Multi-channel eddy current sensor array and data processing method thereof

Abstract

The invention belongs to the technical field of eddy current sensor detection, and discloses a multi-channel eddy current sensor array and a data processing method thereof. The method comprises the steps that a detection area is divided into N detection sub-areas according to a preset method, and M multi-channel eddy current sensors are distributed to each detection sub-area; performing feature extraction based on a sensor signal data set acquired by the multi-channel eddy current sensors corresponding to the N detection sub-regions to obtain sensor signal feature data corresponding to the N detection sub-regions; performing region construction on the N detection sub-regions based on the sensor signal feature data corresponding to the N detection sub-regions to obtain R regions to be finely detected; based on the sensor signal feature data corresponding to the N detection sub-regions, intelligently and adaptively allocating resources and setting parameters for the R regions to be finely detected; according to the method, the sub-region division number, the sensor number and the sampling strategy are dynamically set according to the actual detection requirement of the region, and increasingly complex requirements are met.

Description

Technical Field

[0001] The present invention relates to the technical field of eddy current sensor detection, and more specifically, to a multi-channel eddy current sensor array and its data processing method. Background Art

[0002] With the continuous improvement of the requirements for structural integrity assessment and defect detection accuracy in fields such as high-end manufacturing, aerospace, and energy equipment, the traditional single-channel eddy current detection method has been difficult to meet the rapid detection needs for large-area workpieces and complex-shaped structures. In the industry, a multi-channel eddy current sensor array structure has gradually emerged, which uses multiple channels to simultaneously collect electromagnetic response signals at multiple positions, improving the detection efficiency while enhancing the ability to capture local abnormal responses.

[0003] However, in actual engineering applications, with the expansion of the array scale and the complexity of the detection target structure, the existing technology still faces bottleneck problems. There is an obvious conflict between the arrangement density of the sensor array and the detection resolution: although an increase in the array size can cover a larger area, the resolution of a single channel decreases accordingly, and crosstalk may occur between channels, affecting the measurement stability; if high resolution is maintained, the probe arrangement density surges, resulting in high system manufacturing difficulty, high signal acquisition bandwidth, and a sharp increase in data processing pressure. In addition, in the multi-region simultaneous detection task, most existing solutions adopt a fixed partitioning and static resource allocation strategy, and fail to dynamically set the number of sub-region partitions, the number of sensors, and the sampling strategy according to the actual detection needs of the region, resulting in the inability to meet the increasingly complex requirements.

[0004] Therefore, how to construct a multi-channel eddy current sensor array system that supports heterogeneous region partitioning and intelligent resource scheduling on the basis of ensuring detection accuracy and response integrity, and realize full-process data-driven partition construction and resource allocation has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art and achieve the above object, the present invention provides the following technical solutions: A data processing method for a multi-channel eddy current sensor array, including: Dividing the detection area into N detection sub-regions according to a preset method, and allocating M multi-channel eddy current sensors to each detection sub-region; Performing feature extraction on the sensor signal data set collected by the multi-channel eddy current sensors corresponding to the N detection sub-regions to obtain sensor signal feature data corresponding to the N detection sub-regions; Performing region construction on the N detection sub-regions based on the sensor signal feature data corresponding to the N detection sub-regions to obtain R regions to be precisely inspected; Based on the sensor signal feature data corresponding to N detection sub-regions, resources and parameter settings are intelligently and adaptively allocated for R regions to be precisely inspected.

[0006] Further, the method for intelligently and adaptively allocating resources and parameter settings for R regions to be precisely inspected includes: S600: Let the initial value of r be 1, and the value range of r is from 1 to R; S601: Denote the number of detection sub-regions in the r-th region to be precisely inspected as ; Input the sensor signal feature data corresponding to detection sub-regions into the region demand assessment model respectively to obtain the corresponding region demand levels; the region demand levels include low demand, medium demand, and high demand; S602: Construct the corresponding resource allocation feature data from each detection sub-region and the corresponding region demand level in the r-th region to be precisely inspected, and construct a resource allocation feature data set; S603: Input the resource allocation feature data set of the r-th region to be precisely inspected into the resource configuration setting model to obtain a resource configuration pool; the resource configuration pool includes the allocation quantity of multi-channel eddy current sensors, scanning time, sampling frequency, resolution, and compression ratio; S604: Allocate resources and set parameters for the r-th region to be precisely inspected according to the corresponding resource configuration pool; S605: Let r = r + 1. If r is less than or equal to R, continue to execute S601 to S604; if r is greater than R, complete the intelligent and adaptive allocation of resources and parameter settings for R regions to be precisely inspected, and end the current process.

[0007] Further, the method for obtaining the sensor signal feature data corresponding to N detection sub-regions includes: S100: Let the initial value of n be 1, and the value range of n is from 1 to N; S101: According to the spatial arrangement order of the probe units of the multi-channel eddy current sensors, the real part matrix of impedance, the imaginary part matrix of impedance, the amplitude matrix of impedance, and the phase matrix, obtain H groups of sensor signal data for the n-th detection sub-region; Map the real part of impedance in the H groups of sensor signal data into the real part matrix of impedance; Map the imaginary part of impedance in the H groups of sensor signal data into the imaginary part matrix of impedance; Map the amplitude of impedance in the H groups of sensor signal data into the amplitude matrix of impedance; Map the phase in the H groups of sensor signal data into the phase matrix; S102: Analyze and process the phase based on the phase matrix to obtain the phase feature data of the nth detection sub-region; analyze and process the real part of the impedance based on the real part matrix of the impedance to obtain the real part feature data of the impedance of the nth detection sub-region; analyze and process the imaginary part and the magnitude of the impedance based on the imaginary part matrix and the magnitude matrix of the impedance to obtain the fusion feature data; S103: Construct the sensor signal feature data of the nth detection sub-region from the phase feature data, the real part feature data of the impedance, and the fusion feature data; S104: Let n = n + 1. If n is less than or equal to N, then continue to execute S101 to S103; if n is greater than N, then obtain the sensor signal feature data corresponding to N detection sub-regions and end the current process.

[0008] Furthermore, the method for obtaining the phase feature data includes: S200: Let the initial value of h be 1, and the value range of h is from 1 to H; calculate the phase mean value based on the phases in the phase matrix; let the initial value of the phase mutation quantity be 0; S201: Obtain the hth phase from the phase matrix, obtain the adjacent phase to the hth phase from the phase matrix, denote it as the adjacent phase, and denote the number of adjacent phases as ; Take the absolute value of the difference between the hth phase and adjacent phases respectively to obtain phase gradients; if there is a phase gradient greater than or equal to the preset phase gradient threshold among the phase gradients, then mark the hth phase as a mutation phase point and let the phase mutation quantity increase by 1; S202: Let h = h + 1. If h is less than or equal to H, then continue to execute S201; if h is greater than H, then execute S203; S203: Denote the number of mutation phase points as TBXW, calculate the phase mutation degree based on the phase mean value and the phases of TBXW mutation phase points; divide the phase mutation quantity by H to obtain the phase mutation ratio; S204: Construct the phase feature data from the phase mutation degree and the phase mutation ratio and end the current process.

[0009] Furthermore, the method for obtaining the real part feature data of the impedance includes: S300: Let the initial value of h be 1, and the value range of h is from 1 to H; calculate the real part mean value of the impedance based on the real part of the impedance in the real part matrix of the impedance; let the initial value of the real part mutation quantity of the impedance be 0; S301: Obtain the h-th real part of impedance from the real part matrix of impedance, obtain the adjacent real part of impedance adjacent to the h-th real part of impedance from the real part matrix of impedance, denote it as the adjacent real part of impedance, and denote the number of adjacent real parts of impedance as ; Subtract the h-th real part of impedance from each of the adjacent real parts of impedance and take the absolute value to obtain real part gradients of impedance; if there is a real part gradient of impedance greater than or equal to the preset real part gradient threshold among the real part gradients of impedance, then mark the h-th real part of impedance as a mutation real part point of impedance, and increment the real part mutation count by 1; S302: Let h = h + 1. If h is less than or equal to H, continue to execute S301; if h is greater than H, execute S303; S303: Denote the number of mutation real part points of impedance as TBSB, and calculate the degree of real part mutation of impedance based on the mean value of the real part of impedance and the real parts of impedance of the TBSB mutation real part points of impedance; Divide the real part mutation count by H to obtain the real part mutation ratio; S304: Construct the real part mutation degree and the real part mutation ratio into the real part feature data, and end the current process.

[0010] Furthermore, the method for obtaining the fusion feature data includes: S400: Let the initial value of h be 1, and the value range of h is from 1 to H; Calculate the mean value of the imaginary part of impedance based on the imaginary part of impedance in the imaginary part matrix of impedance; Calculate the mean value of the amplitude of impedance based on the amplitude of impedance in the amplitude matrix of impedance; S401: Obtain the h-th imaginary part of impedance from the imaginary part matrix of impedance, obtain the h-th amplitude of impedance from the amplitude matrix of impedance, and construct the h-th imaginary part of impedance and the amplitude of impedance into the h-th fusion feature pair; Add the fusion feature pair to the fusion feature pair set; S402: Calculate the impedance imaginary part - amplitude ratio and the combined response amplitude corresponding to the h-th fusion feature pair based on the h-th fusion feature pair; S403: Add the impedance imaginary part - amplitude ratio to the impedance imaginary part - amplitude ratio set; Add the combined response amplitude to the combined response amplitude set; S404: Let h = h + 1. If h is less than or equal to H, continue to execute S401 to S403; if h is greater than H, execute S405; S405: Calculate the fusion feature correlation coefficient based on the mean value of the imaginary part of impedance, the mean value of the amplitude of impedance, the impedance imaginary part - amplitude ratio set, and the combined response amplitude set; S406: Construct the impedance imaginary part - amplitude ratio set, the combined response amplitude set, and the fusion feature correlation coefficient into the fusion feature data, and end the current process.

[0011] Further, the method for obtaining R regions to be precisely inspected includes: S500: Let the initial value of n be 1, and the value range of n is from 1 to N; initialize the region construction flags of the N detection sub-regions as unconstructed; let the initial value of the region count variable R for regions to be precisely inspected be 1; S501: If the region construction flag of the nth detection sub-region is unconstructed, then execute S502; if the region construction flag of the nth detection sub-region is constructed, then execute S505; S502: Construct the detection sub-regions adjacent to the nth detection sub-region and with unconstructed region construction flags into the region to be processed corresponding to the nth detection sub-region; convert the sensor signal feature data of the nth detection sub-region and the corresponding region to be processed into sensor signal feature vectors, and construct them into the Rth sensor signal feature vector set; S503: Apply a clustering algorithm to the Rth sensor signal feature vector set for classification processing; add the detection sub-regions corresponding to the sensor signal feature vectors in the same clustering cluster as the nth detection sub-region in the clustering result to the Rth set to be constructed; Determine whether there are detection sub-regions adjacent to the detection sub-regions in the Rth set to be constructed and with unconstructed region construction flags; if there are, then convert the sensor signal feature data of the corresponding detection sub-regions into sensor signal feature vectors, and add them to the Rth sensor signal feature vector set, and continue to execute S503; if not, then execute S504; S504: Construct the detection sub-regions in the set to be constructed into the Rth region to be precisely inspected, and let R = R + 1; set the region construction flags of the detection sub-regions in the Rth set to be constructed as constructed; S505: Let n = n + 1. If n is less than or equal to N, then continue to execute S501 to S504; if n is greater than N, then obtain R regions to be precisely inspected, and end the current process.

[0012] Further, divide the detection region into N detection sub-regions according to a preset method, and allocate M multi-channel eddy current sensors to each detection sub-region, including: Pre-collect the effective sensing range of the multi-channel eddy current sensors, the total area of the detection region, the material type of the detection region, and the structural type of the detection region; the material type of the detection region includes aluminum, copper, magnesium, titanium, and steel; the structural type of the detection region includes regular plane, curved structure, concave-convex surface, and multi-layer composite structure; Numericalize the material type of the detection region to obtain the material type value of the detection region; numericalize the structural type of the detection region to obtain the structural type value of the detection region; Input the effective induction range, total detection area, numerical value of the material type of the detection area, and numerical value of the structural type of the detection area of the multi-channel eddy current sensor into the sub-region division parameter setting model to obtain the number of detection sub-regions and the number of multi-channel eddy current sensors. The number of detection sub-regions is N; the number of multi-channel eddy current sensors is M. Divide the detection area equally into N detection sub-regions, and allocate M multi-channel eddy current sensors to each detection sub-region.

[0013] Furthermore, the training method of the sub-region division parameter setting model includes: Pre-construct a sub-region division parameter data set, which includes sub-region division parameter data of the HF group and the number of detection sub-regions and the number of multi-channel eddy current sensors corresponding to the sub-region division parameter data of the HF group. HF is a positive integer greater than 0. The sub-region division parameter data includes the effective induction range of the multi-channel eddy current sensor, the total detection area, the numerical value of the material type of the detection area, and the numerical value of the structural type of the detection area; divide the sub-region division parameter data set into a sub-region division parameter data training set and a sub-region division parameter data verification set. The sub-region division parameter data training set is used for parameter learning of the sub-region division parameter setting model, and the sub-region division parameter data verification set is used to evaluate the generalization ability of the sub-region division parameter setting model in real time. During the training process of the sub-region division parameter setting model, the sub-region division parameter setting model adopts a support vector machine model or a decision tree model, converts the sub-region division parameter data into a feature vector as the input, extracts the non-linear features in the data through the hidden layer, and finally generates the probability distribution of the number of detection sub-regions and the number of multi-channel eddy current sensors by using the softmax activation function in the output layer, and outputs the number of detection sub-regions and the number of multi-channel eddy current sensors corresponding to the maximum probability as the final prediction result; the training process aims to minimize the cross-entropy loss function, and at the same time introduces an early stopping strategy to monitor the performance of the sub-region division parameter data verification set. When the prediction accuracy on the sub-region division parameter data verification set reaches the preset accuracy, it is considered that the sub-region division parameter setting model has converged and the training stops immediately.

[0014] Furthermore, the sensor signal data set includes H groups of sensor signal data, where H = M × B, and B is the number of probe units of the multi-channel eddy current sensor, that is, the number of channels; the sensor signal data includes the real part of the impedance, the imaginary part of the impedance, the amplitude of the impedance, and the phase.

[0015] A multi-channel eddy current sensor array for implementing the data processing method of the multi-channel eddy current sensor array includes: The first partitioning module divides the detection area into N detection sub-areas according to a preset method, and assigns M multi-channel eddy current sensors to each detection sub-area; The first processing module extracts features based on the set of sensor signal data collected by the multi-channel eddy current sensors corresponding to the N detection sub-areas, and obtains the sensor signal feature data corresponding to the N detection sub-areas; The second partitioning module constructs regions for the N detection sub-areas based on the sensor signal feature data corresponding to the N detection sub-areas, and obtains R regions to be precisely inspected; The resource allocation module intelligently and adaptively allocates resources and sets parameters for the R regions to be precisely inspected based on the sensor signal feature data corresponding to the N detection sub-areas.

[0016] Compared with the prior art, the technical effects and advantages of a multi-channel eddy current sensor array and its data processing method of the present invention are as follows: A multi-channel eddy current sensor array and its data processing method provided by the present invention integrate array signal acquisition, detection area construction based on feature driving, resource adaptive allocation and priority scheduling strategies, and have an efficient, flexible and intelligent overall detection ability.

[0017] By adaptively partitioning the detection area according to material type, structure type and the effective sensing range of the sensor, multiple detection sub-areas are obtained, and multi-dimensional signal features such as the real part of impedance, the imaginary part of impedance, impedance amplitude and phase are extracted by combining the data collected by the multi-channel sensors in each sub-area to construct a regional feature matrix. Based on the regional feature matrix, a clustering algorithm is executed to complete the dynamic construction of sub-areas with similar features, and further aggregated into regions to be precisely inspected with controllable precision and reasonable distribution, significantly improving the physical consistency and detection effectiveness of the area partitioning. In the stage of precise inspection resource allocation, the system generates a matching resource configuration pool according to the set of resource allocation feature data of each region to be precisely inspected, in combination with a resource evaluation model, to achieve intelligent and adaptive resource allocation and parameter setting. That is, the number of sub-area partitions, the number of sensors and the sampling strategy are dynamically set according to the actual detection requirements of the area, so as to meet the increasingly complex requirements.

[0018] In addition, in the scenario of limited device resources, the present invention further introduces a comprehensive priority scoring mechanism to sort and schedule the regions to be precisely inspected, and preferentially ensure the detection integrity of the regions to be precisely inspected with high comprehensive priority scores; and dynamically schedule the occupied multi-channel eddy current sensors through a task release and recycling mechanism to achieve the maximization of resource utilization and the improvement of the overall system performance. Overall, the present invention can significantly improve the eddy current detection accuracy, efficiency and adaptive ability under multi-region heterogeneous structures without increasing the additional hardware burden, and has good engineering feasibility and practical application value. Description of the Drawings

[0019] Figure 1 Schematic diagram of a multi-channel eddy current sensor array according to Embodiment 1 of the present invention; Figure 2 Flowchart of a data processing method for a multi-channel eddy current sensor array according to Embodiment 2 of the present invention; Figure 3 Flowchart of a data processing method for a multi-channel eddy current sensor array according to Embodiment 3 of the present invention; Figure 4 Flowchart of a method for intelligently and adaptively allocating resources and setting parameters for R areas to be precisely inspected; Figure 5 Schematic diagram of the software and hardware structure of a multi-channel eddy current sensor array according to Embodiment 1 of the present invention. Detailed implementation manners

[0020] The technical solutions in the embodiments of the present invention will be described in detail, clearly and completely with reference to the accompanying drawings in the embodiments of the present invention. It should be specifically noted that the following described specific embodiments are only used to better illustrate and explain the technical solutions of the present invention, aiming to enable those skilled in the art to better understand and implement the present invention, and should not be construed as a limitation on the protection scope of the present invention. Without departing from the spirit and essence of the present invention, those skilled in the art can modify, adjust or make equivalent replacements according to the content disclosed in the present invention, and these should all be regarded as the protection scope of the present invention.

[0021] Embodiment 1

[0022] Please refer to Figure 1 As shown, this embodiment discloses a multi-channel eddy current sensor array, including a first partitioning module, a first processing module, a second partitioning module, and a resource allocation module. Each module is connected by wire and / or wirelessly to achieve data transmission.

[0023] The first partitioning module divides the detection area into N detection sub-areas according to a preset method, and allocates M multi-channel eddy current sensors to each detection sub-area.

[0024] The method for dividing the detection area into N detection sub-areas and setting M multi-channel eddy current sensors for each detection sub-area includes: Pre-collect the effective induction range of the multi-channel eddy current sensor, the total area of the detection area, the material type of the detection area, and the structural type of the detection area; the material type of the detection area includes aluminum, copper, magnesium, titanium, steel, etc.; the structural type of the detection area includes regular plane, curved structure, concave-convex surface, multi-layer composite structure, etc.; Numericalize the material type of the detection area to obtain the material type value of the detection area; numericalize the structure type of the detection area to obtain the structure type value of the detection area. For example, aluminum can be numericalized to 1, copper to 2, magnesium to 3, titanium to 4, and steel to 5. A regular plane can be numericalized to 11, a curved structure to 12, a concave-convex surface to 13, and a multi-layer composite structure to 14.

[0025] Input the effective induction range of the multi-channel eddy current sensor, the total area of the detection area, the material type value of the detection area, and the structure type value of the detection area into the sub-region division parameter setting model to obtain the number of detection sub-regions and the number of multi-channel eddy current sensors. The number of detection sub-regions is N; the number of multi-channel eddy current sensors is M. Divide the detection area into N detection sub-regions equally, and allocate M multi-channel eddy current sensors to each detection sub-region.

[0026] The training method of the sub-region division parameter setting model includes: Pre-construct a sub-region division parameter data set, which includes HF group sub-region division parameter data and the corresponding number of detection sub-regions and multi-channel eddy current sensors. HF is a positive integer greater than 0. The sub-region division parameter data includes the effective induction range of the multi-channel eddy current sensor, the total area of the detection area, the material type value of the detection area, and the structure type value of the detection area. Divide the sub-region division parameter data set into a sub-region division parameter data training set and a sub-region division parameter data verification set. The sub-region division parameter data training set is used for parameter learning of the sub-region division parameter setting model, and the sub-region division parameter data verification set is used to evaluate the generalization ability of the sub-region division parameter setting model in real time.

[0027] During the training process of the sub-region division parameter setting model, the sub-region division parameter setting model adopts a support vector machine model or a decision tree model. Convert the sub-region division parameter data into a feature vector as the input, extract the non-linear features in the data through the hidden layer, and finally generate the probability distribution of the number of detection sub-regions and multi-channel eddy current sensors using the softmax activation function in the output layer. Output the number of detection sub-regions and multi-channel eddy current sensors corresponding to the maximum probability as the final prediction result. The training process aims to minimize the cross-entropy loss function, and at the same time introduce an early stopping strategy to monitor the performance of the sub-region division parameter data verification set. When the prediction accuracy on the sub-region division parameter data verification set reaches the preset accuracy, it is considered that the sub-region division parameter setting model has converged, and the training stops immediately.

[0028] The first processing module extracts features from the set of sensor signal data collected by the multi-channel eddy current sensors corresponding to N detection sub-regions to obtain the sensor signal feature data corresponding to the N detection sub-regions. The set of sensor signal data includes H sets of sensor signal data, where H = M × B, and B is the number of probe units of the multi-channel eddy current sensor, that is, the number of channels; the sensor signal data includes the real part of impedance, the imaginary part of impedance, the amplitude of impedance, and the phase; the real part of impedance refers to the resistance component, and the imaginary part of impedance refers to the inductive reactance component.

[0029] It should be noted that there are M multi-channel eddy current sensors, and each multi-channel eddy current sensor contains B independent-channel probe units. The value of B is determined by the hardware attributes of the multi-channel eddy current sensor itself. The system can obtain M × B sets of sensor signal data in each detection sub-region.

[0030] The method for obtaining the sensor signal feature data corresponding to N detection sub-regions includes: S100: Let the initial value of n be 1, and the value range of n is from 1 to N; S101: According to the spatial arrangement order of the probe units of the multi-channel eddy current sensor, the real part of impedance matrix, the imaginary part of impedance matrix, the amplitude of impedance matrix, and the phase matrix, obtain the H sets of sensor signal data of the nth detection sub-region; map and fill the real part of impedance in the H sets of sensor signal data into the real part of impedance matrix; map and fill the imaginary part of impedance in the H sets of sensor signal data into the imaginary part of impedance matrix; map and fill the amplitude of impedance in the H sets of sensor signal data into the amplitude of impedance matrix; map and fill the phase in the H sets of sensor signal data into the phase matrix; S102: Analyze and process the phase based on the phase matrix to obtain the phase feature data of the nth detection sub-region; analyze and process the real part of impedance based on the real part of impedance matrix to obtain the real part of impedance feature data of the nth detection sub-region; analyze and process the imaginary part of impedance and the amplitude of impedance based on the imaginary part of impedance matrix and the amplitude of impedance matrix to obtain the fusion feature data; S103: Construct the sensor signal feature data of the nth detection sub-region with the phase feature data, the real part of impedance feature data, and the fusion feature data;

[0031] S104: Let n = n + 1. If n is less than or equal to N, continue to execute S101 to S103; if n is greater than N, obtain the sensor signal feature data corresponding to N detection sub-regions and end the current process.

[0032] The method for obtaining the phase feature data includes: S200: Let the initial value of h be 1, and the value range of h is from 1 to H; calculate the phase mean based on the phase in the phase matrix; let the initial value of the number of phase mutations be 0; The method for obtaining the phase mean value includes: ; wherein, is the phase mean value, h is the index variable in the summation formula, is the h-th phase in the phase matrix.

[0033] S201: Obtain the h-th phase from the phase matrix, obtain the phase adjacent to the h-th phase from the phase matrix, denote it as the adjacent phase, and denote the number of adjacent phases as ; Take the absolute value of the difference between the h-th phase and each of the adjacent phases to obtain phase gradients; if there is a phase gradient greater than or equal to a preset phase gradient threshold among the phase gradients, then mark the h-th phase as a mutant phase point and increment the phase mutation count by 1; It should be noted that the phase gradient threshold can be obtained based on the statistical distribution of the phase gradients in the defect-free samples, and for example, the 95% or 99% quantile is extracted as the threshold.

[0034] S202: Let h = h + 1. If h is less than or equal to H, then continue to execute S201; if h is greater than H, then execute S203; S203: Denote the number of mutant phase points as TBXW, and calculate the phase mutation degree based on the phase mean value and the phases of the TBXW mutant phase points; divide the phase mutation count by H to obtain the phase mutation ratio; The method for obtaining the phase mutation degree includes: ; wherein, is the phase mutation degree, is the index variable of the summation formula, is the phase corresponding to the i-th mutant phase point.

[0035] S204: Construct phase feature data from the phase mutation degree and the phase mutation ratio, and end the current process.

[0036] It should be noted that the present invention further introduces two characteristic indexes, namely the phase mutation degree and the phase mutation ratio, to characterize the region with a drastic change in the local response in the phase matrix.

[0037] Among them, the degree of phase mutation is used to represent the magnitude of the phase difference between a certain channel and its adjacent channels, and can reflect the characteristics of local electromagnetic response delay caused by conductivity change, thickness mutation or crack propagation; the phase mutation ratio is used to count the proportion of channels exceeding the preset phase gradient threshold, and can reflect the extensiveness and aggregation trend of the defect area. Introducing the degree of phase mutation and the phase mutation ratio can further improve the system's recognition ability for concealed defects (such as internal cracks and deep material heterogeneity), and enhance the spatial integrity of regional feature expression and the accuracy of anomaly recognition.

[0038] The method for obtaining the real part of impedance characteristic data includes: S300: Let the initial value of h be 1, and the value range of h is from 1 to H; calculate the mean value of the real part of impedance based on the real part of impedance in the real part of impedance matrix; let the initial value of the number of real part of impedance mutations be 0; The method for obtaining the mean value of the real part of impedance includes: ; Among them, is the mean value of the real part of impedance, h is the index variable in the summation formula, is the h-th real part of impedance in the real part of impedance matrix.

[0039] S301: Obtain the h-th real part of impedance from the real part of impedance matrix, obtain the adjacent real part of impedance adjacent to the h-th real part of impedance from the real part of impedance matrix, denote it as the adjacent real part of impedance, and denote the number of adjacent real parts of impedance as ; Subtract the h-th real part of impedance from each of the adjacent real parts of impedance and take the absolute value to obtain real part of impedance gradients; if there is a real part of impedance gradient greater than or equal to the preset real part of impedance gradient threshold among the real part of impedance gradients, then mark the h-th real part of impedance as a mutated real part of impedance point, and let the number of real part of impedance mutations increase by 1; It should be noted that the real part of impedance gradient threshold can be obtained based on the statistical distribution of the real part of impedance gradient in the defect-free sample, and extract, for example, the 95% or 99% quantile as the threshold.

[0040] S302: Let h = h + 1. If h is less than or equal to H, then continue to execute S301; if h is greater than H, then execute S303; S303: Denote the number of mutated real part of impedance points as TBSB, calculate the degree of real part of impedance mutation based on the mean value of the real part of impedance and the real part of impedance of TBSB mutated real part of impedance points; divide the number of real part of impedance mutations by H to obtain the real part of impedance mutation ratio; The method for obtaining the degree of real part of impedance mutation includes: ; wherein, is the mutation degree of the real part of the impedance, is the index variable of the summation formula, is the real part of the impedance corresponding to the th mutation impedance real part point.

[0041] S304: Construct impedance real part feature data from the mutation degree of the real part of the impedance and the mutation ratio of the real part of the impedance, and end the current process.

[0042] It should be noted that in the present invention, by analyzing the response differences between channels in the real part of the impedance matrix, two characteristic indexes, namely the mutation degree of the real part of the impedance and the mutation ratio of the real part of the impedance, are constructed to characterize the local energy dissipation anomaly in the detection area.

[0043] Among them, the mutation degree of the real part of the impedance is used to measure the maximum difference in the real part of the impedance value between a certain channel in the detection area and its adjacent channels, and can effectively reflect structural mutation positions such as crack tips and corrosion pit edges; the mutation ratio of the real part of the impedance statistically represents the proportion of the number of channels meeting the mutation conditions in the detection sub-area, and is used to reflect the overall non-uniformity degree and the spatial distribution density of the abnormal area. Introducing the mutation degree of the real part of the impedance and the mutation ratio of the real part of the impedance as features helps to quickly lock high-risk sub-areas and reasonably schedule fine inspection resources, improving detection accuracy and efficiency.

[0044] The method for obtaining the fused feature data includes: S400: Set the initial value of h to 1, and the value range of h is from 1 to H; calculate the mean value of the imaginary part of the impedance based on the imaginary part of the impedance in the imaginary part of the impedance matrix; calculate the mean value of the amplitude of the impedance based on the amplitude of the impedance in the amplitude of the impedance matrix; S401: Obtain the hth imaginary part of the impedance from the imaginary part of the impedance matrix, obtain the hth amplitude of the impedance from the amplitude of the impedance matrix, and construct the hth fused feature pair from the hth imaginary part of the impedance and the amplitude of the impedance; add the fused feature pair to the fused feature pair set; It should be noted that the channel positions corresponding to the hth imaginary part of the impedance and the hth amplitude of the impedance are the same.

[0045] S402: Calculate the impedance imaginary part - amplitude ratio and the joint response amplitude corresponding to the hth fused feature pair based on the hth fused feature pair; S403: Add the impedance imaginary part - amplitude ratio to the impedance imaginary part - amplitude ratio set; add the joint response amplitude to the joint response amplitude set; S404: Let h = h + 1, if h is less than or equal to H, then continue to execute S401 to S403; if h is greater than H, then execute S405; S405: Calculate the correlation coefficient of the fusion feature based on the mean value of the imaginary part of impedance, the mean value of the impedance amplitude, the set of imaginary part - amplitude ratios of impedance, and the set of combined response amplitudes; S406: Construct the set of imaginary part - amplitude ratios of impedance, the set of combined response amplitudes, and the correlation coefficient of the fusion feature into the fusion feature data, and end the current process.

[0046] The method for obtaining the mean value of the imaginary part of impedance includes: ; where, is the mean value of the imaginary part of impedance, h is the index variable in the summation formula, is the h - th imaginary part of impedance in the imaginary part of impedance matrix.

[0047] The method for obtaining the mean value of the impedance amplitude includes: ; where, is the mean value of the impedance amplitude, h is the index variable in the summation formula, is the h - th impedance amplitude in the impedance amplitude matrix.

[0048] The method for calculating the imaginary part - amplitude ratio of impedance includes: ; where, is the imaginary part - amplitude ratio corresponding to the h - th fusion feature pair; the imaginary part - amplitude ratio of impedance is used to reflect the proportion of the inductive reactance component in the overall response under the corresponding channel. An increase in the imaginary part - amplitude ratio of impedance indicates phenomena such as thickness thinning and magnetic coupling enhancement, and has the significance of defect indication.

[0049] The method for calculating the combined response amplitude includes: ; where, is the h - th combined response amplitude. The combined response amplitude can be regarded as the modulus value of a two - dimensional vector composed of the inductive reactance characteristics and the overall response intensity, and is used to comprehensively reflect the change trend of the intensity and structural characteristics of the corresponding channel in the electromagnetic response. Compared with a single physical quantity, the combined response amplitude has stronger sensitivity and discrimination ability in identifying composite defects, boundary mutations, and material coupling changes, etc.; The method for calculating the correlation coefficient of the fusion feature includes: ; where, It is the correlation coefficient of the fusion feature. By centering the imaginary part and the magnitude of the impedance (subtracting the average value) respectively, the offset interference caused by different amplitude bases between different detection sub-regions can be eliminated; the denominator is normalized by the standard deviation, making it more concerned about the "trend of change" rather than the absolute value difference, which is suitable for cross-detection sub-region feature analysis. If the imaginary part and the magnitude of the impedance have a consistent fluctuation trend, it indicates that the response of this region is stable; if the fluctuations of the two are decoupled or the differences are significant, it means that there are structural abnormalities, signal disturbances or defects.

[0050] It should be noted that there are H groups of sensor signal data in the detection sub-region. The corresponding real part matrix of impedance, imaginary part matrix of impedance, magnitude matrix of impedance and phase matrix are obtained through the H groups of sensor signal data. Therefore, when processing the feature data, h is used to maintain consistency.

[0051] Based on the construction of the imaginary part matrix of impedance and the magnitude matrix of impedance, the present invention adopts a combined fusion extraction method to extract the fusion features of the two types of matrices. The imaginary part matrix of impedance represents the distribution of the inductive reactance component in each channel during the detection process, mainly reflecting the changes in the magnetic permeability, thickness and excitation coupling degree of the material; the magnitude matrix of impedance reflects the comprehensive response intensity (impedance modulus value) of each channel, comprehensively reflecting the conductivity, structural integrity and defect influence. Since there is a strong correlation between the imaginary part and the magnitude of the impedance in the detection physical mechanism, it is suitable for joint analysis to improve the feature interpretation ability. Through the fusion extraction of the imaginary part and the magnitude data of the impedance, the present invention can more comprehensively describe the composite change trend of the material properties and structural characteristics inside the detection region.

[0052] The correlation coefficient of the fusion feature can effectively measure the change consistency between the imaginary part and the magnitude of the impedance in the detection sub-region, and be used as a reference basis for the regional response stability, structural integrity and defect determination. When the correlation coefficient of the fusion feature of the detection sub-region decreases significantly, it indicates that the response characteristics of the detection sub-region are disordered, and there are problems of composite defects, signal disturbances or material discontinuity.

[0053] The second partitioning module constructs regions for the N detection sub-regions based on the sensor signal feature data corresponding to the N detection sub-regions, and obtains R regions to be refined for inspection; The method for obtaining the R regions to be refined for inspection includes: S500: Let the initial value of n be 1, and the value range of n is from 1 to N; initialize the region construction flag of the N detection sub-regions as not constructed; let the initial value of the region count variable R for the regions to be refined for inspection be 1; S501: If the region construction flag of the nth detection sub-region is not constructed, then execute S502; if the region construction flag of the nth detection sub-region is constructed, then execute S505; S502: Construct the detection sub-regions adjacent to the nth detection sub-region and marked as unconstructed in terms of region construction into the to-be-processed region corresponding to the nth detection sub-region; convert the sensor signal feature data of the nth detection sub-region and the corresponding to-be-processed region into sensor signal feature vectors, and construct them into the Rth sensor signal feature vector set. S503: Apply a clustering algorithm to classify the Rth sensor signal feature vector set; add the detection sub-regions corresponding to the sensor signal feature vectors belonging to the same clustering cluster as the nth detection sub-region in the clustering result to the Rth to-be-constructed set; the clustering algorithms include K-Means and DBSCAN. Determine whether there are detection sub-regions adjacent to the detection sub-regions in the Rth to-be-constructed set and marked as unconstructed in terms of region construction; if so, convert the sensor signal feature data of the corresponding detection sub-regions into sensor signal feature vectors, and add them to the Rth sensor signal feature vector set, and continue to execute S503; if not, execute S504. S504: Construct the detection sub-regions in the to-be-constructed set into the Rth to-be-precisely-inspected region, and let R = R + 1; set the region construction marks of the detection sub-regions in the Rth to-be-constructed set to constructed. S505: Let n = n + 1. If n is less than or equal to N, continue to execute S501 to S504; if n is greater than N, obtain R to-be-precisely-inspected regions, and end the current process.

[0054] It should be noted that the Rth sensor signal feature vector set and the Rth to-be-constructed set described in this specification are both used to represent the relevant data set and the detection object set corresponding to the "Rth to-be-precisely-inspected region", and there is a one-to-one correspondence among the three. Among them, the "Rth sensor signal feature vector set" represents the data input set participating in the clustering analysis, the "Rth to-be-constructed set" represents the set of detection sub-regions belonging to the same cluster after the clustering analysis, and finally the "Rth to-be-precisely-inspected region" is constructed. Therefore, the Rth to-be-precisely-inspected region, its corresponding feature vector set, and the to-be-constructed set maintain consistency and uniqueness during the process execution and have a clear corresponding relationship.

[0055] By introducing a recursive spatial expansion mechanism, it is ensured that all detection sub-regions with consistent features and adjacent in space are included in the same to-be-precisely-inspected region, effectively avoiding the region splitting phenomenon caused by boundary jumps or scattered critical points; at the same time, through the combination of adjacent expansion and clustering attribution, the adaptive decision-making and minimum manual intervention in the region construction process are realized, which is applicable to the data partitioning task of large-scale and high-resolution sensor arrays. The method for constructing the to-be-precisely-inspected region not only has higher accuracy and robustness in the response space structure partitioning, but also provides a more targeted and controllable region input basis for subsequent defect location and resource scheduling.

[0056] A resource allocation module, based on the sensor signal feature data corresponding to N detection sub-regions, intelligently and adaptively allocates resources and sets parameters for R regions to be precisely inspected.

[0057] As Figure 4 shown, the method for intelligently and adaptively allocating resources and setting parameters for R regions to be precisely inspected includes: S600: Let the initial value of r be 1, and the value range of r is from 1 to R; S601: Denote the number of detection sub-regions in the r-th region to be precisely inspected as ; Input the sensor signal feature data corresponding to detection sub-regions into the region demand assessment model respectively to obtain the corresponding region demand levels; the region demand levels include low demand, medium demand, and high demand; S602: Construct the corresponding resource allocation feature data for each detection sub-region and the corresponding region demand level in the r-th region to be precisely inspected, and construct a resource allocation feature data set; S603: Input the resource allocation feature data set of the r-th region to be precisely inspected into the resource configuration setting model to obtain a resource configuration pool; the resource configuration pool includes the allocation quantity of multi-channel eddy current sensors, scanning time, sampling frequency, resolution, and compression ratio; S604: Allocate resources and set parameters for the r-th region to be precisely inspected according to the corresponding resource configuration pool; S605: Let r = r + 1. If r is less than or equal to R, continue to execute S601 to S604; if r is greater than R, complete the intelligent and adaptive allocation of resources and parameter setting for R regions to be precisely inspected, and end the current process.

[0058] It should be noted that a schematic diagram of the software and hardware structure of a multi-channel eddy current sensor array is as Figure 5 shown. The multi-channel eddy current sensor array has a hardware layer and an algorithm layer. The hardware layer of the multi-channel eddy current sensor array includes a multi-channel array probe, a multi-channel acquisition unit, a precision displacement sensor, and a multi-axis adjustment bracket; the algorithm layer of the multi-channel eddy current sensor array includes a dynamic region division algorithm, a multi-dimensional feature extraction module, an adaptive clustering analysis, and a resource allocation optimization model; exemplarily, the multi-channel acquisition unit can use a 16-channel acquisition unit, and the multi-axis adjustment bracket can use a 6-axis adjustment bracket. The region demand level is used to determine the resource guarantee level required by the region to be precisely inspected during the actual detection process. The higher the region demand level, the more significant the structural interference, signal interference enhancement, or defect masking risk in the region to be precisely inspected. Therefore, more sensor channels need to be allocated, the sampling frequency needs to be increased, the scanning time needs to be extended, or a low compression ratio (retaining more details) needs to be adopted.

[0059] The resolution is used to control the spatial resolution of the response data collected or generated, the signal reconstruction accuracy, or the clarity level of the subsequent image visualization result during the detection process in the area to be precisely inspected. When performing multi-area detection tasks, it can intelligently control the spatial resolution of data collection according to the actual detection requirements of the target area, thereby effectively compressing redundant data while ensuring detection accuracy, and improving the overall system operation efficiency, computing resource utilization rate, and response speed.

[0060] By introducing a regional demand level evaluation and feature-driven resource evaluation mechanism, based on the feature differences and demand differences of the internal detection sub-areas in each area to be precisely inspected, this method realizes the refined and intelligent allocation of detection resources. Compared with the traditional static mean allocation method, this method can dynamically match the number of sensors, scanning strategies, and data processing capabilities according to the complexity of the detection task, effectively improving the adaptability and resource utilization efficiency of the detection system in multi-area and complex target environments.

[0061] In addition, the multi-channel eddy current sensors in each area to be precisely inspected are constructed into a multi-channel eddy current sensor array. Each multi-channel eddy current sensor array has the physical characteristics of being detachable or spliceable, facilitating free combination according to the workpiece size or detection area. Ordinary density probes can be used for large-range and rapid detection areas, and smaller probes with higher density can be used for key areas requiring high resolution.

[0062] The multi-channel eddy current sensors in the multi-channel eddy current sensor array are assembled on an adjustable support base, and fine adjustment of several degrees of freedom (height / tilt angle, etc.) is achieved through mechanisms such as springs, lead screws, or micro-motors. The support mechanism can automatically adapt to the workpiece surface with different curvatures during the installation stage or the detection process, reducing the gap difference between the probes of the multi-channel eddy current sensors and the workpiece.

[0063] The training method of the regional demand evaluation model includes: Pre-collect a regional risk assessment data set, which includes FX groups of regional risk assessment data and the corresponding regional demand levels for the FX groups of regional risk assessment data. FX is a positive integer greater than 0. The regional risk assessment data includes sensor signal feature data; divide the regional risk assessment data set into a training set and a validation set, where the training set is used to train the regional demand evaluation model, and the validation set is used to evaluate the generalization performance of the regional demand evaluation model; During the training process of the regional demand assessment model, the minimization of the cross-entropy loss function is used as the optimization objective, and the early stopping strategy is used to monitor the performance of the validation set. By continuously adjusting the network parameters, the model performance is optimized. When the prediction accuracy rate on the validation set reaches the expected accuracy rate, it is considered that the regional demand assessment model has converged, and the training is stopped. The regional demand assessment model is trained based on the Naive Bayes model. Convert the regional risk assessment data into feature vectors. The input layer of the regional demand assessment model receives the feature vectors, extracts the non-linear relationships in the data through the hidden layer, and finally the output layer of the regional demand assessment model calculates the probability distribution of the regional demand levels through the softmax activation function, and outputs the regional demand level corresponding to the maximum probability as the final prediction result.

[0064] The training method of the resource allocation setting model includes: Pre-collect a resource allocation setting data set, which includes SD group resource allocation setting data and the resource allocation pool corresponding to the SD group resource allocation setting data, where SD is a positive integer greater than 0. The resource allocation setting data includes a set of resource allocation feature data. Divide the resource allocation setting data set into a training set and a validation set, where the training set is used to train the resource allocation setting model, and the validation set is used to evaluate the generalization performance of the resource allocation setting model. During the training process of the resource allocation setting model, the minimization of the cross-entropy loss function is used as the optimization objective, and the early stopping strategy is used to monitor the performance of the validation set. By continuously adjusting the network parameters, the model performance is optimized. When the prediction accuracy rate on the validation set reaches the expected accuracy rate, it is considered that the resource allocation setting model has converged, and the training is stopped. The resource allocation setting model is trained using the Random Forest model. Convert the resource allocation setting data into feature vectors. The input layer of the resource allocation setting model receives the feature vectors, extracts the non-linear relationships in the data through the hidden layer, and finally the output layer of the resource allocation setting model calculates the probability distribution of the resource allocation pool through the softmax activation function, and outputs the resource allocation pool corresponding to the maximum probability as the final prediction result.

[0065] Embodiment 2

[0066] Please refer to Figure 2 As shown, this embodiment provides a data processing method for a multi-channel eddy current sensor array, including: Divide the detection area into N detection sub-areas according to a preset method, and allocate M multi-channel eddy current sensors to each detection sub-area. Based on the set of sensor signal data collected by the multi-channel eddy current sensors corresponding to the N detection sub-areas, feature extraction is performed to obtain the sensor signal feature data corresponding to the N detection sub-areas. Based on the sensor signal feature data corresponding to N detection sub-regions, N detection sub-regions are constructed to obtain R regions to be precisely inspected; Based on the sensor signal feature data corresponding to N detection sub-regions, resources and parameter settings are intelligently and adaptively allocated for the R regions to be precisely inspected.

[0067] Embodiment 3

[0068] Please refer to Figure 3 As shown, this embodiment provides a data processing method for a multi-channel eddy current sensor array, further including: When the number of multi-channel eddy current sensor resources is limited, priority scheduling and allocation are performed on the multi-channel eddy current sensors in the R regions to be precisely inspected.

[0069] The method for priority scheduling and allocation of the multi-channel eddy current sensors in the R regions to be precisely inspected includes: Step 1: Input the resource allocation feature data set and the resource configuration pool of the R regions to be precisely inspected into the comprehensive priority evaluation model respectively to obtain the comprehensive priority scores corresponding to the R regions to be precisely inspected; sort the R regions to be precisely inspected in descending order according to the comprehensive priority scores to form a region priority queue; execute Step 2; Step 2: Count the total number of currently available multi-channel eddy current sensors, denoted as the currently available quantity; when the region priority queue is not empty, execute Step 3; when the region priority queue is empty, it means that all regions to be precisely inspected have been allocated, then execute Step 5; Step 3: Determine whether the currently available quantity is greater than or equal to the number of multi-channel eddy current sensors to be allocated required by the region to be precisely inspected corresponding to the head of the region priority queue; if the condition is met, set the required number of multi-channel eddy current sensors in the resource configuration pool corresponding to the head of the region priority queue, remove the head of the region priority queue from the region priority queue, and execute Step 2; if the condition is not met, execute Step 4; Step 4: The system enters the waiting state; when the multi-channel eddy current sensor resources are released by the region to be precisely inspected that has completed the task, execute Step 2; Step 5: Complete the priority scheduling and allocation process of the multi-channel eddy current sensors in the R regions to be precisely inspected, and end the current scheduling process.

[0070] The training method of the comprehensive priority evaluation model includes: Pre-collect a comprehensive priority evaluation data set, which includes the comprehensive priority evaluation data of the ZH group and the corresponding comprehensive priority scores, where ZH is a positive integer greater than 0. The comprehensive priority evaluation data includes a set of resource allocation feature data and a resource configuration pool; divide the comprehensive priority evaluation data set into a training set and a validation set, where the training set is used to train the comprehensive priority evaluation model, and the validation set is used to evaluate the generalization performance of the comprehensive priority evaluation model; During the training process of the comprehensive priority evaluation model, minimize the cross-entropy loss function as the optimization goal, use the early stopping strategy to monitor the performance of the validation set, and optimize the model performance by continuously adjusting the network parameters; when the prediction accuracy on the validation set reaches the expected accuracy, it is considered that the comprehensive priority evaluation model has converged and stop training; the comprehensive priority evaluation model is trained using a support vector machine model; Convert the comprehensive priority evaluation data into feature vectors; the input layer of the comprehensive priority evaluation model receives the feature vectors, extracts the non-linear relationships in the data through the hidden layer, and finally the output layer of the comprehensive priority evaluation model calculates the probability distribution of the comprehensive priority scores through the softmax activation function, and outputs the comprehensive priority score corresponding to the maximum probability as the final prediction result.

[0071] As described above, only the specific embodiments of the present invention are provided, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

[0072] Finally: The above are only the preferred embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should all be included in the protection scope of the present invention.

Claims

1. A data processing method for a multi-channel eddy current sensor array, characterized in that Including: Dividing the detection area into N detection sub-areas according to a preset method, and allocating M multi-channel eddy current sensors to each detection sub-area; Performing feature extraction based on the set of sensor signal data collected by the multi-channel eddy current sensors corresponding to the N detection sub-areas to obtain the sensor signal feature data corresponding to the N detection sub-areas; Performing area construction on the N detection sub-areas based on the sensor signal feature data corresponding to the N detection sub-areas to obtain R areas to be precisely inspected; Based on the sensor signal feature data corresponding to the N detection sub-areas, intelligently and adaptively allocating resources and setting parameters for the R areas to be precisely inspected.

2. The data processing method of a multi-channel eddy current sensor array according to claim 1, characterized in that, The method for intelligently and adaptively allocating resources and setting parameters for the R areas to be precisely inspected includes: S600: Let the initial value of r be 1, and the value range of r is from 1 to R; S601: Denote the number of detection sub-regions in the r-th region to be precisely inspected as ; and input the sensor signal feature data corresponding to detection sub-regions into the region demand assessment model respectively to obtain the corresponding region demand levels; the region demand levels include low demand, medium demand, and high demand; S602: Construct the corresponding resource allocation feature data for each detection sub-area and the corresponding area requirement level in the r-th area to be precisely inspected, and construct a resource allocation feature data set; S603: Input the resource allocation feature data set of the r-th area to be precisely inspected into the resource configuration setting model to obtain a resource configuration pool; the resource configuration pool includes the allocated number of multi-channel eddy current sensors, scanning time, sampling frequency, resolution, and compression ratio; S604: Allocate resources and set parameters for the r-th area to be precisely inspected according to the corresponding resource configuration pool; S605: Let r = r + 1. If r is less than or equal to R, then continue to execute S601 to S604; if r is greater than R, then complete the intelligent and adaptive allocation of resources and parameter setting for the R areas to be precisely inspected, and end the current process.

3. The data processing method of a multi-channel eddy current sensor array according to claim 1, characterized in that The method for obtaining the sensor signal feature data corresponding to the N detection sub-areas includes: S100: Let the initial value of n be 1, and the value range of n is from 1 to N; S101: According to the spatial arrangement order of the probe units of the multi-channel eddy current sensors, the real part matrix of impedance, the imaginary part matrix of impedance, the amplitude matrix of impedance, and the phase matrix, obtain H groups of sensor signal data for the n-th detection sub-area; map the real part of impedance in the H groups of sensor signal data into the real part matrix of impedance; map the imaginary part of impedance in the H groups of sensor signal data into the imaginary part matrix of impedance; map the amplitude of impedance in the H groups of sensor signal data into the amplitude matrix of impedance; map the phase in the H groups of sensor signal data into the phase matrix; S102: Based on the phase matrix, analyze and process the phase to obtain the phase feature data for the n-th detection sub-area; based on the real part matrix of impedance, analyze and process the real part of impedance to obtain the real part feature data of impedance for the n-th detection sub-area; based on the imaginary part matrix of impedance and the amplitude matrix of impedance, analyze and process the imaginary part of impedance and the amplitude of impedance to obtain the fusion feature data; S103: Construct the sensor signal feature data for the n-th detection sub-area from the phase feature data, the real part feature data of impedance, and the fusion feature data; S104: Let n = n + 1. If n is less than or equal to N, then continue to execute S101 to S103; if n is greater than N, then obtain the sensor signal feature data corresponding to the N detection sub-areas, and end the current process.

4. A data processing method for a multi-channel eddy current sensor array according to claim 3, characterized in that The method for obtaining the phase feature data includes: S200: Let the initial value of h be 1, and the value range of h is from 1 to H; calculate the phase mean value based on the phases in the phase matrix; let the initial value of the phase mutation quantity be 0; S201: Obtain the h-th phase from the phase matrix, obtain the phase adjacent to the h-th phase from the phase matrix, denote it as the adjacent phase, and denote the number of adjacent phases as ; The h-th phase is respectively subtracted from adjacent phases, and the absolute value is taken to obtain phase gradients; if there is a phase gradient greater than or equal to a preset phase gradient threshold among the phase gradients, then the h-th phase is marked as a mutation phase point, and the phase mutation count is incremented by 1; S202: Let h = h + 1. If h is less than or equal to H, then continue to execute S201; if h is greater than H, then execute S203; S203: Denote the number of mutation phase points as TBXW, calculate the phase mutation degree based on the phase mean value and the phases of the TBXW mutation phase points; divide the phase mutation quantity by H to obtain the phase mutation ratio; S204: Construct the phase mutation degree and the phase mutation ratio into phase feature data, and end the current process.

5. A data processing method for a multi-channel eddy current sensor array according to claim 3, characterized in that, The method for obtaining the real part of impedance feature data includes: S300: Let the initial value of h be 1, and the value range of h is from 1 to H; calculate the real part of impedance mean value based on the real part of impedance in the real part of impedance matrix; let the initial value of the real part of impedance mutation quantity be 0; S301: Obtain the h-th real part of impedance from the real part matrix of impedance, obtain the real part of impedance adjacent to the h-th real part of impedance from the real part matrix of impedance, denote it as the adjacent real part of impedance, and denote the number of adjacent real parts of impedance as ; The real part of the h-th impedance is respectively subtracted from the real parts of adjacent impedances, and the absolute value is taken to obtain the gradients of the real parts of the impedance; if there is a gradient of the real part of the impedance greater than or equal to the preset threshold of the gradient of the real part of the impedance among the gradients of the real parts of the impedance, then the real part of the h-th impedance is marked as a mutation point of the real part of the impedance, and the number of mutations of the real part of the impedance is incremented by 1; S302: Let h = h + 1. If h is less than or equal to H, then continue to execute S301; if h is greater than H, then execute S303; S303: Denote the number of mutation real part of impedance points as TBSB, calculate the real part of impedance mutation degree based on the real part of impedance mean value and the real part of impedance of the TBSB mutation real part of impedance points; divide the real part of impedance mutation quantity by H to obtain the real part of impedance mutation ratio; S304: Construct the real part of impedance mutation degree and the real part of impedance mutation ratio into the real part of impedance feature data, and end the current process.

6. A data processing method for a multi-channel eddy current sensor array according to claim 3, characterized in that, The method for obtaining the fusion feature data includes: S400: Let the initial value of h be 1, and the value range of h is from 1 to H; calculate the imaginary part of impedance mean value based on the imaginary part of impedance in the imaginary part of impedance matrix; calculate the amplitude of impedance mean value based on the amplitude of impedance in the amplitude of impedance matrix; S401: Obtain the h-th imaginary part of impedance from the imaginary part of impedance matrix, obtain the h-th amplitude of impedance from the amplitude of impedance matrix, and construct the h-th imaginary part of impedance and the amplitude of impedance into the h-th fusion feature pair; add the fusion feature pair to the fusion feature pair set; S402: Calculate the impedance imaginary part - amplitude ratio and the joint response amplitude corresponding to the h-th fusion feature pair based on the h-th fusion feature pair; S403: Add the impedance imaginary part - amplitude ratio to the impedance imaginary part - amplitude ratio set; add the joint response amplitude to the joint response amplitude set; S404: Let h = h + 1. If h is less than or equal to H, then continue to execute S401 to S403; if h is greater than H, then execute S405; S405: Calculate the fusion feature correlation coefficient based on the imaginary part of impedance mean value, the amplitude of impedance mean value, the impedance imaginary part - amplitude ratio set and the joint response amplitude set; S406: Construct the impedance imaginary part - amplitude ratio set, the joint response amplitude set and the fusion feature correlation coefficient into the fusion feature data, and end the current process.

7. A data processing method for a multi-channel eddy current sensor array according to claim 1, characterized in that, The method for obtaining R regions to be precisely inspected includes: S500: Let the initial value of n be 1, and the value range of n is from 1 to N; initialize the region construction identifier of the N detection sub-regions as not constructed; let the initial value of the region to be precisely inspected counting variable R be 1; S501: If the region construction flag of the nth detection sub-region is not constructed, then execute S502; if the region construction flag of the nth detection sub-region is already constructed, then execute S505; S502: Construct the detection sub-regions adjacent to the nth detection sub-region and with the region construction flag not constructed into the to-be-processed region corresponding to the nth detection sub-region; convert the sensor signal feature data of the nth detection sub-region and the corresponding to-be-processed region into sensor signal feature vectors, and construct them into the Rth sensor signal feature vector set; S503: Apply a clustering algorithm to classify the Rth sensor signal feature vector set; add the detection sub-regions corresponding to the sensor signal feature vectors that belong to the same clustering cluster as the nth detection sub-region in the clustering result to the Rth to-be-constructed set; Judge whether there are detection sub-regions adjacent to the detection sub-regions in the Rth to-be-constructed set and with the region construction flag not constructed; if there are, then convert the sensor signal feature data of the corresponding detection sub-regions into sensor signal feature vectors, and add them to the Rth sensor signal feature vector set, and continue to execute S503; if not, then execute S504; S504: Construct the detection sub-regions in the to-be-constructed set into the Rth to-be-precisely-inspected region, and let R = R + 1; set the region construction flag of the detection sub-regions in the Rth to-be-constructed set to already constructed; S505: Let n = n + 1, if n is less than or equal to N, then continue to execute S501 to S504; if n is greater than N, then obtain R to-be-precisely-inspected regions, and end the current process.

8. A data processing method for a multi-channel eddy current sensor array according to claim 1, characterized in that, Divide the detection region into N detection sub-regions according to a preset method, and allocate M multi-channel eddy current sensors to each detection sub-region, including: Pre-collect the effective induction range of the multi-channel eddy current sensors, the total area of the detection region, the material type of the detection region, and the structural type of the detection region; the material type of the detection region includes aluminum, copper, magnesium, titanium, and steel; the structural type of the detection region includes regular plane, curved structure, concave-convex surface, and multi-layer composite structure; Numericalize the material type of the detection region to obtain the material type value of the detection region; numericalize the structural type of the detection region to obtain the structural type value of the detection region; Input the effective induction range of the multi-channel eddy current sensors, the total area of the detection region, the material type value of the detection region, and the structural type value of the detection region into the sub-region division parameter setting model to obtain the number of detection sub-regions and the number of multi-channel eddy current sensors, where the number of detection sub-regions is N; the number of multi-channel eddy current sensors is M; Divide the detection region into N equal detection sub-regions, and allocate M multi-channel eddy current sensors to each detection sub-region.

9. A data processing method for a multi-channel eddy current sensor array according to claim 8, characterized in that The training method of the sub-region division parameter setting model includes: Pre-build a sub-region division parameter data set, where the sub-region division parameter data set includes sub-region division parameter data for the HF group and the number of detection sub-regions and the number of multi-channel eddy current sensors corresponding to the sub-region division parameter data for the HF group. HF is a positive integer greater than 0. The sub-region division parameter data includes the effective induction range of the multi-channel eddy current sensor, the total area of the detection region, the numerical value of the material type of the detection region, and the numerical value of the structural type of the detection region; divide the sub-region division parameter data set into a sub-region division parameter data training set and a sub-region division parameter data validation set, where the sub-region division parameter data training set is used for parameter learning of the sub-region division parameter setting model, and the sub-region division parameter data validation set is used to evaluate the generalization ability of the sub-region division parameter setting model in real time; During the training process of the sub-region division parameter setting model, the sub-region division parameter setting model adopts a support vector machine model or a decision tree model, converts the sub-region division parameter data into a feature vector as input, extracts non-linear features in the data through the hidden layer, and finally generates the probability distribution of the number of detection sub-regions and the number of multi-channel eddy current sensors in the output layer using the softmax activation function, and outputs the number of detection sub-regions and the number of multi-channel eddy current sensors corresponding to the maximum probability as the final prediction result; the training process aims to minimize the cross-entropy loss function, and at the same time introduces an early stopping strategy to monitor the performance of the sub-region division parameter data validation set. When the prediction accuracy on the sub-region division parameter data validation set reaches the preset accuracy, it is considered that the sub-region division parameter setting model has converged, and the training stops immediately.

10. A data processing method for a multi-channel eddy current sensor array according to claim 1, characterized in that, The sensor signal data set includes H groups of sensor signal data, where H = M × B, and B is the number of probe units of the multi-channel eddy current sensor, that is, the number of channels; the sensor signal data includes the real part of the impedance, the imaginary part of the impedance, the amplitude of the impedance, and the phase.

11. A multi-channel eddy current sensor array for implementing the data processing method of the multi-channel eddy current sensor array according to any one of claims 1-10, characterized in that, Including: The first division module divides the detection region into N detection sub-regions according to a preset method, and assigns M multi-channel eddy current sensors to each detection sub-region; The first processing module extracts features based on the sensor signal data set collected by the multi-channel eddy current sensors corresponding to the N detection sub-regions to obtain sensor signal feature data corresponding to the N detection sub-regions; The second division module constructs regions for the N detection sub-regions based on the sensor signal feature data corresponding to the N detection sub-regions to obtain R regions to be refined; The resource allocation module intelligently and adaptively allocates resources and parameter settings for the R regions to be refined based on the sensor signal feature data corresponding to the N detection sub-regions.

Images