A method for detecting a weld seam of a runner blade

Abstract

The present application relates to the technical field of nondestructive testing, in particular to a kind of rotating blade weld detection method;Contain dynamic scanning, trajectory fitting, projection transformation and defect evaluation module;System is excited by double frequency by flexible eddy current array probe, and impedance response is collected;Its core is to use high-frequency component fitting to lift trajectory, calculate phase rotation matrix, and carry out orthogonal projection transformation to low-frequency component to extract pure defect characteristics;When signal amplitude exceeds preset threshold, maintenance strategy is generated;The present application constructs digital filter for lift-off effect, overcomes the hysteresis of hardware compensation, realizes the accurate extraction of weak deep defect under strong noise interference.

Description

Technical Field

[0001] This invention relates to the field of nondestructive testing technology, specifically to a method for inspecting the weld seams of turbine blades. Background Technology

[0002] During the manufacturing and maintenance of turbine runner blades, the weld area typically has complex hyperbolic geometry, requiring the use of eddy current probes for close-fitting scanning to obtain information about its internal structure. Existing inspection methods generally rely on passive mechanical following or hardware circuit compensation to suppress lift-off noise. However, when performing dynamic inspections on such complex irregular curved surfaces, the mechanical hysteresis and probe stiffness limit the probe's ability to maintain a constant, minute distance from the surface being measured. Such gap changes caused by surface undulations or scanning jitter will generate strong lift-off interference signals, and these interference signals often overlap with weak deep defect signals in the vector space. Traditional signal processing methods are difficult to achieve effective separation of defect signals under the influence of strong background noise and edge effects, which can easily lead to missed detection of deep buried defects or false alarms of non-defect areas. Furthermore, they lack the ability to perform real-time quantitative inversion and automatic classification decision-making for defect depth. Therefore, how to achieve accurate decoupling and intelligent hierarchical evaluation of interference and defect signals in complex curved surface dynamic scanning environments has become an urgent technical problem to be solved. Summary of the Invention

[0003] To solve the above-mentioned technical problems, the present invention provides a method for inspecting the weld seam of a turbine blade. Specifically, the technical solution of the present invention includes: Step 1: Control the flexible eddy current array probe to perform dynamic scanning in close contact with the weld surface of the impeller blade, and apply a mixed excitation current containing a first frequency and a second frequency to the flexible eddy current array probe, collect the complex impedance response signal of each array unit, and construct the original impedance dataset. Step 2: Establish an impedance plane coordinate system, extract the second impedance component based on the second frequency from the original impedance dataset, and use the second impedance component to fit and generate the current lift-off effect trajectory in the impedance plane coordinate system; Step 3: Extract the first impedance component based on the first frequency from the original impedance dataset, calculate the rotation angle of the lift-off effect trajectory relative to the real axis of the impedance plane coordinate system, and construct a phase rotation matrix; use the phase rotation matrix to perform an orthogonal projection transformation on the first impedance component, and extract the imaginary part signal orthogonal to the lift-off effect trajectory as a pure defect feature signal; Step 4: Preset a defect judgment threshold set, compare the amplitude of the pure defect feature signal with the defect judgment threshold set to obtain the weld defect evaluation result, and generate a corresponding maintenance strategy signal based on the weld defect evaluation result.

[0004] Preferably, step one includes: S11. A flexible eddy current array probe based on a flexible PCB substrate is used to cover the weld seam area of ​​the impeller blade with a double curvature feature. S12. Configure the frequency parameters of the hybrid excitation current, wherein the first frequency is set to a low frequency band to penetrate the weld metal medium to obtain deep structural information; the second frequency is set to a high frequency band to sense the surface state of the weld to obtain lift-off interference information, and the value of the first frequency is specified to be less than the value of the second frequency. S13. The induced voltage of the flexible eddy current array probe is synchronously demodulated using the FPGA multi-frequency demodulation module to obtain the first complex impedance corresponding to the first frequency and the second complex impedance corresponding to the second frequency, and the data of all array units are summarized to construct the original impedance dataset.

[0005] Preferably, step two includes: S21. Map the second complex impedance to the impedance plane coordinate system to obtain its real part coordinates and imaginary part coordinates; S22. Due to the high-frequency characteristics of the second frequency, the second complex impedance is not sensitive to deep defects but only sensitive to the gap change between the probe and the measured surface. Data points of the second complex impedance within a continuous time window are collected. S23. The least squares method is used to perform curve fitting on the data points to generate a curve that reflects the surface undulation state of the current scanning area, and the curve is defined as the lift-off effect trajectory.

[0006] Preferably, step three includes: S31. Calculate the tangent slope of the lift-off effect trajectory at the current operating point, and determine the rotation angle required to rotate the lift-off effect trajectory to be parallel to the real axis of the impedance plane coordinate system based on the tangent slope. S32. Construct the phase rotation matrix based on the rotation angle; S33. Treat the first complex impedance as a vector, and multiply it by the phase rotation matrix on the left to obtain the new impedance vector after rotation; S34. Extract the component of the new impedance vector in the imaginary axis direction. Since the lift-off effect trajectory has been rotated to the real axis direction, define the component in the imaginary axis direction as the decoupled pure defect feature signal.

[0007] Preferably, step four includes: S41. Extract the amplitude of the pure defect feature signal; S42. The defect determination threshold set is preset, including a first safety threshold and a second critical threshold, and the first safety threshold is defined to be less than the second critical threshold; S43. Perform interval judgment on the amplitude to obtain the weld defect assessment result: If the amplitude is less than or equal to the first safety threshold, the current detection area is determined to be in a defect-free state, and the weld defect assessment result is labeled as qualified. If the amplitude is greater than the first safety threshold and less than or equal to the second critical threshold, it is determined that there is a slight shallow disturbance in the current detection area, and the weld defect assessment result is labeled as a warning. If the amplitude is greater than the second critical threshold, it is determined that there is a buried defect with a depth greater than or equal to 1 mm in the current detection area, and the weld defect assessment result is labeled as unqualified.

[0008] Preferably, step four further includes: S44. Establish an amplitude-depth mapping database, which stores the correspondence between the amplitude of the pure defect feature signal measured on the standard test block and the known defect depth. S45. In response to the weld defect assessment result being labeled as unqualified, the amplitude is input into the amplitude-depth mapping database for linear interpolation retrieval, and the quantified defect depth value of the current detection point is output. S46. Associate the quantized defect depth value with the physical position coordinates of the current array unit to generate a three-dimensional distribution map of internal defects in the weld.

[0009] Preferably, step four further includes: S47. When the edge region of the runner blade is detected, the lift-off effect trajectory is used as the reference signal of the edge effect to perform edge compensation correction on the pure defect feature signal to generate the corrected edge defect feature signal. S48. Substitute the corrected edge defect feature signal into step S43 for re-determination to eliminate false alarms caused by edge effects.

[0010] Preferably, step four further includes: S49. Generate the maintenance strategy signal based on the weld defect assessment results: In response to the tag being deemed valid, a command to continue scanning is generated, and the current path is recorded as a safe path; In response to the label being a warning, a marking instruction is generated, the warning area is highlighted in yellow in a preset digital model, and it is recommended to re-check it in the next maintenance cycle; In response to the label being deemed unqualified, a shutdown review instruction and a polishing repair suggestion are generated. The area is highlighted in red in the preset digital model, and the quantified defect depth value is output to guide the polishing depth.

[0011] Compared with the prior art, the present invention has the following beneficial effects: 1. This method utilizes the physical response difference of dual-frequency eddy current signals in the frequency domain, combined with phase rotation and orthogonal projection techniques in vector space, to construct a digital filter targeting the lift-off effect. By fitting the lift-off trajectory with high-frequency components and performing coordinate rotation, the lift-off interference is forcibly aligned to the real axis, thereby separating the pure defect signal in the imaginary axis direction. This mechanism overcomes the mechanical lag of traditional hardware compensation, enabling the extraction of weak deep defect signals from strong background noise during dynamic scanning when the flexible probe cannot perfectly fit the surface of the hyperbolic blade, thus ensuring the accuracy of the detection results. 2. This method establishes a mapping database between amplitude and depth, upgrading a single alarm signal into quantified depth data. Through linear interpolation retrieval technology, it achieves accurate inversion of the depth of invisible buried defects. This non-destructive measurement capability avoids traditional destructive slicing inspection, significantly reducing the time and economic cost of quality assessment. At the same time, the system integrates the quantified depth value with the physical location coordinates to generate a three-dimensional distribution map of defects inside the weld, intuitively displaying the spatial morphology of the defects and providing accurate digital guidance for subsequent grinding and repair work. 3. This method utilizes the geometric similarity between edge effect and lift-off effect on the impedance plane response trajectory to propose an edge compensation correction method based on the lift-off trajectory benchmark. By monitoring the impedance change gradient, edge regions are automatically identified, and correlation analysis is used to correct the edge signal, eliminating signal distortion caused by abrupt geometric changes at the edge. This strategy effectively solves the industry problem of inaccurate evaluation of blade edges in traditional eddy current detection, ensuring the integrity of full coverage detection of runner blades and greatly reducing the risk of false positives in edge regions. 4. This method, by setting dual thresholds, subdivides the detection results into three levels: qualified, warning, and unqualified, thus achieving quantitative grading of defect severity. The system automatically generates differentiated maintenance strategy signals based on the evaluation results, such as continuing scanning, highlighting digital models, or triggering shutdown review instructions. This closed-loop control logic not only avoids frequent shutdowns caused by oversensitivity but also ensures that major structural hazards are dealt with promptly and accurately, achieving a leap from simple non-destructive testing to intelligent operation and maintenance decision-making. Attached Figure Description

[0012] The present invention will be further explained below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0014] Example 1 Please see Figure 1 A method for inspecting the weld seam of a turbine blade, comprising the following steps: Step 1: Control the flexible eddy current array probe to perform dynamic scanning in close contact with the surface of the rotor blade weld, and apply a mixed excitation current containing the first frequency and the second frequency to the flexible eddy current array probe, collect the complex impedance response signal of each array unit, and construct the original impedance dataset. Step 2: Establish an impedance plane coordinate system, extract the second impedance component based on the second frequency from the original impedance dataset, and use the second impedance component to fit and generate the current lift-off effect trajectory in the impedance plane coordinate system. Step 3: Extract the first impedance component based on the first frequency from the original impedance dataset, calculate the rotation angle of the lift-off effect trajectory relative to the real axis of the impedance plane coordinate system, and construct the phase rotation matrix; use the phase rotation matrix to perform orthogonal projection transformation on the first impedance component, and extract the imaginary part signal orthogonal to the lift-off effect trajectory as the pure defect feature signal. Step 4: Preset a defect judgment threshold set, compare the amplitude of the pure defect feature signal with the defect judgment threshold set to obtain the weld defect evaluation result, and generate the corresponding maintenance strategy signal based on the weld defect evaluation result.

[0015] This embodiment details the overall execution flow of the above-mentioned turbine blade weld inspection method. This flow aims to solve the physical problem of separating noise and defect signal aliasing in complex curved surface inspection. The system performs a data acquisition step by controlling the end effector of the robotic arm to drive the flexible eddy current array probe to move along a predetermined path. During this process, the signal generator continuously outputs a mixed excitation current injected into the probe coil, inducing an eddy current field in the conductive medium of the weld using the principle of electromagnetic induction. The system performs signal decoupling preprocessing, mapping the acquired complex impedance response signal to an impedance plane coordinate system, which has the resistance component as the real axis and the inductance component as the imaginary axis. The core execution step... The vector orthogonalization process utilizes the skin effect characteristic, where high-frequency components are highly sensitive to lift-off gaps but ignore deep defects, to construct a lift-off effect trajectory containing only non-defect interference. The deviation angle between this trajectory and the coordinate system's reference axis is determined by calculating its tangential vector. The system executes a phase rotation algorithm to rigidly rotate the impedance vector corresponding to the low-frequency components in mathematical space, forcing the lift-off interference components to align with the real axis direction, thereby separating the pure defect characteristic signal unaffected by lift-off along the orthogonal imaginary axis direction. Based on the comparison between the signal amplitude and a preset safety benchmark, the system automatically outputs graded maintenance strategy signals. This embodiment utilizes the physical response difference of dual-frequency eddy current signals in the frequency domain, combined with geometric transformation technology in vector space, to construct a digital filter targeting the lift-off effect at the algorithm level. This orthogonal projection method cleverly avoids the mechanical lag of traditional hardware compensation, enabling the extraction of weak deep defect signals from strong lift-off background noise even during dynamic scanning when the probe cannot perfectly fit the surface of the hyperbolic blade. This ensures high signal-to-noise ratio detection of the internal structural integrity of the turbine runner blade weld.

[0016] Example 2 Step one includes: S11. A flexible eddy current array probe based on a flexible PCB substrate is used to cover the weld seam area of ​​the impeller blade with double curvature characteristics. S12. Configure the frequency parameters of the hybrid excitation current, wherein the first frequency is set to the low frequency band to penetrate the weld metal medium to obtain deep structural information; the second frequency is set to the high frequency band to sense the surface state of the weld to obtain lift-off interference information, and the value of the first frequency is specified to be less than the value of the second frequency. S13. Use the FPGA multi-frequency demodulation module to synchronously demodulate the induced voltage of the flexible eddy current array probe, obtain the first complex impedance corresponding to the first frequency and the second complex impedance corresponding to the second frequency, and summarize the data of all array units to construct the original impedance dataset.

[0017] This embodiment further specifies the data acquisition and raw dataset construction steps in Embodiment 1. In the hardware deployment stage, a flexible probe based on polyimide thin film technology is selected. This probe has extremely low bending stiffness and can adapt to the complex hyperbolic geometry of the impeller blades to generate elastic deformation, thereby minimizing the average gap between the probe coil and the measured surface at the physical level. In the excitation parameter configuration stage, the skin depth formula for electromagnetic waves in specific metallic materials is used: , in, Standard penetration depth, unit: ; Pi Excitation frequency, unit: ; The permeability of free space is a constant. ; The relative permeability of the material; Electrical conductivity of the material, unit: ; First frequency The source is preset parameters, and the setting principle is to make... Greater than or equal to the weld thickness, physically meaning the penetration depth covering the low-frequency carrier wave at the weld root, measured in units of... ; Second frequency The source is preset parameters, and the setting principle is to make... Less than Physically, it refers to a high-frequency carrier wave with energy concentrated on the surface of a material; its unit is 1. ; The FPGA multi-frequency demodulation module performs parallel signal processing, employing digital down-conversion technology to perform spectrum shifting and filtering on the induced voltage signal within the same physical channel, simultaneously separating the low-frequency impedance component carrying deep information and the high-frequency impedance component carrying surface state information. The system structures and stores the dual-frequency impedance data of all array units in the time series, forming the original impedance dataset. It should be noted that, in this invention, for ease of subsequent digital signal processing and quantization evaluation, the complex impedance change is equivalently represented as a voltage amplitude signal after demodulation. The unit for signal amplitude and threshold determination in this paper is uniformly expressed in volts, denoted by . ; This embodiment constructs a high-fidelity data acquisition front-end through a combination of flexible physical bonding and dual-frequency signal separation. The application of a flexible PCB substrate significantly reduces signal distortion caused by the bridging effect of rigid probes, while strict frequency band division ensures that defect signals and interference signals have a physical basis for separation at the source, providing high-quality input variables for subsequent algorithm decoupling.

[0018] Example 3: Step two includes: S21. Map the second complex impedance to the impedance plane coordinate system to obtain its real part coordinates and imaginary part coordinates; S22. Due to the high-frequency characteristics of the second frequency, the second complex impedance is not sensitive to deep defects but only to the change in the gap between the probe and the measured surface. Data points of the second complex impedance are collected within a continuous time window. S23. Use the least squares method to fit the data points to generate a curve that reflects the surface undulation of the current scanned area. Define the curve as the lift-off effect trajectory.

[0019] This embodiment further specifies the lift-off effect trajectory fitting steps in Embodiment 1; the system performs coordinate mapping operations, analyzes the real and imaginary parts of the second complex impedance, and transforms it into a point set in the Cartesian coordinate system; the system uses a sliding time window to capture a continuous impedance data stream, which physically corresponds to the process of the probe scanning a section of the weld surface; since the second frequency is set in a high-frequency band with extremely shallow skin depth, the eddy current field in this band cannot reach the buried defect, so its impedance change vector is mainly controlled by the change in the distance between the coil and the conductor surface; for the data point set within this window, the system calls the least squares algorithm to solve the regression equation and fits an optimal approximation curve; specifically, to accurately describe the nonlinear characteristics of the lift-off effect on the impedance plane, the system constructs a quadratic polynomial regression model: , in, and These are the real and imaginary parts of the second complex impedance, respectively. These are the fitting coefficients obtained using the least squares method; Lift-off effect trajectory The source is the above fitting calculation result. The physical meaning is the impedance change path caused only by the change in lift-off distance under the current material conductivity and magnetic permeability conditions. The unit is a dimensionless vector. This trajectory essentially depicts the distribution law of background noise on the impedance plane. This embodiment utilizes the physical shielding properties of high-frequency signals on deep defects to successfully extract a pure lift-off interference model. This dynamic fitting mechanism based on real-time data overcomes the limitations of traditional methods that rely on static standard test blocks for calibration. It can adaptively compensate for reference drift caused by uneven surface roughness or microscopic material inhomogeneity of the blade, significantly improving the robustness of interference modeling.

[0020] Example 4: Step three includes: S31. Calculate the tangent slope of the lift-off effect trajectory at the current operating point, and determine the rotation angle required to rotate the lift-off effect trajectory to be parallel to the real axis of the impedance plane coordinate system based on the tangent slope. S32. Construct a phase rotation matrix based on the rotation angle; S33. Treat the first complex impedance as a vector, multiply it by the phase rotation matrix on the left to obtain the new impedance vector after rotation; S34. Extract the component of the new impedance vector in the imaginary axis direction. Since the lift-off effect trajectory has been rotated to the real axis direction, the component in the imaginary axis direction is defined as the decoupled pure defect characteristic signal.

[0021] This embodiment is a further specification of the signal orthogonal projection and decoupling steps in Embodiment 1; the regression model coefficients determined in system call step S23. Extract the real part of the second complex impedance at the current moment as the current operating point. Perform differentiation to calculate the slope of the tangent to the lift-off trajectory. : , The rotation angle is then calculated based on inverse trigonometric functions. This angle represents the angle between the direction of interference removal and the real axis of the coordinate system; the system constructs a two-dimensional phase rotation matrix. This matrix, acting as a linear transformation operator, is used to rotate the impedance vector clockwise to eliminate the liftoff angle. , The system introduces a first complex impedance containing information about deep defects. ;in, and The real and imaginary components of the impedance at the first frequency are respectively treated as two-dimensional column vectors, and matrix multiplication is performed on them. During this process, what was originally mixed in The lift-off interference component is forcibly rotated to the real axis direction of the new coordinate system. According to electromagnetic field theory, the direction of impedance change caused by the defect has a phase difference with the lift-off direction; therefore, the defect signal component is retained in the imaginary axis direction. The system directly extracts the imaginary part of the new impedance vector as the output, i.e.: , Pure defect characteristic signal The source is the vector projection calculation mentioned above, and its physical meaning is the defect response intensity after removing lift-off interference, with the unit being volts; This embodiment achieves geometric orthogonal separation of signal and noise through mathematical coordinate rotation transformation; the algorithm transforms complex nonlinear signal processing problems into simple linear algebra operations, achieving a very high signal-to-noise ratio improvement with extremely low computational cost, and eliminates common-mode interference caused by probe jitter by using the same source signal from the same probe at the same time for self-calibration.

[0022] Example 5: Step four includes: S41. Extract the amplitude of the pure defect feature signal; S42. A preset defect judgment threshold set, including a first safety threshold and a second critical threshold, and defining the first safety threshold as less than the second critical threshold; S43. Perform range judgment on the amplitude to obtain the weld defect assessment results: If the amplitude is less than or equal to the first safety threshold, the current detection area is determined to be in a defect-free state, and the weld defect assessment result is labeled as qualified. If the amplitude is greater than the first safety threshold and less than or equal to the second critical threshold, it is determined that there is a slight shallow disturbance in the current detection area, and the weld defect assessment result is labeled as a warning. If the amplitude is greater than the second critical threshold, it is determined that there is a buried defect with a depth of 1 mm or more in the current detection area, and the weld defect assessment result is labeled as unqualified.

[0023] This embodiment further specifies the defect classification and evaluation steps in Embodiment 1; the system performs envelope detection on the decoupled time-domain signal to extract the absolute amplitude of the pure defect feature signal. The system loads a set of defect judgment thresholds preset in non-volatile memory. This set of thresholds is derived from statistical analysis of experimental data from a large number of standard test blocks. To ensure the scientific validity and feasibility of the threshold settings, the following statistical methods are used for specific definition. First safety threshold The source is a calculated value based on statistical analysis of on-site background noise. The specific setup steps are as follows: perform dynamic scanning in the defect-free weld area, collect the background noise signal sequence, and calculate its mean. with standard deviation ,set up The physical meaning of is the statistical limit of the background noise of the system, and the unit is volt; Second critical threshold The source is the standard test block calibration value. The specific setting steps are as follows: Obtain the characteristic signal amplitude on a standard test block containing an artificial defect with a flat bottom hole of 1 mm depth. ,set up The physical meaning of is the alarm threshold corresponding to a critical size crack, and the unit is volts; The system performs a three-level interval logic judgment: responding to amplitude. Less than or equal to The system determines it to be qualified and filters out electronic noise; response to amplitude Greater than and less than or equal to The system issues a warning, indicating the possible presence of non-destructive surface defects; the response is based on the amplitude. Greater than The system determines that the item is unqualified, confirming the presence of a hazardous buried defect; the system writes the corresponding label into the detection log; This embodiment establishes a hierarchical control mechanism based on signal energy intensity. By setting dual thresholds, it not only achieves qualitative detection of defects but also quantitative classification of defect severity, effectively avoiding false positives caused by oversensitivity. It also ensures zero tolerance for major structural hazards, providing logical support for subsequent differentiated maintenance decisions.

[0024] Example 6: Step four also includes: S44. Establish an amplitude-depth mapping database, which stores the correspondence between the amplitude of the pure defect characteristic signal measured on the standard test block and the known defect depth. S45. In response to the weld defect assessment result being labeled as unqualified, the amplitude is input into the amplitude-depth mapping database for linear interpolation retrieval, and the quantitative defect depth value of the current detection point is output. S46. Associate the quantized defect depth value with the physical location coordinates of the current array unit to generate a three-dimensional distribution map of defects inside the weld.

[0025] This embodiment is a depth quantification extension of the defect assessment process in Embodiment 5. In the calibration stage before detection, the system is used to scan standard test blocks with artificial grooves of different depths to build an amplitude-depth mapping database. During real-time detection, in response to the system's judgment result being unqualified, the control unit immediately calls the lookup table program and inputs the currently measured signal amplitude as the index key value into the database. Linear interpolation is performed using two adjacent standard data points to calculate the estimated depth of the current defect. Quantitative Defect Depth Value The source is interpolation calculation, and its physical meaning is the estimated vertical depth of the defect, in millimeters; Based on this, the system synchronously acquires the probe coordinates fed back by the external position sensor and converts the depth value... The system fuses data with planar coordinates to generate point cloud data containing spatial location and depth information; a visualization rendering engine then generates a 3D distribution map that intuitively displays the morphology of internal defects in the weld; specifically, the system obtains the physical location coordinates of the current array element. And obtain the surface normal vector at that location based on the robot arm end-effector pose data or the digital model of the rotor blades. Using the formula Calculate the actual coordinates of the defect in three-dimensional space, where Let be the unit normal vector perpendicular to the blade surface and pointing outwards at that point, where To quantify the defect depth, the generated point cloud data are correlated to construct a three-dimensional distribution map of defects inside the weld. This embodiment upgrades a single alarm signal into visualized three-dimensional quantitative data; through database mapping and linear interpolation technology, it achieves accurate inversion of the depth of invisible buried defects. This non-destructive depth measurement capability directly replaces the traditional destructive slicing inspection, greatly reducing the time and economic costs of quality assessment.

[0026] Example 7: Step four also includes: S47. When the edge region of the runner blade is detected, the lift-off effect trajectory is used as the reference signal of the edge effect to perform edge compensation correction on the pure defect feature signal and generate the corrected edge defect feature signal. S48. Substitute the corrected edge defect feature signal into step S43 for re-judgment to eliminate false alarms caused by edge effects.

[0027] This embodiment is an adaptive optimization of the signal processing flow in Embodiment 5 for a specific region; the system identifies whether the probe is close to the physical edge of the impeller blade by monitoring the abrupt change slope of the real part of the impedance. The specific identification logic is as follows: real-time calculation of the projection component of the new impedance vector obtained after the phase rotation matrix transformation in step S33 on the real axis. Discrete first-order difference values In response to Exceeding the preset gradient threshold , defined here The value is set to three times the standard deviation of the background noise in the stable region, indicating that the probe has entered the edge influence zone. At this point, the system immediately pauses the fitting of the lift-off effect trajectory coefficients in step S23. Real-time updates, locking and recalling the moment before entering the edge of the influence zone. The fitting coefficients of the cache are used as a fixed benchmark; In the edge region, due to the sharp reduction in conductive volume, the edge effect caused by the distortion of the eddy current field exhibits a highly geometrical similarity to the lift-off effect on the impedance plane. The system reuses the lift-off effect trajectory generated in step S23 as a reference. An edge correction coefficient is introduced. This coefficient originates from an edge calibration experiment. The specific operational procedure of this experiment is as follows: The flexible eddy current array probe is controlled to start from the center of the defect-free blade substrate area, scanning outwards at a uniform speed and completely crossing the physical edge into the air domain, thereby acquiring a complete edge response data stream. The specific calculation reuses the least squares closed-form formula. In the aforementioned defect-free edge calibration experiment, the system forcibly uses fixed lift-off trajectory parameters of the blade center region for rotational projection to simulate the parameter locking state in actual detection. Data acquisition... Group, The preset number of sampling points, and The real projection component after rotation With the corresponding imaginary part signal Data pairs The ratio of their covariance to their variance is calculated as the correlation slope: , in, This represents the index of the sampling point in the summation operation, with a value range of 1. to , and These are the arithmetic mean of the real and imaginary parts of the calibration dataset, respectively; calculate the projection component of the impedance vector along the real axis after the current rotation. The calculation formula is as follows: , Based on this, a linear correction formula is established: , in, This is the imaginary part of the original decoupled signal. This is the corrected edge defect feature signal; this step physically means subtracting the residual component caused by edge effects and proportionally coupled to the imaginary axis direction; the system will then use the corrected signal. Resubmit the threshold determination logic and perform a second evaluation; This embodiment innovatively utilizes the physical homology of edge effect and lift-off effect, and adopts a strategy based on the cancellation of homogeneous interference features, using a lift-off suppression algorithm to simultaneously suppress edge noise. This mechanism effectively solves the industry problem of blind spots in traditional eddy current detection at workpiece edges, ensuring the integrity of full coverage detection of the impeller blades and eliminating the risk of false edge alarms.

[0028] Example 8: Step four also includes: S49. Generate maintenance strategy signals based on weld defect assessment results: In response to the tag being deemed valid, a command to continue scanning is generated, and the current path is recorded as a safe path; In response to the label being a warning, a marking instruction is generated, and the warning area is highlighted in yellow in the preset digital model, with a suggestion to re-check it in the next maintenance cycle; In response to the label being unqualified, a shutdown review instruction and grinding repair suggestions are generated. The area is highlighted in red in the preset digital model, and a quantitative defect depth value is output to guide the grinding depth.

[0029] This embodiment further specifies the maintenance strategy generation steps in Embodiment 5; based on the evaluation labels output by S43, the system drives the industrial control system to execute differentiated decision logic; in response to a qualified label, the system kernel generates a continue scanning instruction to maintain the movement state of the automated equipment and marks the current coordinate path as a green zone in the database; in response to a warning label, the system generates a soft marking instruction, rendering a yellow block at the corresponding coordinates of the digital twin model of the impeller blade and adding the area to the T+1 cycle key attention list, without interrupting the current operation; in response to a disqualified label, the system immediately triggers the highest level... Prioritized shutdown and review instructions interrupt the scanning process to prevent missed detections. Simultaneously, a red warning area is rendered in the digital model, and the quantified defect depth value calculated by S45 is directly invoked to generate a repair recommendation containing precise grinding depth, which is then sent to the maintenance terminal. This embodiment achieves a closed-loop transition from non-destructive testing to intelligent operation and maintenance. By directly mapping the detection results to specific engineering instructions, the system not only acts as a quality inspector but also assumes the function of a decision-maker. This dynamic response mechanism based on risk grading maximizes the operational availability of the turbine unit while ensuring that major safety hazards are addressed promptly and accurately.

[0030] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for inspecting weld seams on turbine blades, characterized in that, The specific steps include: Step 1: Control the flexible eddy current array probe to perform dynamic scanning in close contact with the weld surface of the impeller blade, and apply a mixed excitation current containing a first frequency and a second frequency to the flexible eddy current array probe, collect the complex impedance response signal of each array unit, and construct the original impedance dataset. Step 2: Establish an impedance plane coordinate system, extract the second impedance component based on the second frequency from the original impedance dataset, and use the second impedance component to fit and generate the current lift-off effect trajectory in the impedance plane coordinate system; Step 3: Extract the first impedance component based on the first frequency from the original impedance dataset, calculate the rotation angle of the lift-off effect trajectory relative to the real axis of the impedance plane coordinate system, and construct a phase rotation matrix; use the phase rotation matrix to perform an orthogonal projection transformation on the first impedance component, and extract the imaginary part signal orthogonal to the lift-off effect trajectory as a pure defect feature signal; Step 4: Preset a defect judgment threshold set, compare the amplitude of the pure defect feature signal with the defect judgment threshold set to obtain the weld defect evaluation result, and generate a corresponding maintenance strategy signal based on the weld defect evaluation result.

2. The method for inspecting the weld seam of a turbine blade according to claim 1, characterized in that: Step one includes: S11. A flexible eddy current array probe based on a flexible PCB substrate is used to cover the weld seam area of ​​the impeller blade with a double curvature feature. S12. Configure the frequency parameters of the hybrid excitation current, wherein the first frequency is set to a low frequency band to penetrate the weld metal medium to obtain deep structural information; the second frequency is set to a high frequency band to sense the surface state of the weld to obtain lift-off interference information, and the value of the first frequency is specified to be less than the value of the second frequency. S13. The induced voltage of the flexible eddy current array probe is synchronously demodulated using the FPGA multi-frequency demodulation module to obtain the first complex impedance corresponding to the first frequency and the second complex impedance corresponding to the second frequency, and the data of all array units are summarized to construct the original impedance dataset.

3. The method for inspecting the weld seam of a turbine blade according to claim 2, characterized in that: Step two includes: S21. Map the second complex impedance to the impedance plane coordinate system to obtain its real part coordinates and imaginary part coordinates; S22. Due to the high-frequency characteristics of the second frequency, the second complex impedance is not sensitive to deep defects but only sensitive to the gap change between the probe and the measured surface. Data points of the second complex impedance within a continuous time window are collected. S23. The least squares method is used to perform curve fitting on the data points to generate a curve that reflects the surface undulation state of the current scanning area, and the curve is defined as the lift-off effect trajectory.

4. The method for inspecting the weld seam of a turbine blade according to claim 3, characterized in that: Step three includes: S31. Calculate the tangent slope of the lift-off effect trajectory at the current operating point, and determine the rotation angle required to rotate the lift-off effect trajectory to be parallel to the real axis of the impedance plane coordinate system based on the tangent slope. S32. Construct the phase rotation matrix based on the rotation angle; S33. Treat the first complex impedance as a vector, and multiply it by the phase rotation matrix on the left to obtain the new impedance vector after rotation; S34. Extract the component of the new impedance vector in the imaginary axis direction. Since the lift-off effect trajectory has been rotated to the real axis direction, define the component in the imaginary axis direction as the decoupled pure defect feature signal.

5. The method for inspecting the weld seam of a turbine blade according to claim 4, characterized in that: Step four includes: S41. Extract the amplitude of the pure defect feature signal; S42. The defect determination threshold set is preset, including a first safety threshold and a second critical threshold, and the first safety threshold is defined to be less than the second critical threshold; S43. Perform interval judgment on the amplitude to obtain the weld defect assessment result: If the amplitude is less than or equal to the first safety threshold, the current detection area is determined to be in a defect-free state, and the weld defect assessment result is labeled as qualified. If the amplitude is greater than the first safety threshold and less than or equal to the second critical threshold, it is determined that there is a slight shallow disturbance in the current detection area, and the weld defect assessment result is labeled as a warning. If the amplitude is greater than the second critical threshold, it is determined that there is a buried defect with a depth greater than or equal to 1 mm in the current detection area, and the weld defect assessment result is labeled as unqualified.

6. The method for inspecting the weld seam of a turbine blade according to claim 5, characterized in that: Step four also includes: S44. Establish an amplitude-depth mapping database, which stores the correspondence between the amplitude of the pure defect feature signal measured on the standard test block and the known defect depth. S45. In response to the weld defect assessment result being labeled as unqualified, the amplitude is input into the amplitude-depth mapping database for linear interpolation retrieval, and the quantified defect depth value of the current detection point is output. S46. Associate the quantized defect depth value with the physical position coordinates of the current array unit to generate a three-dimensional distribution map of internal defects in the weld.

7. The method for inspecting the weld seam of a turbine blade according to claim 5, characterized in that: Step four also includes: S47. When the edge region of the runner blade is detected, the lift-off effect trajectory is used as the reference signal of the edge effect to perform edge compensation correction on the pure defect feature signal to generate the corrected edge defect feature signal. S48. Substitute the corrected edge defect feature signal into step S43 for re-determination to eliminate false alarms caused by edge effects.

8. The method for inspecting the weld seam of a turbine blade according to claim 6, characterized in that: Step four also includes: S49. Generate the maintenance strategy signal based on the weld defect assessment results: In response to the tag being deemed valid, a command to continue scanning is generated, and the current path is recorded as a safe path; In response to the label being a warning, a marking instruction is generated, the warning area is highlighted in yellow in a preset digital model, and it is recommended to re-check it in the next maintenance cycle; In response to the label being deemed unqualified, a shutdown review instruction and a polishing repair suggestion are generated. The area is highlighted in red in the preset digital model, and the quantified defect depth value is output to guide the polishing depth.

Images