Metal part detection method and system based on pulsed eddy current and storage medium

Abstract

The invention relates to a metal part detection method and system based on pulsed eddy current and a storage medium, and relates to the technical field of defect detection.The metal part detection method comprises the steps that detection data packets of all metal parts are obtained one by one according to timestamps and analyzed, and corresponding original eddy current detection data are obtained; correcting and converting the original eddy current detection data according to a preset baseline value to obtain a two-dimensional detection array containing a vector point X and a vector point Y; generating an eddy current detection signal according to the two-dimensional detection array, extracting defect feature data by combining a preset frequency threshold, and generating an impedance diagram and a time sequence diagram; performing multi-dimensional fusion judgment on the defect feature data, the impedance diagram and the time sequence diagram, and identifying the defects of the metal part; signal restoration and accurate feature extraction are realized through high-pass filtering and sliding window processing, the problems of high-frequency boundary effect and low-frequency track loss are solved, and the completeness and anti-interference capability of defect identification are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of defect detection, in particular to a metal component detection method and system based on pulsed eddy current and a storage medium. BACKGROUND

[0002] Eddy current testing (ET) is a non-destructive testing method based on electromagnetic induction principle, and its detection object is the surface and near-surface defects of conductive materials. Pulsed eddy current (PEC) technology is rapidly developed on the basis of conventional eddy current testing. Unlike the traditional technology which uses a single frequency as excitation, a periodic rectangular pulse is used. The electromagnetic field of the pulse excitation current excites a pulse eddy current in the conductor, and the electromagnetic field generated by the pulse eddy current is received by the sensor. If there is a defect in the conductor, the pulse eddy current in the conductor and the electromagnetic field generated thereby will be different. Since the pulse excitation contains a very wide frequency spectrum from direct current to high frequency (high frequency is generally tens of kHz to hundreds of kHz), the eddy currents formed by different frequencies have different penetration depths in the metal, so that defects at different depths of multi-layer structures can be detected at one time.

[0003] Patent No. CN202410368981.9 discloses a metal component crack depth quantitative detection method and system based on pulsed eddy current and a storage medium. The method comprises the following steps: step S1. Constructing a defect detection model of the metal component to be detected, adjusting the material properties and crack parameters of the metal component in the defect detection model, respectively applying pulse current excitation in the excitation coil of the flexible electromagnetic sensor, simulating and testing the pulse eddy current field generated by the metal component under different detection conditions, extracting the time-frequency domain features of the pulse eddy current detection signal, and forming a time-frequency domain feature dataset; step S2. Training a neural network model to obtain a crack depth detection model; step S3. Real-time detection of the metal component to be detected, inputting the real-time extracted time-frequency domain features into the crack depth detection model for detection to obtain the crack depth value. The above-mentioned application has the advantages of simple operation, low complexity, high detection efficiency and accuracy.

[0004] The prior art in the above has the following defects:

[0005] 1. In the traditional method, when processing high-frequency signals, the feature appears extremely short and may cross the data packet boundary, which easily leads to the truncation of the feature waveform.

[0006] 2. Low-frequency detection lacks adaptive rules, and it is difficult to balance the data volume and the integrity preservation of complex features, resulting in insufficient sensitivity of deep crack detection. SUMMARY

[0007] In view of the deficiencies of the prior art, the present application automatically intercepts feature segments through high-pass filtering and sliding window, combines frame header deduplication and conditional sorting algorithm to ensure the integrity of cross-packet features, improve crack depth detection accuracy, and make up for the deficiencies of traditional time-domain feature methods.

[0008] The following technical solutions are adopted:

[0009] A metal part detection method based on pulse eddy current, comprising:

[0010] According to the timestamp, the detection data packet of each metal part is obtained one by one and parsed to obtain the corresponding original eddy current detection data;

[0011] According to the preset baseline value, the original eddy current detection data is corrected and converted to obtain a two-dimensional detection array containing vector points X and vector points Y;

[0012] According to the two-dimensional detection array, the eddy current detection signal is generated, and the corresponding defect feature data is extracted in combination with the preset frequency threshold, and the corresponding impedance diagram and time sequence diagram are generated in combination with the timestamp;

[0013] The defect feature data, the impedance diagram and the time sequence diagram are fused and judged to obtain the metal part defect corresponding to each detection data packet.

[0014] By adopting the above technical solutions, the efficient acquisition and analysis of metal part detection data packets are realized by customizing a 12-byte data frame protocol (frame header (4 bytes), phase rotation voltage Ur (4 bytes), original voltage Ux (4 bytes)), a two-dimensional detection array containing vector points X and vector points Y is generated in combination with baseline correction (supporting max_min / mean / median decentralization) and big-endian byte analysis (each group of 6 bytes is analyzed as seq, data1, data2), and then based on the preset frequency threshold (>100kHz triggers high-frequency channel, filter window (>0 filters low-frequency part, =0 mainly uses low-frequency signal) dynamically selects the signal processing path: 2000-point window segmentation (ensures the integrity of the transient feature across the data packet) is used for high-frequency signal, and the defect feature is reserved through 4096-point buffer and 8-shaped track double-crossing rule (1x1 square window) for low-frequency signal; finally, the phase distortion is eliminated through zero-phase filtering (second-order Butterworth high-pass filter), and the defect visualization diagnosis is performed by fusing the feature data, impedance spectrum and time sequence diagram. The high-pass filtering and sliding window solve the problem of easy truncation of high-frequency transient features, effectively reduce the risk of missed detection, and realize the closed-loop analysis from signal acquisition to industrial decision-making.

[0015] The present application further provides that: the specific steps of obtaining the detection data packet of each metal part one by one according to the timestamp and parsing to obtain the corresponding original eddy current detection data comprise:

[0016] initializing the Ethernet communication, setting the Internet protocol address and port number of the defect detection end;

[0017] According to the port number, the defect detection end and the data analysis end establish and monitor a plurality of data transmission channels and transmit data streams through the communication protocol;

[0018] According to the preset number of data points, the data analysis end divides the data stream into a plurality of data packets;

[0019] The data packets are parsed to obtain a packet header, a command field, a data length and valid data;

[0020] According to the command field and the packet header, the corresponding data transmission channel is triggered and selected for transmission of the data packet;

[0021] According to the preset data frame format and the data length, the valid data is divided to obtain a frame header, a real part and an imaginary part;

[0022] The frame header is repeatedly detected, if the current frame header is unique, it is determined that the current data packet is not repeated, and the real part and the imaginary part are integrally parsed to generate an original data pair;

[0023] According to the timestamp of the data packet, the corresponding original data pair is sorted to obtain original eddy current detection data.

[0024] By adopting the above technical scheme, the data transmission channel between the detection end and the data analysis end is established through Ethernet initialization (setting IP / port), the data stream is divided into data packets based on the preset number of data points, the double deduplication mechanism (data serial number difference comparison and frame header uniqueness hash check) is adopted to remove repeated frames, the original voltage pair is parsed according to the 12-byte frame format (frame header 4B, real part 4B, imaginary part 4B), and finally the continuous and non-redundant original eddy current detection data is generated according to the nanosecond timestamp. Through the dynamic packet strategy (such as 4096 points per group) and real-time channel monitoring (heartbeat packet feedback), it is ensured that there is no data transmission loss in the high-noise environment of industrial field, and the hardware timestamp synchronization technology is used to solve the timing dislocation problem of multiple devices.

[0025] The application further sets that the specific steps of correcting and converting the original eddy current detection data according to the preset baseline value to obtain a two-dimensional detection array containing vector point X and vector point Y include:

[0026] The original eddy current detection data is calculated by difference with the preset baseline value to obtain standard detection data;

[0027] group the standard detection data according to a preset number of data points, to obtain a plurality of groups of data points;

[0028] repeat detection is performed on the previous group of data points and the current group of data points, to obtain a plurality of non-repeated data points;

[0029] interpolation operation is performed on all the non-repeated data points, and is expanded into a new group of data points with the same number of data points as the preset number of data points;

[0030] the new group of data points is spliced with the current group of data points, and all the standard detection data is operated and aggregated with a preset gain coefficient, to obtain a byte array;

[0031] the byte array is decomposed according to a preset byte bit extraction mechanism in big-endian order, to obtain a plurality of signed integer arrays including a sequence number, a first integer array and a second integer array;

[0032] the byte bit extraction mechanism is to skip the first M bytes of the byte array, and start from the (M+1)th byte, take S bytes as a group of data each time; the remaining number of bytes of the byte array is judged with S, if the remaining number of bytes is less than S, the bit extraction is stopped.

[0033] difference operation and abnormal position marking are performed on the sequence numbers of adjacent data points in the first integer array and the second integer array; if the difference is greater than 1, it indicates that the sequence number of a certain adjacent data point in the current integer array jumps, and the jump position is marked as a starting position; if the difference is less than 0 and greater than a negative span, it indicates that the sequence number of a certain adjacent data point in the current integer array has a fault, and the fault position is marked as a starting position, and the number of cycles without abnormality is counted;

[0034] the sequence number of the abnormal position data point in the current integer array is corrected according to the starting position, to form a new sequence number, and an index array is generated;

[0035] each integer array is sorted according to a sorting type, and each data point in all the integer arrays is decentered according to a data processing type, to obtain a corresponding corrected integer array;

[0036] convolution smoothing processing is performed on each data point in each of the corrected integer arrays, to obtain a corresponding smoothed integer array;

[0037] an amplification factor is calculated according to a gain coefficient, and is operated and merged with each data point in all the smoothed integer arrays, to obtain a vector point X and a vector point Y;

[0038] the vector point X and the vector point Y are combined to generate a two-dimensional detection array.

[0039] By adopting the technical scheme, through the sliding window, the big-endian byte analysis (grouping according to S bytes after skipping M bytes), the adaptive exception correction (jump mark and cycle count based on the difference value of adjacent data points), and the dual-path signal reconstruction (vector point X square root combined with the phase solution of the angle compensation), robust processing and high-precision feature extraction of the eddy current detection data are realized.

[0040] The application is further provided as follows: the specific steps of generating an eddy current detection signal according to the two-dimensional detection array, extracting corresponding defect feature data in combination with a preset frequency threshold, and generating a corresponding impedance graph and timing graph in combination with the timestamp include:

[0041] Feature extraction is performed on the vector point X and the vector point Y in the two-dimensional detection array to obtain a modulus and a phase angle, and an eddy current detection signal is generated;

[0042] The eddy current detection signal is decomposed according to a preset frequency threshold to obtain a high-frequency signal and a low-frequency signal, and a corresponding signal dedicated processing channel is triggered;

[0043] The signal dedicated processing channel performs data extraction on the high-frequency signal according to a preset fixed window parameter to obtain a plurality of groups of signal values;

[0044] On the one hand, superposition operation is performed on each group of signal values according to a preset dynamic step to obtain overlapping signal data;

[0045] Amplitude operation and comparison are performed on the overlapping signal data to obtain a maximum amplitude;

[0046] An impedance graph and a timing graph are generated according to the overlapping signal data in combination with a finite impulse response function;

[0047] On the other hand, the value fluctuation of the last T bits of the signal value is calculated and compared with a preset fluctuation threshold, and if the value fluctuation is greater than the fluctuation threshold, it indicates that the signal value contains defect feature data, and the defect feature data is segmented;

[0048] An impedance graph and a timing graph are generated according to the defect feature data in combination with a finite impulse response function;

[0049] The signal dedicated processing channel performs in-out frequency statistics on the low-frequency signal within a preset window point number through a square window, and compares the in-out frequency with a preset in-out threshold to determine whether the signal feature corresponding to the square window is defect feature data; if the in-out frequency is greater than the in-out threshold, the defect feature data is segmented to generate the impedance graph.

[0050] By adopting the technical scheme, the complex eddy current detection signal is generated by performing operation on the vector point X and the vector point Y in the two-dimensional detection array, and the signal is high and low frequency decomposed based on a preset frequency threshold (such as high frequency characteristics are extracted by high pass filtering, and low frequency characteristics are extracted by low pass filtering) : the data is extracted by using a preset window parameter for the high frequency signal, and the impedance graph and the time sequence graph are generated by combining the finite impulse response function with the sliding window; the low frequency signal is counted in and out within a preset point number by a square window (such as a 1*1 window), and if the number is greater than a threshold, it is determined that the defect feature exists, and the impedance graph is generated. The signal is restored and the feature is accurately extracted by high pass filtering and sliding window processing, effectively solving the problems of high frequency boundary effect and low frequency track loss, and improving the integrity and anti-interference ability of defect recognition.

[0051] The application further provides that: the specific steps of fusing the defect feature data, the impedance graph and the time sequence graph to obtain the metal part defect corresponding to each detection data packet include:

[0052] According to the pulse frequency f and the material permeability μ, the penetration depth δ is calculated;

[0053]

[0054] Wherein, σ is the conductivity, and k is the frequency calibration coefficient;

[0055] According to the skin effect and the shape of the material, the impedance function of different shape materials is calculated;

[0056] If the shape of the material is a semi-infinite flat plate, the impedance function is

[0057] Z=(1+j)*(A / (σ*δ*l))

[0058] Wherein, Z is the impedance function, A is the cross-sectional area of the eddy current path, and l is the length of the eddy current path;

[0059] If the shape of the material is a flat plate with a finite thickness, the impedance function is

[0060] Z=(1+j) / (σ*δ)*coth[(1+j)d / δ]

[0061] Wherein, coth is the hyperbolic cotangent function, and d is the thickness;

[0062] If the shape of the material is a cylinder, the impedance function is

[0063] Z=(j*ω*μ / (2π*a))*(J0(n*a) / J1(n*a))

[0064] Wherein, ω is the angular frequency, a is the radius, n is the wave number, J0 is the zero order Bessel function, J1 is the first order Bessel function;

[0065] According to the phase change of the phase angle, the impedance graph is characterized, the impedance trajectory is obtained, and the area of the impedance trajectory is calculated;

[0066] If the phase change is zero, that is, the phase angle is constant, then the impedance trajectory is a radial offset;

[0067] If the phase change is less than or equal to x degrees, then the impedance trajectory is stretched in the real axis direction;

[0068] If the phase change is greater than x degrees and less than or equal to y degrees, then the impedance trajectory is offset or compressed in the imaginary axis direction;

[0069] If the phase change is greater than y degrees, then the impedance trajectory is a whole trajectory translation or phase lag;

[0070] The time sequence graph is analyzed, and the defect length L is calculated;

[0071] L=v·Δt

[0072] Wherein, v is the detection speed, and Δt is the abnormal duration;

[0073] According to the defect type, the feature sensitivity of the impedance trajectory area, the phase change, the abnormal duration and the waveform steepness is obtained, each feature is assigned a weight, and a feature threshold interval is set; The waveform steepness is the maximum change rate of the modulus in the defect section;

[0074] The feature weights are weighted and normalized to obtain a defect score;

[0075] The defect score is compared with a preset defect determination threshold, the dominant feature weight is extracted, and the corresponding feature threshold interval is compared, if the defect score is greater than or equal to the defect determination threshold, and the dominant feature weight is located in the feature threshold interval, then the defect type is directly determined;

[0076] Otherwise, the secondary feature weight is extracted and compared with the corresponding feature threshold interval, if the secondary feature weight is located in the feature threshold interval, then the defect type is comprehensively determined; Otherwise, it is marked as an unknown defect;

[0077] Further comprising a scheme: if the phase change amount is greater than y degrees, determining that the defect type is a deep defect; otherwise, when the abnormal duration is greater than a preset time threshold, and the impedance trajectory area is greater than a preset trajectory area threshold, determining that the defect type is a surface crack, and determining the crack boundary in combination with the waveform steepness, otherwise determining that the defect type is a conductivity anomaly; when the abnormal duration is less than or equal to the time threshold, determining that the defect type is a pitting or small hole defect.

[0078] By adopting the technical scheme, four-dimensional features of a phase change amount driven impedance trajectory form (radial deviation, real axis stretching, virtual axis compression, and overall translation), a defect segment abnormal duration, an impedance trajectory area, and a waveform steepness are extracted by establishing a penetration depth model and an impedance function based on material magnetic permeability, pulse frequency, and a shape factor (semi-infinite plate, finite thickness plate, and cylinder); a hierarchical weighted decision mechanism (dominant feature threshold interval matching combined with defect score threshold comparison) is adopted to realize high-precision classification of surface cracks (long time, large area, and steep boundary), deep defects (large phase lag), pitting / small holes (short time), and conductivity anomalies, and to deeply integrate physical models, feature sensitivity, and adaptive decision logic, thereby significantly improving the robustness and generalization ability of defect identification.

[0079] In a second aspect, the present application also provides a metal component detection system based on pulsed eddy current, which adopts the following technical scheme:

[0080] A metal component detection system based on pulsed eddy current, comprising:

[0081] A connection configuration module is configured to connect a defect detection end and a data analysis end, establish and monitor a data transmission channel;

[0082] A signal acquisition module is configured to acquire and parse detection data packets of each metal component one by one according to a timestamp through the data transmission channel, and obtain corresponding original eddy current detection data;

[0083] A signal processing module is configured to correct and convert the original eddy current detection data according to a preset baseline value, and construct a two-dimensional detection array;

[0084] A feature extraction module is configured to generate an eddy current detection signal according to the two-dimensional detection array, and extract defect feature data;

[0085] An impedance decomposition module is configured to separate and process the phase and amplitude of the eddy current detection signal, analyze the impedance characteristics of the metal component, and generate an impedance diagram and a timing diagram;

[0086] A defect identification module is configured to determine the defect of the metal component by the impedance diagram according to a preset frequency threshold, and determine the position of the defect by the timing diagram.

[0087] By adopting the technical scheme, the data transmission channel between the defect detection end and the data analysis end is established and monitored in real time by the connection configuration module; the signal acquisition module acquires and analyzes the detection data packet of the metal part, and extracts the original eddy current detection data; the signal processing module converts the data by applying a baseline correction algorithm, and constructs a two-dimensional detection array; the feature extraction module extracts defect feature data from the eddy current detection signal in combination with a preset frequency threshold, including impedance trajectory morphology analysis based on a phase change amount, impedance trajectory area quantization energy loss, defect length calculation based on an abnormal duration, and waveform steepness evaluation of boundary sharpness; the impedance decomposition module separates the phase and amplitude of the signal, and generates an impedance graph and a timing graph by using a penetration depth algorithm and an impedance function for different material shapes; the defect recognition module implements multi-dimensional fusion judgment, calculates a defect score by a weighted summation normalization algorithm, and performs hierarchical decision based on a feature threshold interval (such as a phase change amount threshold and a time threshold) (when the dominant feature meets the threshold, the defect type is directly determined, otherwise, the secondary feature is reviewed or marked as unknown), so that the detection accuracy and efficiency are significantly improved, and the defect types such as surface cracks, pitting, and conductivity abnormalities are effectively distinguished, and the misjudgment rate is reduced and the system robustness is enhanced.

[0088] In a third aspect, the present application also provides an electronic device, which adopts the following technical scheme:

[0089] An electronic device comprises:

[0090] one or more processors;

[0091] a memory for storing one or more programs;

[0092] When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described above.

[0093] By adopting the above technical scheme, the above metal part detection method based on pulsed eddy current is presented in the form of computer readable code and stored in the memory, and when the processor runs the computer readable code in the memory, the steps of the above metal part detection method based on pulsed eddy current are executed to obtain the effect of reducing labor intensity and improving the degree of automation.

[0094] In a fourth aspect, the present application also provides a computer storage medium, which adopts the following technical scheme:

[0095] A computer storage medium has a computer program stored thereon, and the program is executed by a processor to implement the method as described above.

[0096] In summary, the beneficial technical effects of the present application are:

[0097] 1. By high-pass filtering and sliding window automatic interception of feature segments, combined with frame header deduplication and conditional sorting algorithm to ensure the integrity of cross-packet features, to improve the accuracy of crack depth detection.

[0098] 2. By square window and double-crossing rule to identify special-shaped trajectories, to realize significant separation of subsurface defects and corrosion defects, and to make up for the shortcomings of traditional time-domain feature methods.

[0099] 3. By high-pass filtering and sliding window processing to realize signal restoration and accurate feature extraction, effectively solving the problems of high-frequency boundary effect and low-frequency trajectory loss, and improving the integrity and anti-interference ability of defect recognition. BRIEF DESCRIPTION OF DRAWINGS

[0100] Figure 1 is a flowchart of a metal part detection method according to an embodiment of the present application.

[0101] Figure 2 is a flowchart of a metal part detection method according to an embodiment of the present application.

[0102] Figure 3 is a flowchart of a metal part detection method according to an embodiment of the present application.

[0103] Figure 4 is a structural diagram of an impedance decomposition module according to an embodiment of the present application. DETAILED DESCRIPTION

[0104] The present application will be further described in detail below with reference to the accompanying drawings.

[0105] REFERENCE Figure 1 A metal part detection method based on pulsed eddy current is disclosed, comprising:

[0106] S1: Obtain and parse the detection data packet of each metal part according to the timestamp to obtain the corresponding original eddy current detection data;

[0107] S2: Correct and convert the original eddy current detection data according to the preset baseline value to obtain a two-dimensional detection array containing vector points X and vector points Y;

[0108] S3: According to the two-dimensional detection array, generate an eddy current detection signal, and combine a preset frequency threshold to extract corresponding defect feature data, and combine the timestamp to generate a corresponding impedance diagram and a timing diagram;

[0109] S4: Fuse and judge the defect feature data, the impedance diagram and the timing diagram to obtain the metal part defect corresponding to each detection data packet.

[0110] The implementation principle of the embodiment is: the amplitude / phase original data of a metal part is synchronously collected by a multi-frequency eddy current probe, after environmental interference is eliminated by a temperature compensation module, a two-dimensional vector point X-vector point Y array is generated through dynamic baseline correction; a dedicated processing channel is triggered based on a preset frequency threshold, electromagnetic noise is stripped and defect time-frequency features are extracted by wavelet transform, and an impedance plane graph and a time sequence fluctuation graph are synchronously generated; the above features are input into a pre-trained CNN model for joint analysis of spatial topology and time sequence evolution, the depth, length and type probability of the defect are quantified; finally, real-time data and a historical defect library are fused for confidence evaluation, detection parameters are dynamically optimized through a closed-loop feedback mechanism, and an intelligent diagnosis report containing a three-dimensional defect graph and repair suggestions is output.

[0111] Embodiment two:

[0112] The specific steps of step S1 include:

[0113] Ethernet communication is initialized, and the Internet protocol address and port number of the defect detection end are set;

[0114] According to the communication protocol, the defect detection end and the data analysis end establish and monitor a plurality of data transmission channels, and transmit data streams according to the port number;

[0115] According to the preset number of data points, the data analysis end divides the data stream into a plurality of data packets;

[0116] In this embodiment, two groups of data sequence numbers data1 and data2 are input; data1 is traversed to find the first position greater than data2[i]; the difference num=data1[i]-data2[i] is calculated; if num>340, it means no repetition, and 340 is returned; otherwise, the actual difference num is returned.

[0117] The data packets are parsed to obtain a packet header, a command field, a data length and valid data;

[0118] According to the command field and the packet header, the corresponding data transmission channel is triggered and selected, and the data packet is transmitted;

[0119] According to the preset data frame format and the data length, the valid data is divided to obtain a frame header, a real component and an imaginary component;

[0120] In this embodiment, the data frame format includes 12 bytes: frame header: 4 bytes, (ADI acquisition conversion) phase rotation 90 degree voltage waveform discrete value (Ur): 4 bytes and (ADI acquisition conversion) original voltage waveform discrete value (Ux): 4 bytes;

[0121] Each TCP packet contains 340 groups of data, and the four bytes of the frame header are used to check the data and ensure that no packets are lost.

[0122] The current data packet is determined to be non-repetitive if the current frame header is unique, and the real part and the imaginary part are analyzed as a whole to generate an original data pair;

[0123] The corresponding original data pairs are sorted according to the timestamps of the data packets to obtain original eddy current detection data.

[0124] The implementation principle of the embodiment is as follows: first, the communication connection between the defect detection end and the data analysis end is initialized through the TCP protocol, the IP address and the port number are set, and the heartbeat monitoring mechanism is established; during data transmission, the sending end packages the original voltage waveform (Ur and Ux obtained after AD conversion) collected by the eddy current sensor according to 12 bytes / group (frame header 4 bytes containing checksum, Ur 4 bytes, Ux 4 bytes), and each 340 groups form a TCP packet and are attached with a timestamp and a serial number; the receiving end ensures reliability through ACK confirmation and timeout retransmission, performs CRC check and frame header uniqueness detection on the received data packets, discards repetitive packets and buffers out-of-order packets; after reorganization according to the serial number, the data is sorted according to the timestamp, and the data continuity is verified through difference calculation, and finally the original data pair (Ur, Ux) without repetition and arranged in chronological order is output, providing a high-integrity data basis for subsequent eddy current signal processing.

[0125] Embodiment Three

[0126] The specific steps of step S2 include:

[0127] The original eddy current detection data is subjected to difference calculation with the preset baseline value to obtain standard detection data;

[0128] The standard detection data is grouped according to the preset number of data points to obtain a plurality of groups of data points;

[0129] The previous group of data points and the current group of data points are subjected to repetitive detection to obtain a plurality of non-repetitive data points;

[0130] The non-repetitive data points are subjected to interpolation operation and are expanded into a new group of data points with the same number of data points as the data points;

[0131] The new group of data points and the current group of data points are spliced, and the entire standard detection data and the preset gain coefficient are operated and aggregated to obtain a byte array;

[0132] In the embodiment, the original eddy current detection data is subtracted by the baseline values base_x and base_y;

[0133] Group processing, 340 points per group; the first group is not interpolated; from the second group, use overlap_num() to find the number of non-repeating points with the previous group; linear interpolation is performed on these points, which are expanded to 340 points; the interpolated data is spliced with the current group; finally, all data is multiplied by factor / (10^2.3); return the processed complete data set.

[0134] According to the preset byte bit extraction mechanism, the byte array is decomposed in big-endian sequence to obtain a plurality of signed integer arrays including a sequence number, a first integer array and a second integer array;

[0135] The byte bit extraction mechanism is to skip the first M bytes of the byte array, and start from the M+1th byte, take S bytes as a group of data each time; judge the remaining byte number of the byte array with S, if the remaining byte number is less than S, stop taking bits.

[0136] Difference operation and abnormal position marking are performed on the sequence numbers of adjacent data points in the first integer array and the second integer array; if the difference is greater than 1, it indicates that the sequence number of a certain adjacent data point in the current integer array jumps, and the jump position is marked as the starting position; if the difference is less than 0 and greater than the negative span, it indicates that the sequence number of a certain adjacent data point in the current integer array has a fault, and the fault position is marked as the starting position, and the number of cycles without abnormality is counted;

[0137] Extract data from the byte array, and parse each group of 6 bytes as three 16-bit signed integers (seq, data1, data2) in big-endian sequence; input a byte array byte_array.

[0138] Each group of data is parsed in order: seq: the first 2 bytes (big-endian) → 16-bit signed integer; data1: the middle 2 bytes (big-endian) → 16-bit signed integer; data2: the last 2 bytes (big-endian) → 16-bit signed integer; if the remaining byte number is less than 6, stop processing; save all parsing results as a tuple list of (seq, data1, data2) and return.

[0139] The improved sorting index adjustment adds cycle count statistics to more accurately determine whether a fault or mutation occurs. Traverse the difference between adjacent data points in data; if a jump or fault is found, record the starting position start; count the number of cycles without abnormality; if there is an abnormality (i.e. the number of cycles is not equal to m-1): reconstruct the index order; otherwise, keep the original order; return the rearranged index array.

[0140] Sort the index of the input one-dimensional array data to handle sequence number mutation or fault problems. Traverse the difference between adjacent data points in data:

[0141] If the difference is greater than 1, it indicates that there is a sequence number jump (such as jumping from 32767 to 0), and the start position is recorded.

[0142] If the difference is <0 and >-32767, it indicates a sequence number break (e.g., jumping from 3000 to 2800), and the position "start" is recorded; otherwise, continue traversing.

[0143] If start is found: construct a new index order: first take start+1 to the end, then continue from the beginning to start; otherwise, keep the original order; return the rearranged index array.

[0144] Perform a second-order Butterworth high-pass filter on the signal s. Set the sampling rate fs and the cutoff frequency f; use scipy.signal.butter to generate the second-order high-pass filter coefficients; apply zero-phase filtering (filtfilt) to the signal s to avoid phase distortion; return the filtered signal.

[0145] Based on the starting position, the sequence number of the abnormal position data point in the current integer group is corrected to form a new sequence number, and a corresponding index array is generated;

[0146] Sort each integer group according to the sorting type, and decentralize each data point in all integer groups according to the data processing type to obtain the corresponding corrected integer group;

[0147] In this embodiment, the amplification factor factor = 10^(db / 20) is calculated based on the gain coefficient db; the data is divided into groups of 340 rows each; and each group of data is processed as follows:

[0148] If the sort type is "normal": sort in ascending order by serial number;

[0149] If the sort type is "specific": use sort() to rearrange the indices;

[0150] If the data processing type is "max_min": subtract the average of the maximum and minimum values;

[0151] If the data processing type is "mean": subtract the mean;

[0152] If the data processing type is "median": subtract the median;

[0153] Perform convolution smoothing on each data point in each of the corrected integer groups to obtain the corresponding smoothed integer groups;

[0154] The amplification factor is calculated according to the gain coefficient, and is operated and combined with each data point in all the smooth integer arrays to obtain vector point X and vector point Y;

[0155] In the embodiment, if the smooth is enabled (smooth=True), a moving average (convolution) is applied to each data point in the smooth integer array respectively; finally, the data is multiplied by factor / (10^2.25); and all the processed data is combined into a two-dimensional array and returned.

[0156] The vector point X is combined with the vector point Y to generate a two-dimensional detection array.

[0157] In the embodiment, the input is a two-dimensional array data, each row containing two numerical values (such as x and y); for each row: calculate the vector point X: sqrt(x^2+y^2); calculate the vector point Y: if y≥0: directly use arctan2(y,x); if y<0: add 360° to ensure that the vector point Y is in the range of [0, 360); return a new array containing the vector point X and the vector point Y.

[0158] The implementation principle of the embodiment is as follows: first, the original eddy current detection data is differentiated from the baseline value to obtain standard data, and the data continuity is ensured through dynamic overlapping grouping and interpolation algorithm (optional linear / spline mode); then, an abnormality detection mechanism with dynamic tolerance is used to correct the serial number jump / fault, and a Butterworth high-pass filter with a configurable order is used to eliminate low-frequency noise; then, convolution smoothing and uniform gain compensation are performed on the gain normalized data to generate a two-dimensional smooth array; finally, a two-dimensional detection array containing amplitude and phase characteristics is output through vector point X square root operation and vector point Y calculation (dynamic K value anti-winding) with adaptive angle, and the data boundary and memory safety are strictly checked in the process, so that the high-precision extraction of eddy current defect features is realized.

[0159] Embodiment four:

[0160] The specific steps of step S3 include:

[0161] The vector point X and the vector point Y in the two-dimensional detection array are subjected to feature extraction to obtain a modulus and a phase angle, and an eddy current detection signal is generated;

[0162] According to the preset frequency threshold, the eddy current detection signal is decomposed to obtain a high-frequency signal and a low-frequency signal, and a corresponding signal special processing channel is triggered;

[0163] The signal special processing channel extracts data from the high-frequency signal according to a preset fixed window parameter to obtain a plurality of groups of signal values.

[0164] In one aspect, the signal values of each group are superimposed according to a preset dynamic step to obtain overlapped signal data;

[0165] The overlapped signal data is subjected to amplitude operation and comparison to obtain a maximum amplitude;

[0166] The overlapped signal data is combined with a finite impulse response function to generate an impedance graph and a timing graph;

[0167] In another aspect, the numerical fluctuation of the last T bits of the signal values is calculated and compared with a preset fluctuation threshold, and if the numerical fluctuation is greater than the fluctuation threshold, it indicates that the signal values contain defect feature data, and the defect feature data is segmented;

[0168] The defect feature data is combined with a finite impulse response function to generate an impedance graph and a timing graph;

[0169] In this embodiment, according to the detection workpiece mode, there are high-frequency and low-frequency parts; during the detection process, the high-frequency feature appears for a very short time, so a window with a length of 2000 is used to display the feature, which can generally cover it; as just mentioned, the feature can be distributed in any adjacent group of bytes, so there is a preprocessing link in the algorithm, when the data with a length of 2000 is obtained, the numerical fluctuation of the last 1 / 4 bits is checked, if the fluctuation is large, it indicates that the feature data is contained, and this part of feature data is cut off for processing in the next window, which can ensure the integrity of the feature; z=R+jX; only the low-frequency part needs to be filtered out, a second-order high-pass filter is used, the sampling frequency is 300k, and the cutoff frequency is given by the filter parameters; parameter filtering: indicates the lower limit of the window filtering, if the parameter value is greater than 100, it basically shows the high-frequency feature; if it is 0, it shows the low-frequency feature.

[0170] The signal-specific processing channel counts the number of times of entering and exiting within a preset window point number through a square window for the low-frequency signal, and compares it with a preset entering and exiting threshold to determine; if the number of times of entering and exiting is greater than the entering and exiting threshold, the signal feature corresponding to the square window is determined as defect feature data, and the defect feature data is segmented to generate the impedance graph;

[0171] In this embodiment, the span of the low-frequency feature depends on the speed of the workpiece passing through the probe, and generally a window of 4096 can cover it; there is no preprocessing link in this, because there is a higher degree of freedom for feature drawing, so a set of rules is set to ensure that the complete feature can be drawn; the screen is cleared once every 4096 points, and some features are generally 8-shaped, so a 1*1 square window is set, and the entering and exiting window is greater than twice within 4096 points, and the graph is retained.

[0172] The implementation principle of the embodiment is: after Euler transformation is performed on a two-dimensional detection array to generate an eddy current signal, the eddy current signal is decomposed into high-frequency and low-frequency signals according to a frequency threshold value:

[0173] The high-frequency channel extracts a signal through a dynamic window (length = sampling rate * characteristic time length), detects fluctuations in the whole window by using a sliding standard deviation (threshold value is dynamically updated), combines a ring buffer to divide defect characteristics, and corrects a finite impulse response function based on a skin depth model to generate an anti-aliasing impedance graph and a time sequence graph.

[0174] The low-frequency channel uses a multi-scale square window array (0.5*0.5-2*2) to count the number of crossings and the proportion of stopping points, combines adjacent defect characteristics through continuity verification, and generates an impedance graph based on electromagnetic theory.

[0175] Both channels introduce real-time optimization strategies (FFT template / skip sampling) and GPU acceleration to realize millisecond-level defect feature extraction and accurate reconstruction of physical properties.

[0176] Embodiment five:

[0177] The specific steps of step S4 include:

[0178] According to the pulse frequency f and the magnetic permeability μ of the material, the penetration depth δ is calculated.

[0179]

[0180] Where σ is the electrical conductivity, and k is the frequency calibration coefficient.

[0181] According to the skin effect and the shape of the material, the impedance function of the material of different shapes is calculated.

[0182] If the shape of the material is a semi-infinite flat plate, the impedance function is

[0183] Z = (1 + j) * (A / (σ * δ * l))

[0184] Where Z is the impedance function, A is the cross-sectional area of the eddy current path, and l is the length of the eddy current path.

[0185] If the shape of the material is a flat plate with a finite thickness, the impedance function is

[0186] Z = (1 + j) / (σ * δ) * coth [(1 + j) d / δ]

[0187] Where coth is the hyperbolic cotangent function, and d is the thickness.

[0188] If the shape of the material is a cylinder, the impedance function is

[0189] Z = (j * omega * mu / (2 * pi * a)) * (J0(n * a) / J1(n * a))

[0190] wherein omega is an angular frequency, a is a radius, n is a wave number, J0 is a zero-order Bessel function, and J1 is a first-order Bessel function;

[0191] According to the phase change of the phase angle, the impedance graph is characterized, the impedance trajectory is obtained, and the area of the impedance trajectory is calculated;

[0192] If the phase change is zero, that is, the phase angle is constant, then the impedance trajectory is a radial offset;

[0193] If the phase change is less than or equal to x degrees, then the impedance trajectory is a real axis direction stretching;

[0194] If the phase change is greater than x degrees and less than or equal to y degrees, then the impedance trajectory is a virtual axis direction offset or compression;

[0195] If the phase change is greater than y degrees, then the impedance trajectory is a whole trajectory translation or phase lag;

[0196] The time sequence graph is analyzed, and the defect length L is calculated;

[0197] L = v * Delta t

[0198] wherein v is a detection speed, and Delta t is an abnormal duration;

[0199] According to the defect type, the feature sensitivity of the impedance trajectory area, the phase change, the abnormal duration, and the waveform steepness is obtained, each feature is assigned a weight, and a feature threshold interval is set; the waveform steepness is a maximum change rate of a modulus in a defect section;

[0200] The feature weights are weighted and normalized to obtain a defect score;

[0201] The defect score is compared with a preset defect determination threshold, the dominant feature weight is compared with the corresponding feature threshold interval, if the defect score is greater than or equal to the defect determination threshold, and the dominant feature weight is located in the feature threshold interval, then the defect type is directly determined;

[0202] Otherwise, the secondary feature weight is compared with the corresponding feature threshold interval, if the secondary feature weight is located in the feature threshold interval, then the defect type is comprehensively determined; otherwise, it is marked as an unknown defect;

[0203] Also included in the scheme: if the phase change amount is greater than y degrees, the defect type is determined to be a deep defect; otherwise, when the abnormal duration is greater than a preset time threshold, and the impedance trajectory area is greater than a preset trajectory area threshold, the defect type is determined to be a surface crack, and the crack boundary is determined in combination with the waveform steepness, otherwise the defect type is determined to be a conductivity anomaly; when the abnormal duration is less than or equal to the time threshold, the defect type is determined to be a pitting or small hole defect.

[0204] The embodiment also includes:

[0205] Referring to Figure 2 Ethernet configuration starts, after data reception, deduplication / conversion, feature data extraction and impedance diagram display are performed for high-frequency signals greater than 100 kHz, and a zero-phase filter Sfiltered=(H·(H·S) is designed through a transfer function. rev ) rev Signal optimization is realized, and finally feature data, impedance diagrams and timing diagrams are output;

[0206] Referring to Figure 3 Ethernet communication configuration as a starting point, if it fails, it is reinitialized, if it succeeds, the original detection data is received and denoising and physical quantity conversion are performed; through frequency threshold judgment (>100kHz) to trigger high-frequency signal exclusive processing channel, automatically extract characteristic values such as crack depth / corrosion rate and generate impedance spectrum; simultaneously start adaptive filter design vector point X block—according to the filter type selected by the user (such as Butterworth / chebyshev), combined with the preset order, sampling rate and frequency upper and lower limit parameters, dynamically generate the finite impulse response function h(n)=hdn(n)×w(n) (where hdn(n) is the ideal filter coefficient, w(n) is the window function coefficient), and then eliminate phase distortion through zero-phase filtering technology; finally, multi-dimensional defect visualization diagnosis is performed by fusing feature data, impedance diagrams and timing diagrams, realizing closed-loop analysis from signal acquisition to industrial decision-making.

[0207] The implementation principle of the embodiment is: based on the pulse frequency f, the magnetic permeability μ and the conductivity σ, the dynamic penetration depth δ is calculated (including the frequency calibration coefficient k and the shape factor η), the impedance function Z is constructed for semi-infinite plate / finite thickness plate / cylinder respectively, and the edge effect, thickness tolerance and curvature attenuation correction are introduced; the phase change amount Δφ is analyzed in four stages: when Δφ=0, it is judged as radial offset (uniform corrosion), Δφ≤30° corresponds to real axis stretching (shallow crack), 30°<Δφ≤90° triggers virtual axis compression (subsurface inclusion), and Δφ>90° is identified as trajectory translation and hysteresis (deep defect); in the defect classification stage, the weights of impedance trajectory area, phase change amount, abnormal duration Td and waveform steepness Sd are dynamically allocated according to the defect type, the normalized score is obtained by weighted summation, and the defect type is determined in combination with the dominant feature threshold interval; at the same time, the signal is optimized by using the Hann window finite impulse response (FIR) and bidirectional zero-phase filtering technology, and finally the three-dimensional fusion visualization results of impedance diagram, time sequence diagram and skin depth thermal diagram are generated, realizing the closed-loop processing from electromagnetic parameter calculation to defect intelligent diagnosis.

[0208] Embodiment six:

[0209] A metal part detection system based on pulse eddy current, comprising:

[0210] The connection configuration module is used for connecting the defect detection end and the data analysis end, establishing and monitoring the data transmission channel;

[0211] The signal acquisition module is used for acquiring and analyzing the detection data packet of each metal part according to the timestamp through the data transmission channel, and obtaining the corresponding original eddy current detection data;

[0212] The signal processing module is used for correcting and converting the original eddy current detection data according to the preset baseline value, and constructing a two-dimensional detection array;

[0213] The feature extraction module is used for generating an eddy current detection signal and extracting defect feature data according to the two-dimensional detection array;

[0214] The impedance decomposition module is used for separating and processing the phase and amplitude of the eddy current detection signal, analyzing the impedance characteristics of the metal part, and generating an impedance diagram and a time sequence diagram;

[0215] In the embodiment:

[0216] Reference Figure 4, the eddy current probe emits a high-frequency electromagnetic field to the measured object, and receives an eddy current induced signal caused by changes in the conductivity, magnetic conductivity, etc. of the measured object, denoted as Ux; the signal Ux first passes through a preamplifier, which is used to amplify the weak induced signal to meet the processing requirements of the subsequent circuit. The carrier oscillator generates a reference signal, which usually includes: Uo: the original carrier signal; UR: the in-phase carrier for reference; Us: another reference signal for subsequent processing; Urs: a modulation or control signal; the 90° phase shifter performs orthogonal phase separation on the reference signal to generate a reference signal with a phase difference of 90° from Uo; the signals Ux and the reference signal (UR or the signal after 90° phase shift) are respectively input into a phase-sensitive detector, and through phase comparison, the in-phase component (In-phase, I) and the quadrature component (Quadrature, Q) of the signal are extracted. By adjusting the phase of the reference signal through the adjustable phase shifter, the phase difference of the measured signal can be accurately matched, and the measurement accuracy can be improved. By combining the I and Q components, the impedance amplitude and phase of the measured object can be further calculated to realize the analysis of the material properties (such as conductivity, magnetic conductivity, etc.).

[0217] A defect identification module is configured to determine the defect of the metal part by judging the impedance diagram according to a preset frequency threshold and determining the position of the defect by a time sequence diagram.

[0218] The implementation principle of the embodiment is: a stable link between the detection end and the data analysis end is established through a dual-channel redundant connection configuration module (TCP main channel + UDP backup channel), real-time channel health is calculated, and automatic switching is triggered; the signal acquisition module dynamically adjusts the sampling rate based on the phase change amount Δφ (Δφ > 90°, fs = 10fmax), and uses Hamming error correction code to ensure data integrity; the signal processing module performs temperature compensation baseline correction on the original eddy current signal and outputs a two-dimensional detection array; the feature extraction module extracts defect features in combination with a frequency threshold of > 100 kHz; the impedance decomposition module generates high-precision impedance diagrams and time sequence diagrams through multi-frequency point weighted fusion and phase unwrapping algorithms; the defect identification module realizes multi-dimensional defect classification according to a three-dimensional decision matrix (impedance trajectory area logarithmic ratio, phase-time length square root relationship, waveform steepness speed constraint), and finally forms a closed-loop system from signal acquisition to intelligent diagnosis.

[0219] An electronic device comprising:

[0220] one or more processors;

[0221] a memory storing one or more programs;

[0222] When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described above.

[0223] A computer storage medium having stored thereon a computer program which, when executed by a processor, implements the method as described above.

[0224] The embodiments of the specific implementation are the preferred embodiments of the present application, not limited to the protection scope of the present application, so: any equivalent changes made according to the structure, shape, principle of the present application should be covered within the protection scope of the present application.

Claims

1. A method for detecting metal components based on pulsed eddy currents, characterized in that, include: The detection data packets for each metal component are obtained one by one according to the timestamp and parsed to obtain the corresponding raw eddy current detection data; The original eddy current detection data is corrected and transformed according to a preset baseline value to obtain a two-dimensional detection array containing vector point X and vector point Y; Based on the two-dimensional detection array, an eddy current detection signal is generated, and combined with a preset frequency threshold, the corresponding defect feature data is extracted. Combined with the timestamp, the corresponding impedance diagram and time sequence diagram are generated. The defect feature data, the impedance diagram, and the timing diagram are fused and judged to obtain the metal component defect corresponding to each detection data packet.

2. The method for detecting metal components based on pulsed eddy currents according to claim 1, characterized in that, The specific steps for obtaining and parsing the detection data packets for each metal component based on timestamps to obtain the corresponding raw eddy current detection data include: Initialize Ethernet communication by setting the Internet Protocol address and port number of the defect detection terminal; According to the port number and communication protocol, the defect detection terminal and the data analysis terminal establish and monitor several data transmission channels and transmit data streams; Based on the preset number of data points, the data analysis terminal divides the data stream into several data packets; The data packet is parsed to obtain the packet header, command field, data length, and valid data; The data packet is transmitted by triggering and selecting the corresponding data transmission channel based on the command field and the packet header; The effective data is divided according to the preset data frame format and the data length to obtain the frame header, real part and imaginary part; Perform a duplicate check on all the frame headers. If the current frame header is unique, determine that the current data packet is not duplicated. Then, parse the real part and the virtual part as a whole to generate the original data pair. The original data pairs are sorted according to the timestamps of the data packets to obtain the original eddy current detection data.

3. The method for detecting metal components based on pulsed eddy currents according to claim 1, characterized in that, The specific steps of correcting and transforming the original eddy current detection data according to a preset baseline value to obtain a two-dimensional detection array containing vector points X and Y include: The difference between the original eddy current test data and the preset baseline value is calculated to obtain the standard test data; The standard detection data is grouped according to a preset number of data points to obtain several groups of data points; The previous set of data points is compared with the current set of data points to detect duplicates, resulting in a number of non-duplicate data points. Interpolate all the non-repeating data points and expand them into a new group of data points with the same number of data points. The new group of data points is concatenated with the current group of data points, and all standard detection data are calculated and aggregated with the preset gain coefficient to obtain a byte array; According to the preset byte bitwise mechanism, the byte array is decomposed in big-endian order to obtain several signed integer groups including the sequence number, the first integer group, and the second integer group; Perform difference calculations and mark abnormal positions for the indices of adjacent data points in the first integer group and the second integer group; if the difference is greater than 1, it indicates that the indices of an adjacent data point in the current integer group have changed, and the change position is marked as the starting position; if the difference is less than 0 and greater than the negative span, it indicates that the indices of an adjacent data point in the current integer group have broken, and the break position is marked as the starting position, while counting the number of loops without abnormalities. Based on the starting position, the sequence number of the abnormal position data point in the current integer group is corrected to form a new sequence number, and a corresponding index array is generated.

4. The method for detecting metal components based on pulsed eddy currents according to claim 3, characterized in that, The specific steps of correcting and transforming the original eddy current detection data according to a preset baseline value to obtain a two-dimensional detection array containing vector points X and Y further include: Sort each integer group according to the sorting type, and decentralize each data point in all integer groups according to the data processing type to obtain the corresponding corrected integer group; Perform convolution smoothing on each data point in each of the corrected integer groups to obtain the corresponding smoothed integer groups; The amplification factor is calculated based on the gain coefficient and then calculated with each data point in all the smoothed integer groups to obtain vector point X and vector point Y. The vector points X and Y are combined to generate a two-dimensional detection array.

5. The method for detecting metal components based on pulsed eddy currents according to claim 1, characterized in that, The specific steps of generating eddy current detection signals based on the two-dimensional detection array, extracting corresponding defect feature data by combining with a preset frequency threshold, and generating corresponding impedance diagrams and time series diagrams by combining with the timestamps include: Feature extraction is performed on vector points X and Y in the two-dimensional detection array to obtain the magnitude and phase angle, and an eddy current detection signal is generated. The eddy current detection signal is decomposed according to a preset frequency threshold to obtain high-frequency and low-frequency signals, and the corresponding dedicated signal processing channel is triggered. The dedicated signal processing channel extracts data from the high-frequency signal according to preset fixed window parameters to obtain several sets of signal values. The signal values ​​of each group are superimposed according to a preset dynamic step size to obtain overlapping signal data; The amplitude of the overlapping signal data is calculated and compared to obtain the maximum amplitude value; Based on the overlapping signal data and the finite impulse response function, an impedance diagram and a timing diagram are generated; The dedicated signal processing channel counts the number of times the low-frequency signal enters and exits within a preset number of points using a square window, and compares this count with a preset entry and exit threshold. If the number of entries and exits exceeds the threshold, the signal feature corresponding to the square window is identified as defect feature data, and the defect feature data is segmented to generate the impedance diagram.

6. The method for detecting metal components based on pulsed eddy currents according to claim 1, characterized in that, The specific steps for fusing and judging the defect feature data, the impedance diagram, and the timing diagram to obtain the metal component defect corresponding to each detection data packet include: The impedance map is feature extracted based on the phase change of the phase angle to obtain the impedance trajectory, and the area of ​​the impedance trajectory is calculated. If the phase change is zero, that is, the phase angle remains unchanged, then the impedance trajectory is radially offset. If the phase change is less than or equal to x degrees, then the impedance trajectory is stretched in the real axis direction; If the phase change is greater than x degrees and less than or equal to y degrees, then the impedance trajectory is offset or compressed in the imaginary axis direction; If the phase change is greater than y degrees, then the impedance trajectory is either a complete trajectory shift or a phase lag. Analyze the timing diagram and calculate the defect length L; L=v·Δt Where v is the detection velocity and Δt is the duration of the anomaly.

7. The method for detecting metal components based on pulsed eddy currents according to claim 6, characterized in that, The specific steps for fusing and judging the defect feature data, the impedance diagram, and the timing diagram to obtain the metal component defect corresponding to each detection data packet further include: Based on the characteristic sensitivity of the defect type to the impedance trajectory area, phase change, anomaly duration, and waveform steepness, a weight is assigned to each feature, and a feature threshold range is set; the waveform steepness is the maximum rate of change of the modulus within the defect segment. The defect score is obtained by weighted summation and normalization of the weights of each feature. The defect score is compared with a preset defect determination threshold, and the dominant feature weight is extracted and compared with the corresponding feature threshold interval. If the defect score is greater than or equal to the defect determination threshold and the dominant feature weight is located within the feature threshold interval, the defect type is directly determined. Otherwise, the secondary feature weights are extracted and compared with the corresponding feature threshold intervals. If the secondary feature weights are within the feature threshold intervals, the defect type is determined comprehensively; otherwise, it is marked as an unknown defect.

8. A pulsed eddy current metal component detection system, characterized in that, include: The connection configuration module is used to connect the defect detection end and the data analysis end, and to establish and monitor the data transmission channel. The signal acquisition module is used to acquire and parse the detection data packets of each metal component one by one according to the timestamp through the data transmission channel to obtain the corresponding raw eddy current detection data. The signal processing module is used to correct and transform the original eddy current detection data according to a preset baseline value, and construct a two-dimensional detection array; The feature extraction module is used to generate eddy current detection signals and extract defect feature data based on the two-dimensional detection array; The impedance decomposition module is used to separate and process the phase and amplitude of the eddy current detection signal, analyze the impedance characteristics of the metal component, and generate impedance diagrams and timing diagrams. The defect identification module is used to determine the defects of metal parts by impedance diagram based on preset frequency thresholds and to determine the location of the defects by timing diagram.

9. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-7.

10. A storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-7.

Images