Eddy current impedance-based shaft body surface crack on-line detection method and system

Abstract

This invention belongs to the field of shaft crack detection technology, and relates to an online detection method and system for shaft surface cracks based on eddy current impedance. The method includes: constructing a multi-frequency high-dimensional impedance array and performing spatial registration; projecting the array onto a decoupled feature space using a pre-constructed decoupling transformation matrix to obtain the coordinate values ​​of the directional feature dimension and the depth feature dimension; performing linear discrimination based on the directional feature distribution of each category in the sample library to determine whether the crack orientation is axial, circumferential, or oblique; and generating a continuous index characterizing the depth change trend through the mapping relationship between depth features and depth development trends, outputting both as detection results. This invention achieves feature decoupling of crack direction and depth through multi-frequency synchronous excitation and phase-locked space registration, elevating the traditional binary threshold discrimination to a joint evaluation of orientation classification and continuous depth quantification, helping to overcome the problems of uneven sensitivity to orientation and unquantifiable depth in single-frequency detection.

Description

Technical Field

[0001] This invention belongs to the field of shaft crack detection technology, and relates to an online detection method and system for shaft surface cracks based on eddy current impedance. Background Technology

[0002] As the core load-bearing component of rotating machinery, shafts are prone to fatigue crack initiation on their surfaces under alternating stress. Eddy current testing, due to its non-contact and fast response characteristics, is widely used for in-service detection of surface cracks in shafts.

[0003] In existing technologies, eddy current detection of cracks on shaft surfaces typically employs the following approach: an eddy current probe excited at a single frequency scans the shaft surface axially or circumferentially; a rotary encoder synchronously triggers and acquires impedance signals; the deviation of the impedance amplitude or phase relative to the defect-free area is extracted; and a binary determination of the presence or absence of a crack is output after comparison with a preset alarm threshold. Some improved schemes utilize the phase rotation angle on the impedance plane to roughly classify crack orientation, or obtain a comprehensive response at different depths by time-division switching of multiple frequencies.

[0004] The existing practices described above have the following drawbacks in practical applications:

[0005] 1. Single-frequency excitation can only obtain the comprehensive electromagnetic response within a single skin depth. It cannot resolve the impedance characteristics of different depth levels at the same spatial point. This results in the phase shift caused by the crack direction and the amplitude attenuation caused by the crack depth being highly coupled on the impedance plane, making it difficult to independently distinguish the crack direction and depth trend.

[0006] 2. In the time-division multi-frequency scheme, the sampling times of each frequency data are different. Under the online rotational motion state, speed fluctuations and probe lift-off changes will cause the responses of different frequencies to actually correspond to different spatial positions on the shaft surface, resulting in spatial mismatch of multi-frequency information. After superimposed edge effects, the frequency response is further distorted, making it difficult to extract stable crack geometric features.

[0007] 3. Existing methods use discrete thresholds or phase intervals for qualitative discrimination, which can only provide a rough classification of the presence or orientation of cracks. They cannot output quantitative indicators of the continuous change of crack depth along the circumferential direction of the shaft. The detection sensitivity for cracks with different orientations is severely uneven, and it is easy to miss dangerous axial cracks or generate excessive alarms for harmless surface indentations. They cannot meet the needs of dynamic assessment of crack development trends under online operating conditions. Summary of the Invention

[0008] In view of this, in order to solve the problems mentioned in the background art, an online detection method and system for shaft surface cracks based on eddy current impedance is proposed.

[0009] The objective of this invention can be achieved through the following technical solutions: The first embodiment of this invention provides an online detection method for surface cracks of shafts based on eddy current impedance, comprising: synchronously applying a multi-frequency excitation signal composed of multiple frequency components superimposed on the surface of the shaft, receiving an eddy current induced response signal, separating the complex impedance at the same detection position at each excitation frequency through frequency division multiplexing, and forming a multi-frequency high-dimensional impedance array;

[0010] The circumferential position pulse signal during the rotation of the shaft is acquired. The pulse edge is used as the trigger reference to perform synchronous phase-locked trigger acquisition of complex impedances at various frequencies, so that impedance data at different frequencies correspond to the same circumferential spatial point, thus achieving spatial registration.

[0011] The spatially registered high-dimensional impedance array is projected onto the decoupled feature space through a pre-constructed decoupling transformation matrix to obtain the coordinate values ​​of the directional feature dimension and the coordinate values ​​of the depth feature dimension; wherein, the decoupling transformation matrix is ​​determined by kernel principal component analysis of a multi-frequency impedance array of a sample library containing known crack direction labels and depth labels;

[0012] Based on the directional feature dimension coordinate values, linear discrimination is performed according to the directional feature distribution of different crack orientation categories in the sample library to determine whether the crack orientation is axial, circumferential, or oblique.

[0013] Based on the coordinate values ​​of the depth feature dimension, a continuous index representing the trend of crack depth change is generated through the mapping relationship between depth features and depth development trend.

[0014] The crack orientation and continuity index are output as the crack state detection results at the current detection location.

[0015] The second embodiment of the present invention provides an online detection system for surface cracks of shafts based on eddy current impedance, including: a multi-frequency excitation acquisition module, a position synchronization module, a decoupling projection module, a crack orientation discrimination module, a crack depth evaluation module, and a result output module.

[0016] The multi-frequency excitation acquisition module is connected to the position synchronization module, and the position synchronization module is connected to the decoupled projection module. The decoupled projection module is connected to the crack orientation discrimination module and the crack depth evaluation module, respectively, and is used to output the directional feature dimension coordinate values ​​to the crack orientation discrimination module and the depth feature dimension coordinate values ​​to the crack depth evaluation module. The crack orientation discrimination module is connected to the crack depth evaluation module and is used to output the crack orientation category to the crack depth evaluation module. Both the crack orientation discrimination module and the crack depth evaluation module are connected to the result output module and are used to output the crack orientation discrimination result and the continuity index to the result output module.

[0017] The multi-frequency excitation acquisition module synchronously applies a multi-frequency excitation signal composed of multiple frequency components to the surface of the shaft, receives the eddy current induction response signal, and separates the complex impedance at the same detection position at each excitation frequency through frequency division multiplexing to form a multi-frequency high-dimensional impedance array.

[0018] The position synchronization module acquires the circumferential position pulse signal during the rotation of the shaft, and uses the pulse edge as the trigger reference to perform synchronous phase-locked trigger acquisition of complex impedances at various frequencies, so that impedance data at different frequencies correspond to the same circumferential spatial point, thereby achieving spatial registration.

[0019] The decoupling projection module projects the spatially registered high-dimensional impedance array onto the decoupling feature space through a pre-constructed decoupling transformation matrix to obtain the coordinate values ​​of the directional feature dimension and the coordinate values ​​of the depth feature dimension; wherein, the decoupling transformation matrix is ​​determined by kernel principal component analysis of a multi-frequency impedance array of a sample library containing known crack direction labels and depth labels;

[0020] The crack orientation discrimination module performs linear discrimination based on the directional feature dimension coordinate values ​​and the directional feature distribution of different crack orientation categories in the sample library to determine whether the crack orientation is axial, circumferential, or oblique.

[0021] The crack depth assessment module generates a continuous index characterizing the crack depth change trend based on the coordinate values ​​of the depth feature dimension and the mapping relationship between the depth feature and the depth development trend.

[0022] The results output module outputs the crack orientation and continuity index as the crack state detection results at the current detection location.

[0023] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0024] (1) This invention obtains a multi-frequency high-dimensional impedance array containing multiple frequency complex impedances at the same detection position by synchronously applying multi-frequency excitation signals and separating frequency division multiplexing. After the decoupling transformation matrix determined by kernel principal component analysis is projected onto the decoupling feature space, the coordinate values ​​of the direction feature dimension corresponding to the phase rotation change and the coordinate values ​​of the depth feature dimension corresponding to the amplitude attenuation change are obtained respectively. This decouples the crack direction and depth on a single impedance plane and realizes the independent characterization of the two types of features.

[0025] (2) This invention utilizes the edge of the circumferential position pulse to perform synchronous phase-locked trigger acquisition of complex impedances at various frequencies, so that impedance data of different frequencies correspond to the same circumferential spatial point on the shaft surface, which helps to eliminate the spatial mismatch of multi-frequency information caused by sampling time deviation under online rotational motion, and provides a spatial alignment basis for multi-frequency joint analysis at the same detection point.

[0026] (3) The present invention performs linear discrimination based on the coordinate values ​​of the directional feature dimension and outputs the crack orientation classification results of axial, circumferential or oblique direction; based on the coordinate values ​​of the depth feature dimension, a continuous index is generated through the mapping relationship to output a continuous quantitative characterization of crack depth change, avoiding the problem of uneven sensitivity of the threshold binary judgment to orientation, and providing a continuous quantitative basis for the dynamic evaluation of crack development trend. Attached Figure Description

[0027] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a schematic diagram illustrating the implementation steps of the method of the present invention;

[0029] Figure 2 This is a schematic diagram of the system module connections of the present invention. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] Please see Figure 1 As shown, the first embodiment of the present invention provides an online detection method for surface cracks of shafts based on eddy current impedance, including: S1. Simultaneously applying a multi-frequency excitation signal composed of multiple frequency components to the surface of the shaft, receiving the eddy current induction response signal, and separating the complex impedance at the same detection position at each excitation frequency through frequency division multiplexing to form a multi-frequency high-dimensional impedance array.

[0032] In this embodiment, the multi-frequency high-dimensional impedance array includes the following construction process:

[0033] At least three excitation frequencies are selected, which cover different skin depths of the shaft material under the current working conditions, forming a frequency sequence from low frequency to high frequency.

[0034] The selection of the excitation frequency is performed through the following process:

[0035] First, the relative permeability and electrical conductivity of the material used in the shaft are retrieved from the material database. If no existing record is found in the material database, the values ​​are obtained through actual measurement as follows: a sample block made of the same material as the shaft is cut out, the electrical conductivity is measured using the four-probe method, and the DC permeability is measured using the solenoid method or a vibrating sample magnetometer and converted to relative permeability.

[0036] Secondly, the allowable crack depth limit specified in the shaft maintenance procedure is taken as the upper limit of the depth of interest, and the skin depth corresponding to the highest usable excitation frequency of the multi-frequency eddy current probe in the current shaft material is taken as the lower limit of the depth of interest. The lower limit of the depth of interest is directly calculated by the following relationship, the physical basis of which is: the higher the excitation frequency, the closer the eddy current is to the surface, and the skin depth corresponding to the highest frequency is the shallowest depth layer that the current detection system can detect in principle. Below this depth, the eddy current density has decayed to the surface value. the following.

[0037] Special Note: If the calculated lower limit of the depth of interest is less than the sum of the probe lift-off distance and the shaft surface roughness, it indicates that the nominal highest usable excitation frequency cannot effectively penetrate the lift-off gap to reach the shaft surface under the current detection conditions. The original highest frequency will be removed from the frequency sequence, and the second highest frequency will be used as the new highest frequency. The lower limit of the depth of interest will be recalculated until it is greater than or equal to the sum of the probe lift-off distance and the shaft surface roughness. The probe lift-off distance is obtained from the probe installation dimension chain and refers to the design gap between the end face of the multi-frequency eddy current probe and the shaft surface. The shaft surface roughness is obtained from the shaft machining process card or by taking the average of at least three measured points along the axial and circumferential directions on the shaft surface using a surface roughness meter.

[0038] Then, based on the skin effect of eddy current detection, the excitation frequency is established. Skin depth of eddies in shaft materials The relationship between them is specifically expressed as follows:

[0039] ;

[0040] In the formula, Let be the vacuum permeability, and take . , Let be the relative permeability of the axial material. Pi denoted as , where is the electrical conductivity of the shaft material.

[0041] The physical meaning of the formula is that, for a given material electromagnetic parameter, in order for the eddy current to penetrate to a specified depth and still have a detectable current density at the specified depth, the excitation frequency should be less than or equal to the value calculated by this formula.

[0042] Next, the lower and upper limits of the focus depth are substituted into the relational expression to obtain the required highest and lowest frequencies, thereby defining the frequency range.

[0043] Finally, the number of frequencies and their specific values ​​are determined within the frequency range. Number of frequencies The method for determining it is: calculate the ratio of the highest frequency to the lowest frequency. ,like Then take ,like Then take ,like Then take Divide the frequency range into equal parts on a logarithmic coordinate system. The frequency interval is divided into (N-1) segments, and a frequency sequence is formed by taking the frequency values ​​at the endpoints of each segment. In addition, when R exceeds 1000, the frequency interval is divided into (N-1) segments in logarithmic coordinates, and a frequency point is added at the midpoint of each segment so that the total number of frequencies does not exceed 15.

[0044] The standard sinusoidal waveforms of each frequency in the frequency sequence are superimposed point-by-point in the time domain to obtain a digital waveform sequence of multi-frequency composite excitation. This sequence is then converted into an analog signal by a digital-to-analog converter and used to drive the excitation coil of the multi-frequency eddy current probe through a power amplifier. The multi-frequency eddy current probe is in close contact with the shaft surface. The composite alternating magnetic field generated by the excitation coil induces eddy currents on the shaft surface. The eddy currents are distorted at the crack, and this distortion information is superimposed on the induced magnetic field, which is received by the detection coil as a response voltage signal.

[0045] The received response voltage signal is amplified by a preamplifier and then filtered by a low-pass anti-aliasing filter. It is then converted from analog to digital at a sampling rate no less than a preset multiple of the highest excitation frequency to obtain a digital response sequence. The preset multiple is determined as follows: Before formal testing, pre-sampling is performed at sampling rates of 2, 3, 4, and 5 times the highest excitation frequency. A Fast Fourier Transform is then performed on the digital response sequences at each sampling rate to calculate the ratio of the main lobe spacing between adjacent excitation frequencies to the 3dB bandwidth of each main lobe. When this ratio is greater than 2 for the first time, it indicates that the main lobes of adjacent frequencies are sufficiently separated, and this multiple is taken as the preset multiple. If none of the above multiples meet the condition of a ratio greater than 2, then 5 times is taken as the default preset multiple.

[0046] Perform a fast Fourier transform on the digital response sequence to locate the spectral peaks corresponding to each excitation frequency in the spectrum. Extract the response voltage amplitude and phase difference relative to the excitation current at each frequency, and construct the impedance values ​​at each frequency in complex form, where the real part is the resistance component and the imaginary part is the reactance component.

[0047] The impedance values ​​at the same detection location at various frequencies are arranged in order of frequency from low to high to form a multi-frequency high-dimensional impedance array for the current detection location.

[0048] S2. Acquire the circumferential position pulse signal during the rotation of the shaft, and use the pulse edge as the trigger reference to perform synchronous phase-locked acquisition of complex impedances at each frequency, so that impedance data of different frequencies correspond to the same circumferential spatial point, thereby achieving spatial registration.

[0049] In this embodiment, the synchronous phase-locked loop triggering acquisition of complex impedances at various frequencies using the pulse edge as the trigger reference includes:

[0050] The position pulse generated by the position encoder at preset circumferential angle intervals is received, and the circumferential spatial point corresponding to the pulse edge is extracted.

[0051] The process of determining the preset circumferential angle interval is as follows: obtain the nominal diameter of the shaft from the shaft design drawing. Extract the minimum crack length to be detected from the shaft maintenance procedure or crack tolerance standard. Circumferential angle interval The arc length corresponding to a single angular interval on the surface of the shaft is determined based on the following relationship: It should be less than or equal to This ensures that at least two sampling points along the circumferential direction fall within the minimum crack length range, i.e. Record this upper limit value as In the commercial encoder specification sequence, the single-pulse angle interval of each encoder specification is calculated sequentially in ascending order of single-turn pulse count. The single-pulse angle interval is the circumferential angle divided by the single-turn pulse count. The first encoder with a single-pulse angle interval less than or equal to... is selected. The encoder specifications are used to select the single-pulse angle interval as the preset circumferential angle interval.

[0052] At the extracted edge moment, the real-time calculated values ​​of complex impedance at each frequency are synchronously latched. The latching execution method is as follows: in the pulse edge interrupt service routine, the latest set of impedance values ​​at each frequency stored in the output register of the frequency division multiplexing module is read, and the pulse number at the time of reading is used as the spatial index and stored together with the impedance value in the data buffer.

[0053] The latched full-band impedance data is associated with the circumferential space point corresponding to the pulse edge that triggered this latch and stored.

[0054] For the complex impedances of the same frequency obtained by latching multiple times at the same point on a circular spatial axis over multiple consecutive turns, a vector averaging operation is performed. The specific execution logic of the vector averaging operation is as follows: the arithmetic mean of the real parts of the multiple recorded complex impedances of the same frequency is obtained. The arithmetic mean of the imaginary parts is obtained. Synthetic average complex impedance .

[0055] If data for a certain point in a certain circle is missing, that point in that circle will not participate in the vector averaging calculation. When the number of valid circles is less than the preset minimum number of valid circles (e.g., 3 circles), the detection result of that point will be marked as insufficient data, and the result is for reference only.

[0056] Since the position pulses are generated by the encoder based on the actual rotation angle and are independent of the instantaneous rotation speed of the shaft, even if the rotation speed fluctuates, the edges of each pulse still precisely correspond to the preset circumferential spatial points. The frequency impedance data acquired by the synchronous phase-locked trigger always remain synchronously aligned in space.

[0057] S3. The spatially registered high-dimensional impedance array is projected onto the decoupled feature space through a pre-constructed decoupling transformation matrix to obtain the coordinate values ​​of the directional feature dimension and the coordinate values ​​of the depth feature dimension; wherein, the decoupling transformation matrix is ​​determined by the multi-frequency impedance array of the sample library containing known crack direction labels and depth labels through kernel principal component analysis.

[0058] In this embodiment, the decoupling transformation matrix is ​​pre-constructed in the following manner:

[0059] Step 1, Sample Preparation: Simulated cracks of different orientations and depths are fabricated on the surface of the shaft material specimen using wire electrical discharge machining (EDM). The crack orientation covers the axial direction (angle between the crack's long axis and the shaft axis). ), oblique (angle) ) and circumferential (angle) Three categories. Each category involves processing cracks of multiple depth levels: minimum depth... Take 0.05mm as the starting depth that can be stably controlled in electrical discharge machining, and the maximum depth. Taking the upper limit of the attention depth calculated above, the intermediate depth levels are set as follows: The crack is divided into at least four equal segments on a logarithmic coordinate system, and the depth values ​​at the endpoints of each segment are used to machine the corresponding crack. Logarithmic division ensures sufficient calibration sample density for both shallow and deep crack regions.

[0060] The second step, sample scanning and annotation: Each sample is mounted on a rotating fixture, and using the same multi-frequency eddy current probe and signal acquisition parameters as in online testing, at least 36 angular positions are evenly distributed along the circumference of the sample (i.e., every...). Multi-frequency eddy current scanning is performed at one location, and the multi-frequency high-dimensional impedance array at each location is recorded. Each impedance array is labeled with the corresponding crack direction label (axial / oblique / circumferential) and crack depth label (the measured value of the actual processing depth). All calibrated samples constitute a sample library.

[0061] The third step is to construct the nuclear matrix: This involves extracting all samples from the sample library. The multi-frequency high-dimensional impedance arrays of each sample are arranged in rows to form The original data matrix, Let be the dimension of the impedance array. Calculate the impedance for any two samples. and The square of the Euclidean distance between them Kernel function scaling parameter The method for determining the median is as follows: calculate the squared Euclidean distance of all sample pairs and take the median. ,make Replacing the mean with the median is to avoid extreme distance pairs between samples. To mitigate the distortion effect of values, when the number of sample pairs is even, the median is taken as the arithmetic mean of the two middle values; the kernel matrix elements are calculated pairwise. ,get kernel matrix, where, For row index, For column indexes, , , For the first A multi-frequency high-dimensional impedance array of a sample For the first A multi-frequency high-dimensional impedance array of samples.

[0062] Step 4, Centering and Eigenvalue Decomposition: Center the kernel matrix by calculating the mean of each row to form a row mean vector. The mean of each column forms a column mean vector. The global mean is .

[0063] The elements of the centralized kernel matrix are: , The first kernel matrix represents the centered kernel matrix. Line number Column elements, The first kernel matrix represents the first kernel matrix. Line number Column elements, The first row of the mean vector represents the first row of the mean vector. One portion, The first column of the mean vector represents the first column of the mean vector. Each component.

[0064] right Perform eigenvalue decomposition to obtain eigenvalues. and corresponding feature vectors ,Pick corresponding As the first eigenvector, corresponding As the second eigenvector.

[0065] The fifth step is to compare the loading differences of samples with different directional labels on the first feature vector, and the loading differences of samples with different depth labels on the second feature vector. The first and second feature vectors are then designated as the directional dimension projection vector and the depth dimension projection vector, respectively. The specific designation process is as follows:

[0066] Samples with the same crack depth label but different direction labels were selected from the sample library and grouped by depth label. Each group contained at least three samples in each of the three directions: axial, oblique, and circumferential, forming a direction differentiation evaluation set. Simultaneously, samples with the same crack direction label but different depth label were selected from the sample library and grouped by direction label. Each group contained at least four samples of different depth levels from the lower to the upper limit of the depth of interest, forming a depth correlation evaluation set.

[0067] Directional dimension confirmation: For each depth group in the directional differentiation evaluation set, calculate the first feature vector of each sample within the group. Projection values ​​on ,in For the centered kernel row vector corresponding to the sample, group the projection values ​​by direction label and calculate the mean and standard deviation. If the following three conditions are met simultaneously, then it is confirmed. Projection vector of direction dimension:

[0068] I. The absolute value of the difference between the projected means of adjacent direction categories is greater than the sum of their standard deviations. The adjacent direction category refers to the categories that are adjacent in the order of axial, oblique, and circumferential directions. That is, axial and oblique are a pair of adjacent direction categories, and oblique and circumferential are another pair of adjacent direction categories.

[0069] II. The signs of the differences in the projection mean values ​​of adjacent directional categories remain consistent, that is, in the order of axial, oblique, and circumferential, the differences between the axial and oblique mean values, and the differences between the oblique and circumferential mean values ​​are both positive or both negative.

[0070] III. Plot the impedance data of each sample in the depth group on the complex plane along with the direction labels, and obtain the statistical median of the impedance vector phase angle corresponding to the three types of crack directions: axial, oblique, and circumferential. The three statistical medians show a monotonically increasing or monotonically decreasing trend along the label order of axial, oblique, and circumferential directions, which represents the monotonically changing impedance vector phase angle with the direction label.

[0071] Depth dimension confirmation: For each directional group of the deep association evaluation set, calculate the second feature vector of each sample within the group. Projection values ​​on Arrange the projected values ​​in ascending order of depth. If both of the following conditions are met, then... Confirmed as a depth dimension projection vector:

[0072] Condition 1: The signs of the projection differences between adjacent depth levels are all consistent: from the minimum depth to the maximum depth, the signs of the projection differences between every two adjacent depth levels are consistent. The differences are either all positive or all negative, with no local reversal.

[0073] Condition 2, the relationship between the projection value and the depth label The absolute value of the rank correlation coefficient is greater than 0.8; The rank correlation coefficient is calculated as follows: assign ranks to the depth labels and projection values ​​according to their magnitude, calculate the sum of squared differences between the two sets of ranks, and substitute them into... In the formula This represents the number of samples contained in the depth group. for Rank correlation coefficient, with a range of values ​​of , For the first The difference in rank between two variables for each sample , For the first The depth label of each sample in the whole The rank of a sample is the integer order assigned after sorting the depth labels by their numerical values. For the first The projection values ​​of each sample onto the first or second feature vector are in all The rank of a sample is the integer order assigned after the projected values ​​are sorted by numerical value. This refers to the sample number within the direction group. , For all The sum of the squared differences in the rank of each sample.

[0074] The rank correlation coefficient does not require the data to follow a normal distribution; it only measures whether there is a monotonic association between two variables. This corresponds to a statistically strong monotonic relationship. Further calculations are made to determine the relationship between the projected value and the impedance amplitude attenuation of the corresponding sample. The rank correlation coefficient, if also greater than 0.8, indicates a significant positive correlation between the increase in depth dimension projection value and the decrease in impedance amplitude, confirming... The physical meaning is consistent with expectations; wherein, the impedance amplitude attenuation refers to the reduction of the arithmetic mean of the impedance magnitudes of each frequency of the sample relative to the mean of the defect-free sample.

[0075] If the depth dimension confirmation process finds The projection value changes monotonically with depth, while If the projected values ​​satisfy conditions I and II in the direction dimension confirmation process, then it is determined that the direction and depth are interchanged on the feature vector. At this point, the positions of two columns in the decoupling transformation matrix are swapped: the original corresponding... The column was changed to ,correspond The column was changed to .

[0076] The sixth step is to merge the direction dimension projection vector and the depth dimension projection vector column by column to form a decoupling transformation matrix.

[0077] In this embodiment, the orientation feature dimension coordinates and depth feature dimension coordinates are obtained in the following way:

[0078] For the spatially registered multi-frequency high-dimensional impedance array at the current location Calculate the relationship between each sample and every sample in the sample library. kernel function value ,Will The kernel function values ​​are arranged as follows: kernel row vector .

[0079] Will Subtract the row mean vector stored during the build phase Subtract The corresponding terms are used to achieve the centralization of the kernel row vector.

[0080] The centered kernel row vector is then multiplied by the projection vectors of the direction and depth dimensions of the decoupling transformation matrix to obtain the coordinate values ​​of the direction and depth features, respectively.

[0081] S4. Based on the directional feature dimension coordinate values, perform linear discrimination according to the directional feature distribution of different crack orientation categories in the sample library to determine whether the crack orientation is axial, circumferential, or oblique.

[0082] In this embodiment, the linear discrimination based on the directional characteristic distribution of different crack orientation categories in the sample library includes:

[0083] The directional feature dimension coordinates of all samples of axial, oblique, and circumferential crack types were extracted sequentially from the sample library to form a set. , , .

[0084] Calculate the sets separately , , The sample mean of the feature dimension coordinates along the midline is denoted as . Calculate the sum of squared deviations of the sample values ​​in each set relative to the mean of their respective classes, and sum the sums of squared deviations of the three classes to obtain the within-class dispersion. .

[0085] Calculate the global mean of the directional feature dimension coordinates of the three types of samples. , ,in These represent the number of samples for axial, oblique, and circumferential crack types, respectively.

[0086] Calculate the inter-class scatter , .

[0087] Solving the generalized characteristic equation The projection direction is determined. ;in, For the generalized eigenvalue corresponding to the discriminative projection direction, it represents the ratio of inter-class scatter to intra-class scatter along that projection direction. When the coordinate value of the direction feature dimension is a one-dimensional scalar, the intra-class scatter is... Inter-class dispersion All are degenerated into scalars to determine the projection direction. It degenerates to scalar 1, directly using the coordinate values ​​of the directional feature dimension as the discriminant variable, without the need for additional projection transformation.

[0088] Will Sort by value from smallest to largest, denoted as ;Pick and The mean of the values ​​is taken as the first dividing value. and The mean of the values ​​is used as the second dividing value.

[0089] The directional feature dimension coordinates of the current detection location are compared with the first and second boundary values: if they are less than or equal to the first boundary value, the crack orientation is determined to be the crack orientation corresponding to the category with the smallest mean; if they are greater than the first boundary value and less than or equal to the second boundary value, the crack orientation is determined to be the crack orientation corresponding to the category with the middle mean; if they are greater than the second boundary value, the crack orientation is determined to be the crack orientation corresponding to the category with the largest mean.

[0090] S5. Based on the coordinate values ​​of the depth feature dimension, a continuous index representing the trend of crack depth change is generated through the mapping relationship between depth features and depth development trend.

[0091] In this embodiment, the mapping relationship between the depth features and the depth development trend is pre-established in the following way:

[0092] Extract all samples from the sample library whose orientation labels belong to the current crack orientation category, and sort them in ascending order by depth label to obtain the calibration point sequence. ,in For the first Crack depth label for each sample For the first The projection value of each sample onto the depth dimension projection vector. This represents the number of calibration points for the current crack orientation category.

[0093] from to Connect adjacent calibration points in sequence and ,constitute Line segments; among which, connecting and line segments The slope is line segment In the depth range The linear mapping relationship between the depth feature dimension coordinates and the crack depth is defined above.

[0094] For all Monotonicity check of line segments: Check slope If all slopes are either positive or negative, the monotonicity check passes. If there are slopes with opposite signs, locate the depth interval corresponding to the opposite-signed line segments, and supplement at least two new depth levels of samples within the opposite-signed depth interval by dividing it into equal parts according to the logarithmic depth values. Perform multi-frequency eddy current scanning and kernel principal component analysis on the supplemented samples again until all slopes are of the same sign.

[0095] If, after three consecutive replenishment of samples, slopes with different signs still exist in the same depth range, then the direction with the smallest absolute value of the slope within that range is taken as the mapping direction for that segment, and a mapping direction marker is output to indicate that the confidence level of that segment is lower than that of other segments.

[0096] Will pass the monotonicity check The broken line formed by the line segments is used to determine the mapping relationship between the depth characteristics and the depth development trend under the current crack orientation category.

[0097] In this embodiment, the continuous index is generated in the following manner:

[0098] Obtain the depth feature dimension coordinates of the current detection location and retrieve the calibration point intervals that fall within the piecewise linear mapping line.

[0099] If the coordinate value of the depth feature dimension is equal to the depth feature value of a certain calibration point, then the crack depth label of that calibration point is taken as the continuous index.

[0100] If it is located between two adjacent calibration points ( Then, based on the ratio of the distances from the coordinate values ​​to the depth feature values ​​of the two calibration points... A continuous exponent is obtained by linear interpolation between the corresponding depth labels, specifically: .

[0101] If the depth exceeds the coverage area of ​​the calibration point, the depth label of the nearest calibration point is used for extrapolation, and an extrapolation mark is added.

[0102] S6. Output the crack orientation and continuity index as the crack state detection results at the current detection location.

[0103] Reference Figure 2 As shown, the second embodiment of the present invention provides an online detection system for surface cracks of shafts based on eddy current impedance, including: a multi-frequency excitation acquisition module, a position synchronization module, a decoupling projection module, a crack orientation discrimination module, a crack depth evaluation module, and a result output module.

[0104] The multi-frequency excitation acquisition module is connected to the position synchronization module, and the position synchronization module is connected to the decoupled projection module. The decoupled projection module is connected to the crack orientation discrimination module and the crack depth evaluation module, respectively, and is used to output the directional feature dimension coordinate values ​​to the crack orientation discrimination module and the depth feature dimension coordinate values ​​to the crack depth evaluation module. The crack orientation discrimination module is connected to the crack depth evaluation module and is used to output the crack orientation category to the crack depth evaluation module. Both the crack orientation discrimination module and the crack depth evaluation module are connected to the result output module and are used to output the crack orientation discrimination result and the continuity index to the result output module.

[0105] The multi-frequency excitation acquisition module synchronously applies a multi-frequency excitation signal composed of multiple frequency components to the surface of the shaft, receives the eddy current induction response signal, and separates the complex impedance at the same detection position at each excitation frequency through frequency division multiplexing to form a multi-frequency high-dimensional impedance array.

[0106] The position synchronization module acquires the circumferential position pulse signal during the rotation of the shaft, and uses the pulse edge as the trigger reference to perform synchronous phase-locked trigger acquisition of complex impedances at various frequencies, so that impedance data at different frequencies correspond to the same circumferential spatial point, thereby achieving spatial registration.

[0107] The decoupling projection module projects the spatially registered high-dimensional impedance array onto the decoupling feature space through a pre-constructed decoupling transformation matrix to obtain the coordinate values ​​of the directional feature dimension and the coordinate values ​​of the depth feature dimension; wherein, the decoupling transformation matrix is ​​determined by kernel principal component analysis of a multi-frequency impedance array of a sample library containing known crack direction labels and depth labels;

[0108] The crack orientation discrimination module performs linear discrimination based on the directional feature dimension coordinate values ​​and the directional feature distribution of different crack orientation categories in the sample library to determine whether the crack orientation is axial, circumferential, or oblique.

[0109] The crack depth assessment module generates a continuous index characterizing the crack depth change trend based on the coordinate values ​​of the depth feature dimension and the mapping relationship between the depth feature and the depth development trend.

[0110] The results output module outputs the crack orientation and continuity index as the crack state detection results at the current detection location.

[0111] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. An online detection method for surface cracks in shafts based on eddy current impedance, characterized in that, include: A multi-frequency excitation signal composed of multiple frequency components is synchronously applied to the surface of the shaft, and the eddy current induction response signal is received. The complex impedance at the same detection position at each excitation frequency is separated by frequency division multiplexing to form a multi-frequency high-dimensional impedance array. The circumferential position pulse signal during the rotation of the shaft is acquired. The pulse edge is used as the trigger reference to perform synchronous phase-locked trigger acquisition of complex impedances at various frequencies, so that impedance data at different frequencies correspond to the same circumferential spatial point, thus achieving spatial registration. The spatially registered high-dimensional impedance array is projected onto the decoupled feature space through a pre-constructed decoupling transformation matrix to obtain the coordinate values ​​of the directional feature dimension and the coordinate values ​​of the depth feature dimension; wherein, the decoupling transformation matrix is ​​determined by kernel principal component analysis of a multi-frequency impedance array of a sample library containing known crack direction labels and depth labels; Based on the directional feature dimension coordinate values, linear discrimination is performed according to the directional feature distribution of different crack orientation categories in the sample library to determine whether the crack orientation is axial, circumferential, or oblique. Based on the coordinate values ​​of the depth feature dimension, a continuous index representing the trend of crack depth change is generated through the mapping relationship between depth features and depth development trend. The crack orientation and continuity index are output as the crack state detection results at the current detection location.

2. The online detection method for shaft surface cracks based on eddy current impedance according to claim 1, characterized in that, The multi-frequency high-dimensional impedance array includes the following construction process: Select at least three excitation frequencies to cover different skin depths of the shaft material and form a frequency sequence; The sinusoidal waves in the frequency sequence are superimposed, and after digital-to-analog conversion and power amplification, they drive the excitation coil. The received response signal is amplified and filtered against anti-aliasing. Then, it is converted from analog to digital at a sampling rate not lower than a preset multiple of the highest excitation frequency to obtain a digital response sequence. Perform a fast Fourier transform on the digital response sequence to extract the response amplitude and phase difference relative to the excitation current at each frequency, and construct the impedance value in complex form; The impedance values ​​at the same detection location at different frequencies are arranged from low to high frequency to form a multi-frequency high-dimensional impedance array.

3. The online detection method for shaft surface cracks based on eddy current impedance according to claim 1, characterized in that, The synchronous phase-locked loop triggering acquisition of complex impedances at various frequencies using the pulse edge as the trigger reference includes: The position pulse generated by the position encoder at preset circumferential angle intervals is received, and the circumferential spatial point corresponding to the edge of the pulse is extracted. At the extraction edge moment, the real-time calculated value of the complex impedance at each frequency is synchronously latched; The latched full-band impedance data is associated with the circumferential space point corresponding to the pulse edge that triggered this latch and stored accordingly. For the complex impedance of the same frequency obtained by latching multiple times in a continuous circular space point, perform vector averaging.

4. The online detection method for shaft surface cracks based on eddy current impedance according to claim 1, characterized in that, The decoupling transformation matrix is ​​pre-constructed in the following manner: Prepare axial specimens with axial, circumferential, and oblique crack orientations and multiple depth levels; Each sample was subjected to multi-frequency eddy current scanning at multiple circumferential angular positions. The impedance array was recorded and the crack direction and depth labels were marked to form a sample library. The impedance arrays of all samples in the sample library are arranged in rows to form the original data matrix, with each row corresponding to one sample and each column corresponding to one frequency-impedance component. The radial basis kernel function is selected to calculate the kernel matrix. After centering, eigenvalue decomposition is performed, and the eigenvectors corresponding to the largest and second largest eigenvalues ​​are taken as the first and second eigenvectors. Compare the load differences of samples with different directional labels on the first feature vector and the load differences of samples with different depth labels on the second feature vector, and designate the first feature vector and the second feature vector as the directional dimension projection vector and the depth dimension projection vector, respectively. The orientation dimension projection vector and the depth dimension projection vector are merged column-wise to form a decoupling transformation matrix.

5. The online detection method for shaft surface cracks based on eddy current impedance according to claim 4, characterized in that, The step of designating the first feature vector and the second feature vector as the orientation dimension projection vector and the depth dimension projection vector, respectively, includes: Keeping the depth label unchanged, the distribution of the projection values ​​of the statistical samples on the maximum eigenvector with the direction label is analyzed. If a regular offset occurs and the offset direction is consistent with the impedance phase rotation direction, the maximum eigenvector is marked as the direction dimension projection vector. Keeping the orientation label unchanged, the distribution of the projection value of the sample onto the second largest feature vector with respect to the depth label is statistically analyzed. If it shows a monotonic change and the change amplitude is positively correlated with the attenuation of the impedance amplitude, then the second largest feature vector is marked as the depth dimension projection vector. If the eigenvectors corresponding to the direction and depth are interchanged, then the column positions of the direction dimension projection vector and the depth dimension projection vector in the decoupling transformation matrix are swapped.

6. The online detection method for shaft surface cracks based on eddy current impedance according to claim 4, characterized in that, The orientation feature dimension coordinates and depth feature dimension coordinates are obtained in the following way: Calculate the high-dimensional impedance array to be projected and the radial basis kernel function values ​​of all samples in the sample library, and arrange them as kernel row vectors; The kernel row vectors are centered. The centered kernel row vector is then multiplied by the projection vectors of the direction and depth dimensions of the decoupling transformation matrix to obtain the coordinate values ​​of the direction and depth features, respectively.

7. The online detection method for shaft surface cracks based on eddy current impedance according to claim 1, characterized in that, The linear discrimination based on the directional characteristic distribution of different crack orientation categories in the sample library includes: The directional feature dimension coordinates of three types of samples—axial cracks, circumferential cracks, and oblique cracks—are extracted from the sample library. The mean vector and intra-class dispersion matrix of the directional feature dimension coordinates of each type of sample are calculated respectively. Based on the mean vectors of various samples, the inter-class scatter matrix is ​​calculated, and the generalized eigenvalue problem of the inter-class scatter matrix and the intra-class scatter matrix is ​​solved to obtain the discriminant projection direction; Project the mean vector of each type of sample along the discriminant projection direction to obtain the discriminant center value of each category, and calculate the midpoint of adjacent discriminant center values ​​as the category boundary value; For the directional feature dimension coordinates of the current detection position, the discriminant projection value is obtained by projecting along the discriminant projection direction. The discriminant projection value is compared with the category boundary value, and the crack orientation is determined to be axial, circumferential or oblique based on the interval in which the projection value falls.

8. The online detection method for shaft surface cracks based on eddy current impedance according to claim 1, characterized in that, The mapping relationship between the depth features and the depth development trend is established in advance through the following methods: In the sample library, for the same crack orientation category, the depth feature dimension coordinate values ​​of each depth label sample are extracted, and a depth-feature calibration point sequence is formed with the crack depth label as the horizontal axis and the depth feature dimension coordinate values ​​as the vertical axis. From the start point to the end point of the depth-feature calibration point sequence, adjacent calibration points are connected sequentially to form a piecewise linear mapping line; Monotonicity verification is performed on the mapping relationship represented by the piecewise linear mapping line: if the coordinate values ​​of the depth feature dimension do not show a monotonically increasing or monotonically decreasing trend with crack depth, the coverage of depth labels in the sample library is expanded and the kernel principal component analysis is performed again. The piecewise linear mapping line that passes the monotonicity check is determined as the mapping relationship between depth features and depth development trend.

9. The online detection method for shaft surface cracks based on eddy current impedance according to claim 8, characterized in that, The continuous index is generated in the following manner: Obtain the depth feature dimension coordinates of the current detection location and retrieve the calibration point intervals that fall within the piecewise linear mapping polyline; If the coordinate value of the depth feature dimension is equal to the depth feature value of a certain calibration point, then the crack depth label of that calibration point is taken as the continuous index. If it is located between two adjacent calibration points, the continuous exponent is obtained by linear interpolation between the corresponding depth labels based on the ratio of the distance between the coordinate value and the depth feature value of the two calibration points. If the depth exceeds the coverage area of ​​the calibration point, the depth label of the nearest calibration point is used for extrapolation, and an extrapolation mark is added.

10. An online detection system for surface cracks in shafts based on eddy current impedance, characterized in that, The system for implementing the method as described in any one of claims 1 to 9 comprises: The multi-frequency excitation acquisition module synchronously applies a multi-frequency excitation signal composed of multiple frequency components to the surface of the shaft, receives the eddy current induction response signal, and separates the complex impedance at the same detection position at each excitation frequency through frequency division multiplexing to form a multi-frequency high-dimensional impedance array. The position synchronization module acquires the circumferential position pulse signal during the rotation of the shaft, and uses the pulse edge as the trigger reference to perform synchronous phase-locked trigger acquisition of complex impedances at various frequencies, so that impedance data at different frequencies correspond to the same circumferential spatial point, thereby achieving spatial registration. The decoupling projection module projects the spatially registered high-dimensional impedance array onto the decoupling feature space through a pre-constructed decoupling transformation matrix to obtain the coordinate values ​​of the directional feature dimension and the coordinate values ​​of the depth feature dimension; wherein, the decoupling transformation matrix is ​​determined by kernel principal component analysis of a multi-frequency impedance array of a sample library containing known crack direction labels and depth labels; The crack orientation discrimination module performs linear discrimination based on the directional feature dimension coordinate values ​​and the directional feature distribution of different crack orientation categories in the sample library to determine whether the crack orientation is axial, circumferential, or oblique. The crack depth assessment module generates a continuous index characterizing the crack depth change trend based on the coordinate values ​​of the depth feature dimension and the mapping relationship between the depth feature and the depth development trend. The results output module outputs the crack orientation and continuity index as the crack state detection results at the current detection location.

Images