitude of the second-layer factor is used to correct the threshold judgment benchmark of the first-layer factor. This balancing mechanism ensures that the system can maintain optimal detection sensitivity when facing different types of defects.

[0025] Specifically, the system classifies the collected fine wire data into layers: The first layer of detection factors (absolute channel): The system extracts the signal amplitude and phase changes of the absolute coil; for fine wire, this layer of factors mainly reflects the overall cross-sectional dimensions (such as slight diameter reduction) and material consistency of the fine wire; if the fine wire undergoes uneven heat treatment during production, resulting in a gradual change in magnetic permeability, this will manifest as smooth low-frequency fluctuations in the first layer of factors.

[0026] The second layer of detection factors (differential channel): The system extracts the signal peak value and zero-crossing characteristics of the differential coil. This layer of factors mainly focuses on capturing local discontinuities on the surface of the wire. For example, longitudinal cracks caused by die scratches on the surface of the wire, or pits formed by oxide scale peeling off, these defects will excite high-amplitude, narrow-width pulse signals in the second layer of factors.

[0027] Suppose there is a 500mm long wear defect (diameter slowly decreasing) on ​​the fine wire, accompanied by a short crack 1mm deep; Balanced long gradient monitoring (first layer dominant): For the 500mm long wear, the differential coil (second layer factor) has no obvious output because the change is too gradual; At this time, the system's multi-factor combination method will significantly increase the weight of the first layer factor; The absolute channel captures the 500mm long baseline voltage drop, and the system determines that there is a long gradient dimensional loss in the fine wire, realizing wide-area monitoring.

[0028] Balanced surface defect monitoring (second layer dominant): For the short crack with a depth of 1mm, the absolute coil (first layer factor) is overwhelmed by noise because the rate of change of cross-sectional area is too small; at this time, it automatically switches to high gain mode to read the second layer factor; the differential channel generates a violent pulse response to the instantaneous magnetic field distortion of the crack, and the system identifies the surface crack accordingly.

[0029] When two signals are input simultaneously, the system compares them on the same time axis: the first layer of factors provides the "background wall thickness / material information" at the defect location, and the second layer of factors provides the "sharpness information" of the defect; combined with the two, the system can not only detect defects, but also distinguish whether it is a "surface crack on intact material" (first layer normal, second layer alarm) or a "surface crack on a severely worn base" (first layer alarm, second layer also alarm), thereby greatly improving the accuracy of defect classification of fine wire.

[0030] refer to Figure 3 In step S12, the specific steps are as follows: S121: Synchronous detection of the absolute coil channel and the differential coil channel is performed, and the timestamp synchronization is achieved using a field-programmable gate array to determine multiple fine wire detection data. At the same time, the multiple fine wire detection data are time-aligned based on interpolation to calibrate the phase deviation caused by the delay of the analog front-end circuit or the difference in sampling rate. S122: Based on the aligned data, wavelet packet decomposition technology is applied to extract transient pulse features in the time domain and capture signal abrupt changes caused by crack edges. At the same time, higher-order spectral analysis is used to extract defect frequency response features for different defect types in the frequency domain and analyze the depth attributes of the defects. The transient pulse features in the time domain and the defect frequency response features in the frequency domain are subjected to multi-factor weighted fusion analysis. The validity of eddy current detection events is determined based on the consistency and significance of the features, and noise caused by mechanical vibration or electromagnetic interference is filtered out.

[0031] In the embodiments of this application, the absolute coil channel and the differential coil channel are synchronously detected, and the timestamp synchronization is achieved using a field-programmable gate array to determine multiple fine wire detection data. At the same time, the multiple fine wire detection data are time-aligned based on interpolation to calibrate the phase deviation caused by the delay of the analog front-end circuit or the difference in the sampling rate. This introduces a calibration method for the phase deviation caused by the delay of the analog front-end circuit or the difference in the sampling rate.

[0032] At this point, the parallel logic resources inside the field programmable gate array (FPGA) are used to configure independent analog-to-digital converter (ADC) controllers for the absolute coil channel and the differential coil channel respectively; the FPGA generates a global clock source through its internal phase-locked loop (PLL) to drive the two ADCs to sample simultaneously, ensuring strict alignment of the sampling trigger edges.

[0033] After each sampling, the FPGA packages the current system count value (i.e., high-precision timestamp) with the sampled data. Regardless of the subsequent processing flow, the data from both channels carries a physically meaningful "detection time". This mechanism eliminates the nondeterministic jitter caused by the operating system or software layer scheduling, ensuring the consistency of the time base of the data stream.

[0034] Although sampling is synchronous, the analog front-end circuitry (such as filter order, amplifier bandwidth, and cable length) of the absolute and differential coils often differ, which can cause different group delays in signal transmission or slight frequency deviations (sampling rate differences) between the sampling clocks of the two channels.

[0035] To correct these deviations, the system employs interpolation methods (such as Sinc interpolation or Farrow filters) in the digital domain. Using the time axis of one channel (e.g., the absolute channel) as a reference, the theoretical value of the other channel (differential channel) at the reference time point is reconstructed through mathematical calculations. This process is equivalent to fine-tuning the signal on the time axis by "stretching" or "translating," thereby eliminating phase mismatch caused by circuit delays or sampling rate differences.

[0036] Specifically, assuming the fine wire passes through the detection probe at a high speed of 10 meters per second, the system requires an extremely high sampling rate to capture minute defects. The FPGA inside the detector starts parallel logic to control both the absolute channel ADC and the differential channel ADC to sample synchronously at a frequency of 1MHz. When a tiny rust spot on the fine wire enters the coil area, the FPGA accurately records the data value of the absolute channel at that moment as V1, with a timestamp of T1, and simultaneously records the data value of the differential channel as V2, with the timestamp also being T1. This nanosecond-level hardware synchronization ensures that the two channels capture the magnetic field information of the fine wire in the same microsecond state.

[0037] The analog filter of the absolute coil channel is a fourth-order low-pass filter, while the differential coil channel uses a second-order band-pass filter. This causes the signal of the absolute channel to be delayed by 3 sampling points (i.e., 3 microseconds) compared to the differential channel. If no calibration is performed, when the system fuses the data, the crack peak detected by the differential channel at time T will be superimposed with the background data of the absolute channel at time T (which actually corresponds to a position slightly later than the fine wire), resulting in feature matching errors.

[0038] The system reconstructs the data of the absolute channel using interpolation. The system shifts the data stream of the absolute channel "backward" on the time axis, or calculates the precise value at time T+3μs according to the interpolation method, so as to align the data of the differential channel at time T. After calibration, for the same physical defect point on the wire (such as a scratch with a depth of 0.5mm), the maximum peak phase point of the differential channel output perfectly coincides with the phase point of the impedance change caused by the scratch in the absolute channel. This high-precision alignment lays a solid foundation for the accurate fusion of "transient pulse characteristics" and "defect frequency response characteristics" in the subsequent step S12.

[0039] Furthermore, based on the aligned data, wavelet packet decomposition technology is applied to extract transient pulse features in the time domain and capture signal abrupt changes caused by crack edges. Simultaneously, higher-order spectral analysis is used to extract defect frequency response features for different defect types in the frequency domain, analyzing the depth attributes of the defects. The transient pulse features in the time domain and the defect frequency response features in the frequency domain are subjected to multi-factor weighted fusion analysis. Valid eddy current detection events are comprehensively determined based on the consistency and significance of the features, and noise caused by mechanical vibration or electromagnetic interference is filtered out. At the same time, absolute coil channels and differential coil channels are introduced, and multiple precision wire detection data are controlled, realizing multi-factor fusion of transient pulse features and defect frequency response features, thus improving the accuracy of eddy current detection events.

[0040] At this point, wavelet packet decomposition is an optimization of multi-resolution analysis. It decomposes not only the low-frequency part but also the high-frequency part, providing a more refined frequency division. In the time domain, defect signals often manifest as non-stationary transient pulses. By using wavelet packet transform to decompose the signal into different sub-bands, the high-frequency components containing defect edge information can be effectively separated from the low-frequency background noise.

[0041] By selecting appropriate wavelet basis functions (such as the Db series or Sym series) and performing modulus maxima analysis on the decomposed coefficients, the time points at which the signal amplitude and phase change abruptly can be accurately captured, which correspond to the edge moment of crack formation. This method has extremely strong robustness to background signals.

[0042] Traditional power spectrum analysis (second-order statistics) cannot preserve the phase information of a signal and is insensitive to Gaussian noise. Higher-order spectrum analysis (such as bispectral or trispectral) is a Fourier transform of the third or fourth-order cumulants of the signal. At the same time, defects of different depths will cause different degrees of nonlinear distortion in the eddy current field. Higher-order spectra are extremely sensitive to this nonlinear characteristic and can suppress Gaussian noise. By calculating the higher-order spectrum of the signal, the frequency components and their coupling relationships unique to different defect types can be extracted. These frequency response characteristics are closely related to the geometry, depth and material properties of the defect, and can distinguish between shallow surface scratches and deep internal cracks.

[0043] The system assigns weights to the pulse amplitude and pulse width in the time domain, as well as the peak values ​​of higher-order spectra and the energy distribution ratio of specific frequency components in the frequency domain. A comprehensive feature vector is constructed through a weighting method (such as DS evidence theory or adaptive weighting). The existence of a valid eddy current detection event is determined based on the "consistency" (whether anomalies occur simultaneously in the time and frequency domains) and "significance" (whether the feature amplitude exceeds the dynamic threshold) of the comprehensive feature vector. If there is only mechanical vibration, it usually only causes low-frequency fluctuations in the time domain and lacks nonlinear higher-order features in the frequency domain. If it is electromagnetic interference, the frequency domain features are often chaotic and lack specific pulse patterns in the time domain. Through this dual verification, noise can be effectively filtered out.

[0044] Specifically, suppose there is a tiny transverse crack on the surface of the fine wire; when the crack passes through the differential coil at high speed, an extremely narrow pulse will be superimposed on the original signal; when executing S122, the system uses wavelet packet decomposition (e.g., using the Daubechies4 wavelet basis) to perform multi-scale decomposition on the aligned data; in the high-frequency subband of the detail coefficients, the system accurately captures the two reverse mode maxima points generated when the crack's "front edge" and "back edge" pass through the coil. These two points constitute the transient pulse feature, which not only confirms the existence of the defect, but also accurately marks the width of the crack in the axial direction of the fine wire.

[0045] Meanwhile, the system performs bispectral analysis on this signal segment; if it is a slight scratch on the surface of the fine wire, the eddy current penetration depth is shallow, the nonlinear distortion is weak, and it appears as a low-energy fundamental frequency component in the higher-order spectrum; if it is a deep crack (such as a depth exceeding 5% of the diameter), the reflection of the eddy current at the bottom will produce complex nonlinear interference, and the higher-order spectrum will show significant non-zero peaks on specific frequency slices, i.e., defect frequency response characteristics; based on the strength of this peak, the system quantitatively analyzes that the crack belongs to "deep crack" rather than "surface stain".

[0046] The system performs a fusion judgment on the above features; Scenario simulation (mechanical vibration): Suppose that the conveyor belt of the detection mechanism experiences a vibration at this time; the time domain signal fluctuates due to the lift-off effect, but wavelet packet decomposition shows that this fluctuation is a smooth low-frequency change, lacking sharp edge pulses; at the same time, high-order spectrum analysis shows that the signal follows a Gaussian distribution and has no nonlinear coupling characteristics; the system judges it as "inconsistent" and uses it as noise filtering, without triggering an eddy current detection event; Scenario simulation (real defect): For the aforementioned transverse crack, the time domain captures sharp transient pulses, and the frequency domain simultaneously detects high-intensity nonlinear high-order spectrum features; the two are highly consistent and significant in the judgment of "existence of anomaly"; based on this, the system comprehensively judges that this is a valid eddy current detection event and marks it as a key object that needs further analysis in step S13.

[0047] refer to Figure 4 In step S13, the specific steps are as follows: S131: Real-time monitoring of eddy current detection events, marking the detection data combinations corresponding to the eddy current detection events, identifying the detection data combinations, and decomposing continuous detection signals into multiple sub-eddy current detection projects according to temporal and spatial distribution. Each sub-eddy current detection project marks the corresponding potential abnormal region, and performs isolation analysis on each potential abnormal region; marking the time-domain waveform segments and impedance plane trajectories of each sub-detection project, extracting high-dimensional feature maps using principal component analysis and manifold learning, and characterizing the geometric morphology and electromagnetic response characteristics of wire defects through feature maps; S132: Input each feature map into a convolutional neural network and perform convolutional layer analysis to output the feature map of each sub-eddy current detection project. At the same time, based on the traceability of the precision wire detector, determine the preset standard feature map, and perform feature matching and comparison between the feature map of each sub-eddy current detection project and the preset standard feature map. By calculating the Euclidean distance and structural similarity index between the maps, and combining nonlinear mapping to quantify the feature deviation, multiple defect features can be identified, and surface stains can be effectively distinguished from real scratches.

[0048] In the embodiments of this application, eddy current detection events are monitored in real time, the detection data combinations corresponding to the eddy current detection events are marked, the detection data combinations are identified, and the continuous detection signals are decomposed into multiple sub-eddy current detection projects according to temporal and spatial distribution. Each sub-eddy current detection project is marked with a corresponding potential abnormal region, and each potential abnormal region is isolated and analyzed. The time-domain waveform segments and impedance plane trajectories of each sub-detection project are marked, and high-dimensional feature maps are extracted using principal component analysis and manifold learning. The feature maps characterize the geometric morphology and electromagnetic response characteristics of the fine wire defects, thus introducing the feature maps to characterize the geometric morphology and electromagnetic response characteristics of the fine wire defects.

[0049] At this point, the system performs windowing processing on the eddy current detection event confirmed in S122, extracting the absolute and differential channel data within a certain time window before and after the event to form a "detection data combination". Based on the temporal correlation (temporal continuity) and spatial distribution of the signal (physical location inferred from the wire speed), the system cuts a long signal segment containing complex defect morphologies (such as continuously distributed wear or clustered cracks) into multiple independent "sub-eddy current detection projects". Each project represents an independent "potential anomaly area". An independent analysis context is established for each sub-project to prevent signals from adjacent areas from interfering with each other (e.g., the edge effect of a large crack covers up a small pit next to it).

[0050] The time-domain waveform segments (amplitude varying over time) and impedance plane trajectories (plane trajectories formed by the real and imaginary parts after phase-sensitive detection) of each sub-item are extracted. Simultaneously, Principal Component Analysis (PCA) and manifold learning (such as LLE, Isomap, and t-SNE) are introduced. PCA removes redundant information and linear noise from the data, compressing the original high-dimensional signal data into a few main linear components. Manifold learning (such as LLE, Isomap, and t-SNE) uncovers the nonlinear intrinsic structure within the data. The trajectory of eddy current signals on the impedance plane is often a complex low-dimensional manifold embedded in a high-dimensional space; manifold learning can "unfold" these trajectories, extracting "high-dimensional feature maps" that characterize the essence of defects (such as geometry and electromagnetic response).

[0051] Specifically, suppose the wire passes through an area caused by mold wear, resulting in a dense cluster of longitudinal scratches approximately 200mm in length, interspersed with an isolated deep pitting corrosion. After the system monitors the eddy current detection event in this area, it decomposes it into three "sub-eddy current detection projects" based on the spatial distribution of the signal. Projects 1 and 2 correspond to the continuous scratch cluster at the front end, while project 3 corresponds to the pitting corrosion at the rear end. The system isolates the detection data of project 3 separately to avoid interference from background magnetic field fluctuations generated by the scratch clusters in projects 1 and 2 on the parameter calculation of the pitting corrosion, ensuring that the analysis of each sub-project is pure.

[0052] For Project 3 (pitting pits): The system marks the typical "positive then negative" zero-crossing pulse waveform of the differential channel; on the complex plane, an arc-shaped trajectory is marked; the starting point of the trajectory represents the equilibrium point when the coil is unaffected, and the ending point represents the point of maximum impedance change caused by the defect. The curvature of the trajectory reflects the direction of the defect.

[0053] The original waveform data had 1000 sampling points. The system compressed it into three principal components (PC1, PC2, PC3) using PCA, removing the linear noise component caused by wire vibration. The system used manifold learning (such as t-SNE) to analyze the nonlinear trajectory of the pitting on the impedance plane. The analysis results generated a high-dimensional feature map. In this map, the pitting appears as a specific "manifold cluster", and its map coordinates directly correspond to the geometric shape of the defect (the horizontal axis of the map represents the opening width of the defect, and the vertical axis represents the erosion depth of the defect). For item 1 (shallow scratch), the generated feature map is morphologically very different from item 3, showing a slender and low-energy distribution. By comparing these feature maps, the system can accurately distinguish between "shallow scratches on the surface of the wire" and "threatening deep pitting" at the geometric and electromagnetic levels, and can complete the in-depth characterization of the features without manually setting thresholds.

[0054] Furthermore, each feature map is input into a convolutional neural network and subjected to convolutional layer analysis to output the feature maps of each sub-eddy current detection item. Simultaneously, a preset standard feature map is determined based on the traceability of the precision wire detector. The feature maps of each sub-eddy current detection item are matched and compared one by one with the preset standard feature map. By calculating the Euclidean distance and structural similarity index between the maps and combining nonlinear mapping to quantify feature deviation, multiple defect features are identified, and surface stains are effectively distinguished from real scratches. This approach is compatible with the overall consideration of the traceability of the precision wire detector and ensures the accuracy of the preset standard feature map.

[0055] At this point, the feature map is used as the input image, and sliding window operations are performed through multiple convolutional kernels. The convolutional layers can automatically extract deep features such as local texture, edge gradients, and topological structures from the map. These features are often more abstract and discriminative than manually designed features. After processing by pooling layers and fully connected layers, the CNN outputs a high-dimensional feature vector of the map. This vector condenses the essential attributes of the sub-eddy current detection project (i.e., potential defects) for subsequent comparative analysis.

[0056] Based on the specifications, materials, and historical testing data of the fine wire, the system traces and retrieves the "preset standard feature maps" from the knowledge base. The standard map library contains feature templates of fine wire under various known conditions, such as "defect-free standard templates", "crack typical templates", "rust templates", and "standard porosity templates". These maps are samples collected in a controlled environment and annotated by experts.

[0057] Euclidean distance and structural similarity (SSIM) are calculated. Euclidean distance is used to calculate the straight-line distance between the feature vector of the test image and the standard image vector in space; the smaller the distance, the closer the overall features are. Structural similarity (SSIM) compares the similarity of two images from three dimensions: brightness, contrast, and structure, and is highly sensitive to differences in perceived structure. Since eddy current signals are nonlinear, a simple linear distance is insufficient to characterize complex differences. The system introduces nonlinear mapping (such as kernel function methods) to map the data to a high-dimensional space for quantization, calculating a more accurate "feature deviation value." Surface stains (such as sludge and oxide scale) typically cause slow changes or amplitude attenuation in eddy current signals. Their feature maps have relatively blurred texture representations in CNNs and extremely low structural similarity to standard "scar" images, resulting in a large Euclidean distance. Real scars (such as cracks and folds) cause abrupt phase changes and specific distortions in the impedance plane trajectory. Their feature maps have clear edges and specific topological manifolds, highly matching the standard "scar" images.

[0058] Specifically, the system inputs the high-dimensional feature map generated in S131 for "Project 3 (deep pitting)" into the trained CNN model; the CNN convolutional layer scans the map, extracts the unique deep features such as "ring texture" and "central high energy point" of the map, and outputs a specific 1024-dimensional feature vector, which mathematically uniquely represents the physical anomaly detected at present.

[0059] Based on the parameters "fine wire, 5mm diameter, 65Mn material", the system retrieves preset standard maps from the standard library, including: Standard map A: typical fatigue crack characteristics of fine wire; Standard map B: oil stain adhesion characteristics on the surface of fine wire; Standard map C: pitting characteristics of fine wire; The system matches the feature vector of the current project with maps A, B, and C respectively.

[0060] Optionally, the Euclidean distance between the current spectrum and the standard spectrum B (oil stain) is 2.85 (far greater than the threshold of 0.5, which is the preset value), and the SSIM index is only 0.42 (which is the preset value). This indicates that although oil stains can also change the eddy current signal, their spectrum structure is significantly different from that of the current detection object. The Euclidean distance between the current spectrum and the standard spectrum C (pitting pit) is only 0.15, and the SSIM is as high as 0.92. Through kernel mapping analysis, it was found that the current signal is completely consistent with the nonlinear characteristics of pitting pit in terms of phase change rate, while oil stains mainly exhibit linear amplitude decay and do not have this nonlinear phase characteristic. The system comprehensively judges that the feature deviation value of the current detection object is within the confidence interval of "true damage", successfully identifies the defect feature as "pitting pit", and effectively eliminates false alarms caused by residual lubricating oil or rust-preventive oil (surface stains).

[0061] refer to Figure 5 In step S14, the specific steps are as follows: S141: Mark the feature location of each defect feature, and invert its corresponding feature shape by combining electromagnetic inversion method to quantify the length, depth and direction of the defect. At the same time, call the structural shape of the fine wire, and trigger the multi-dimensional mechanical analysis of the fine wire by combining the feature location and corresponding feature shape of each defect feature, and comprehensively evaluate the defect level of each defect feature. S142: Using a sliding weighting method based on a time window, the defect levels of each defect feature within a preset length range are dynamically superimposed along the extension direction of the fine wire, and the distribution density of the defect features is comprehensively considered to obtain a dynamic early warning level that reflects the current quality status of the fine wire in real time. S143: When the dynamic warning level reaches the preset warning level threshold, the wire detector will issue a dynamic warning for the wire and mark the comprehensive defect content within the preset length range. Based on the matching of the comprehensive defect content and the defect maintenance list, the corresponding defect maintenance measures will be determined.

[0062] In the embodiments of this application, the feature positions of each defect feature are marked, and their corresponding feature shapes are inverted by electromagnetic inversion to quantify the length, depth and direction of the defect. At the same time, the structural shape of the fine wire is invoked, and the multi-dimensional mechanical analysis of the fine wire is triggered by combining the feature positions and corresponding feature shapes of each defect feature. The defect level of each defect feature is comprehensively evaluated, and the defect level of each defect feature is comprehensively evaluated.

[0063] At this point, based on the previous time alignment (S121) and spatial decomposition (S131), the system accurately records the absolute coordinates of each defect feature along the axial direction of the wire (e.g., X meters from the head of the wire) and its circumferential position (if a rotating probe or multi-coil layout is used); a forward model (e.g., a coil-wire impedance model established through finite element simulation FEM) is established as a reference; using optimization methods (e.g., gradient descent or genetic methods), the defect geometric parameters (length, depth, orientation, angle) in the model are adjusted so that the simulation signal output by the model is infinitely close to the actual detected feature map; when the error between the simulation signal and the measured signal is minimal, the corresponding model parameters are the inversion result, thereby accurately quantifying the geometric morphology of the defect.

[0064] The system acquires the basic structural parameters of the wire rod to be tested, including material grade (e.g., 65Mn, 70#), tensile strength, diameter (nominal size), moment of inertia of the section, and the current preload or working load. It then triggers multi-dimensional mechanical analysis, including: stress concentration analysis: based on the defect morphology obtained from the inversion (e.g., the aspect ratio of the crack and the radius of curvature at the tip), the stress concentration factor (Kt) at the root of the defect is calculated; residual strength assessment: the effective bearing area after deducting the defect section is calculated, and the static strength residual safety factor of the wire rod is assessed; fracture assessment: for crack-type defects, the stress intensity factor (KI) is calculated using fracture mechanics theory to determine whether the crack is in a critical propagation state.

[0065] The final defect level is determined by integrating geometric dimensions and mechanical analysis results using multiple indicators; the single "size exceeding the limit" judgment logic is abandoned, and a comprehensive scoring model of "geometry + mechanics" is adopted; for example, a sharp crack with a small size but located in a high-stress area has a higher level than a surface scratch with a large size but located in a low-stress area; based on the scoring results, defects are divided into several levels (e.g., Level I - minor / concern, Level II - significant / requires recording, Level III - serious / requires alarm, Level IV - fatal / requires cutting off).

[0066] Specifically, when identifying defects in high-strength galvanized wire (assuming a diameter of 12mm, used for bridge cables), the system marked a defect feature 150 meters from the head of the wire. The system called up the electromagnetic simulation model of "12mm wire - surface crack". By adjusting the model parameters, it was found that when a crack with a length of 10mm and a depth of 1.2mm (approximately 10% of the diameter) perpendicular to the axis was set in the model, the output impedance trajectory perfectly matched the measured spectrum. The system accurately deduced that the defect was a 1.2mm deep transverse surface crack and marked its specific location on the circumference of the wire (e.g., at the 12 o'clock position).

[0067] The system marked a defect feature 150 meters from the head of the fine wire. The system called the electromagnetic simulation model of "12mm fine wire - surface crack". By adjusting the model parameters, it was found that when a crack with a length of 10mm, a depth of 1.2mm (about 10% of the diameter) and perpendicular to the axis was set in the model, the output impedance trajectory perfectly coincided with the measured spectrum.

[0068] The system accurately identified the defect as a 1.2mm deep transverse surface crack and marked its specific location on the circumference of the wire (e.g., at the 12 o'clock position). Stress concentration calculation: Due to the 1.2mm deep notch, the stress concentration factor (Kt) at the crack root is as high as 4.5, which means that although the cross-sectional loss is not large, the local stress peak at the root will reach an extremely high level. Fracture assessment: Combining crack depth and material fracture toughness (KIC), the system calculation found that the stress intensity factor of the crack under working tension is close to the material's critical value, indicating an extremely high risk of propagation.

[0069] Taking into account the geometric dimensions (depth 1.2mm, although not exceeding the conventional scrap threshold such as 5%), high stress concentration factor, and fracture risk, the system determined that the defect belonged to "Level III (Severe)". In contrast, another defect was found 200 meters away from the head, with a length of 50mm and a depth of only 0.2mm. The inversion showed that its bottom was flat and the stress concentration factor was only 1.2. Although it was long, the mechanical risk was low, and the system determined it to be "Level I (Slight)". Through S141, the system no longer just reported "crack found", but directly informed the engineer: "A fatal defect was found at 150 meters. It is recommended to remove it immediately", which achieved a highly valuable dynamic early warning.

[0070] Furthermore, a sliding weighted method based on a time window is adopted to dynamically superimpose the defect levels of various defect features within a preset length range along the extension direction of the refined wire. Taking into account the distribution density of defect features, a dynamic early warning level reflecting the current quality status of the refined wire is obtained in real time. This introduces a dynamic early warning level that reflects the current quality status of the refined wire in real time.

[0071] At this point, the system sets a fixed evaluation window that slides along the timeline as the wire moves; in physical space, this corresponds to a "preset length range" (e.g., a 1-meter or 5-meter length of wire) along the direction of the wire's extension; as the window slides, the system retrieves all defect features falling within the current window in real time; based on the various defect levels obtained in S141 (e.g., Level I = 1 point, Level II = 3 points, Level III = 10 points), a weighted summation is performed; to reflect the different degrees of harm, the weight of higher-level defects is usually much higher than that of lower-level defects; for example, the cumulative contribution of a Level III fatal defect requires ten Level I minor defects to be equivalent.

[0072] Simple defect level superposition ignores the clustering effect of defects. This step introduces spatial distribution density as a correction factor. If there are many defects in the preset length interval and the spacing is very close (high density), even if the individual defect level is not high, the superposition effect will be amplified. The system will calculate the number of defects per unit length (e.g., number of defects per meter) and use it as a gain coefficient to apply to the total superposition score. The weighted superposition score is combined with the density correction coefficient and mapped to a dynamic warning threshold table. The system outputs the current "dynamic warning level" in real time, which reflects the overall quality status of the wire section in the current window (e.g., safe, attention, warning, danger).

[0073] Specifically, when identifying defects in fine wire, the system sets a preset length range of 1 meter; currently, the detection probe is scanning the fine wire, and the time window moves with the fine wire; Scenario A (sparse large defects): within the current 1-meter window, the system detected 1 Class III defect (deep crack) and 2 Class I defects (minor scratches); the system performs weighted summation: Score = (1×10) + (2×1) = 12 points.

[0074] Scenario B (Dense Small Defects): Within another 1-meter window, the system did not detect any Level III defects, but detected 15 densely distributed Level I defects (continuous tearing caused by mold adhesion); the system performs a basic weighting: Score = (15 × 1) = 15 points; due to the extremely high distribution density of 15 defects per meter, the system triggers a density correction coefficient (e.g., coefficient is 1.5), and the final corrected score is: Scorefinal = 15 × 1.5 = 22.5 points.

[0075] The system determines the quality status based on real-time scores: For scenario A, the score is 12 points, exceeding the "warning" threshold (10 points) but below the "danger" threshold (20 points). The system triggers a yellow dynamic warning level, alerting the operator to a major defect in the section of wire. For scenario B, the score is 22.5 points, far exceeding the "danger" threshold. Although individual defects are minor, the high-density continuous tearing severely weakens the effective cross-sectional area of ​​the wire, making it extremely prone to breakage. The system immediately triggers a red dynamic warning level for scenario B and can directly instruct the cutting device at the back end to cut off the section of wire, preventing the wire with severe and dense defects from flowing into the next process.

[0076] Therefore, when the dynamic warning level reaches the preset warning level threshold, the wire detector issues a dynamic warning for the wire and marks the comprehensive defect content within the preset length range. Based on the matching of the comprehensive defect content and the defect maintenance list, the corresponding defect maintenance measures are determined. This takes into account the overall consideration of matching the comprehensive defect content and the defect maintenance list, ensuring the accuracy of the corresponding defect maintenance measures. At the same time, multiple defect features are controlled, and the defect levels of each defect feature are superimposed, improving the phased dynamic detection of the wire and the accuracy of the dynamic warning level, thus realizing the dynamic warning of the wire detector for the wire.

[0077] At this time, the system monitors the dynamic warning level output by S142 in real time; once the value reaches or exceeds the preset threshold (e.g., triggering the red danger level), the system immediately sends a control signal to the actuator (e.g., audible and visual alarm, shutdown relay, inkjet marker) to implement dynamic warning for the fine wire.

[0078] When a warning is triggered, the system will not simply issue an "alarm" signal, but will package and mark all key data within the preset length range that caused the warning. This includes: the start / end coordinates of the defect, the highest defect level, the total number of defects, the dominant defect type (such as crack clusters or wear zones), and the mechanical assessment summary in S141.

[0079] The system utilizes an expert decision-making system to achieve standardized and automated operation and maintenance. The defect maintenance list is a pre-built knowledge base that establishes a mapping relationship between the feature patterns of "comprehensive defect content" and "maintenance measures." The list covers various disposal strategies ranging from "monitoring only" to "must be scrapped." The system compares the currently marked comprehensive defect content with the rules in the maintenance list (using a rule engine or fuzzy matching method). For example: Rule 1: If (defect level = severe and type = fatigue crack) > measure = cut off and scrap; Rule 2: If (defect level = medium and type = surface wear) > measure = downgrade or grind and repair.

[0080] Specifically, in the previous S142 step, the system detected dense surface scratches on the fine wire in section L, and calculated that the dynamic warning level reached "Red Danger Level (Level 4)"; trigger action: the fine wire detector immediately controlled the production line to stop, and triggered the inkjet device at the L section position of the fine wire to spray a conspicuous red marking line, physically isolating the defective product in this section.

[0081] The system automatically generates a record in the console log: "Location: Reel #03, length 1250m-1251m; Comprehensive defect content: 12 Class I surface scratches were found, with extremely high distribution density (high damage area), inversion average depth 0.3mm, mechanical assessment shows the remaining strength has dropped to below 95%."

[0082] The system calls the "Defect Maintenance List" and matches it according to the above characteristics: Matching process: The system retrieves a rule: "If the fine wire has more than 10 dense surface defects within 1 meter, and the inversion depth does not exceed 0.5mm (not damaging the core of the matrix)"; The corresponding measure for this rule is: "Partial removal" - that is, the damaged section is removed, and the remaining part is downgraded to "Grade II" for use, rather than being scrapped entirely; The system outputs the determined defect maintenance measures and prompts the operator on the screen: "Handling suggestion: Remove the 1250m-1251m section; the remaining fine wire is re-inspected and put into storage as Grade B product."

[0083] Please see Figure 6 , Figure 6 This is a schematic diagram of the structural composition of a precision wire defect identification system based on eddy current detection according to an embodiment of the present invention; the precision wire defect identification system based on eddy current detection is applied to the above-mentioned precision wire defect identification method based on eddy current detection; the precision wire defect identification system based on eddy current detection includes: The multi-factor combination module 21 is used to determine the dual-channel detection architecture based on the detection of the fine wire detector, and to combine the absolute coil channel and the differential coil channel in a multi-factor manner, so that the absolute coil channel and the differential coil channel can perform wide-area detection in different dimensions; Eddy current detection module 22 is used to determine multiple fine wire detection data based on the synchronous detection of the absolute coil channel and the differential coil channel, align the multiple fine wire detection data to extract transient pulse features in the time domain and defect frequency response features in the frequency domain, and determine eddy current detection events by multi-factor fusion based on transient pulse features and defect frequency response features. The defect module 23 is used to identify multiple sub-eddy current detection items based on the identification of the eddy current detection event in the eddy current detection event, mark the feature map of each sub-eddy current detection item, and determine multiple defect features by comparing the feature map of each sub-eddy current detection item with the preset standard feature map. The dynamic early warning module 24 is used to determine the defect level of each defect feature based on its feature location, corresponding feature shape, and structural shape of the wire, and to determine the dynamic early warning level by superimposing the defect levels of each defect feature, so as to trigger the wire detector to issue a dynamic early warning for the wire.

[0084] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A method for identifying defects in a fine wire based on eddy current detection, characterized by, include: The dual-channel detection architecture is determined based on the detection of the fine wire detector, and the absolute coil channel and the differential coil channel are combined with multiple factors. The absolute coil channel and the differential coil channel perform wide-area detection in different dimensions. Multiple fine wire detection data are determined based on the synchronous detection of the absolute coil channel and the differential coil channel. The multiple fine wire detection data are aligned to extract transient pulse features in the time domain and defect frequency response features in the frequency domain. Eddy current detection events are determined by multi-factor fusion based on transient pulse features and defect frequency response features. In this eddy current detection event, multiple sub-eddy current detection items are identified based on the identification of the eddy current detection event, and the feature maps of each sub-eddy current detection item are marked. Multiple defect features are determined by comparing the feature maps of each sub-eddy current detection item with the preset standard feature maps. Based on the characteristic location, corresponding characteristic shape, and structural shape of each defect feature, the defect level of each defect feature is determined, and the dynamic warning level is determined by superimposing the defect levels of each defect feature, so as to trigger the dynamic warning of the fine wire detector on the fine wire.

2. The eddy current based fine wire defect identification method according to claim 1, characterized in that, The dual-channel detection architecture is determined based on the detection of the precision wire detector, and the absolute coil channel and differential coil channel are combined by multiple factors. The absolute coil channel and differential coil channel perform wide-area detection in different dimensions, including: In the fine wire detector, multiple detection channels of the fine wire detector are marked, and a corresponding dual-channel detection architecture is constructed based on the multiple detection channels and the fine wire to be detected. In the dual-channel detection architecture, the absolute coil channel and the differential coil channel are determined, and the absolute coil channel and the differential coil channel are triggered to detect the same fine wire.

3. The method of claim 2, wherein the method is characterized by: The dual-channel detection architecture, determined based on the detection of a precision wire detector, combines the absolute coil channel and the differential coil channel using multiple factors. The absolute coil channel and the differential coil channel perform wide-area detection in different dimensions. This also includes: The first layer of detection factors for the absolute coil channel and the second layer of detection factors for the differential coil channel are marked. The first layer of detection factors and the second layer of detection factors are combined in a multi-factor combination, and the detection performance of the absolute coil channel and the differential coil channel at the same position of the fine wire is balanced. At this time, the absolute coil channel is responsible for wide-area monitoring of long gradient defects and material abrupt changes; the differential coil channel is responsible for wide-area monitoring of surface defects.

4. The method for identifying defects in precision wire based on eddy current detection according to claim 1, characterized in that, The process involves determining multiple precision wire detection data points based on synchronous detection of the absolute coil channel and the differential coil channel, aligning these multiple precision wire detection data points to extract transient pulse features in the time domain and defect frequency response features in the frequency domain, and determining eddy current detection events through multi-factor fusion based on the transient pulse features and defect frequency response features. This includes: Synchronous detection is performed on the absolute coil channel and the differential coil channel, and the timestamp synchronization is achieved using a field-programmable gate array to determine multiple fine wire detection data. At the same time, the multiple fine wire detection data are time-aligned based on interpolation to calibrate the phase deviation caused by the delay of the analog front-end circuit or the difference in sampling rate.

5. The method for identifying defects in fine wire based on eddy current detection according to claim 4, characterized in that, The method of determining multiple precision wire detection data based on synchronous detection of the absolute coil channel and the differential coil channel, aligning the multiple precision wire detection data to extract transient pulse features in the time domain and defect frequency response features in the frequency domain, and determining eddy current detection events by multi-factor fusion based on transient pulse features and defect frequency response features, further includes: Based on the aligned data, wavelet packet decomposition is applied to extract transient pulse features in the time domain and capture signal abrupt changes caused by crack edges. At the same time, higher-order spectral analysis is used to extract defect frequency response features for different defect types in the frequency domain and analyze the depth attributes of the defects. The transient pulse features in the time domain and the defect frequency response features in the frequency domain are subjected to multi-factor weighted fusion analysis. The validity of eddy current detection events is determined based on the consistency and significance of the features, and noise caused by mechanical vibration or electromagnetic interference is filtered out.

6. The method for identifying defects in precision wire based on eddy current detection according to claim 1, characterized in that, In this eddy current detection event, multiple sub-eddy current detection items are identified based on the identification of the eddy current detection event, and the feature maps of each sub-eddy current detection item are marked. Multiple defect features are determined by comparing the feature maps of each sub-eddy current detection item with preset standard feature maps, including: The system monitors eddy current detection events in real time, marks the detection data combinations corresponding to each event, identifies these data combinations, and decomposes continuous detection signals into multiple sub-eddy current detection items according to temporal and spatial distribution. Each sub-eddy current detection item is marked with a corresponding potential anomaly region, and each potential anomaly region is isolated and analyzed. The system also marks the time-domain waveform segments and impedance plane trajectories of each sub-detection item, and extracts high-dimensional feature maps using principal component analysis and manifold learning. These feature maps characterize the geometric morphology and electromagnetic response characteristics of defects in the precision wire.

7. The method for identifying defects in precision wire based on eddy current detection according to claim 6, characterized in that, In this eddy current detection event, multiple sub-eddy current detection items are identified based on the identification of the eddy current detection event, and the feature maps of each sub-eddy current detection item are marked. Multiple defect features are determined by comparing the feature maps of each sub-eddy current detection item with preset standard feature maps. The method also includes: Each feature map is input into a convolutional neural network and analyzed by convolutional layers to output the feature maps of each sub-eddy current detection project. At the same time, based on the traceability of the precision wire detector, a preset standard feature map is determined. The feature maps of each sub-eddy current detection project are matched and compared with the preset standard feature map one by one. By calculating the Euclidean distance and structural similarity index between the maps and combining nonlinear mapping to quantify feature deviation, multiple defect features can be identified and surface stains can be effectively distinguished from real scratches.

8. The method for identifying defects in precision wire based on eddy current detection according to claim 1, characterized in that, The method involves determining the defect level of each defect feature based on its location, corresponding shape, and structural morphology, and then determining a dynamic warning level by superimposing the defect levels of each feature to trigger a dynamic warning from the wire detector. This includes: The characteristic locations of each defect feature are marked, and their corresponding characteristic morphology is inverted using electromagnetic inversion to quantify the length, depth, and direction of the defect. At the same time, the structural morphology of the wire is invoked, and the multi-dimensional mechanical analysis of the wire is triggered by combining the characteristic locations and corresponding characteristic morphologies of each defect feature. The defect level of each defect feature is then comprehensively evaluated.

9. The method for identifying defects in fine wire based on eddy current detection according to claim 8, characterized in that, The method of determining the defect level of each defect feature based on its characteristic location, corresponding characteristic shape, and structural shape of the wire, and determining the dynamic warning level by superimposing the defect levels of each defect feature, so as to trigger the dynamic warning of the wire detector for the wire, further includes: Using a sliding weighting method based on a time window, the defect levels of each defect feature within a preset length range are dynamically superimposed along the extension direction of the fine wire, and the distribution density of the defect features is comprehensively considered to obtain a dynamic early warning level that reflects the current quality status of the fine wire in real time. When the dynamic warning level reaches the preset warning level threshold, the wire detector issues a dynamic warning to the wire and marks the comprehensive defect content within the preset length range. Based on the matching of the comprehensive defect content and the defect maintenance list, the corresponding defect maintenance measures are determined.

10. A precision wire defect identification system based on eddy current detection, characterized in that, The eddy current-based precision wire defect identification system is applied to the eddy current-based precision wire defect identification method as described in any one of claims 1-9; the eddy current-based precision wire defect identification system comprises: The multi-factor combination module is used to determine the dual-channel detection architecture based on the detection of the fine wire detector, and to combine the absolute coil channel and the differential coil channel in a multi-factor combination, so that the absolute coil channel and the differential coil channel can perform wide-area detection in different dimensions; The eddy current detection module is used to determine multiple fine wire detection data based on the synchronous detection of the absolute coil channel and the differential coil channel. The multiple fine wire detection data are aligned to extract transient pulse features in the time domain and defect frequency response features in the frequency domain. The eddy current detection event is determined by multi-factor fusion based on the transient pulse features and defect frequency response features. The defect module is used to identify multiple sub-eddy current detection items based on the identification of the eddy current detection event in the eddy current detection event, mark the feature map of each sub-eddy current detection item, and determine multiple defect features by comparing the feature map of each sub-eddy current detection item with the preset standard feature map. The dynamic early warning module is used to determine the defect level of each defect feature based on its characteristic location, corresponding characteristic shape, and structural shape of the wire. It then determines the dynamic early warning level by superimposing the defect levels of each defect feature, thereby triggering the wire detector to issue a dynamic early warning for the wire.

Images