A large workpiece-based ultrasonic sensitivity detection method

Abstract

The application provides a large workpiece-based ultrasonic sensitivity detection method, relates to the technical field of detection, and aims to solve the problems of inconsistent sensitivity in the full-thickness range and insufficient detection precision of deep defects in traditional ultrasonic detection. The method sets multiple groups of depth-adaptive equivalent reflectors on the test block, dynamically adjusts the gain value in combination with the fuzzy neural network controller, realizes sensitivity adaptive compensation in the full-thickness range, and ensures that the detection error is stable within ±1%. Based on the dynamic reflector interval design of the material attenuation coefficient and the acoustic parameter, the reflector distribution is optimized, and the complexity of artificial calibration is significantly reduced. The TCG curve is fitted by the piecewise weighted least square method and simulated to generate a high-precision DAC curve, realizing high-confidence output of defect quantitative evaluation. The sensitivity, accuracy and engineering applicability of ultrasonic detection of large workpieces are improved, and the method is particularly suitable for defect detection of complex curvature workpieces.

Description

Technical Field

[0001] This invention relates to the field of detection technology, specifically to an ultrasonic sensitivity detection method for large workpieces. Background Technology

[0002] Ultrasonic testing technology, as an important method of industrial non-destructive testing, is widely used in the detection of internal defects in large workpieces in fields such as nuclear power and aerospace. Its core principle is to use an ultrasonic probe to emit high-frequency sound waves and receive the reflected signals from inside the workpiece to identify defects.

[0003] However, existing technologies still have significant limitations in practical applications. The fixed reflector depth intervals of traditional calibration blocks cannot adapt to the dynamic changes in the acoustic properties of different materials, leading to an imbalance in sensitivity between shallow and deep layers. This problem is particularly prominent in workpieces with non-uniform materials or significant thickness variations, severely affecting calibration accuracy. Furthermore, in deep regions, factors such as sound beam diffusion, enhanced noise interference, and superimposed interface reflections make it difficult for existing fitting algorithms to effectively suppress data fluctuations, directly impacting the accuracy of quantitative defect assessment. Summary of the Invention

[0004] To achieve the above objectives, the present invention provides the following technical solution: an ultrasonic sensitivity detection method for large workpieces, comprising:

[0005] a. Multiple sets of reflectors are placed on the test block, each set containing at least three equivalent reflectors at different depths, covering the full thickness range of the workpiece under test. The equivalent reflector's weight is adaptively adjusted according to the acoustic properties of the material under test, with the depth interval between adjacent reflectors based on a depth interval calculation formula. Dynamically determined, among which α is the depth interval, γ is the material attenuation coefficient, Z is the acoustic impedance, f is the probe frequency, and α and β are calibration factors optimized by inversion from scanning experimental data.

[0006] b. The test block is automatically scanned by an ultrasonic probe to obtain the reflected echo amplitude of the equivalent reflector at different depths in real time, and the scanning position, echo amplitude and corresponding depth are recorded simultaneously.

[0007] c. Based on the dynamic deviation between the echo amplitude and the preset target amplitude, the gain value is dynamically adjusted using a gain controller to generate a dynamic gain compensation dataset.

[0008] d. Input the dynamic gain compensation dataset into the sound field propagation-based fitting module to generate TCG curves and DAC curves;

[0009] e. Integrate the TCG curve and DAC curve into the system, dynamically adjust the gain value according to the depth and local curvature of the current scanning position, normalize the echo amplitude of the defect, and output the quantitative size and confidence assessment of the defect.

[0010] Multiple sets of reflectors are placed on the test block, each set containing at least three equivalent reflectors at different depths, the depth range of which should cover the full thickness of the workpiece under inspection. The equivalent of the reflectors is adaptively adjusted according to the acoustic characteristics of the material under inspection, ensuring that the reflected echoes accurately reflect the acoustic characteristics at different depths. The depth intervals of the reflectors are determined by... Confirmed, among which Let α be the depth interval, γ be the material attenuation coefficient, Z be the acoustic impedance, f be the probe frequency, and α and β be calibration factors optimized through inversion from scanned experimental data. First, initial gain compensation data for equivalent reflectors at different depths are obtained through scanned experiments. The scanned experimental data is then segmented by depth, and each segment is processed independently to calculate its linear regression coefficients, yielding preliminary estimates of α and β. Based on the regression coefficients of each segment, the estimated values ​​of α and β are dynamically adjusted to ensure a smooth transition between different depth segments. Through iterative adjustments, until the regression coefficients of each segment become consistent, the final optimized values ​​of α and β are output and input into the depth interval calculation formula to achieve dynamic adaptation of the reflector distribution.

[0011] An ultrasonic probe is used to automatically scan the test block. During the scanning process, the amplitude of the reflected echo from the equivalent reflector at different depths is acquired in real time, and the scanning position, echo amplitude, and corresponding depth are recorded simultaneously. During the scanning process, the position and angle of the probe are controlled by a high-precision robotic arm to ensure the uniformity and consistency of the scanning. The scanning speed is dynamically adjusted according to the material and thickness of the workpiece being inspected to ensure that a stable reflected echo signal can be obtained at different depths.

[0012] The scanning experiment was conducted in a constant temperature environment to avoid the influence of temperature changes on acoustic characteristics. During the scanning process, the contact pressure and coupling status of the probe were monitored in real time to ensure good contact between the probe and the test block. The scanning data was recorded and sampled in real time, and the sampling frequency was adjusted according to the probe frequency and scanning speed to ensure the integrity and accuracy of the data.

[0013] Based on the echo amplitude acquired during the scanning process, the real-time difference between the current echo amplitude and the preset target amplitude is calculated. According to the amplitude and rate of change of the difference, the gain value is dynamically adjusted using a fuzzy neural network controller until the echo amplitude stabilizes within ±1% of the target amplitude error range. The input variables of the fuzzy neural network controller include echo amplitude deviation, deviation rate of change, and scanning acceleration, and the output is the gain adjustment amount. The network weights are updated in real time through an online learning algorithm, and the learning rate is inversely proportional to the scanning speed to ensure rapid convergence at different scanning speeds.

[0014] The fuzzy neural network controller is implemented by designing a three-layer neural network structure. The input layer contains three nodes, corresponding to the echo amplitude deviation, the rate of change of deviation, and the scanning acceleration, respectively. The hidden layer contains five nodes and uses the sigmoid activation function. The output layer contains one node, which outputs the gain adjustment. The network weights are updated through the backpropagation algorithm, and the learning rate is dynamically adjusted according to the scanning speed. The faster the scanning speed, the lower the learning rate, to ensure the stability of the system.

[0015] The backpropagation algorithm calculates the gradient of each weight, which is the partial derivative of the loss function with respect to the weight. The loss function is typically defined as the square of the echo amplitude deviation, and the goal is to minimize this deviation. During each scan, the error of the output layer is calculated based on the current echo amplitude deviation, the rate of change of the deviation, and the scan acceleration. This error is then backpropagated to the hidden and input layers, and the gradient of each weight is calculated layer by layer. The learning rate is dynamically adjusted based on the scan speed; the faster the scan speed, the lower the learning rate, to ensure the stability of the system during high-speed scans. Specifically, the relationship between the learning rate η and the scan speed v is as follows: Where η0 is the initial learning rate and k is the adjustment coefficient, an initial value is first set, and then the convergence speed and stability of the system are observed at different scan speeds; information such as the fluctuation of echo amplitude and the timeliness of gain adjustment is recorded; based on the observation results, the value of k is adjusted; if the system is unstable during high-speed scanning, the value of k is decreased; if convergence is too slow, the value of k is increased. In this way, weight updates can converge quickly during low-speed scanning, while maintaining a small step size during high-speed scanning to avoid system oscillation or instability; finally, the weight update formula is: , where L is the loss function and w is the weight.

[0016] During the gain adjustment process, the gain value and corresponding depth at the steady state are recorded in real time to generate a dynamic gain compensation dataset. This dataset contains gain compensation values ​​at different depths, which are used to generate subsequent TCG and DAC curves.

[0017] The dynamic gain compensation dataset is input into the sound field propagation-based fitting module to generate TCG and DAC curves. The fitting module operates in the following steps: First, the dataset is pre-corrected by combining the beam spread angle and material inhomogeneity parameters to eliminate near-field effects of the probe and interface reflection interference. Then, a piecewise weighted least squares method is used to nonlinearly interpolate the gain values ​​in different depth ranges to generate TCG curves. The weighting function in the piecewise weighted least squares method is: in, The weight at depth d, λ and λ are adjustment parameters based on the material attenuation coefficient and probe frequency, used to suppress noise interference in deep data.

[0018] Based on the generated TCG curve, the DAC curve is derived in reverse, and the self-consistency of the hyperbola is optimized through cross-validation of experimental data. 1000 sets of simulated defect data are generated through Monte Carlo simulation, covering the full depth range and different equivalent gradients. The normalized amplitude of the simulated defect is calculated based on the TCG curve and the DAC curve. If the average deviation between the simulated amplitude and the theoretical amplitude exceeds 2%, the fitting process is iterated again until the curve converges.

[0019] During the detection process, based on the depth and local curvature of the current scanning position, the gain compensation value corresponding to the TCG curve is called, and the curvature correction factor is superimposed to dynamically adjust the gain; the defect echo amplitude is normalized based on the DAC curve to eliminate the influence of depth difference on the amplitude; by comparing the normalized amplitude with the preset defect equivalent threshold, combined with the spatial distribution density of the defect and the sound field energy distribution characteristics, the quantitative size and confidence assessment of the defect are output.

[0020] The equivalent size of the defect is initially determined by comparing the normalized amplitude with the preset defect equivalent threshold. Then, the actual size of the defect is calculated by combining the spatial distribution density of the defect. Finally, the confidence level of the defect is evaluated based on the characteristics of the sound field energy distribution. The confidence level evaluation is based on the stability and consistency of the defect echo amplitude. If the defect echo amplitude remains consistent at different scanning positions, the confidence level is high; otherwise, the confidence level is low.

[0021] This invention provides an ultrasonic sensitivity detection method for large workpieces, which has the following advantages:

[0022] 1. This invention solves the problem of sensitivity attenuation in deep regions by dynamically adjusting the gain value through a fuzzy neural network controller, which is a problem of traditional fixed gain mode. It significantly improves the consistency of detection sensitivity and ensures more stable and reliable defect detection across the entire thickness range.

[0023] 2. This invention optimizes the distribution of reflectors based on the dynamic reflector spacing design of material attenuation coefficient and acoustic parameters, maximizes the detection coverage, and reduces the complexity of manual calibration, making the detection process more efficient and intelligent.

[0024] 3. This invention uses a sound field propagation-based fitting module to pre-correct the dynamic gain compensation dataset by combining the sound beam spread angle and material non-uniformity parameters. It uses a piecewise weighted least squares method to generate TCG curves and derives DAC curves in reverse, which effectively improves detection efficiency and quality control level, and reduces the deviation and uncertainty of detection results caused by inaccurate curves. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0026] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] The test block is the foundation of ultrasonic testing, and its material should be similar to the acoustic characteristics of the workpiece being tested. For example, if testing an SAE-1020 carbon steel workpiece, the test block should also be made of the same material to ensure that acoustic parameters such as sound velocity and acoustic impedance are consistent. A flat-bottomed hole is selected as the equivalent reflector because its reflection characteristics are stable and it is easy to process. The size is determined according to the workpiece thickness and the expected sensitivity. For example, for a 300mm thick workpiece, the reflector diameter can be set to φ2mm, φ3mm, φ4mm, etc., and the depth can cover the entire thickness range of the workpiece, from 10mm to 290mm. A group is set every 50mm, and each group contains at least three reflectors of different depths, such as 10mm, 60mm, 110mm, etc.

[0028] Depth interval through The values ​​are dynamically determined, where Δd is the depth interval, γ is the material attenuation coefficient, Z is the acoustic impedance, f is the probe frequency, and α and β are calibration factors. Taking SAE-1020 carbon steel as an example, γ is approximately 0.5 dB / (mm·MHz), and Z is approximately 45 × 10⁻⁶. 6 kg / (m²·s); Initial gain compensation data is obtained by scanning and processing according to depth segments, linear regression coefficients are calculated, α and β are initially estimated, and dynamic adjustment is made to make them transition smoothly in different depth segments. Iterative adjustment is made until the regression coefficients of each segment tend to be consistent, and the optimized values ​​α=0.8 and β=1.2 are output to achieve dynamic adaptation of reflector distribution.

[0029] The ultrasonic probe is selected based on the workpiece thickness and testing requirements. For example, for a 300mm thick SAE-1020 carbon steel workpiece, a 5MHz, φ20mm longitudinal wave straight probe is selected. The scanning equipment adopts a high-precision automatic scanning device, equipped with a data acquisition system, and the robotic arm movement accuracy is not less than 0.1mm to ensure scanning uniformity and consistency. The scanning speed is dynamically adjusted according to the workpiece material and thickness. For example, for an SAE-1020 carbon steel workpiece, the scanning speed is set to 100mm / s to 200mm / s. The sampling frequency is adjusted according to the probe frequency and scanning speed. For example, for a 5MHz probe, the sampling frequency is set to 10MHz to 20MHz.

[0030] The scanning was conducted in a constant temperature environment, with the ambient temperature controlled within the range of 20℃±2℃ to avoid temperature changes affecting the acoustic characteristics. Glycerin or a special coupling agent was used between the probe and the test block to maintain good contact, and the coupling pressure was monitored in real time within the range of 1N to 3N. A robotic arm was used to perform automated scanning of the test block, starting from the surface and gradually penetrating in 5mm intervals until all reflector depths were covered. The amplitude of the reflected echo was acquired in real time, and the scanning position, echo amplitude, and corresponding depth were recorded simultaneously. The collected data underwent preliminary processing to remove noise and interference signals. For example, filtering algorithms were used to filter the echo signals to eliminate high-frequency noise and low-frequency interference, thereby improving the signal-to-noise ratio.

[0031] Based on the echo amplitude obtained from the scan, the real-time difference between the actual echo amplitude and the preset target amplitude is calculated. For example, if the preset target amplitude is 80% of the screen height and the actual echo amplitude is 75%, the difference is -5%. The gain controller dynamically adjusts the gain value based on this difference. The input variables of the fuzzy neural network controller include the echo amplitude deviation, the rate of change of the deviation, and the scan acceleration. The output is the gain adjustment amount. The network weights are updated in real time through the backpropagation algorithm. The learning rate is inversely proportional to the scan speed. For example, the initial learning rate... The baseline scanning speed v0 is 100 mm / s. When the scanning speed is 150 mm / s, the adjustment coefficient k is 1 and the learning rate η≈0.067. The gain value is dynamically adjusted by the gain controller until the echo amplitude is stable within ±1% of the target amplitude. The gain value and corresponding depth at steady state are recorded to generate a dynamic gain compensation dataset.

[0032] The dynamic gain compensation dataset is input into the sound field propagation-based fitting module. Combined with beam spread angle and material inhomogeneity parameters, the dataset is pre-corrected to eliminate near-field effects and interface reflection interference (e.g., the beam spread angle of a 5MHz probe is approximately 60°). A piecewise weighted least squares method is used to nonlinearly interpolate the gain values ​​in different depth ranges, generating TCG curves. The weighting function is... , where w(d) is the weight at depth d and λ is the adjustment parameter; based on the generated TCG curve, the DAC curve is derived in reverse, and the self-consistency of the hyperbola is optimized by cross-validation of experimental data. For example, 1000 sets of simulated defect data are generated by Monte Carlo simulation, covering the full depth range and different equivalent gradients. The normalized amplitude of the simulated defect is calculated based on the TCG curve and the DAC curve. If the average deviation between the simulated amplitude and the theoretical amplitude exceeds 2%, the fitting process is iterated again until the curve converges.

[0033] During the inspection process, based on the depth and local curvature of the current scanning position, the gain compensation value corresponding to the TCG curve is called, and a curvature correction factor is superimposed to dynamically adjust the gain. For example, for workpieces with complex curvature, when scanning areas with small curvature radii, the gain compensation value is increased. The defect echo amplitude is normalized based on the DAC curve to eliminate the influence of depth differences on the amplitude. For example, if the echo amplitude of a certain defect is 70% of the screen height at a depth of 100mm, after normalization by the DAC curve, it is converted into the amplitude of a standard reflector with an equivalent diameter of φ3mm. By comparing the normalized amplitude with the preset defect equivalent threshold, combined with the spatial distribution density of the defect and the sound field energy distribution characteristics, the quantitative size and confidence assessment of the defect are output. For example, if the preset defect equivalent threshold is φ2mm, and the normalized amplitude corresponds to an equivalent diameter of φ3mm, and the spatial distribution density of the defect in the workpiece is low and the sound field energy distribution is uniform, then the quantitative size of the defect is judged to be φ3mm, and the confidence assessment is high.

[0034] Taking a large, complex curvature SAE-1020 carbon steel workpiece as an example, with a thickness of 300mm and a curvature radius ranging from 500mm to 1000mm, defects of different depths and sizes are artificially pre-fabricated on the workpiece, such as flat-bottomed holes with a depth of 50mm and a diameter of 3mm, and flat-bottomed holes with a depth of 150mm and a diameter of 4mm. Using the detection method of this invention, the pre-fabricated defects are accurately identified, and the error between the quantitative defect size and the actual size is within ±1%. The detection stability is high, and the echo amplitude fluctuation range is less than ±1% under different scanning speeds and coupling states. The gain adjustment is fast and accurate; compared with manual calibration, the time is reduced by about 70%, improving detection efficiency and intelligence. This method effectively solves the shortcomings of ultrasonic detection through depth-adaptive equivalent reflector setting, dynamic gain adjustment of fuzzy neural network controller, and TCG and DAC curve fitting based on sound field propagation. Especially in the detection of workpieces with complex curvature, it can dynamically adjust the gain according to the local curvature, ensuring the accuracy and reliability of defect detection, reducing manual intervention, improving detection efficiency and intelligence, and has high engineering application value.

[0035] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for ultrasonic sensitivity testing of large workpieces, characterized in that, Includes the following steps: a. Multiple sets of reflectors are placed on the test block, each set containing at least three equivalent reflectors at different depths, covering the full thickness range of the workpiece under test. The equivalent reflector's weight is adaptively adjusted according to the acoustic properties of the material under test, with the depth interval between adjacent reflectors based on a depth interval calculation formula. Dynamically determined, among which α is the depth interval, γ is the material attenuation coefficient, Z is the acoustic impedance, f is the probe frequency, and α and β are calibration factors optimized by inversion from scanning experimental data. b. The test block is automatically scanned by an ultrasonic probe to obtain the reflected echo amplitude of the equivalent reflector at different depths in real time, and the scanning position, echo amplitude and corresponding depth are recorded simultaneously. c. Based on the dynamic deviation between the echo amplitude and the preset target amplitude, the gain value is dynamically adjusted using a gain controller to generate a dynamic gain compensation dataset. d. Input the dynamic gain compensation dataset into the sound field propagation-based fitting module to generate TCG curves and DAC curves; e. Integrate the TCG curve and DAC curve into the system, dynamically adjust the gain value according to the depth and local curvature of the current scanning position, normalize the echo amplitude of the defect, and output the quantitative size and confidence assessment of the defect.

2. The ultrasonic sensitivity detection method for large workpieces according to claim 1, characterized in that: The optimization method for calibration factors α and β includes: obtaining initial gain compensation data of equivalent reflectors at different depths through scanning experiments; segmenting the scanning experiment data by depth and processing each segment independently; calculating the linear regression coefficient of each segment to obtain preliminary estimates of α and β; dynamically adjusting the estimates of α and β based on the regression coefficients of each segment to allow them to transition between different depth segments; iteratively adjusting until the regression coefficients of each segment become consistent, and outputting the final optimized values ​​of α and β; and inputting the optimized α and β into the depth interval calculation formula to achieve dynamic adaptation of reflector distribution.

3. The ultrasonic sensitivity detection method for large workpieces according to claim 1, characterized in that: The dynamic adjustment process of the gain controller in step c includes: calculating the real-time difference between the current echo amplitude and the target amplitude for each equivalent reflector at each depth; dynamically adjusting the gain value using a fuzzy neural network-based controller based on the amplitude and rate of change of the difference until the echo amplitude stabilizes within ±1% of the target amplitude error range; recording the gain value and corresponding depth when steady state is reached, and generating a dynamic gain compensation dataset.

4. The ultrasonic sensitivity detection method for large workpieces according to claim 3, characterized in that: The input variables of the fuzzy neural network controller include echo amplitude deviation, deviation change rate and scanning acceleration, and the output is gain adjustment. The network weights are updated in real time through backpropagation algorithm, and the learning rate is inversely proportional to the scanning speed.

5. The ultrasonic sensitivity detection method for large workpieces according to claim 1, characterized in that: The operation of the fitting module in step d includes: pre-correcting the dataset by combining the beam spread angle and material inhomogeneity parameters to eliminate near-field effects of the probe and interference from interface reflections; using piecewise weighted least squares method to nonlinearly interpolate the gain values ​​in different depth ranges to generate TCG curves; deriving DAC curves in reverse based on the TCG curves, and optimizing the self-consistency of the hyperbola through cross-validation of experimental data.

6. The ultrasonic sensitivity detection method for large workpieces according to claim 5, characterized in that: In the piecewise weighted least squares method, the weight function is: In the formula, w(d) represents the weight at depth d. λ and λ are adjustment parameters based on the material attenuation coefficient and probe frequency, used to suppress noise interference in deep data.

7. The ultrasonic sensitivity detection method for large workpieces according to claim 1, characterized in that: The dynamic adjustment and normalization process described in step e includes: calling the gain compensation value corresponding to the TCG curve based on the depth and local curvature of the current scanning position, and dynamically adjusting the gain by superimposing a curvature correction factor; normalizing the defect echo amplitude based on the DAC curve to eliminate the influence of depth differences on the amplitude; and outputting the quantitative size and confidence assessment of the defect by comparing the normalized amplitude with the preset defect equivalent threshold and combining the spatial distribution density of the defect and the sound field energy distribution characteristics.

Images