Intelligent defect recognition and data analysis system based on phased array ultrasound

Abstract

This invention discloses a phased array ultrasonic defect intelligent identification and data analysis system, relating to the field of PAUT (Polymerization and Angiography). It includes: an acquisition module for driving a phased array ultrasonic sensor array to perform weld seam scanning according to preset scanning rules, simultaneously acquiring and outputting ultrasonic echo time-series data across the entire depth range of the weld seam; and a preprocessing module for receiving the ultrasonic echo time-series data, performing data correction, noise filtering, and initial screening of effective signals, while simultaneously receiving feature anchoring parameters output by a mapping module, refining the signals in the corresponding regions, and outputting the results. This invention enables continuous, blind-zone-free scanning across the entire thickness and width range of the weld seam, simultaneously completing accurate echo signal correction and defect feature extraction, improving defect identification accuracy through closed-loop adaptive processing, accurately determining defect types and calculating their spatial location, size, and other core parameters, and simultaneously completing defect data statistical analysis and welding process correlation.

Description

Technical Field

[0001] This invention relates to the field of PAUT technology, specifically to a defect intelligent identification and data analysis system based on phased array ultrasound. Background Technology

[0002] In the field of industrial non-destructive testing, phased array ultrasonic technology, with its flexibility in electronically controlled beam deflection and focusing, has become a core means of detecting defects in key components such as special equipment, steel structures, and pipelines.

[0003] Patent application No. 202511735975.3 discloses an intelligent analysis system for ultrasonic phased array inspection data of welds. This application aims to solve the problem that "most common analysis methods rely on manual experience or rule-based algorithms to interpret the inspection images, which is not only inefficient but also easily affected by human subjectivity, resulting in serious problems of missed detections and misjudgments. At the same time, existing methods cannot fully mine the deep feature information in the original data of multi-channel and multi-angle phased arrays, resulting in weak ability to identify small defects in complex environments, especially in welds of thick-walled components with a wall thickness greater than 30mm, where small cracks and other weak defects are extremely difficult to identify reliably."

[0004] However, existing weld inspection technologies often rely on manual interpretation of phased array ultrasonic data or can only identify surface defects in welds, failing to accurately identify internal defects such as lack of fusion or micro-cracks. To address this, we have proposed a new intelligent defect identification and data analysis system based on phased array ultrasound. Summary of the Invention

[0005] In view of the above-mentioned shortcomings of the existing technology, the present invention provides a defect intelligent identification and data analysis system based on phased array ultrasound, which can effectively solve the problems of the existing technology.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions; This invention discloses a defect intelligent identification and data analysis system based on phased array ultrasound, comprising: The acquisition module drives the phased array ultrasonic sensor array to perform weld seam scanning according to preset scanning rules, simultaneously acquiring and outputting ultrasonic echo time-series data across the entire depth range of the weld seam. The preprocessing module receives the ultrasonic echo time-series data, performs data correction, noise filtering, and initial screening of valid signals before outputting the data. Simultaneously, it receives feature anchoring parameters output by the mapping module, performs upsampling and refinement processing on the echo signals of the spatial region corresponding to the feature anchoring parameters, obtains refined valid signals, and outputs them to the mapping module. The preprocessing module and the mapping module form a closed-loop adaptive refinement processing link, achieving accurate extraction of defect features through iterative refinement. The mapping module receives the valid signals output by the preprocessing module and the refined valid signals, performs multi-dimensional feature encoding, and constructs a weld seam region feature mapping matrix. The preprocessing module then processes the features in the mapping matrix... Abnormal regions are clustered, and defect feature vectors with discrete parameters are extracted based on the adaptive boundaries of the clustered regions. Feature anchoring parameters are generated and output to the preprocessing module simultaneously. After the closed-loop adaptive refinement process converges, a stable weld region feature mapping matrix and defect feature vector are output. The parsing module receives the stable feature mapping matrix and defect feature vector output by the mapping module, performs classification and parsing to determine the defect type, and calculates the spatial location, size, and orientation parameters of the defect within the weld. The analysis module associates and stores the classification and parsing results from the parsing module, and performs statistical and correlation comparisons of weld defect data based on normalized relative deviations. The output module obtains the statistical and correlation comparison results from the analysis module, performs structured encoding on the result data, and outputs corresponding visualization maps and inspection reports. The acquisition module is interconnected with a preprocessing module via a wireless network. The mapping module is interconnected with a preprocessing module via a wireless network. The preprocessing module and the mapping module are interconnected with a parsing module and an analysis module via a wireless network. The parsing module and the analysis module are interconnected with an output module via a wireless network.

[0007] Furthermore, the acquisition module includes a phased array ultrasonic sensor array unit, a timing synchronization drive unit, and an echo acquisition unit; The timing synchronization drive unit generates multi-channel parallel phase modulation trigger signals according to preset scanning rules, driving the corresponding array elements in the phased array ultrasonic sensor array to emit ultrasonic beams according to preset timing. The echo acquisition unit keeps clock synchronized with the phase modulation trigger signals, and synchronously samples the ultrasonic echoes received by each array element at a preset sampling frequency, outputting ultrasonic echo timing data of the full depth range of the weld that corresponds one-to-one with the emission timing.

[0008] Furthermore, when the preprocessing module performs data correction, noise filtering, and initial screening of valid signals, it processes the received ultrasonic echo timing data. Perform timing and amplitude joint correction to obtain the corrected signal: ; In the formula: Indicates the corrected signal. The channel number of the phased array ultrasonic sensor array; Sampling time; For the first Theoretical propagation delay of the ultrasonic beam in the channel; For the first The actual propagation path of the ultrasonic beam within the weld seam; Preset reference sound path; The ultrasonic attenuation calibration coefficient for the weld base material; Coherent noise filtering is performed on the corrected full-channel signal to obtain a denoised signal set; based on a preset signal energy threshold, the effective signal set is initially screened, and signal segments exceeding the preset signal energy threshold are retained and output to the mapping module; Simultaneously, it receives the feature anchoring parameters output by the mapping module, performs upsampling and refinement processing on the echo signal of the spatial region corresponding to the feature anchoring parameters, and repeatedly performs correction, noise reduction and initial screening on the refined signal before outputting it to the mapping module.

[0009] Furthermore, when the mapping module performs multi-dimensional feature encoding, weld area feature mapping matrix construction and defect feature vector extraction, it extracts the time domain, frequency domain and wave domain multi-dimensional feature set of the effective signal output by the preprocessing module. The multi-dimensional feature set includes echo peak amplitude, peak arrival time, spectral centroid, wave packet duration and phase change amplitude. Using the transverse scanning position of the weld as the row and the depth direction as the column, a two-dimensional grid coordinate system is constructed for the weld region. Feature normalization encoding is performed on the multi-dimensional feature set corresponding to each grid node to obtain the feature encoding value of each grid node: ; In the formula: express Feature encoding value, The coordinates of the transverse scan position of the weld; The coordinates are in the direction of weld depth. For feature item indices of a multi-dimensional feature set, , This represents the total number of items in the multi-dimensional feature set. coordinates The corresponding grid node's first The unnormalized values ​​of the features; This is the minimum value of this feature across the entire weld area; This is the maximum value of this feature across the entire weld area; Based on the feature encoding values ​​of all grid nodes in the entire weld area, a weld area feature mapping matrix is ​​constructed; continuous regions in the feature mapping matrix whose feature encoding values ​​exceed a preset feature threshold are clustered, and feature distribution parameters and dispersion parameters of the clustered regions are extracted to generate defect feature vectors; the dispersion parameters include the variance, range, and coefficient of variation of the feature encoding values; at the same time, feature anchoring parameters are generated based on the adaptive spatial boundary of the clustered regions and output to the preprocessing module.

[0010] Furthermore, in the stage of generating feature anchoring parameters based on the adaptive spatial boundary of the clustered region, the feature distribution center of the clustered region in the lateral scanning position and depth direction in the two-dimensional grid coordinate system is extracted. Using the feature distribution center as a reference, the adaptive boundary is obtained by adaptive expansion according to a preset edge margin, and the lateral scanning start and end coordinates of the adaptive boundary are determined. and the start and end coordinates of the depth direction ; Based on the element arrangement parameters and preset scanning rules of the phased array ultrasonic sensor array, the start and end coordinates of the lateral scan are mapped to the corresponding phased array channel number range. Based on the ultrasonic velocity of the weld base material and the start and end coordinates of the depth direction, the round-trip propagation time delay interval of the ultrasonic beam within the corresponding depth range is calculated using the depth-time delay mapping formula. The depth-delay mapping formula is as follows: ; In the formula: For weld depth The corresponding round-trip propagation delay of the ultrasonic beam; 1 represents the ultrasonic longitudinal wave velocity of the weld base material; 2 represents the round-trip path coefficient of the ultrasonic wave from transmission to reception. The minimum propagation delay is calculated based on the depth start and end coordinates. With maximum propagation delay This is mapped to the sampling time window of the echo signal within the corresponding channel. ,in and , and Each clock should be kept synchronized. Finally, the feature anchoring parameters, consisting of the channel number range, sampling time window, and preset upsampling factor, are output to the preprocessing module.

[0011] Furthermore, the execution logic of the closed-loop adaptive refinement processing link is as follows: After the mapping module outputs the feature anchoring parameters for the first time, the preprocessing module performs upsampling and refinement processing on the corresponding region signal and outputs it to the mapping module; the mapping module repeatedly performs feature encoding and feature mapping matrix update on the refined signal; if the fluctuation amplitude of the feature encoding value of the corresponding region in the updated feature mapping matrix exceeds the preset fluctuation threshold, the feature anchoring parameters are regenerated and output to the preprocessing module until the fluctuation amplitude of the feature encoding value is within the preset fluctuation threshold, and the closed-loop refinement processing is terminated. When the closed loop terminates, the feature mapping matrix and the defect feature vector after the final iteration convergence are used as the final result and output to the parsing module.

[0012] Furthermore, the analysis module, based on a preset defect type feature library, performs multi-dimensional matching and classification on the defect feature vectors output by the mapping module to determine the defect type. Based on the coordinate distribution of the clustering regions corresponding to the defect feature vectors in the weld region feature mapping matrix, and combined with the element arrangement parameters of the phased array ultrasonic sensor array and the acoustic parameters of the weld base material, it calculates the three-dimensional spatial position, projection size, extension length, and orientation angle parameters of the defect within the weld. The orientation angle includes the horizontal orientation angle within the weld horizontal plane. Vertical inclination angle within the vertical section ; The analysis module has a built-in weld defect database. It associates and binds the defect type, spatial location, size, and orientation parameters output by the analysis module with the corresponding base material properties, welding process parameters, and scanning environment parameters of the weld, and stores them in the weld defect database. The statistical and correlation comparison performed by the analysis module includes time-series comparison of defect data from different scanning batches of the same weld, distribution statistics of defect data from different welds of the same batch and process, and normalized relative deviation correlation analysis of defect parameters and welding process parameters, generating corresponding statistical and comparison results.

[0013] Furthermore, the correlation analysis of the normalized relative deviations between defect parameters and welding process parameters follows the following order: This includes the individual correlation between defect parameters and welding process parameters. With overall correlation ; set up Here is a sequence of defect feature parameters, where For the first The defect characteristic parameters include the defect projection size, extension length, and horizontal orientation angle. Burial depth; definition This is the sequence of welding process parameters for the corresponding weld, where For the first Welding process parameters, including welding heat input, welding current, welding speed, and interpass temperature; For the first The preset baseline value of the defect characteristic parameter. For the first The preset rated values ​​of the welding process parameters; Normalized relative deviation calculation: ; in, This represents the normalized relative deviation of the k-th defect feature parameter. This represents the normalized relative deviation of the k-th welding process parameter; One-to-one correlation: ; in, Indicates the degree of correlation between individual items; Overall relevance: ; In the formula: This represents the number of matching items between defect characteristic parameters and welding process parameters.

[0014] Furthermore, the output module performs structured encoding on the statistical and correlation comparison results output by the analysis module to generate a standardized defect dataset; based on the defect dataset and the weld area feature mapping matrix, it generates a visualization atlas containing two-dimensional cross-sectional imaging of the weld, three-dimensional spatial positioning imaging of the defect, and annotation of defect feature parameters; at the same time, based on a preset report template, it automatically generates an inspection report containing basic weld information, scanning parameters, defect details, statistical analysis results, and compliance judgment conclusions.

[0015] Compared with the known prior art, the technical solution provided by this invention has the following beneficial effects: This invention utilizes phased array ultrasonic scanning of the entire weld seam without blind spots and synchronous acquisition and processing of echo timing data to complete signal timing amplitude correction, noise filtering, and initial screening of effective signals. Through closed-loop adaptive processing of feature anchoring and signal refinement, it further realizes the encoding and mapping of multi-dimensional features of the weld seam area and the extraction of defect features. It achieves accurate classification of defect types and calculation of core parameters such as spatial location, size, and orientation. Simultaneously, through the associated storage, statistical comparison, and correlation analysis with welding process parameters of defect data, it finally outputs structured visualization maps and standardized inspection reports. This effectively improves the stability and defect identification accuracy of ultrasonic weld inspection, reduces the subjective bias of manual inspection, and provides traceable inspection data and data analysis support for welding quality control. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0017] Figure 1 This is a schematic diagram of a defect intelligent identification and data analysis system based on phased array ultrasound. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0019] The present invention will be further described below with reference to embodiments.

[0020] Example: The defect intelligent identification and data analysis system based on phased array ultrasound in this embodiment, such as Figure 1 As shown, it includes: The acquisition module is used to drive the phased array ultrasonic sensor array to perform weld seam scanning according to the preset scanning rules, and simultaneously acquire and output ultrasonic echo timing data of the full depth range of the weld seam. The acquisition module includes a phased array ultrasonic sensor array unit, a timing synchronization drive unit, and an echo acquisition unit; The timing synchronization drive unit generates multi-channel parallel phase modulation trigger signals according to preset scanning rules, driving the corresponding array elements in the phased array ultrasonic sensor array to emit ultrasonic beams according to preset timing. The echo acquisition unit keeps the clock synchronized with the phase modulation trigger signals, and synchronously samples the ultrasonic echoes received by each array element at a preset sampling frequency, outputting ultrasonic echo timing data of the full depth range of the weld that corresponds one-to-one with the emission timing. The preprocessing module receives ultrasonic echo timing data, performs data correction, noise filtering, and initial screening of effective signals before outputting the data. Simultaneously, it receives the feature anchoring parameters output by the mapping module, performs upsampling and refinement processing on the echo signals of the spatial region corresponding to the feature anchoring parameters, obtains the refined effective signals, and outputs them to the mapping module. The preprocessing module and the mapping module form a closed-loop adaptive refinement processing link. When the preprocessing module performs data correction, noise filtering, and initial screening of valid signals, it processes the received ultrasonic echo timing data. Perform timing and amplitude joint correction to obtain the corrected signal: ; In the formula: Indicates the corrected signal. The channel number of the phased array ultrasonic sensor array; Sampling time; For the first Theoretical propagation delay of the ultrasonic beam in the channel; For the first The actual propagation path of the ultrasonic beam within the weld seam; Preset reference sound path; The ultrasonic attenuation calibration coefficient for the weld base material; The above formula makes dual corrections for timing and amplitude for the propagation differences of different channels of the phased array. It eliminates the signal reception time deviation caused by the spatial position of the array elements through time delay compensation. It corrects the amplitude by combining the ratio of the actual sound path to the reference sound path and the sound attenuation coefficient, so that the ultrasonic echo signals of different channels are at a unified reference standard. This can more realistically reflect the actual echo situation inside the weld and lay accurate basic data for subsequent signal processing. Coherent noise filtering is performed on the corrected full-channel signal to obtain a denoised signal set; based on a preset signal energy threshold, the effective signal set is initially screened, and signal segments exceeding the preset signal energy threshold are retained and output to the mapping module; Simultaneously, it receives the feature anchoring parameters output by the mapping module, performs upsampling and refinement processing on the echo signal of the spatial region corresponding to the feature anchoring parameters, and repeatedly performs correction, noise reduction and initial screening on the refined signal before outputting it to the mapping module. The mapping module receives the valid signal and the refined valid signal output by the preprocessing module, performs multi-dimensional feature encoding, and constructs a feature mapping matrix for the weld area. It then clusters the abnormal areas in the feature mapping matrix, extracts the defect feature vector containing discrete parameters based on the adaptive boundary of the clustered areas, and synchronously generates and outputs the feature anchoring parameters to the preprocessing module. When the mapping module performs multi-dimensional feature encoding, weld area feature mapping matrix construction and defect feature vector extraction, it extracts the time domain, frequency domain and wave domain multi-dimensional feature set of the effective signal output by the preprocessing module. The multi-dimensional feature set includes echo peak amplitude, peak arrival time, spectral centroid, wave packet duration and phase change amplitude. Using the transverse scanning position of the weld as the row and the depth direction as the column, a two-dimensional grid coordinate system is constructed for the weld region. Feature normalization encoding is performed on the multi-dimensional feature set corresponding to each grid node to obtain the feature encoding value of each grid node: ; In the formula: express Feature encoding value, The coordinates of the transverse scan position of the weld; The coordinates are in the direction of weld depth. For feature item indices of a multi-dimensional feature set, , This represents the total number of items in the multi-dimensional feature set. coordinates The corresponding grid node's first The unnormalized values ​​of the features; This is the minimum value of this feature across the entire weld area; This is the maximum value of this feature across the entire weld area; The above formula uses the arctangent function to perform a nonlinear transformation on the normalized eigenvalues, mapping the eigenvalues ​​of the entire weld area to a fixed interval. This not only achieves the normalized unified encoding of features with different dimensions and units, but also highlights the relative differences of eigenvalues ​​through the nonlinear characteristics of the arctangent, avoiding the eigenvalue compression problem that is prone to occur in linear normalization. This allows the eigenvalue encoding of the grid nodes to more accurately represent the actual eigenstate of the corresponding weld position, so as to facilitate the subsequent construction of the feature mapping matrix and the identification of defect areas. To further improve the accuracy of feature encoding in representing different types of defects, the mapping module can construct a multi-angle echo feature set for each two-dimensional grid node (x,z) when performing feature encoding on the grid nodes. Based on the detection echoes of phased array ultrasound at different incident angles, the time-domain, frequency-domain, and wave-domain feature parameters at the corresponding angles are extracted respectively. At the same time, a prior library of angle-defect type sensitivity is pre-stored. According to the sensitivity of different incident angles to the identification of various defects such as porosity, slag inclusion, incomplete penetration, and cracks, dynamic weight coefficients are assigned to the echo features of different incident angles. The normalized encoded values ​​of each angle feature are weighted and fused to finally generate the final feature encoding value of the grid node (x,z). Through multi-angle feature fusion and dynamic weight allocation, the feature representation capability of defect-sensitive angles is strengthened, and the feature interference of non-sensitive angles is weakened, thereby further improving the defect discrimination and recognition robustness of the weld area feature mapping matrix. Based on the feature encoding values ​​of all grid nodes in the entire weld area, a feature mapping matrix for the weld area is constructed. Continuous regions in the feature mapping matrix whose feature encoding values ​​exceed a preset feature threshold are clustered, and the feature distribution parameters and dispersion parameters of the clustered regions are extracted to generate defect feature vectors. The dispersion parameters include the variance, range, and coefficient of variation of the feature encoding values. Simultaneously, feature anchoring parameters are generated based on the adaptive spatial boundary of the clustered regions and output to the preprocessing module. In the stage of generating feature anchoring parameters based on the adaptive spatial boundary of clustered regions, the feature distribution centers of the clustered regions in the lateral scanning position and depth direction in the two-dimensional grid coordinate system are extracted. Using the feature distribution centers as a reference, the adaptive boundary is obtained by adaptively expanding according to a preset edge margin, and the lateral scanning start and end coordinates of the adaptive boundary are determined. and the start and end coordinates of the depth direction ; Based on the element arrangement parameters and preset scanning rules of the phased array ultrasonic sensor array, the start and end coordinates of the lateral scan are mapped to the corresponding phased array channel number range. Based on the ultrasonic velocity of the weld base material and the start and end coordinates of the depth direction, the round-trip propagation time delay interval of the ultrasonic beam within the corresponding depth range is calculated using the depth-time delay mapping formula. The depth-delay mapping formula is as follows: ; In the formula: For weld depth The corresponding round-trip propagation delay of the ultrasonic beam; 1 represents the ultrasonic longitudinal wave velocity of the weld base material; 2 represents the round-trip path coefficient of the ultrasonic wave from transmission to reception. Based on the propagation characteristics of ultrasound in the weld base material, the above formula establishes a linear correspondence between weld depth and round-trip propagation time delay by utilizing the longitudinal wave velocity of ultrasound. This transforms the spatial physical quantity of depth into the signal acquisition quantity of sampling time, achieving a precise mapping between weld spatial depth and ultrasonic echo sampling time. This provides a direct acoustic theoretical basis for subsequently determining the sampling time window for feature anchoring, allowing time-domain signal processing to accurately correspond to the actual position in the weld depth domain. The minimum propagation delay is calculated based on the depth start and end coordinates. With maximum propagation delay This is mapped to the sampling time window of the echo signal within the corresponding channel. ,in and , and Each clock should be kept synchronized. Finally, the feature anchoring parameters, consisting of the channel number range, sampling time window, and preset upsampling rate, are output to the preprocessing module. The execution logic of the closed-loop adaptive refinement processing link is as follows: After the mapping module outputs the feature anchoring parameters for the first time, the preprocessing module performs upsampling and refinement processing on the corresponding region signal and outputs it to the mapping module; the mapping module repeatedly performs feature encoding and feature mapping matrix update on the refined signal; if the fluctuation amplitude of the feature encoding value of the corresponding region in the updated feature mapping matrix exceeds the preset fluctuation threshold, the feature anchoring parameters are regenerated and output to the preprocessing module until the fluctuation amplitude of the feature encoding value is within the preset fluctuation threshold, and the closed-loop refinement processing is terminated. When the closed loop terminates, the feature mapping matrix and the defect feature vector after the convergence of the last iteration are used as the final result and output to the parsing module. The parsing module is used to receive the stable feature mapping matrix and defect feature vector output by the mapping module, perform classification and parsing, determine the defect type, and calculate the spatial location, size, and orientation parameters of the defect inside the weld. The analysis module, based on a pre-defined defect type feature library, performs multi-dimensional matching and classification on the defect feature vectors output by the mapping module to determine the defect type. Then, based on the coordinate distribution of the clustered regions corresponding to the defect feature vectors in the weld region feature mapping matrix, and combined with the element arrangement parameters of the phased array ultrasonic sensor array and the acoustic parameters of the weld base material, it calculates the three-dimensional spatial position, projection size, extension length, and orientation angle parameters of the defect within the weld. The orientation angle includes the horizontal orientation angle within the weld horizontal plane. Vertical inclination angle within the vertical section ; Furthermore, when classifying and analyzing defect feature vectors, the parsing module employs a two-way verification method combining feature vector similarity matching and spatial geometric features to determine the defect type. First, based on a pre-defined defect type feature library, a cosine similarity algorithm is used to match the feature vector of the defect to be identified with the standard defect feature vector, completing a preliminary defect type determination. Then, the preliminary determination is cross-validated with the calculated spatial geometric features such as the defect's three-dimensional spatial location, projected dimensions, extension length, and orientation angle. This is combined with the spatial geometric distribution patterns of various typical weld defects, such as porosity, slag inclusions, incomplete penetration, lack of fusion, and cracks, to verify and correct the preliminary type. The final defect type is output only when the dual verification results are consistent; otherwise, the defect feature extraction and matching parsing process is re-executed. This multi-dimensional two-way verification effectively reduces the probability of misjudgment based on a single feature match, improving the accuracy and reliability of defect type identification. The analysis module has a built-in weld defect database. It associates and binds the defect type, spatial location, size, and orientation parameters output by the analysis module with the corresponding base material properties, welding process parameters, and scanning environment parameters of the weld, and stores them in the weld defect database. The analysis module performs statistical and correlation comparisons, including time-series comparison of defect data from different scanning batches of the same weld, distribution statistics of defect data from different welds of the same batch and process, and normalized relative deviation correlation analysis of defect parameters and welding process parameters, generating corresponding statistical and comparison results. In the stage of calculating the three-dimensional spatial location, projected size, extension length, and orientation angle parameters of defects within the weld, a three-dimensional rectangular coordinate system is established with the starting point of the phased array ultrasonic sensor scan as the origin, the transverse direction of the weld perpendicular to its length extension as the X-axis, the weld length extension direction as the Y-axis, and the depth direction from the weld base material surface inward as the Z-axis. This represents the set of effective mesh nodes in the three-dimensional coordinate system of the weld, corresponding to the clustered regions of the defect feature vector. for The set of vertices of the convex hull boundary. coordinates Feature encoding value of grid node, To preset feature thresholds, In the convex hull boundary The set of valid vertices, The number of valid vertices, The ultrasonic longitudinal wave velocity in the weld base material. This represents the average propagation delay of the echoes corresponding to the clustered regions; The three-dimensional spatial location of the defect is determined by the centroid coordinates of the defect. The core characterization parameter is calculated using the following formula: ; In the formula, , Valid vertices The corresponding X-axis and Y-axis coordinates; The above formula takes the effective vertices of the convex hull of the defect clustering region as the calculation basis, determines the centroid of the defect in the plane on the X and Y axes by the mean of the effective vertex coordinates, and calculates the centroid of the depth on the Z axis by combining the relationship between the average propagation delay of the ultrasonic wave and the sound speed. By combining the coordinate information of the feature mapping matrix with the acoustic parameters of ultrasonic wave propagation, the accurate conversion from two-dimensional feature mesh to three-dimensional spatial position is realized, so as to objectively characterize the core spatial position of the defect inside the weld. The calculation logic fits the actual data characteristics of phased array ultrasonic testing. definition , for The two effective vertices with the largest spatial distance are taken as the extreme endpoints of the main extension direction of the defect; Defect lateral projection size Depth projection size Length of extension The solution formula is: ; This formula selects the two endpoints with the largest spatial distance among the effective vertices of the convex hull of the defect clustering region. By calculating the coordinate differences of the endpoints on the X, Z, and Y axes, the lateral projection size, depth projection size, and extension length of the defect are obtained respectively. The actual size of the defect is determined directly based on the spatial extreme boundary of the defect feature region, which fits the spatial distribution characteristics of weld defects. The solution method is simple and can accurately reflect the actual extension range of the defect, providing an intuitive calculation method for defect size quantification. The defect orientation angle parameter includes the horizontal orientation angle within the weld horizontal plane. Vertical inclination angle within the vertical section The solution formula is: ; Based on the extreme endpoint coordinates of the main extension direction of the defect, this formula calculates the direction and inclination angle in the horizontal and vertical sections respectively. The distance difference of the coordinate axes is converted into an angle value through the inverse cosine function, which accurately represents the extension direction of the defect in different sections of the weld. The spatial position relationship of the defect is transformed into an intuitive inclination angle parameter, which meets the needs of detecting and judging the direction of weld defects in engineering, and can clearly reflect the spatial extension state of the defect inside the weld. The correlation analysis of the normalized relative deviation between defect parameters and welding process parameters follows the following order: This includes the individual correlation between defect parameters and welding process parameters. With overall correlation ; set up Here is a sequence of defect feature parameters, where For the first The defect characteristic parameters include the defect projection size, extension length, and horizontal orientation angle. Burial depth; definition This is the sequence of welding process parameters for the corresponding weld, where For the first Welding process parameters, including welding heat input, welding current, welding speed, and interpass temperature; For the first The preset baseline value of the defect characteristic parameter. For the first The preset rated values ​​of the welding process parameters; Normalized relative deviation calculation: ; in, This represents the normalized relative deviation of the k-th defect feature parameter. This represents the normalized relative deviation of the k-th welding process parameter; One-to-one correlation: ; in, Indicates the degree of correlation between individual items; This formula transforms the relative deviations of defect feature parameters and welding process parameters by using the arctangent function, then calculates the absolute value of the difference between the two and normalizes it, transforming the numerical difference of parameter deviations into correlation values ​​in the 0-1 range. This not only weakens the influence of extreme deviation values ​​on correlation by using the arctangent function, but also accurately quantifies the matching degree between a single set of defect parameters and process parameters. Overall relevance: ; In the formula: This represents the number of matching items between defect characteristic parameters and welding process parameters; The analysis module is used to associate and store the classification, analysis and solution results of the analysis module, and to perform statistical analysis and correlation comparison of weld defect data based on the normalized relative deviation. The output module is used to obtain the statistical and correlation comparison results from the analysis module, perform structured encoding on the result data, and output the corresponding visualization map and detection report; The output module performs structured encoding on the statistical and correlation comparison results output by the analysis module to generate a standardized defect dataset. Based on the defect dataset and the weld area feature mapping matrix, it generates a visualization map that includes two-dimensional cross-sectional imaging of the weld, three-dimensional spatial positioning imaging of defects, and annotation of defect feature parameters. At the same time, based on a preset report template, it automatically generates an inspection report that includes basic weld information, scanning parameters, defect details, statistical analysis results, and compliance judgment conclusions. The acquisition module is interconnected with the preprocessing module via a wireless network. The mapping module is interconnected with the preprocessing module via a wireless network. The preprocessing module and the mapping module are interconnected with the parsing module and the analysis module via a wireless network. The parsing module and the analysis module are interconnected with the output module via a wireless network.

[0021] In this embodiment, the acquisition module drives the phased array ultrasonic sensor array to perform weld seam scanning according to preset scanning rules, simultaneously acquiring and outputting ultrasonic echo time-series data across the entire depth range of the weld seam. The preprocessing module, running subsequently, receives the ultrasonic echo time-series data, performs data correction, noise filtering, and initial screening of valid signals before outputting the data. Simultaneously, it receives the feature anchoring parameters output by the mapping module, performs upsampling and refinement processing on the echo signals of the spatial region corresponding to the feature anchoring parameters, obtains the refined valid signals, and outputs them to the mapping module. The preprocessing module and the mapping module form a closed-loop adaptive refinement processing link. The mapping module synchronously receives the valid signals output by the preprocessing module and the refined valid signals, performs multi-dimensional feature encoding, and constructs a feature mapping matrix for the weld seam region. The system performs clustering on abnormal regions in the feature mapping matrix, extracts defect feature vectors with discrete parameters based on the adaptive boundaries of the clustered regions, and simultaneously generates and outputs feature anchoring parameters to the preprocessing module. The parsing module further receives the stable feature mapping matrix and defect feature vectors output by the mapping module for classification and analysis to determine the defect type, calculates the spatial location, size, and orientation parameters of the defect inside the weld, and then the analysis module associates and stores the classification and analysis results from the parsing module. Based on the normalized relative deviation, it performs statistical and correlation comparison of weld defect data. Finally, the output module obtains the statistical and correlation comparison results from the analysis module, performs structured encoding on the result data, and outputs the corresponding visualization map and inspection report.

[0022] In the above embodiments, the system can achieve continuous, blind-spot-free scanning of the entire thickness and width of the weld during the on-site implementation of phased array ultrasonic testing of welds. It can simultaneously complete the accurate correction of echo signals and the extraction of defect features. Through closed-loop adaptive processing, it can improve the accuracy of defect identification, accurately determine the defect type and calculate its core parameters such as spatial location and size. It can also simultaneously complete the statistical analysis of defect data and the correlation with welding processes, and automatically generate visual maps and inspection reports. This effectively reduces the difficulty of on-site inspection operations, reduces the subjective bias of manual interpretation, and provides reliable data support for welding quality control.

[0023] Based on the system described in the above embodiments, the following is an application example of the system: This study implemented a routine non-destructive testing (NDT) scenario for the circumferential weld of the steam drum of Unit #2 at a thermal power plant. The test object was the circumferential butt weld of the steam drum shell. The base material of the weld was 12Cr1MoV pearlitic heat-resistant steel, the shell wall thickness was 65mm, and the total length of the circumferential weld was 8.2m. The welding process adopted was narrow-gap submerged arc welding. The test standard was DL / T820 "Technical Specification for Ultrasonic Testing of Pipeline Welded Joints". The entire test was conducted using this system to complete the defect scanning, identification, quantitative analysis, and report output of the entire weld.

[0024] Before testing, the system and phased array ultrasonic equipment were integrated and debugged. A 64-element ultrasonic sensor array was used to match the testing conditions. The system's acquisition module drove the array to perform a full-area scan of the weld according to the preset scanning rules. The preset scanning rules were based on the core of full thickness and full width of the weld without blind spots, and defined four execution dimensions: the array element group division rule set 16 array elements for a single transmission excitation, with 8 array elements overlapping in adjacent array element groups; the beam focusing depth stepping rule set the focusing depth interval between adjacent transmissions to 2mm, covering the full depth range from the base material surface to a wall thickness of 65mm; the scan path stepping rule set the lateral position interval of adjacent weld scans to 1mm, continuously stepping along the circumference of the weld; and the scan angle coverage rule set the ultrasonic beam incident angle range to -45° to +45°. The four rules worked together to achieve full-area scan of the weld without blind spots. During the scanning process, the timing synchronization drive unit generates multi-channel parallel phase modulation trigger signals, which drive the corresponding array elements to emit ultrasonic beams according to a preset timing sequence. The echo acquisition unit keeps the clock synchronized with the trigger signal and synchronously samples the ultrasonic echoes received by each array element at a sampling frequency of 100MHz, outputting ultrasonic echo timing data of the full depth of the weld that corresponds one-to-one with the emission timing sequence.

[0025] After receiving the acquired full-channel echo timing data, the system preprocessing module first performs joint correction of timing and amplitude to compensate for propagation delay and amplitude attenuation caused by differences in sound path between different channels. Based on the material characteristics of the 12Cr1MoV base material and the ultrasonic frequency used in this test, the ultrasonic attenuation calibration coefficient for the base material is set to 0.03 dB / mm. After correction, coherent noise filtering is performed on the full-channel signals to remove electrical noise and base material structural noise introduced during acquisition. Then, based on a preset signal energy threshold, the denoised signal set is initially screened for effective signals, retaining effective echo signal segments exceeding the threshold, which are then output to the mapping module. Subsequently, after receiving the feature anchoring parameters output by the mapping module, the preprocessing module performs upsampling and refinement processing on the echo signals in the corresponding spatial region. The refined signal undergoes the correction, denoising, and initial screening processes again before being output to the mapping module, where it works in conjunction with the mapping module to complete closed-loop adaptive refinement processing.

[0026] The system mapping module receives the preprocessed valid signal and first extracts multi-dimensional feature sets of the signal in the time domain, frequency domain, and wave domain. Specifically, it includes five core features: echo peak amplitude, peak arrival time, spectral centroid, wave packet duration, and phase change amplitude. Then, it constructs a two-dimensional grid coordinate system for the weld area with the weld lateral scanning position as the row and the depth direction as the column. It performs feature normalization encoding on the multi-dimensional feature set corresponding to each grid node to obtain the feature encoding value of each grid node. Based on the feature encoding values ​​of all grid nodes in the entire weld area, it constructs a complete weld area feature mapping matrix. Subsequently, continuous regions in the feature mapping matrix whose feature encoding values ​​exceed the preset feature threshold are clustered, and feature distribution parameters of the clustered regions are extracted to generate defect feature vectors. At the same time, the extreme values ​​of the lateral position and depth direction of the clustered regions are extracted in the two-dimensional grid coordinate system to determine the spatial bounding box of the clustered regions. The corresponding phased array channel number range and echo signal sampling time window are mapped and combined with the preset upsampling ratio to generate feature anchoring parameters, which are output to the preprocessing module. This forms a closed-loop adaptive refinement processing link with the preprocessing module until the fluctuation amplitude of the feature encoding values ​​of the corresponding regions in the feature mapping matrix is ​​within the preset fluctuation threshold. The closed-loop refinement processing is then terminated, and finally, a stable feature mapping matrix and defect feature vectors are output to the parsing module.

[0027] The system analysis module receives the feature mapping matrix and defect feature vector output by the mapping module. Based on a preset defect type feature library, it performs multi-dimensional matching and classification of the defect feature vectors, ultimately determining that two defects were detected in this inspection: one is a planar incomplete penetration defect, and the other is a volumetric slag inclusion defect. Simultaneously, using the starting point of the phased array ultrasonic sensor array scan as the origin, a three-dimensional rectangular coordinate system for the weld is established. Combining the array element arrangement parameters and the acoustic parameters of the base material, the complete parameters of the two defects within the weld are calculated: the three-dimensional coordinates of the centroid of the incomplete penetration defect, its lateral projection size is 3.2 mm, its depth projection size is 1.8 mm, its extension length along the weld length direction is 12.5 mm, its horizontal inclination angle is 87°, its vertical inclination angle is 3°, and its burial depth is 32 mm; the three-dimensional coordinates of the centroid of the slag inclusion defect, its lateral projection size is 2.1 mm, its depth projection size is 1.6 mm, its extension length along the weld length direction is 4.2 mm, its horizontal inclination angle is 22°, its vertical inclination angle is 15°, and its burial depth is 18 mm.

[0028] The system analysis module has a built-in weld defect database. It associates and binds the defect type, spatial location, size, and orientation parameters output by the analysis module with the base metal properties, welding process parameters, and scanning environment parameters corresponding to this inspection, and stores them completely in the weld defect database. Simultaneously, it completes three core statistical and correlation comparison tasks: First, it performs a time-series comparison of the current inspection data for this weld with the scanning data from the previous maintenance cycle, confirming that both defects are newly detected within this cycle, with no extended historical defects; second, it performs distribution statistics on the defect data of the other three steam drum ring welds constructed in the same batch and with the same welding process in the same unit, clarifying that the probability of this weld defect occurring is at a medium level within the same batch; third, it performs correlation analysis between defect characteristic parameters and welding process parameters, ultimately determining that the factor with the highest single correlation is welding heat input, and the comprehensive correlation between defect parameters and welding process parameters is 0.82, clarifying that fluctuations in welding heat input are the main influencing factor for the defects in this weld.

[0029] After acquiring the statistical and correlation comparison results from the analysis module, the system output module performs structured encoding on the result data to generate a standardized defect dataset. Based on the defect dataset and the weld area feature mapping matrix, it generates a visual atlas containing two-dimensional cross-sectional imaging of the weld, three-dimensional spatial positioning imaging of defects, and annotation of defect feature parameters. Simultaneously, following the preset report template for power plant pressure equipment inspection, it automatically generates a complete inspection report. The report covers basic weld information, scanning parameters, defect details, statistical analysis results, and conformity judgment conclusions that meet industry standards. It also provides suggestions for subsequent defect monitoring and rework. This inspection took 2.5 hours in total. Compared to conventional ultrasonic testing under the same conditions, the inspection efficiency was improved by more than 70%, the defect location error was controlled within 0.5mm, and the defect type identification accuracy was 100%. During subsequent weld rework, the actual defect location, size, and type perfectly matched the system output results.

[0030] It should be noted that: The preset scanning rules of the acquisition module can be selected as follows: the number of array elements emitted in a single transmission is 1 / 4 to 1 / 2 of the total number of array elements in the sensor array and not less than 16; adjacent transmitting array elements are translated by 1 array element step; the beam focusing depth step interval is 1 / 2 of the ultrasonic longitudinal wave wavelength of the weld base material, and the focusing range covers the surface of the base material to 1 plate thickness below the bottom surface of the weld; the transverse step interval of the scanning path is 1 / 3 of the effective coverage width in a single transmission; the ultrasonic beam incident angle range is -60° to +60° with a step ≤1°, which can be finely adjusted according to the bevel shape. The four rules work together to achieve continuous scanning of the entire weld area without blind spots.

[0031] The preset reference sound path for echo signal correction in the preprocessing module is calculated from the surface of the array element. For welds with a nominal thickness ≥ 5 mm, the sound path is taken at 1 / 2 depth of the nominal thickness. For thin-walled welds < 5 mm, the sound path is taken at the corresponding depth of the bottom of the weld. The ultrasonic attenuation calibration coefficient α is calibrated using a standard test block of the same material and heat treatment state. The test block contains at least 3 sets of flat-bottomed hole reflectors of different depths. The echo signal is collected with the same parameters as the actual test, and the unit sound path attenuation value is linearly fitted to obtain α. Its value is limited to 0.002 dB / mm to 0.08 dB / mm.

[0032] In the feature encoding stage of the mapping module, after a single feature item completes normalization encoding, a weight coefficient is assigned to each feature item according to its contribution to defect identification, and the sum of all weights is 1. Among them, the weights of echo peak amplitude and peak arrival time are 0.2~0.3, and the weights of spectral centroid, wave packet duration and phase change amplitude are 0.1~0.2, which can be finely adjusted according to the detection accuracy.

[0033] The preset feature threshold for defect region screening in the mapping module first removes the largest 5% of abnormal data points in the final feature encoding value of the grid nodes in the entire weld area, and takes the maximum value of the remaining 95% of data points as the baseline value of the normal area. The preset feature threshold is set to 1.2 to 1.5 times the baseline value. For low noise and high signal-to-noise ratio scenarios, it is set to 1.2 times, and for high noise and low signal-to-noise ratio scenarios, it is set to 1.5 times. Grid nodes with feature encoding values ​​not less than the preset feature threshold are suspected defect nodes. Clustering of continuous suspected nodes yields defect clustering regions.

[0034] The preset fluctuation threshold of the closed-loop adaptive refinement processing link is set to 1%~3% of the maximum value of the feature encoding value range [0,π / 2]. For high-precision detection, it is set to 1%, and for fast scanning, it is set to 3%. The fluctuation amplitude is the maximum value of the relative deviation of the feature anchoring region encoding value before and after the update. When it is ≤ the threshold, it is determined to be converged. The base value of the upsampling factor is 2 times. If the iteration does not converge, it increases in multiples of 2, with a maximum of no more than 8 times. If it still does not converge after reaching 8 times, the closed loop is terminated, and the current result is taken as the final value.

[0035] The Y-axis coordinate of the weld seam in the analysis module is synchronously bound to the mechanical displacement mechanism of the scanning device. Equally spaced scanning step points are preset along the weld seam length, with the spacing ≤ 1 / 2 of the effective coverage width of the sensor array's Y-axis. Displacement data in the length direction is collected in real time through a high-precision displacement encoder, synchronized with the ultrasonic echo timing data clock. Each set of XZ plane echo data is bound to a unique Y-axis coordinate. The XZ plane encoded value of the same Y-axis coordinate is mapped to the corresponding cross-sectional feature data. All step point data are stitched together to construct a three-dimensional feature mapping matrix. The start and end coordinates of the Y-axis of the defect clustering region are determined by the minimum and maximum Y-axis coordinates of the step points it covers.

[0036] The feature library for defect type identification in the parsing module covers five common defects: porosity, slag inclusion, incomplete penetration, lack of fusion, and cracks. Echo data of various defects under different detection parameters are collected using standard weld test blocks of different base materials, processes, and thicknesses. After preprocessing and feature encoding, standard defect feature vectors are extracted, establishing a mapping relationship between defect types and standard vectors. The standard vectors contain baseline values ​​and allowable fluctuation ranges for five core features. Defect classification uses a cosine similarity algorithm to calculate the similarity between the vector to be identified and the standard vectors in the feature library. The maximum value corresponds to the defect type; when the maximum value is below 0.75, it is determined to be an unknown type of defect, and its feature vector and detection data are stored in the feature library's calibration dataset for iterative optimization.

[0037] The correlation analysis between defect parameters and welding process parameters in the analysis module is performed. The number of matching items K is taken as the smaller value of the number of items in the two types of parameter sequences. In this scheme, the defect characteristic parameters include four items: projection size, extension length, orientation angle, and burial depth. The welding process parameters include four items: welding heat input, welding current, welding speed, and interpass temperature. Therefore, K is fixed at 4. When adding new parameters, it is necessary to ensure that the number of items in both is consistent and the K value is updated synchronously. The parameter matching adopts the causal association rule, that is, the defect projection size corresponds to the welding heat input, the extension length corresponds to the welding current, the orientation angle corresponds to the welding speed, and the burial depth corresponds to the interpass temperature. Based on this, the single-item correlation degree and the comprehensive correlation degree are calculated.

[0038] In summary, the system in the above embodiments completes signal timing amplitude correction, noise filtering, and initial screening of effective signals through phased array ultrasonic scanning of the entire weld seam without blind spots and synchronous acquisition and processing of echo timing data. After closed-loop adaptive processing of feature anchoring and signal refinement, it further realizes the encoding and mapping of multi-dimensional features of the weld seam area and the extraction of defect features, completes the accurate classification of defect types and the calculation of core parameters such as spatial location, size, and orientation. Simultaneously, through the associated storage, statistical comparison, and correlation analysis with welding process parameters of defect data, it finally outputs structured visualization maps and standardized inspection reports, thereby effectively improving the stability of ultrasonic weld seam inspection and the accuracy of defect identification, reducing the subjective bias of manual inspection, and providing traceable inspection data and data analysis support for welding quality control.

[0039] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A defect intelligent identification and data analysis system based on phased array ultrasound, characterized in that, include: The acquisition module is used to drive the phased array ultrasonic sensor array to perform weld seam scanning according to the preset scanning rules, and simultaneously acquire and output ultrasonic echo timing data of the full depth range of the weld seam. The preprocessing module receives the ultrasonic echo timing data, performs data correction, noise filtering, and initial screening of effective signals before outputting the data. Simultaneously, it receives the feature anchoring parameters output by the mapping module, performs upsampling and refinement processing on the echo signals of the spatial region corresponding to the feature anchoring parameters, obtains the refined effective signals, and outputs them to the mapping module. The preprocessing module and the mapping module form a closed-loop adaptive refinement processing link. The mapping module is used to receive the valid signal output by the preprocessing module and the refined valid signal, perform multi-dimensional feature encoding, and construct a feature mapping matrix for the weld area. Clustering is performed on abnormal regions in the feature mapping matrix, and defect feature vectors with discrete parameters are extracted based on the adaptive boundary of the clustered regions. Feature anchoring parameters are generated and output to the preprocessing module simultaneously. The parsing module is used to receive the stable feature mapping matrix and defect feature vector output by the mapping module, perform classification and parsing, determine the defect type, and calculate the spatial location, size, and orientation parameters of the defect inside the weld. The analysis module is used to associate and store the classification, analysis and solution results of the analysis module, and to perform statistical analysis and correlation comparison of weld defect data based on the normalized relative deviation. The output module is used to obtain the statistical and correlation comparison results from the analysis module, perform structured encoding on the result data, and output the corresponding visualization maps and detection reports.

2. The intelligent defect identification and data analysis system based on phased array ultrasound according to claim 1, characterized in that, The acquisition module includes a phased array ultrasonic sensor array unit, a timing synchronization drive unit, and an echo acquisition unit. The timing synchronization drive unit generates multi-channel parallel phase modulation trigger signals according to preset scanning rules, driving the corresponding array elements in the phased array ultrasonic sensor array to emit ultrasonic beams according to preset timing. The echo acquisition unit keeps clock synchronized with the phase modulation trigger signals, and synchronously samples the ultrasonic echoes received by each array element at a preset sampling frequency, outputting ultrasonic echo timing data of the full depth range of the weld that corresponds one-to-one with the emission timing.

3. The intelligent defect identification and data analysis system based on phased array ultrasound according to claim 1, characterized in that, When the preprocessing module performs data correction, noise filtering, and initial screening of valid signals, it processes the received ultrasonic echo time-series data. Perform timing and amplitude joint correction to obtain the corrected signal: ； In the formula: Indicates the corrected signal. The channel number of the phased array ultrasonic sensor array; Sampling time; For the first Theoretical propagation delay of the ultrasonic beam in the channel; For the first The actual propagation path of the ultrasonic beam within the weld seam; Preset reference sound path; The ultrasonic attenuation calibration coefficient for the weld base material; Coherent noise filtering is performed on the corrected full-channel signal to obtain the denoised signal set; Based on a preset signal energy threshold, the denoised signal set is initially screened for effective signals, and signal segments exceeding the preset signal energy threshold are retained and output to the mapping module. Simultaneously, it receives the feature anchoring parameters output by the mapping module, performs upsampling and refinement processing on the echo signal of the spatial region corresponding to the feature anchoring parameters, and repeatedly performs correction, noise reduction and initial screening on the refined signal before outputting it to the mapping module.

4. The intelligent defect identification and data analysis system based on phased array ultrasound according to claim 1, characterized in that, When the mapping module performs multi-dimensional feature encoding, weld area feature mapping matrix construction and defect feature vector extraction, it extracts the time domain, frequency domain and wave domain multi-dimensional feature set of the effective signal output by the preprocessing module. The multi-dimensional feature set includes echo peak amplitude, peak arrival time, spectral centroid, wave packet duration and phase change amplitude. Using the transverse scanning position of the weld as the row and the depth direction as the column, a two-dimensional grid coordinate system is constructed for the weld region. Feature normalization encoding is performed on the multi-dimensional feature set corresponding to each grid node to obtain the feature encoding value of each grid node: ； In the formula: express Feature encoding value, The coordinates of the transverse scan position of the weld; The coordinates are in the direction of weld depth. For feature item indices of a multi-dimensional feature set, , This represents the total number of items in the multi-dimensional feature set. coordinates The corresponding grid node's first The unnormalized values ​​of the features; This is the minimum value of this feature across the entire weld area; This is the maximum value of this feature across the entire weld area; Based on the feature encoding values ​​of all grid nodes in the entire weld region, a feature mapping matrix for the weld region is constructed. Clustering is performed on continuous regions in the feature mapping matrix where the feature encoding values ​​exceed a preset feature threshold. Feature distribution parameters and dispersion parameters of the clustered regions are extracted to generate defect feature vectors. The dispersion parameters include the variance, range, and coefficient of variation of the feature encoding values. At the same time, feature anchoring parameters are generated based on the adaptive spatial boundary of the clustered regions and output to the preprocessing module.

5. The intelligent defect identification and data analysis system based on phased array ultrasound according to claim 4, characterized in that, In the stage of generating feature anchoring parameters based on the adaptive spatial boundary of clustered regions, the feature distribution centers of the clustered regions in the lateral scanning position and depth direction in the two-dimensional grid coordinate system are extracted. Using the feature distribution centers as a reference, the adaptive boundary is obtained by adaptive expansion according to a preset edge margin, and the lateral scanning start and end coordinates of the adaptive boundary are determined. and the start and end coordinates of the depth direction ; Based on the element arrangement parameters and preset scanning rules of the phased array ultrasonic sensor array, the start and end coordinates of the lateral scan are mapped to the corresponding phased array channel number range. Based on the ultrasonic velocity of the weld base material and the start and end coordinates of the depth direction, the round-trip propagation time delay interval of the ultrasonic beam within the corresponding depth range is calculated using the depth-time delay mapping formula. The depth-delay mapping formula is as follows: ； In the formula: For weld depth The corresponding round-trip propagation delay of the ultrasonic beam; 1 represents the ultrasonic longitudinal wave velocity of the weld base material; 2 represents the round-trip path coefficient of the ultrasonic wave from transmission to reception. The minimum propagation delay is calculated based on the depth start and end coordinates. With maximum propagation delay This is mapped to the sampling time window of the echo signal within the corresponding channel. ,in and , and Each clock should be kept synchronized. Finally, the feature anchoring parameters, consisting of the channel number range, sampling time window, and preset upsampling factor, are output to the preprocessing module.

6. The intelligent defect identification and data analysis system based on phased array ultrasound according to claim 4, characterized in that, The execution logic of the closed-loop adaptive refinement processing link is as follows: After the mapping module outputs the feature anchoring parameters for the first time, the preprocessing module performs upsampling and refinement processing on the corresponding region signal and outputs it to the mapping module; the mapping module repeatedly performs feature encoding and feature mapping matrix update on the refined signal; if the fluctuation amplitude of the feature encoding value of the corresponding region in the updated feature mapping matrix exceeds the preset fluctuation threshold, the feature anchoring parameters are regenerated and output to the preprocessing module until the fluctuation amplitude of the feature encoding value is within the preset fluctuation threshold, and the closed-loop refinement processing is terminated. When the closed loop terminates, the feature mapping matrix and the defect feature vector after the final iteration convergence are used as the final result and output to the parsing module.

7. The intelligent defect identification and data analysis system based on phased array ultrasound according to claim 1, characterized in that, The analysis module, based on a pre-defined defect type feature library, performs multi-dimensional matching and classification on the defect feature vectors output by the mapping module to determine the defect type. Based on the coordinate distribution of the clustered regions corresponding to the defect feature vectors in the weld region feature mapping matrix, and combined with the element arrangement parameters of the phased array ultrasonic sensor array and the acoustic parameters of the weld base material, it calculates the three-dimensional spatial position, projection size, extension length, and orientation angle parameters of the defect within the weld. The orientation angle includes the horizontal orientation angle within the weld horizontal plane. Vertical inclination angle within the vertical section ; The analysis module has a built-in weld defect database. It associates and binds the defect type, spatial location, size, and orientation parameters output by the analysis module with the corresponding base material properties, welding process parameters, and scanning environment parameters of the weld, and stores them in the weld defect database. The statistical and correlation comparison performed by the analysis module includes time-series comparison of defect data from different scanning batches of the same weld, distribution statistics of defect data from different welds of the same batch and process, and normalized relative deviation correlation analysis of defect parameters and welding process parameters, generating corresponding statistical and comparison results.

8. The intelligent defect identification and data analysis system based on phased array ultrasound according to claim 7, characterized in that, The correlation analysis of the normalized relative deviation between defect parameters and welding process parameters follows the following order: This includes the individual correlation between defect parameters and welding process parameters. With overall correlation ; set up Here is a sequence of defect feature parameters, where For the first The defect characteristic parameters include the defect projection size, extension length, and horizontal orientation angle. Burial depth; definition This is the sequence of welding process parameters for the corresponding weld, where For the first Welding process parameters, including welding heat input, welding current, welding speed, and interpass temperature; For the first The preset baseline value of the defect characteristic parameter. For the first The preset rated values ​​of the welding process parameters; Normalized relative deviation calculation: ； in, This represents the normalized relative deviation of the k-th defect feature parameter. This represents the normalized relative deviation of the k-th welding process parameter; One-to-one correlation: ； in, Indicates the degree of correlation between individual items; Overall relevance: ； In the formula: This represents the number of matching items between defect characteristic parameters and welding process parameters.

9. The intelligent defect identification and data analysis system based on phased array ultrasound according to claim 1, characterized in that, The output module performs structured encoding on the statistical and correlation comparison results output by the analysis module to generate a standardized defect dataset. Based on the defect dataset and the weld area feature mapping matrix, a visual atlas is generated that includes two-dimensional cross-sectional imaging of the weld, three-dimensional spatial positioning imaging of defects, and annotation of defect feature parameters. At the same time, based on a preset report template, an inspection report is automatically generated that includes basic weld information, scanning parameters, defect details, statistical analysis results, and compliance judgment conclusions.

10. The intelligent defect identification and data analysis system based on phased array ultrasound according to claim 1, characterized in that, The acquisition module is interconnected with a preprocessing module via a wireless network. The mapping module is interconnected with a preprocessing module via a wireless network. The preprocessing module and the mapping module are interconnected with a parsing module and an analysis module via a wireless network. The parsing module and the analysis module are interconnected with an output module via a wireless network.

Images