Automatic recognition and measurement method for defect image based on phased array detection

Abstract

The present application relates to the technical field of nondestructive testing, and more particularly to an automatic identification and measurement method for defect images based on phased array detection, comprising: step S1, collecting a phased array ultrasonic detection original signal and generating a multimodal detection image; step S2, generating a fusion feature image; step S3, generating a preliminary screening defect list and obtaining a real defect list; step S4, extracting a defect geometric feature, calculating a defect measurement parameter, reconstructing a defect three-dimensional profile, and calculating a defect volume; step S5, identifying a defect type and generating a defect disposal decision; and step S6, performing precision compensation and optimization. The present application fuses multimodal image information, realizes accurate geometric measurement and three-dimensional morphological reconstruction of defects, establishes a multidimensional measurement confidence evaluation and optimization mechanism, and realizes self-evolution of a measurement evaluation system, so that the defect identification accuracy and measurement precision meet the requirements of high-demand industrial scenarios.

Description

Technical Field

[0001] This invention relates to the field of nondestructive testing technology, and in particular to a method for automatic identification and measurement of defect images based on phased array detection. Background Technology

[0002] Phased array ultrasonic testing (PAUT) technology, as one of the most advanced ultrasonic non-destructive testing methods, achieves dynamic deflection and focusing of the sound beam by controlling the excitation timing of each element in the transducer array, enabling the generation of high-resolution images of the interior of the workpiece under inspection. With the development of industrial manufacturing towards higher precision and higher reliability, the demand for the detection of internal defects in complex components in fields such as aerospace, nuclear power equipment, and petrochemicals is becoming increasingly urgent. PAUT technology, with its flexible sound beam control capabilities and rich imaging modes, has become a core technical means for the quality inspection of critical components.

[0003] However, PAUT technology faces significant technical bottlenecks in practical applications. First, the multimodal images (A-scan, B-scan, C-scan, S-scan) generated by phased array detection contain massive amounts of data. Traditional manual interpretation relies on the experience of inspectors, leading to strong subjectivity, low efficiency, and susceptibility to missed or misjudged defects. Second, existing automated identification methods mostly analyze single-modal images, failing to fully utilize the complementary information of multimodal images, thus limiting defect identification accuracy. More importantly, current technologies largely remain at the level of qualitative defect identification, lacking sufficient capabilities for precise geometric measurement (depth, length, height, volume) and three-dimensional morphological reconstruction, failing to meet the stringent requirements for quantitative defect assessment in engineering practice. Furthermore, the confidence level of measurement data is difficult to guarantee due to factors such as material properties, surface condition, and coupling conditions during the detection process. It lacks a dynamic optimization mechanism and the self-evolutionary capability of the measurement and evaluation system.

[0004] Chinese Patent Publication (Announcement) No. CN116858938A discloses an intelligent identification and classification method for defects in complex ring components using ultrasonic phased array testing, comprising the following steps: S1, selecting a suitable ultrasonic phased array probe and setting the process parameters for each area of ​​the complex ring component based on the geometry of the raceway, flange, and step of the cross-section and the depth range of the area to be tested; S2, conducting fully immersive ultrasonic phased array testing of the complex ring component based on the selected ultrasonic phased array probe and process parameters; the probe scans along the cross-section of the ring component while the ring component rotates itself to complete a full-coverage scan of the complex ring component; S3, during the full-coverage scan, the ultrasonic phased array software acquires ultrasonic phased array test images in real time and transmits the ultrasonic images to a pre-trained YOLOX-S target detection algorithm model to identify defects and determine the defect type online; S4, generating alarm information through the ultrasonic software to indicate that the complex ring component is of substandard quality. However, this solution still has the problem of using only a single C-scan image as input, without integrating multimodal image information, and only achieving qualitative classification of defects without involving precise geometric measurement and three-dimensional morphological reconstruction of defects. It also lacks a multi-dimensional measurement confidence assessment and optimization mechanism, and fails to achieve self-evolution of the measurement and evaluation system. As a result, the defect identification accuracy and measurement precision are difficult to meet the requirements of high-demand industrial scenarios. Summary of the Invention

[0005] To address this, the present invention provides an automatic identification and measurement method for defect images based on phased array detection, which overcomes the problems in existing technologies such as the lack of integration of multimodal image information, failure to achieve accurate geometric measurement and three-dimensional morphological reconstruction of defects, lack of establishment of multi-dimensional measurement confidence assessment and optimization mechanism, and failure to achieve self-evolution of measurement evaluation system, resulting in the inability to meet the requirements of high-demand industrial scenarios in terms of defect identification accuracy and measurement precision.

[0006] To achieve the above objectives, the present invention provides an automatic identification and measurement method for defect images based on phased array detection, the method comprising: Step S1: Acquire the raw signal of the phased array ultrasonic detection to obtain the raw phased array data, and generate a multimodal detection image based on the raw phased array data; Step S2: Obtain a standard multimodal image based on the multimodal detection image, and generate a fused feature image based on the standard multimodal image; Step S3: Mark the candidate defect regions according to the fused feature image, generate a preliminary defect list according to the candidate defect regions, determine the true nature of the defects according to the preliminary defect list, and obtain the true defect list according to the determination result. Step S4: Extract the geometric features of the defects based on the actual defect list, automatically measure the defect measurement parameters based on the defect geometric features, reconstruct the three-dimensional contour of the defects based on the defect measurement parameters, and calculate the defect volume based on the three-dimensional contour of the defects. Step S5: Automatically identify the defect type based on the three-dimensional contour of the defect and the defect measurement parameters, assess the defect hazard level based on the defect type, and generate a defect handling decision based on the assessment results; Step S6: Determine the measurement confidence level based on the defect measurement parameters, perform accuracy compensation for step S4 based on the judgment result, and optimize the measurement confidence level judgment process based on the number of accuracy compensations.

[0007] Further, step S2, when acquiring the standard multimodal image based on the multimodal detection image, includes: Step P01: Perform image quality assessment on the multimodal detection image to obtain an image quality score T. Compare the image quality score T with a preset image quality score T0. Based on the comparison result, determine whether the quality of the multimodal detection image meets the standard. When T≥T0, the quality is deemed to be up to standard, and the multimodal detection image is used as the multimodal image to be processed. When T < T0, the quality is judged to be substandard. Image enhancement is performed on the multimodal detection image to obtain the enhanced multimodal detection image, and the enhanced multimodal detection image is used as the multimodal image to be processed. Step P02: Perform geometric correction on the multimodal image to be processed to obtain a geometrically corrected multimodal image, and then normalize the grayscale of the geometrically corrected multimodal image to obtain a grayscale-normalized multimodal image. Step P03: Output the grayscale-normalized multimodal image as a standard multimodal image.

[0008] Further, when generating the fused feature image based on the standard multimodal image in step S2, it includes: Step P11: Input the A-scan image, B-scan image, C-scan image and S-scan image from the standard multimodal image into the multimodal feature extraction network to obtain the A-scan feature map, B-scan feature map, C-scan feature map and S-scan feature map respectively; Step P12: Calculate the fused feature image F based on the feature fusion weight coefficient set ω={ωA,ωB,ωC,ωS}, setting F=ωA×FA+ωB×FB+ωC×FC+ωS×FS, where FA is the A-scan feature map, FB is the B-scan feature map, FC is the C-scan feature map, and FS is the S-scan feature map, thus obtaining the fused feature image.

[0009] Further, step S3, which involves marking candidate defect regions based on the fused feature image and generating a preliminary defect list based on the candidate defect regions, includes: Step C01: Input the fused feature image into the defect candidate region detection network to obtain the candidate region confidence Z and candidate region location coordinates output by the defect candidate region detection network; Step C02: Compare the confidence scores of each candidate region in the candidate region confidence score set Z with the preset candidate region confidence scores Z, and mark the candidate defect regions according to the comparison results, wherein: When Z≥Z0, the location coordinates of the candidate region are marked as the candidate defect region, and the candidate defect region is added to the initial screening defect list; When Z < Z0, the location coordinates of the candidate region are not marked as a candidate defect region.

[0010] Furthermore, the step S3, which involves determining the true nature of defects based on the initial defect list, includes: Step V01: Extract multi-dimensional features from the candidate defect regions in the initial defect list to obtain a multi-dimensional feature vector of the defect. The multi-dimensional feature vector of the defect includes gray-scale statistical features, texture features, edge gradient features and frequency domain features. Step V02: Input the multi-dimensional feature vector of the defect into the defect authenticity judgment model to obtain the defect authenticity probability Preal output by the defect authenticity judgment model; Step V03: Compare the actual defect probability Preal with the preset actual defect probability Preal0. Based on the comparison result, determine the true nature of the defect, and obtain the list of actual defects based on the determination result. When Preal≥Preal0, the defect is determined to be a real defect and the candidate defect region is added to the list of real defects. When Preal < Preal0, the defect is determined to be a pseudo-defect, and the candidate defect area is not added to the list of real defects.

[0011] Further, step S4, which extracts the geometric features of defects based on the actual defect list, includes: Step Q01: Extract the timing data of the defect echo signal from the original phased array data based on the location coordinates of the candidate regions in the actual defect list. Step Q02: Calculate the defect depth D, defect length L, and defect height H based on the defect echo signal timing data; Step Q03: Output the defect depth D, defect length L, and defect height H as the defect geometric features; Step S4, which involves automatically measuring the defect measurement parameters based on the defect's geometric characteristics, includes: Step Q11: Calculate the equivalent diameter Deq of the defect based on the defect depth D, defect length L, and defect height H, and set Deq = (D × L × H). (1 / 3） ; Step Q12: Calculate the defect elongation rate E based on the defect depth D and defect length L, and set E=L / D; Step Q13: Calculate the defect height-to-length ratio R based on the defect height H and defect length L, and set R = H / L; Step Q14: Output the defect equivalent diameter Deq, defect elongation E, and defect height-to-length ratio R as defect measurement parameters; Step S4, reconstructing the three-dimensional contour of the defect based on the defect measurement parameters, includes: Step R01: Establish a three-dimensional spatial coordinate system for the defect based on the defect depth D, defect length L, and defect height H. Step R02: Calculate the three-dimensional morphology coefficient α of the defect based on the equivalent diameter Deq and the height-to-length ratio R, setting α = 1 / (1+e^(-qq)). (-k×(R-R0) ); Step R03: Construct the three-dimensional contour surface equation Z of the defect based on the three-dimensional morphology coefficient α of the defect, and set Z = α × Deq / 2 × cos(π × X / L) × cos(π × Y / H); Step R04: Determine the defect volume based on the three-dimensional contour of the defect. Perform calculations and set , The integral region is defined as the range of the defect boundary.

[0012] Furthermore, step S5, which automatically identifies the defect type based on the defect's three-dimensional contour and measurement parameters, includes: Step W01: Input the three-dimensional contour of the defect, the equivalent diameter of the defect Deq, the elongation rate of the defect E, the height-to-length ratio of the defect R, and the volume of the defect V into the defect type identification model to obtain the set of defect type confidence scores output by the defect type identification model. Step W02: Select the maximum value from the set of confidence scores for defect types to obtain the maximum confidence score for a defect type Tmax and the defect type corresponding to the maximum confidence score for a defect type Tmax. Step W03: Compare the maximum defect type confidence score Tmax with the preset defect type confidence score Tmax0, and confirm the defect type identification result based on the comparison result, wherein: When Tmax≥Tmax0, the defect type corresponding to the maximum defect type confidence Tmax is confirmed as the defect type identification result; When Tmax < Tmax0, the defect type identification result is marked as an unidentified type, and the manual review process is triggered.

[0013] Furthermore, step S5, which assesses the hazard level of a defect based on its type and generates a defect handling decision based on the assessment results, includes: Step J01: Determine the defect baseline hazard factor Gbase according to the defect type; Step J02: Calculate the size correction factor β based on the defect equivalent diameter Deq, and set β = 1 + ln(Deq / Deq0), where Deq0 is the reference defect equivalent diameter and ln is the natural logarithm. Step J03: Calculate the position correction coefficient γ based on the defect depth D, and set γ = 1 + η × (D / Dmax), where η is the position sensitivity coefficient and Dmax is the material thickness; Step J04: Calculate the comprehensive hazard index G based on the defect baseline hazard factor Gbase, size correction factor β, and position correction factor γ, and set G = Gbase × β × γ; Step J05: Compare the comprehensive hazard index G with the preset hazard level threshold set {G1, G2, G3}, assess the defect hazard level based on the comparison results, and generate a defect handling decision based on the assessment results, wherein: When G≤G1, the defect hazard level is assessed as Level I, the defect handling decision is generated and archived, and it is included in the scope of routine inspection and monitoring. No special handling is required during this inspection cycle. When G1 < G ≤ G2, the defect hazard level is assessed as Level II, and the defect handling decision is to shorten the detection cycle to 50% of the original cycle, increase the monitoring frequency, assess the defect development trend, and formulate a preventive maintenance plan. When G2 < G ≤ G3, the defect hazard level is assessed as Level III, and the defect handling decision is to immediately initiate a special testing procedure, use other non-destructive testing methods for verification and confirmation, determine the repair or replacement plan based on the verification results, and limit the equipment operating load. When G > G3, the defect hazard level is assessed as Level IV, and the defect handling decision is to immediately stop the equipment operation, initiate the emergency response procedure, isolate the defect area, organize expert assessment, formulate an emergency repair or replacement plan, and not resume operation without handling.

[0014] Furthermore, step S6, which determines the measurement confidence level based on the defect measurement parameters, includes: Step U01: Extract signal quality dimension parameters based on the time series data of the defect echo signal. The signal quality dimension parameters include signal-to-noise ratio (SNR), dynamic range (DR), and echo pulse width (τ). Step U02: Extract geometric measurement dimension parameters based on the defect geometric features. The geometric measurement dimension parameters include depth measurement resolution Rd, length measurement resolution Rl, and height measurement resolution Rh. Step U03: Extract system stability dimension parameters based on the original phased array data. The system stability dimension parameters include gain stability coefficient Gs, time baseline stability coefficient Ts, and probe coupling stability coefficient Cs. Step U04: Calculate the measurement confidence index C based on the signal quality dimension parameters, geometric measurement dimension parameters, and system stability dimension parameters. Set C = ωs × f(SNR,DR,τ) + ωg × f(Rd,Rl,Rh) + ωc × f(Gs,Ts,Cs), where ωs is the signal quality weight coefficient, ωg is the geometric measurement weight coefficient, and ωc is the system stability weight coefficient, satisfying ωs + ωg + ωc = 1. f() is the normalized scoring function, thus obtaining the measurement confidence index C. Step U05: Compare the measurement confidence index C with the preset measurement confidence index C0 and the secondary preset measurement confidence index C1, and determine the measurement confidence level based on the comparison results, wherein: Set C1 < C0; When C≥C0, the measurement confidence level is determined to be Level 1 confidence, and no accuracy compensation is performed in step S4. When C1≤C<C0, the measurement confidence level is determined to be level 2 confidence, and level 1 accuracy compensation is performed on step S4. When C < C1, the measurement confidence level is determined to be level three, and level two accuracy compensation is performed on step S4. Step S6, which performs accuracy compensation for step S4 based on the judgment result, includes: When the measurement confidence level is level 2, level 1 accuracy compensation is performed: Step U11a: Calculate the digital gain compensation coefficient Kd based on the deviation between the signal-to-noise ratio SNR and the reference signal-to-noise ratio SNR0. Set Kd=SNR0 / SNR to obtain the digital gain compensation coefficient. Step U12a: Perform digital gain compensation on the timing data of the defect echo signal according to the digital gain compensation coefficient Kd. Set As'=As×Kd, where As is the original echo amplitude and As' is the echo amplitude after digital gain compensation, and obtain the timing data of the defect echo signal after digital gain compensation. Step U13a: The timing data of the defect echo signal after digital gain compensation is re-inputted into step S4. Step S4 performs a first-level automatic remeasurement of the defect measurement parameters based on the timing data of the defect echo signal after digital gain compensation, and records the number of first-level automatic remeasurements as the first-level accuracy compensation number BCy. When the measurement confidence level is level three, perform level two accuracy compensation: Step U11b: Calculate the analog gain compensation coefficient Ka based on the deviation between the dynamic range DR and the reference dynamic range DR0. Set Ka = DR0 / DR to obtain the analog gain compensation coefficient. Step U12b: Perform analog gain compensation on the receiving link of the phased array ultrasonic testing according to the analog gain compensation coefficient Ka, and set Ga'=Ga×Ka, where Ga is the original analog gain and Ga' is the adjusted analog gain, to obtain the defect echo signal after analog gain adjustment. Step U13b: Calculate the pulse compression coefficient Kp based on the deviation between the echo pulse width τ and the reference pulse width τ0. Set Kp=τ0 / τ to obtain the pulse compression coefficient. Step U14b: Perform pulse compression processing on the defect echo signal after analog gain adjustment according to the pulse compression coefficient Kp, and set τ'=τ / Kp, where τ' is the pulse width after compression, to obtain the timing data of the defect echo signal after pulse compression. Step U15b: The timing data of the defect echo signal after pulse compression is re-input into step S4. Step S4 performs secondary automatic re-measurement of the defect measurement parameters based on the timing data of the defect echo signal after pulse compression, and records the number of secondary automatic re-measurements as the number of secondary accuracy compensations BCa.

[0015] Furthermore, step S6, which optimizes the determination process of measurement confidence level based on the number of accuracy compensations, includes: Step U21: Compare the first-level precision compensation count BCY with the preset first-level compensation threshold BCY0, and compare the second-level precision compensation count BCE with the preset second-level compensation threshold BCE0. Based on the comparison results, decide on the optimization strategy for adjusting the parameters, wherein: When BCy≤BCy0 and BCE=0, the optimization strategy is to maintain the current parameters and not optimize the process of determining the confidence level. When BCy>BCy0 and BCE=0, the optimization strategy is determined to be Level 1 optimization, which optimizes the primary optimization adjustment parameters of the determination process of measurement confidence level. When BCa > 0, the optimization strategy is determined to be a secondary optimization, which optimizes the high-level optimization adjustment parameters of the determination process of measurement confidence level. The optimization of the measurement confidence level determination process based on the number of accuracy compensations in step S6 also includes: Step U22: When the optimization strategy is Level 1 optimization, the primary optimization adjustment parameters are optimized. These primary optimization adjustment parameters include the signal quality weighting coefficient ωs, the geometric measurement weighting coefficient ωg, and the system stability weighting coefficient ωc, where: The deviation of the weight coefficients of each dimension is calculated based on the number of first-level precision compensations BCy. The signal quality deviation δs=|Ns-N0| / N0, the geometric measurement deviation δg=|Ng-N0| / N0, and the system stability deviation δc=|Nc-N0| / N0 are set, where Ns, Ng, and Nc are the number of times the first-level precision compensation is triggered in the signal quality dimension, geometric measurement dimension, and system stability dimension, respectively, and N0 is the baseline value of the expected number of compensations for each dimension. The dominant imbalance dimension ARG is identified based on the principle of maximum deviation. The dominant imbalance dimension ARG is set as argmax(δs,δg,δc) to obtain the dominant imbalance dimension ARG. The weight coefficients are rebalanced and optimized based on the dominant imbalance dimension ARG. When the dominant imbalance dimension ARG is the signal quality dimension, ωs'=ωs×(1-ρ×δs), ωg'=ωg+ρ×δs×ωg / (ωg+ωc), ωc'=ωc+ρ×δs×ωc / (ωg+ωc), where ρ is the weight adjustment rate coefficient. The rebalanced weight coefficient set {ωs',ωg',ωc'} is obtained, and ωs'+ωg'+ωc'=1 is satisfied. When the dominant imbalance dimension is the geometric measurement dimension, we set ωg'=ωg×(1-ρ×δg), ωs'=ωs+ρ×δg×ωs / (ωs+ωc), ωc'=ωc+ρ×δg×ωc / (ωs+ωc), where ρ is the weight adjustment rate coefficient, and obtain the rebalanced weight coefficient set {ωs',ωg',ωc'}, which satisfies ωs'+ωg'+ωc'=1; When the dominant imbalance dimension is the system stability dimension, we set ωc'=ωc×(1-ρ×δc), ωs'=ωs+ρ×δc×ωs / (ωs+ωg), ωg'=ωg+ρ×δc×ωg / (ωs+ωg), where ρ is the weight adjustment rate coefficient, and obtain the set of weight coefficients after rebalancing {ωs',ωg',ωc'}, which satisfies ωs'+ωg'+ωc'=1; The optimization of the measurement confidence level determination process based on the number of accuracy compensations in step S6 also includes: Step U23: When the optimization strategy is secondary optimization, the advanced optimization adjustment parameters are optimized. These advanced optimization adjustment parameters include a preset measurement confidence index C0, a secondary preset measurement confidence index C1, a reference signal-to-noise ratio SNR0, a reference dynamic range DR0, and a reference pulse width τ0, wherein: The pass rate ηc of the measurement confidence index C is calculated based on the number of secondary accuracy compensations. ηc is set as ηc = Npass / (Npass + Nfail), where Npass is the number of times C ≥ C0 and Nfail is the number of times C < C0. The measurement confidence pass rate ηc is obtained. The preset measurement confidence index C0 is optimized based on the measurement confidence achievement rate ηc. C0' is set as C0' = C0 × [1 + σ × (ηc - ηc0)], where σ is the confidence adjustment coefficient and ηc0 is the target achievement rate, thus obtaining the optimized preset measurement confidence index C0'. Based on the optimized preset measurement confidence index C0', the secondary preset measurement confidence index C1 is optimized in conjunction with it. C1' is set to C1×(C0' / C0) to obtain the optimized secondary preset measurement confidence index C1'. Based on the actual resolution improvement effect of the timing data of the defect echo signal after pulse compression, the reference pulse width τ0 is iteratively optimized to obtain the optimized reference pulse width τ0'. τ0' is set as τ0×[1-ξ×(Kp-1) / Kp], where ξ is the resolution convergence coefficient.

[0016] Compared with existing technologies, the beneficial effects of this invention are as follows: the method generates multimodal detection images in step S1, providing a comprehensive data foundation for subsequent analysis; the method performs quality assessment, enhancement, and fusion of the multimodal detection images in step S2, fully utilizing the complementary information of the four modalities (A-scan, B-scan, C-scan, and S-scan) to significantly improve the ability to express defect features; the method screens and verifies the authenticity of candidate defect regions in step S3, effectively eliminating false defect interference and improving the accuracy of defect identification; and the method accurately extracts the geometric features of defects in step S4, and innovatively realizes three-dimensional contour reconstruction and volume calculation of defects, meeting the requirements of engineering inspection for defects. The method employs stringent quantitative assessment requirements. Step S5 combines three-dimensional morphological features and geometric measurement parameters to identify defect types and comprehensively assess hazard levels based on size and location factors. Defect hazards are categorized into four levels, facilitating the development of differentiated treatment strategies and providing a scientific basis for defect management. Step S6 establishes a three-dimensional measurement confidence assessment system. A hierarchical, progressive accuracy compensation mechanism enables dynamic control of measurement quality, while an optimization mechanism for adjusting parameters facilitates the self-evolution of the measurement assessment system, forming a complete closed loop of "assessment-compensation-optimization-reassessment." This comprehensively improves the accuracy, precision, and reliability of automatic identification and measurement of defect images detected by phased array detectors. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the automatic identification and measurement method for defect images based on phased array detection in this embodiment. Figure 2 This is a flowchart illustrating the process of extracting the geometric features of the defect in step S4 of this embodiment. Figure 3 This is a flowchart illustrating the automatic measurement of defect measurement parameters in step S4 of this embodiment. Figure 4 This is a flowchart illustrating the process of reconstructing the three-dimensional contour of the defect in step S4 of this embodiment. Figure 5 This is a flowchart illustrating the automatic identification of defect types in step S5 of this embodiment. Detailed Implementation

[0018] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0019] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0020] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0021] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0022] Please see Figure 1 As shown, this is a flowchart illustrating the automatic identification and measurement method for defect images based on phased array detection in this embodiment. The method includes: Step S1: Acquire the raw signal of the phased array ultrasonic detection to obtain the raw phased array data, and generate a multimodal detection image based on the raw phased array data; Step S2: Obtain a standard multimodal image based on the multimodal detection image, and generate a fused feature image based on the standard multimodal image; Step S3: Mark the candidate defect regions according to the fused feature image, generate a preliminary defect list according to the candidate defect regions, determine the true nature of the defects according to the preliminary defect list, and obtain the true defect list according to the determination result. Step S4: Extract the geometric features of the defects based on the actual defect list, automatically measure the defect measurement parameters based on the defect geometric features, reconstruct the three-dimensional contour of the defects based on the defect measurement parameters, and calculate the defect volume based on the three-dimensional contour of the defects. Step S5: Automatically identify the defect type based on the three-dimensional contour of the defect and the defect measurement parameters, assess the defect hazard level based on the defect type, and generate a defect handling decision based on the assessment results; Step S6: Determine the measurement confidence level based on the defect measurement parameters, perform accuracy compensation for step S4 based on the judgment result, and optimize the measurement confidence level judgment process based on the number of accuracy compensations.

[0023] Specifically, the automatic identification and measurement method for defect images based on phased array detection is applied to industrial non-destructive testing terminals, such as aerospace component inspection systems, nuclear power equipment inspection systems, and petrochemical pipeline inspection systems. The method generates multimodal detection images in step S1, providing a comprehensive data foundation for subsequent analysis. Step S2 performs quality assessment, enhancement, and fusion of the multimodal detection images, fully utilizing the complementary information of the four modes (A-scan, B-scan, C-scan, and S-scan) to significantly improve the ability to express defect features. Step S3 filters and verifies the authenticity of candidate defect regions, effectively eliminating false defect interference and improving the accuracy of defect identification. Step S4 precisely extracts the geometric features of defects and innovatively... The method achieves three-dimensional contour reconstruction and volume calculation of defects, meeting the stringent requirements of engineering inspection for quantitative defect assessment. In step S5, the method identifies defect types by combining three-dimensional morphological features and geometric measurement parameters, and assesses the hazard level by comprehensively considering size and location factors. The hazard level is divided into four categories, facilitating the development of differentiated treatment strategies and providing a scientific basis for defect management. In step S6, the method establishes a three-dimensional measurement confidence assessment system, achieves dynamic control of measurement quality through a hierarchical progressive accuracy compensation mechanism, and realizes the self-evolution of the measurement assessment system through an optimization mechanism that adjusts parameters, forming a complete closed loop of "assessment-compensation-optimization-reassessment." This comprehensively improves the accuracy, precision, and reliability of automatic identification and measurement of defect images detected by phased array detectors.

[0024] Specifically, in step S1, when acquiring the raw phased array ultrasonic detection signal to obtain raw phased array data, and further using this raw phased array data to generate a multimodal detection image, the raw phased array ultrasonic detection signal refers to the ultrasonic echo electrical signal received by each element of the phased array ultrasonic probe. The raw phased array data refers to the digital signal data obtained after analog-to-digital conversion of the raw phased array ultrasonic detection signal. This embodiment does not limit the specific implementation method for acquiring the raw phased array ultrasonic detection signal; those skilled in the art can set it according to actual conditions, such as outputting data from each element of the phased array ultrasonic probe through a multi-channel high-speed data acquisition card. The signal is synchronously sampled at a sampling frequency no less than 5 times the center frequency of the probe. The multimodal detection images include A-scan images, B-scan images, C-scan images, and S-scan images. This embodiment does not limit the specific implementation method for generating multimodal detection images based on the phased array raw data. Those skilled in the art can set it according to the actual situation. For example, the full-matrix capture technology can be used to obtain fully focused data, and A-scan images can be generated by the A-scan display algorithm, B-scan images by the B-scan display algorithm, C-scan images by the C-scan display algorithm, and S-scan images by the S-scan display algorithm. The full matrix capture technology is called Full Matrix Capture, abbreviated as FMC. The A-scan image refers to the waveform display image of the ultrasonic echo amplitude changing with time and depth. The B-scan image refers to the two-dimensional cross-sectional display image unfolded along the scanning direction. The C-scan image refers to the planar projection display image parallel to the detection surface. The S-scan image refers to the angle-depth display image within the fan-shaped scanning range.

[0025] Specifically, step S1 generates multimodal detection images to provide a comprehensive data foundation for subsequent analysis.

[0026] Specifically, step S2, when acquiring the standard multimodal image based on the multimodal detection image, includes: Step P01: Perform image quality assessment on the multimodal detection image to obtain an image quality score T. Compare the image quality score T with a preset image quality score T0. Based on the comparison result, determine whether the quality of the multimodal detection image meets the standard. When T≥T0, the quality is deemed to be up to standard, and the multimodal detection image is used as the multimodal image to be processed. When T < T0, the quality is judged to be substandard. Image enhancement is performed on the multimodal detection image to obtain the enhanced multimodal detection image, and the enhanced multimodal detection image is used as the multimodal image to be processed. Step P02: Perform geometric correction on the multimodal image to be processed to obtain a geometrically corrected multimodal image, and then normalize the grayscale of the geometrically corrected multimodal image to obtain a grayscale-normalized multimodal image. Step P03: Output the grayscale-normalized multimodal image as a standard multimodal image.

[0027] Specifically, the image quality score R refers to a quantitative indicator that comprehensively evaluates the clarity and usability of multimodal detection images, measured in points. In this embodiment, the image quality score is calculated by weighting three sub-indicators: image signal-to-noise ratio Rtx, contrast Rdb, and edge sharpness Rby. R is set to 0.4 × Rtx + 0.3 × Rdb + 0.3 × Rby. The preset image quality score refers to the threshold for determining whether the image quality meets the standard. In this embodiment, the preset image quality score is set to 75 points based on the detection accuracy requirements. Image enhancement refers to the process of improving image quality through digital image processing technology. This embodiment does not limit the specific implementation of image enhancement; those skilled in the art can set it according to actual conditions, such as using an adaptive histogram equalization algorithm to enhance image contrast. Geometric correction refers to the process of eliminating image geometric distortion. In this embodiment, a coordinate mapping relationship is established based on the geometric parameters and scanning motion parameters of the phased array probe, and the image is resampled and corrected. Gray-level normalization refers to the process of unifying the image gray-level values ​​to a standard range. In this embodiment, the image gray-level values ​​are normalized to the range [0, 255].

[0028] Specifically, step S2, when generating the fused feature image based on the standard multimodal image, includes: Step P11: Input the A-scan image, B-scan image, C-scan image and S-scan image from the standard multimodal image into the multimodal feature extraction network to obtain the A-scan feature map, B-scan feature map, C-scan feature map and S-scan feature map respectively; Step P12: Calculate the fused feature image F based on the feature fusion weight coefficient set ω={ωA,ωB,ωC,ωS}, setting F=ωA×FA+ωB×FB+ωC×FC+ωS×FS, where FA is the A-scan feature map, FB is the B-scan feature map, FC is the C-scan feature map, and FS is the S-scan feature map, thus obtaining the fused feature image.

[0029] Specifically, the multimodal feature extraction network refers to a deep residual network that performs deep feature extraction on four types of scanned images. In this embodiment, four deep residual networks with identical structures but independent parameters are used as feature extractors for A-scan, B-scan, C-scan, and S-scan, respectively. The deep residual network is the one proposed by Microsoft Research, containing 50 learnable layers with the following parameters: input layer image size 224×224 pixels, convolution kernel size 7×7, initial number of convolution channels 64, and residual block structure {[1×1,64;3×3,64;1×1,256]×3,[1×1,128;3×3,128;1×1,512]×4,[1×1,256;3×3,256;1×1

[1024] ×6, [1×1,512;3×3,512;1×1,2048]×3}, the global average pooling layer outputs a 2048-dimensional feature vector. In this embodiment, the last fully connected layer is modified to output a 256-dimensional feature map. The training batch size is 32, the learning rate is 0.001, and the training epochs are 100. The set of feature fusion weight coefficients refers to the weight coefficients that measure the importance of each modality feature in the fusion process. In this embodiment, ωA=0.15, ωB=0.25, ωC=0.35, ωS=0.25 are set according to the contribution of each modality to defect detection, satisfying ωA+ωB+ωC+ωS=1. The fused feature image refers to the unified feature representation after integrating the information of the four modalities, which has a stronger defect representation capability.

[0030] Specifically, step S3, which involves marking candidate defect regions based on the fused feature image and generating a preliminary defect list based on the candidate defect regions, includes: Step C01: Input the fused feature image into the defect candidate region detection network to obtain the candidate region confidence Z and candidate region location coordinates output by the defect candidate region detection network; Step C02: Compare the confidence scores of each candidate region in the candidate region confidence score set Z with the preset candidate region confidence scores Z, and mark the candidate defect regions according to the comparison results, wherein: When Z≥Z0, the location coordinates of the candidate region are marked as the candidate defect region, and the candidate defect region is added to the initial screening defect list; When Z < Z0, the location coordinates of the candidate region are not marked as a candidate defect region.

[0031] Specifically, the defect candidate region detection network refers to a deep learning-based target detection network. The parameters of the defect candidate region detection network in this embodiment are as follows: YOLOv8n architecture, input image size 640×640 pixels, backbone network CSPDarknet53, neck network PANet, detection head decoupled head structure, and anchor box scale {(10,13),(16,30),(33,23),(30,61),(62,45),(59,119),(116,90),(156,198),(373,3...} 26)}, confidence threshold 0.6, non-maximum suppression IoU threshold 0.45, training dataset contains 5000 labeled images, validation set 1000 images, test set 1000 images, training epochs 300, learning rate 0.01, optimizer is SGD, momentum 0.937, the candidate region confidence Z refers to the probability value of the region having a defect output by the defect candidate region detection network, the value range is [0,1], the preset candidate region confidence refers to the threshold for determining whether it is a candidate defect region, in this embodiment the preset candidate region confidence is set to 0.6 to reduce the risk of missed detection.

[0032] Specifically, the step S3, which involves determining the true nature of defects based on the initial defect list, includes: Step V01: Extract multi-dimensional features from the candidate defect regions in the initial defect list to obtain a multi-dimensional feature vector of the defect. The multi-dimensional feature vector of the defect includes gray-scale statistical features, texture features, edge gradient features and frequency domain features. Step V02: Input the multi-dimensional feature vector of the defect into the defect authenticity judgment model to obtain the defect authenticity probability Preal output by the defect authenticity judgment model; Step V03: Compare the actual defect probability Preal with the preset actual defect probability Preal0. Based on the comparison result, determine the true nature of the defect, and obtain the list of actual defects based on the determination result. When Preal≥Preal0, the defect is determined to be a real defect and the candidate defect region is added to the list of real defects. When Preal < Preal0, the defect is determined to be a pseudo-defect, and the candidate defect area is not added to the list of real defects.

[0033] Specifically, the multi-dimensional feature vector of defects refers to a vector that quantifies and describes the features of defects from four dimensions: grayscale, texture, edge, and frequency domain. The grayscale statistical features include regional average grayscale, grayscale variance, grayscale skewness, and kurtosis, which are extracted using grayscale histogram statistics. The texture features are extracted using the grayscale co-occurrence matrix to obtain contrast, correlation, energy, and homogeneity. The edge gradient features are extracted using the Sobel operator to obtain edge intensity and orientation histograms. The frequency domain features are extracted using discrete cosine transform to obtain low-frequency and high-frequency energy distributions. The defect authenticity judgment model is a binary classification model that distinguishes between real and fake defects. The parameters of the defect authenticity judgment model described in the example are as follows: a support vector machine classifier is used, the kernel function is a radial basis function, the penalty parameter is 1.0, the kernel coefficient is 0.1, the input feature dimension is 26 dimensions in total: 4-dimensional gray-level statistical features + 4-dimensional texture features + 16-dimensional edge gradient features + 2-dimensional frequency domain features, 2000 real defects and 2000 fake defects are used as training samples, the cross-validation fold number is 5, the model accuracy is greater than or equal to 92%, and it is trained by real defect samples and fake defect samples. The preset defect authenticity probability Preal0 refers to the threshold for judging real defects. In this embodiment, Preal0 is set to 0.75 to ensure the reliability of defect recognition.

[0034] Please see Figure 2 As shown, this is a flowchart illustrating step S4 of this embodiment, which involves extracting the geometric features of defects. Step S4, which extracts the geometric features of defects based on the actual defect list, includes: Step Q01: Extract the timing data of the defect echo signal from the original phased array data based on the location coordinates of the candidate regions in the actual defect list. Step Q02: Calculate the defect depth D, defect length L, and defect height H based on the defect echo signal timing data; Step Q03: Output the defect depth D, defect length L, and defect height H as the defect geometric features.

[0035] Specifically, the defect echo signal time-series data refers to the sequence data of ultrasonic echo amplitude changes over time corresponding to the location coordinates of candidate regions in the actual defect list. The method used in this embodiment to extract the defect echo signal time-series data is as follows: Based on the defect location coordinates (xd, yd) in the actual defect list, locate the corresponding A-scan signal channel index i and time-series sampling point index j in the original phased array data. Set i = xd / Δxd, j = yd / Δyd, where Δxd is the spatial sampling interval and Δyd is the temporal sampling interval. Extract the full-time signal data A(i,t) corresponding to channel index i. M sampling points are taken before and after the timing sampling point index j as timing windows. The timing data of the defect echo signal As(t) is set as A(i,jM:j+M), where M is the timing half-window length, to obtain timing data containing defect echo characteristics. In this embodiment, M=50 is set. The defect depth D refers to the position of the defect in the material thickness direction, which is calculated based on the ultrasonic propagation time. The defect length L refers to the size of the defect in the scanning direction, which is calculated based on the display range of the defect in the scanning path. The defect height H refers to the size of the defect perpendicular to the scanning direction, which is calculated based on the display range in the image resolution direction.

[0036] Please see Figure 3 As shown, this is a flowchart illustrating the automatic measurement of defect measurement parameters in step S4 of this embodiment. Step S4, which involves automatically measuring defect measurement parameters based on defect geometric features, includes: Step Q11: Calculate the equivalent diameter Deq of the defect based on the defect depth D, defect length L, and defect height H, and set Deq = (D × L × H). (1 / 3） ; Step Q12: Calculate the defect elongation rate E based on the defect depth D and defect length L, and set E=L / D; Step Q13: Calculate the defect height-to-length ratio R based on the defect height H and defect length L, and set R = H / L; Step Q14 outputs the defect equivalent diameter Deq, defect elongation E, and defect height-to-length ratio R as defect measurement parameters.

[0037] Specifically, the defect equivalent diameter Deq refers to the diameter when the defect is equivalent to a sphere, and is used to uniformly quantify the size of defects of different shapes. The defect elongation rate E refers to the ratio of the defect length to the depth, reflecting the extension trend of the defect. The defect height-to-length ratio R refers to the ratio of the defect height to the length, reflecting the flatness of the defect.

[0038] Please see Figure 4 As shown, this is a flowchart illustrating step S4 of this embodiment, which involves reconstructing the three-dimensional contour of the defect. Step S4, which reconstructs the three-dimensional contour of the defect based on the defect measurement parameters, includes: Step R01: Establish a three-dimensional spatial coordinate system for the defect based on the defect depth D, defect length L, and defect height H. Step R02: Calculate the three-dimensional morphology coefficient α of the defect based on the equivalent diameter Deq and the height-to-length ratio R, setting α = 1 / (1+e^(-qq)). (-k×(R-R0) ); Step R03: Construct the three-dimensional contour surface equation Z of the defect based on the three-dimensional morphology coefficient α of the defect, and set Z = α × Deq / 2 × cos(π × X / L) × cos(π × Y / H); Step R04: Determine the defect volume based on the three-dimensional contour of the defect. Perform calculations and set , The integral region is defined as the range of the defect boundary.

[0039] Specifically, the defect three-dimensional spatial coordinate system is a local coordinate system with the defect center as the origin. The X-axis is along the defect length direction, the Y-axis is along the defect height direction, and the Z-axis is along the defect depth direction. The defect three-dimensional morphology coefficient α is a coefficient describing the degree to which the three-dimensional morphology of the defect deviates from the standard ellipsoid, and its value range is (0,1]. When R=R0, α=0.5. When R is much greater than R0, α approaches 1, indicating that the defect is flat and disc-shaped. When R is much less than R0, α approaches 0, indicating that the defect is long and thin. The k is a morphology sensitivity coefficient, which controls the sensitivity of α to changes in R. In this embodiment, k=5 and R0=0.5. The defect three-dimensional contour surface equation is a continuous smooth surface constructed based on the cosine function, which can describe a variety of typical defect morphologies. The defect volume V is obtained by performing a triple integral on the surface equation within the defect boundary range, reflecting the actual space occupied by the defect.

[0040] Please see Figure 5 As shown, this is a flowchart illustrating the automatic defect type identification process in step S5 of this embodiment. Step S5, which involves automatically identifying the defect type based on the three-dimensional contour of the defect and defect measurement parameters, includes: Step W01: Input the three-dimensional contour of the defect, the equivalent diameter of the defect Deq, the elongation rate of the defect E, the height-to-length ratio of the defect R, and the volume of the defect V into the defect type identification model to obtain the set of defect type confidence scores output by the defect type identification model. Step W02: Select the maximum value from the set of confidence scores for defect types to obtain the maximum confidence score for a defect type Tmax and the defect type corresponding to the maximum confidence score for a defect type Tmax. Step W03: Compare the maximum defect type confidence score Tmax with the preset defect type confidence score Tmax0, and confirm the defect type identification result based on the comparison result, wherein: When Tmax≥Tmax0, the defect type corresponding to the maximum defect type confidence Tmax is confirmed as the defect type identification result; When Tmax < Tmax0, the defect type identification result is marked as an unidentified type, and the manual review process is triggered.

[0041] Specifically, the defect type identification model refers to a deep learning-based multi-classification model. The parameters of the defect type identification model in this embodiment are as follows: It adopts a 3D ResNet-18 architecture; the input data size is 64×64×64 voxels of the defect's 3D contour data + 5-dimensional geometric measurement parameter vectors; the convolution kernel size is {3×3×3, 3×3×3, 3×3×3, 3×3×3}; the number of channels is {64, 128, 256, 512}; the number of residual block repetitions is {2, 2, 2, 2}; the fully connected layer outputs 5-dimensional classification results, including cracks, porosity, inclusions, lack of fusion, and incomplete penetration; Softmax activation is used; cross-entropy loss function is employed; the training batch size is 16; the learning rate is 0.0001; the training epochs are 200; training stops when the validation accuracy is greater than or equal to 95%; the input data includes the defect's 3D contour, defect equivalent diameter Deq, defect elongation E, defect height-to-length ratio R, and defect volume V. The ResNet-18 architecture refers to a deep learning architecture that extends a two-dimensional residual network to three-dimensional space. It extracts the spatiotemporal features of three-dimensional data through 3D convolutional kernels and includes 3D convolutional layers, 3D batch normalization layers, 3D max pooling layers, and 3D residual blocks. The residual connection method is consistent with that of 2D ResNet, which effectively alleviates the gradient vanishing problem in deep networks and improves the feature extraction capability of three-dimensional data. The preset defect type confidence level Tmax0 refers to the minimum confidence threshold for confirming the defect type. In this embodiment, Tmax0 is set to 0.8. When the maximum confidence level is lower than this threshold, it indicates that the defect features are unclear and manual intervention is required for verification.

[0042] Specifically, step S5 assesses the hazard level of a defect based on its type, and generates a defect handling decision based on the assessment results, including: Step J01: Determine the defect baseline hazard factor Gbase according to the defect type; Step J02: Calculate the size correction factor β based on the defect equivalent diameter Deq, and set β = 1 + ln(Deq / Deq0), where Deq0 is the reference defect equivalent diameter and ln is the natural logarithm. Step J03: Calculate the position correction coefficient γ based on the defect depth D, and set γ = 1 + η × (D / Dmax), where η is the position sensitivity coefficient and Dmax is the material thickness; Step J04: Calculate the comprehensive hazard index G based on the defect baseline hazard factor Gbase, size correction factor β, and position correction factor γ, and set G = Gbase × β × γ; Step J05: Compare the comprehensive hazard index G with the preset hazard level threshold set {G1, G2, G3}, assess the defect hazard level based on the comparison results, and generate a defect handling decision based on the assessment results, wherein: When G≤G1, the defect hazard level is assessed as Level I, the defect handling decision is generated and archived, and it is included in the scope of routine inspection and monitoring. No special handling is required during this inspection cycle. When G1 < G ≤ G2, the defect hazard level is assessed as Level II, and the defect handling decision is to shorten the detection cycle to 50% of the original cycle, increase the monitoring frequency, assess the defect development trend, and formulate a preventive maintenance plan. When G2 < G ≤ G3, the defect hazard level is assessed as Level III, and the defect handling decision is to immediately initiate a special testing procedure, use other non-destructive testing methods for verification and confirmation, determine the repair or replacement plan based on the verification results, and limit the equipment operating load. When G > G3, the defect hazard level is assessed as Level IV, and the defect handling decision is to immediately stop the equipment operation, initiate the emergency response procedure, isolate the defect area, organize expert assessment, formulate an emergency repair or replacement plan, and not resume operation without handling.

[0043] Specifically, the defect baseline hazard coefficient Gbase refers to the baseline value of inherent hazard level determined according to the defect type. In this embodiment, based on engineering experience, Gbase is set to 4.0 for cracks, 3.5 for lack of fusion, 3.0 for incomplete penetration, 2.0 for slag inclusions, and 1.5 for porosity. The size correction coefficient β is a coefficient used to correct the baseline hazard based on the actual size of the defect; the larger the defect, the higher the hazard. Deq0 is the equivalent diameter of the baseline defect; in this embodiment, Deq0 is set to 2mm. The position correction coefficient γ is a coefficient used to correct the baseline hazard based on the depth and position of the defect; surface defects are more hazardy than internal defects. η is the position sensitivity coefficient; in this embodiment, η is set to 0.3. Dmax is the thickness of the material being tested. The preset hazard level threshold set refers to the boundary values ​​for dividing hazard levels. In this embodiment, G1 is set to 2.0, G2 to 4.0, and G3 to 6.0, dividing the defect hazard into four levels to facilitate the development of differentiated treatment strategies.

[0044] Specifically, step S6, which determines the measurement confidence level based on the defect measurement parameters, includes: Step U01: Extract signal quality dimension parameters based on the time series data of the defect echo signal. The signal quality dimension parameters include signal-to-noise ratio (SNR), dynamic range (DR), and echo pulse width (τ). Step U02: Extract geometric measurement dimension parameters based on the defect geometric features. The geometric measurement dimension parameters include depth measurement resolution Rd, length measurement resolution Rl, and height measurement resolution Rh. Step U03: Extract system stability dimension parameters based on the original phased array data. The system stability dimension parameters include gain stability coefficient Gs, time baseline stability coefficient Ts, and probe coupling stability coefficient Cs. Step U04: Calculate the measurement confidence index C based on the signal quality dimension parameters, geometric measurement dimension parameters, and system stability dimension parameters. Set C = ωs × f(SNR,DR,τ) + ωg × f(Rd,Rl,Rh) + ωc × f(Gs,Ts,Cs), where ωs is the signal quality weight coefficient, ωg is the geometric measurement weight coefficient, and ωc is the system stability weight coefficient, satisfying ωs + ωg + ωc = 1. f() is the normalized scoring function, thus obtaining the measurement confidence index C. Step U05: Compare the measurement confidence index C with the preset measurement confidence index C0 and the secondary preset measurement confidence index C1, and determine the measurement confidence level based on the comparison results, wherein: Set C1 < C0; When C≥C0, the measurement confidence level is determined to be Level 1 confidence, and no accuracy compensation is performed in step S4. When C1≤C<C0, the measurement confidence level is determined to be level 2 confidence, and level 1 accuracy compensation is performed on step S4. When C < C1, the measurement confidence level is determined to be level three, and level two accuracy compensation is performed on step S4.

[0045] Specifically, the signal-to-noise ratio (SNR) refers to the ratio of the defect echo signal amplitude to the background noise amplitude in decibels, expressed as SNR = 20 × lg(As / An), where As is the defect echo amplitude and An is the noise amplitude. The dynamic range (DR) refers to the ratio of the largest to the smallest signal that an industrial non-destructive testing terminal can simultaneously resolve, reflecting its ability to handle both strong and weak signals. DR is set to 20 × lg(Amax / Amin), where Amax is the maximum undistorted signal amplitude and Amin is the minimum resolvable signal amplitude. The pulse width τ refers to the duration of the defect echo signal in the time domain, which is directly related to the axial resolution of the industrial non-destructive testing terminal. The depth measurement resolution Rd refers to the minimum distance that the industrial non-destructive testing terminal can resolve in the depth direction, and is set as Rd = c / (2×BW), where c is the material sound velocity and BW is the effective bandwidth of the industrial non-destructive testing terminal. The length measurement resolution Rl refers to the minimum distance that the industrial non-destructive testing terminal can resolve in the scanning direction, and is set as Rl = Δx×(1+ΔL / L), where Δx is the scanning step distance and ΔL is the effective sound beam of the probe. The width, L, represents the defect display length. The height measurement resolution Rh refers to the minimum distance that the industrial non-destructive testing terminal can resolve in the vertical scanning direction, set as Rh = Δy × (1 + ΔH / H), where Δy is the pixel spacing in the vertical direction, ΔH is the beam spread angle of the probe in the height direction, and H is the defect display height. The gain stability coefficient Gs refers to the degree of gain fluctuation of the industrial non-destructive testing terminal during the testing process, set as Gs = 1 - σg / μg, where σg is the gain standard deviation and μg is the gain mean. The time baseline stability coefficient Ts refers to the industrial non-destructive testing terminal's gain stability coefficient. The drift of the detection terminal time reference is set as Ts = 1 - Δtmax / T, where Δtmax is the maximum time drift and T is the detection period. The probe coupling stability coefficient Cs refers to the stability of the coupling state between the probe and the workpiece being detected, and is set as Cs = 1 - Nc / N, where Nc is the number of coupling failure events and N is the total number of detections. The normalized scoring function f() is a standardized function that maps each dimension parameter to the interval [0,1]. In this embodiment, the Sigmoid function is used for normalization, and f(Sigmoid) = 1 / (1+e (-ks×(Sigmoid-Sigmoid0) / Sigmoid0)） Where ks is the curve steepness coefficient, Sigmoid0 is the parameter reference value, and the signal quality weight coefficient ωs, geometric measurement weight coefficient ωg, and system stability weight coefficient ωc are determined according to the degree of influence of each dimension on the measurement accuracy. In this embodiment, ωs=0.4, ωg=0.35, and ωc=0.25 are set. The preset measurement confidence index C0 is the threshold for judging the measurement result as highly reliable. In this embodiment, C0=0.85 is set. The secondary preset measurement confidence index C1 is the threshold for judging the measurement result as needing depth compensation. In this embodiment, C1=0.60 is set.

[0046] Specifically, step S6, which performs precision compensation for step S4 based on the judgment result, includes: When the measurement confidence level is level 2, level 1 accuracy compensation is performed: Step U11a: Calculate the digital gain compensation coefficient Kd based on the deviation between the signal-to-noise ratio SNR and the reference signal-to-noise ratio SNR0. Set Kd=SNR0 / SNR to obtain the digital gain compensation coefficient. Step U12a: Perform digital gain compensation on the timing data of the defect echo signal according to the digital gain compensation coefficient Kd. Set As'=As×Kd, where As is the original echo amplitude and As' is the echo amplitude after digital gain compensation, and obtain the timing data of the defect echo signal after digital gain compensation. Step U13a: The timing data of the defect echo signal after digital gain compensation is re-inputted into step S4. Step S4 performs a first-level automatic remeasurement of the defect measurement parameters based on the timing data of the defect echo signal after digital gain compensation, and records the number of first-level automatic remeasurements as the first-level accuracy compensation number BCy. When the measurement confidence level is level three, perform level two accuracy compensation: Step U11b: Calculate the analog gain compensation coefficient Ka based on the deviation between the dynamic range DR and the reference dynamic range DR0. Set Ka = DR0 / DR to obtain the analog gain compensation coefficient. Step U12b: Perform analog gain compensation on the receiving link of the phased array ultrasonic testing according to the analog gain compensation coefficient Ka, and set Ga'=Ga×Ka, where Ga is the original analog gain and Ga' is the adjusted analog gain, to obtain the defect echo signal after analog gain adjustment. Step U13b: Calculate the pulse compression coefficient Kp based on the deviation between the echo pulse width τ and the reference pulse width τ0. Set Kp=τ0 / τ to obtain the pulse compression coefficient. Step U14b: Perform pulse compression processing on the defect echo signal after analog gain adjustment according to the pulse compression coefficient Kp, and set τ'=τ / Kp, where τ' is the pulse width after compression, to obtain the timing data of the defect echo signal after pulse compression. Step U15b: The timing data of the defect echo signal after pulse compression is re-input into step S4. Step S4 performs secondary automatic re-measurement of the defect measurement parameters based on the timing data of the defect echo signal after pulse compression, and records the number of secondary automatic re-measurements as the number of secondary accuracy compensations BCa.

[0047] Specifically, the reference signal-to-noise ratio (SNR) SNR0 refers to the reference SNR value that the industrial non-destructive testing terminal should achieve under design conditions. In this embodiment, SNR0 is set to 20dB. The digital gain compensation refers to amplifying the amplitude of the acquired signal through digital signal processing algorithms without changing the hardware state, and has the advantage of fast response. The reference dynamic range (DR0) refers to the reference dynamic range value that the industrial non-destructive testing terminal should achieve under design conditions. In this embodiment, DR0 is set to 60dB. The analog gain compensation refers to changing the signal acquisition state by adjusting the gain of the preamplifier in the receiving link. It requires hardware response but has a greater compensation depth. The reference pulse width (τ0) refers to the reference pulse width value that the industrial non-destructive testing terminal should achieve at the designed resolution. In this embodiment, τ0 is set to 0.5μs. The pulse compression refers to the method of compressing the pulse duration and improving the axial resolution through signal processing techniques such as matched filtering. In this embodiment, linear frequency modulation pulse compression technology is used, and the resolution is improved through a compression ratio of Kp=τ0 / τ.

[0048] Specifically, step S6, which optimizes the determination process of measurement confidence level based on the number of accuracy compensations, includes: Step U21: Compare the first-level precision compensation count BCY with the preset first-level compensation threshold BCY0, and compare the second-level precision compensation count BCE with the preset second-level compensation threshold BCE0. Based on the comparison results, decide on the optimization strategy for adjusting the parameters, wherein: When BCy≤BCy0 and BCE=0, the optimization strategy is to maintain the current parameters and not optimize the process of determining the confidence level. When BCy>BCy0 and BCE=0, the optimization strategy is determined to be Level 1 optimization, which optimizes the primary optimization adjustment parameters of the determination process of measurement confidence level. When BCa > 0, the optimization strategy is determined to be a secondary optimization, which optimizes the high-level optimization adjustment parameters of the determination process of measurement confidence level.

[0049] Specifically, the preset first-level compensation threshold BCy0 refers to the threshold for determining that the primary optimization adjustment parameters need to be optimized. In this embodiment, the preset first-level compensation threshold BCy0 is set to 3 times. The preset second-level compensation threshold BCa0 refers to the threshold for determining that the advanced optimization adjustment parameters need to be optimized. In this embodiment, the preset second-level compensation threshold BCa0 is set to 1 time. Maintaining the current parameters means keeping the current set values ​​of each parameter unchanged. The primary optimization adjustment parameters refer to the weight configuration parameters of each dimension in the measurement confidence assessment system. The advanced optimization adjustment parameters refer to the measurement confidence judgment benchmark and compensation benchmark parameters.

[0050] Specifically, step S6, which optimizes the determination process of measurement confidence level based on the number of accuracy compensations, further includes: Step U22: When the optimization strategy is Level 1 optimization, the primary optimization adjustment parameters are optimized. These primary optimization adjustment parameters include the signal quality weighting coefficient ωs, the geometric measurement weighting coefficient ωg, and the system stability weighting coefficient ωc, where: The deviation of the weight coefficients of each dimension is calculated based on the number of first-level precision compensations BCy. The signal quality deviation δs=|Ns-N0| / N0, the geometric measurement deviation δg=|Ng-N0| / N0, and the system stability deviation δc=|Nc-N0| / N0 are set, where Ns, Ng, and Nc are the number of times the first-level precision compensation is triggered in the signal quality dimension, geometric measurement dimension, and system stability dimension, respectively, and N0 is the baseline value of the expected number of compensations for each dimension. The dominant imbalance dimension ARG is identified based on the principle of maximum deviation. The dominant imbalance dimension ARG is set as argmax(δs,δg,δc) to obtain the dominant imbalance dimension ARG. The weight coefficients are rebalanced and optimized based on the dominant imbalance dimension ARG. When the dominant imbalance dimension ARG is the signal quality dimension, ωs'=ωs×(1-ρ×δs), ωg'=ωg+ρ×δs×ωg / (ωg+ωc), ωc'=ωc+ρ×δs×ωc / (ωg+ωc), where ρ is the weight adjustment rate coefficient. The rebalanced weight coefficient set {ωs',ωg',ωc'} is obtained, and ωs'+ωg'+ωc'=1 is satisfied. When the dominant imbalance dimension is the geometric measurement dimension, we set ωg'=ωg×(1-ρ×δg), ωs'=ωs+ρ×δg×ωs / (ωs+ωc), ωc'=ωc+ρ×δg×ωc / (ωs+ωc), where ρ is the weight adjustment rate coefficient, and obtain the rebalanced weight coefficient set {ωs',ωg',ωc'}, which satisfies ωs'+ωg'+ωc'=1; When the dominant imbalance dimension is the system stability dimension, we set ωc'=ωc×(1-ρ×δc), ωs'=ωs+ρ×δc×ωs / (ωs+ωg), ωg'=ωg+ρ×δc×ωg / (ωs+ωg), where ρ is the weight adjustment rate coefficient, and obtain the rebalanced weight coefficient set {ωs',ωg',ωc'}, which satisfies ωs'+ωg'+ωc'=1.

[0051] Specifically, the expected compensation frequency benchmark value N0 for each dimension refers to the expected frequency of compensation triggered by each dimension under ideal detection conditions. In this embodiment, N0 is set to 5% of the total number of detections based on historical statistical data. The deviation refers to the relative deviation between the actual compensation frequency and the expected frequency, reflecting the degree of dominant imbalance of the dimension in the measurement quality assessment. The dominant imbalance dimension is the dimension with the largest deviation, indicating that the current weight configuration of the dimension is too high or too low, resulting in assessment imbalance. The weight adjustment rate coefficient ρ controls the convergence speed of weight adjustment. In this embodiment, ρ is set to 0.3 to ensure that the weight adjustment is both responsive and avoids oscillation. The rebalancing optimization refers to reducing the weight coefficient of the dominant imbalance dimension and distributing the reduction proportionally to other dimensions to achieve the reconstruction and optimization of the assessment system.

[0052] Specifically, step S6, which optimizes the determination process of measurement confidence level based on the number of accuracy compensations, further includes: Step U23: When the optimization strategy is secondary optimization, the advanced optimization adjustment parameters are optimized. These advanced optimization adjustment parameters include a preset measurement confidence index C0, a secondary preset measurement confidence index C1, a reference signal-to-noise ratio SNR0, a reference dynamic range DR0, and a reference pulse width τ0, wherein: The pass rate ηc of the measurement confidence index C is calculated based on the number of secondary accuracy compensations. ηc is set as ηc = Npass / (Npass + Nfail), where Npass is the number of times C ≥ C0 and Nfail is the number of times C < C0. The measurement confidence pass rate ηc is obtained. The preset measurement confidence index C0 is optimized based on the measurement confidence achievement rate ηc. C0' is set as C0' = C0 × [1 + σ × (ηc - ηc0)], where σ is the confidence adjustment coefficient and ηc0 is the target achievement rate, thus obtaining the optimized preset measurement confidence index C0'. Based on the optimized preset measurement confidence index C0', the secondary preset measurement confidence index C1 is optimized in conjunction with it. C1' is set to C1×(C0' / C0) to obtain the optimized secondary preset measurement confidence index C1'. Based on the actual resolution improvement effect of the timing data of the defect echo signal after pulse compression, the reference pulse width τ0 is iteratively optimized to obtain the optimized reference pulse width τ0'. τ0' is set as τ0×[1-ξ×(Kp-1) / Kp], where ξ is the resolution convergence coefficient.

[0053] Specifically, the target compliance rate η0 refers to the expected proportion of measurement confidence levels that meet the standards. In this embodiment, η0 is set to 0.90, meaning that 90% of the detections are expected to reach Level 1 confidence. Simultaneously, when ηc > ηc0, C0 is increased to strengthen quality requirements; when η < η0, C0 is decreased to ensure detection efficiency. The confidence adjustment coefficient σ controls the adjustment range of C0. In this embodiment, σ is set to 0.1 to ensure that the adjustment of C0 reflects changes in the compliance rate while remaining relatively stable. The constant relative proportion means that C1' / C0' = C1 / C0. To ensure that the division ratio of the three confidence levels remains unchanged, thereby maintaining a constant relative ratio between C0' and C1', the resolution convergence coefficient ξ controls the rate at which τ0 approaches the limit resolution of the industrial non-destructive testing terminal. When the pulse compression coefficient Kp is continuously greater than 1, τ0 is gradually reduced to approach the resolution limit of the industrial non-destructive testing terminal. In this embodiment, ξ is set to 0.2. The resolution limit of the industrial non-destructive testing terminal refers to the minimum pulse width that can be achieved under current hardware conditions through pulse compression technology, which is jointly determined by the probe bandwidth and the signal processing algorithm.

[0054] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A method for automatic identification and measurement of defect images based on phased array detection, characterized in that, The method includes: Step S1: Acquire the raw signal of the phased array ultrasonic detection to obtain the raw phased array data, and generate a multimodal detection image based on the raw phased array data; Step S2: Obtain a standard multimodal image based on the multimodal detection image, and generate a fused feature image based on the standard multimodal image; Step S3: Mark the candidate defect regions according to the fused feature image, generate a preliminary defect list according to the candidate defect regions, determine the true nature of the defects according to the preliminary defect list, and obtain the true defect list according to the determination result. Step S4: Extract the geometric features of the defects based on the actual defect list, automatically measure the defect measurement parameters based on the defect geometric features, reconstruct the three-dimensional contour of the defects based on the defect measurement parameters, and calculate the defect volume based on the three-dimensional contour of the defects. Step S5: Automatically identify the defect type based on the three-dimensional contour of the defect and the defect measurement parameters, assess the defect hazard level based on the defect type, and generate a defect handling decision based on the assessment results; Step S6: Determine the measurement confidence level based on the defect measurement parameters, perform accuracy compensation for step S4 based on the judgment result, and optimize the measurement confidence level judgment process based on the number of accuracy compensations.

2. The method for automatic identification and measurement of defect images based on phased array detection according to claim 1, characterized in that, Step S2, which involves acquiring a standard multimodal image based on the multimodal detection image, includes: Step P01: Perform image quality assessment on the multimodal detection image to obtain an image quality score T. Compare the image quality score T with a preset image quality score T0. Based on the comparison result, determine whether the quality of the multimodal detection image meets the standard. When T≥T0, the quality is deemed to be up to standard, and the multimodal detection image is used as the multimodal image to be processed. When T < T0, the quality is judged to be substandard. Image enhancement is performed on the multimodal detection image to obtain the enhanced multimodal detection image, and the enhanced multimodal detection image is used as the multimodal image to be processed. Step P02: Perform geometric correction on the multimodal image to be processed to obtain a geometrically corrected multimodal image, and then normalize the grayscale of the geometrically corrected multimodal image to obtain a grayscale-normalized multimodal image. Step P03: Output the grayscale-normalized multimodal image as a standard multimodal image.

3. The method for automatic identification and measurement of defect images based on phased array detection according to claim 2, characterized in that, When generating the fused feature image based on the standard multimodal image in step S2, it includes: Step P11: Input the A-scan image, B-scan image, C-scan image and S-scan image from the standard multimodal image into the multimodal feature extraction network to obtain the A-scan feature map, B-scan feature map, C-scan feature map and S-scan feature map respectively; Step P12: Calculate the fused feature image F based on the feature fusion weight coefficient set ω={ωA,ωB,ωC,ωS}, setting F=ωA×FA+ωB×FB+ωC×FC+ωS×FS, where FA is the A-scan feature map, FB is the B-scan feature map, FC is the C-scan feature map, and FS is the S-scan feature map, thus obtaining the fused feature image.

4. The method for automatic identification and measurement of defect images based on phased array detection according to claim 1, characterized in that, Step S3, which involves marking candidate defect regions based on the fused feature image and generating a preliminary defect list based on the candidate defect regions, includes: Step C01: Input the fused feature image into the defect candidate region detection network to obtain the candidate region confidence Z and candidate region location coordinates output by the defect candidate region detection network; Step C02: Compare the confidence scores of each candidate region in the candidate region confidence score set Z with the preset candidate region confidence scores Z, and mark the candidate defect regions according to the comparison results, wherein: When Z≥Z0, the location coordinates of the candidate region are marked as the candidate defect region, and the candidate defect region is added to the initial screening defect list; When Z < Z0, the location coordinates of the candidate region are not marked as a candidate defect region.

5. The method for automatic identification and measurement of defect images based on phased array detection according to claim 4, characterized in that, The step S3, which involves determining the true nature of defects based on the initial defect list, includes: Step V01: Extract multi-dimensional features from the candidate defect regions in the initial defect list to obtain a multi-dimensional feature vector of the defect. The multi-dimensional feature vector of the defect includes gray-scale statistical features, texture features, edge gradient features and frequency domain features. Step V02: Input the multi-dimensional feature vector of the defect into the defect authenticity judgment model to obtain the defect authenticity probability Preal output by the defect authenticity judgment model; Step V03: Compare the actual defect probability Preal with the preset actual defect probability Preal0. Based on the comparison result, determine the true nature of the defect, and obtain the list of actual defects based on the determination result. When Preal≥Preal0, the defect is determined to be a real defect and the candidate defect region is added to the list of real defects. When Preal < Preal0, the defect is determined to be a pseudo-defect, and the candidate defect area is not added to the list of real defects.

6. The method for automatic identification and measurement of defect images based on phased array detection according to claim 1, characterized in that, Step S4, which involves extracting the geometric features of defects based on the actual defect list, includes: Step Q01: Extract the timing data of the defect echo signal from the original phased array data based on the location coordinates of the candidate regions in the actual defect list. Step Q02: Calculate the defect depth D, defect length L, and defect height H based on the defect echo signal timing data; Step Q03: Output the defect depth D, defect length L, and defect height H as the defect geometric features; Step S4, which involves automatically measuring the defect measurement parameters based on the defect's geometric characteristics, includes: Step Q11: Calculate the equivalent diameter Deq of the defect based on the defect depth D, defect length L, and defect height H, and set Deq = (D × L × H). (1 / 3） ; Step Q12: Calculate the defect elongation rate E based on the defect depth D and defect length L, and set E=L / D; Step Q13: Calculate the defect height-to-length ratio R based on the defect height H and defect length L, and set R = H / L; Step Q14: Output the defect equivalent diameter Deq, defect elongation E, and defect height-to-length ratio R as defect measurement parameters; Step S4, reconstructing the three-dimensional contour of the defect based on the defect measurement parameters, includes: Step R01: Establish a three-dimensional spatial coordinate system for the defect based on the defect depth D, defect length L, and defect height H. Step R02: Calculate the three-dimensional morphology coefficient α of the defect based on the equivalent diameter Deq and the height-to-length ratio R, setting α = 1 / (1+e^(-qq)). (-k×(R-R0) ); Step R03: Construct the three-dimensional contour surface equation Z of the defect based on the three-dimensional morphology coefficient α of the defect, and set Z = α × Deq / 2 × cos(π × X / L) × cos(π × Y / H); Step R04: Determine the defect volume based on the three-dimensional contour of the defect. Perform calculations and set , The integral region is defined as the range of the defect boundary.

7. The method for automatic identification and measurement of defect images based on phased array detection according to claim 1, characterized in that, Step S5, which automatically identifies the defect type based on the defect's three-dimensional contour and measurement parameters, includes: Step W01: Input the three-dimensional contour of the defect, the equivalent diameter of the defect Deq, the elongation rate of the defect E, the height-to-length ratio of the defect R, and the volume of the defect V into the defect type identification model to obtain the set of defect type confidence scores output by the defect type identification model. Step W02: Select the maximum value from the set of confidence scores for defect types to obtain the maximum confidence score for a defect type Tmax and the defect type corresponding to the maximum confidence score for a defect type Tmax. Step W03: Compare the maximum defect type confidence score Tmax with the preset defect type confidence score Tmax0, and confirm the defect type identification result based on the comparison result, wherein: When Tmax≥Tmax0, the defect type corresponding to the maximum defect type confidence Tmax is confirmed as the defect type identification result; When Tmax < Tmax0, the defect type identification result is marked as an unidentified type, and the manual review process is triggered.

8. The method for automatic identification and measurement of defect images based on phased array detection according to claim 7, characterized in that, Step S5 involves assessing the hazard level of a defect based on its type and generating a defect handling decision based on the assessment results, including: Step J01: Determine the defect baseline hazard factor Gbase according to the defect type; Step J02: Calculate the size correction factor β based on the defect equivalent diameter Deq, and set β = 1 + ln(Deq / Deq0), where Deq0 is the reference defect equivalent diameter and ln is the natural logarithm. Step J03: Calculate the position correction coefficient γ based on the defect depth D, and set γ = 1 + η × (D / Dmax), where η is the position sensitivity coefficient and Dmax is the material thickness; Step J04: Calculate the comprehensive hazard index G based on the defect baseline hazard factor Gbase, size correction factor β, and position correction factor γ, and set G = Gbase × β × γ; Step J05: Compare the comprehensive hazard index G with the preset hazard level threshold set {G1, G2, G3}, assess the defect hazard level based on the comparison results, and generate a defect handling decision based on the assessment results, wherein: When G≤G1, the defect hazard level is assessed as Level I, the defect handling decision is generated and archived, and it is included in the scope of routine inspection and monitoring. No special handling is required during this inspection cycle. When G1 < G ≤ G2, the defect hazard level is assessed as Level II, and the defect handling decision is to shorten the detection cycle to 50% of the original cycle, increase the monitoring frequency, assess the defect development trend, and formulate a preventive maintenance plan. When G2 < G ≤ G3, the defect hazard level is assessed as Level III, and the defect handling decision is to immediately initiate a special testing procedure, use other non-destructive testing methods for verification and confirmation, determine the repair or replacement plan based on the verification results, and limit the equipment operating load. When G > G3, the defect hazard level is assessed as Level IV, and the defect handling decision is to immediately stop the equipment operation, initiate the emergency response procedure, isolate the defect area, organize expert assessment, formulate an emergency repair or replacement plan, and not resume operation without handling.

9. The method for automatic identification and measurement of defect images based on phased array detection according to claim 1, characterized in that, Step S6, which determines the measurement confidence level based on the defect measurement parameters, includes: Step U01: Extract signal quality dimension parameters based on the time series data of the defect echo signal. The signal quality dimension parameters include signal-to-noise ratio (SNR), dynamic range (DR), and echo pulse width (τ). Step U02: Extract geometric measurement dimension parameters based on the defect geometric features. The geometric measurement dimension parameters include depth measurement resolution Rd, length measurement resolution Rl, and height measurement resolution Rh. Step U03: Extract system stability dimension parameters based on the original phased array data. The system stability dimension parameters include gain stability coefficient Gs, time baseline stability coefficient Ts, and probe coupling stability coefficient Cs. Step U04: Calculate the measurement confidence index C based on the signal quality dimension parameters, geometric measurement dimension parameters, and system stability dimension parameters. Set C = ωs × f(SNR,DR,τ) + ωg × f(Rd,Rl,Rh) + ωc × f(Gs,Ts,Cs), where ωs is the signal quality weight coefficient, ωg is the geometric measurement weight coefficient, and ωc is the system stability weight coefficient, satisfying ωs + ωg + ωc = 1. f() is the normalized scoring function, thus obtaining the measurement confidence index C. Step U05: Compare the measurement confidence index C with the preset measurement confidence index C0 and the secondary preset measurement confidence index C1, and determine the measurement confidence level based on the comparison results, wherein: Set C1 < C0; When C≥C0, the measurement confidence level is determined to be Level 1 confidence, and no accuracy compensation is performed in step S4. When C1≤C<C0, the measurement confidence level is determined to be level 2 confidence, and level 1 accuracy compensation is performed on step S4. When C < C1, the measurement confidence level is determined to be level three, and level two accuracy compensation is performed on step S4. Step S6, which performs accuracy compensation for step S4 based on the judgment result, includes: When the measurement confidence level is level 2, level 1 accuracy compensation is performed: Step U11a: Calculate the digital gain compensation coefficient Kd based on the deviation between the signal-to-noise ratio SNR and the reference signal-to-noise ratio SNR0. Set Kd=SNR0 / SNR to obtain the digital gain compensation coefficient. Step U12a: Perform digital gain compensation on the timing data of the defect echo signal according to the digital gain compensation coefficient Kd. Set As'=As×Kd, where As is the original echo amplitude and As' is the echo amplitude after digital gain compensation, and obtain the timing data of the defect echo signal after digital gain compensation. Step U13a: The timing data of the defect echo signal after digital gain compensation is re-inputted into step S4. Step S4 performs a first-level automatic remeasurement of the defect measurement parameters based on the timing data of the defect echo signal after digital gain compensation, and records the number of first-level automatic remeasurements as the first-level accuracy compensation number BCy. When the measurement confidence level is level three, perform level two accuracy compensation: Step U11b: Calculate the analog gain compensation coefficient Ka based on the deviation between the dynamic range DR and the reference dynamic range DR0. Set Ka = DR0 / DR to obtain the analog gain compensation coefficient. Step U12b: Perform analog gain compensation on the receiving link of the phased array ultrasonic testing according to the analog gain compensation coefficient Ka, and set Ga'=Ga×Ka, where Ga is the original analog gain and Ga' is the adjusted analog gain, to obtain the defect echo signal after analog gain adjustment. Step U13b: Calculate the pulse compression coefficient Kp based on the deviation between the echo pulse width τ and the reference pulse width τ0. Set Kp=τ0 / τ to obtain the pulse compression coefficient. Step U14b: Perform pulse compression processing on the defect echo signal after analog gain adjustment according to the pulse compression coefficient Kp, and set τ'=τ / Kp, where τ' is the pulse width after compression, to obtain the timing data of the defect echo signal after pulse compression. Step U15b: The timing data of the defect echo signal after pulse compression is re-input into step S4. Step S4 performs secondary automatic re-measurement of the defect measurement parameters based on the timing data of the defect echo signal after pulse compression, and records the number of secondary automatic re-measurements as the number of secondary accuracy compensations BCa.

10. The method for automatic identification and measurement of defect images based on phased array detection according to claim 1, characterized in that, The optimization of the measurement confidence level determination process based on the number of accuracy compensations in step S6 includes: Step U21: Compare the first-level precision compensation count BCY with the preset first-level compensation threshold BCY0, and compare the second-level precision compensation count BCE with the preset second-level compensation threshold BCE0. Based on the comparison results, decide on the optimization strategy for adjusting the parameters, wherein: When BCy≤BCy0 and BCE=0, the optimization strategy is to maintain the current parameters and not optimize the process of determining the confidence level. When BCy>BCy0 and BCE=0, the optimization strategy is determined to be Level 1 optimization, which optimizes the primary optimization adjustment parameters of the determination process of measurement confidence level. When BCa > 0, the optimization strategy is determined to be a secondary optimization, which optimizes the high-level optimization adjustment parameters of the determination process of measurement confidence level. The optimization of the measurement confidence level determination process based on the number of accuracy compensations in step S6 also includes: Step U22: When the optimization strategy is Level 1 optimization, the primary optimization adjustment parameters are optimized. These primary optimization adjustment parameters include the signal quality weighting coefficient ωs, the geometric measurement weighting coefficient ωg, and the system stability weighting coefficient ωc, where: The deviation of the weight coefficients of each dimension is calculated based on the number of first-level precision compensations BCy. The signal quality deviation δs=|Ns-N0| / N0, the geometric measurement deviation δg=|Ng-N0| / N0, and the system stability deviation δc=|Nc-N0| / N0 are set, where Ns, Ng, and Nc are the number of times the first-level precision compensation is triggered in the signal quality dimension, geometric measurement dimension, and system stability dimension, respectively, and N0 is the baseline value of the expected number of compensations for each dimension. The dominant imbalance dimension ARG is identified based on the principle of maximum deviation. The dominant imbalance dimension ARG is set as argmax(δs,δg,δc) to obtain the dominant imbalance dimension ARG. The weight coefficients are rebalanced and optimized based on the dominant imbalance dimension ARG. When the dominant imbalance dimension ARG is the signal quality dimension, ωs'=ωs×(1-ρ×δs), ωg'=ωg+ρ×δs×ωg / (ωg+ωc), ωc'=ωc+ρ×δs×ωc / (ωg+ωc), where ρ is the weight adjustment rate coefficient. The rebalanced weight coefficient set {ωs',ωg',ωc'} is obtained, and ωs'+ωg'+ωc'=1 is satisfied. When the dominant imbalance dimension is the geometric measurement dimension, we set ωg'=ωg×(1-ρ×δg), ωs'=ωs+ρ×δg×ωs / (ωs+ωc), ωc'=ωc+ρ×δg×ωc / (ωs+ωc), where ρ is the weight adjustment rate coefficient, and obtain the rebalanced weight coefficient set {ωs',ωg',ωc'}, which satisfies ωs'+ωg'+ωc'=1; When the dominant imbalance dimension is the system stability dimension, we set ωc'=ωc×(1-ρ×δc), ωs'=ωs+ρ×δc×ωs / (ωs+ωg), ωg'=ωg+ρ×δc×ωg / (ωs+ωg), where ρ is the weight adjustment rate coefficient, and obtain the set of weight coefficients after rebalancing {ωs',ωg',ωc'}, which satisfies ωs'+ωg'+ωc'=1; The optimization of the measurement confidence level determination process based on the number of accuracy compensations in step S6 also includes: Step U23: When the optimization strategy is secondary optimization, the advanced optimization adjustment parameters are optimized. These advanced optimization adjustment parameters include a preset measurement confidence index C0, a secondary preset measurement confidence index C1, a reference signal-to-noise ratio SNR0, a reference dynamic range DR0, and a reference pulse width τ0, wherein: The pass rate ηc of the measurement confidence index C is calculated based on the number of secondary accuracy compensations. ηc is set as ηc = Npass / (Npass + Nfail), where Npass is the number of times C ≥ C0 and Nfail is the number of times C < C0. The measurement confidence pass rate ηc is obtained. The preset measurement confidence index C0 is optimized based on the measurement confidence achievement rate ηc. C0' is set as C0' = C0 × [1 + σ × (ηc - ηc0)], where σ is the confidence adjustment coefficient and ηc0 is the target achievement rate, thus obtaining the optimized preset measurement confidence index C0'. Based on the optimized preset measurement confidence index C0', the secondary preset measurement confidence index C1 is optimized in conjunction with it. C1' is set to C1×(C0' / C0) to obtain the optimized secondary preset measurement confidence index C1'. Based on the actual resolution improvement effect of the timing data of the defect echo signal after pulse compression, the reference pulse width τ0 is iteratively optimized to obtain the optimized reference pulse width τ0'. τ0' is set as τ0×[1-ξ×(Kp-1) / Kp], where ξ is the resolution convergence coefficient.

Images