A calibration method for ultrasonic flaw detector verification device

Abstract

The present invention relates to a calibration method for an ultrasonic flaw detector calibration device. The method comprises obtaining and emitting an ultrasonic wave having a theoretical frequency matching that of a material to be detected, obtaining the actual propagation path of the ultrasonic wave and performing real-time dynamic correction, generating calibration parameters for a frequency-delay compensation coefficient, compensating the emitted ultrasonic wave based on the calibration parameters, and collecting a time-baseline signal after the compensated ultrasonic wave enters the material to be detected. The signal is then input into a trained deviation calculation model to calculate the time axis scale deviation, adjusting the time-baseline linearity based on the time axis scale deviation, generating a stepped ultrasonic attenuation signal, and collecting the amplitude value of the flaw detector echo signal. The amplitude value of the echo signal is compared with a preset amplitude threshold to obtain a dynamic difference between the two, and adjusting the attenuation amount in real time based on the dynamic difference. The method solves the problems of insufficient calibration efficiency and consistency, poor dynamic parameter adaptability, limited sensitivity calibration accuracy, and the ability to simulate complex working conditions in the prior art.

Description

Technical Field

[0001] The invention belongs to the technical field of ultrasonic flaw detection equipment calibration, and in particular relates to a calibration method for an ultrasonic flaw detector verification device. Background Art

[0002] As core equipment for industrial nondestructive testing, the performance of ultrasonic flaw detectors is directly related to the accuracy and reliability of material defect detection. To ensure the traceability of flaw detector measurement results, key parameters (such as time-base linearity, sensitivity margin, and dynamic range) must be regularly calibrated using a calibration device.

[0003] Existing calibration methods mainly rely on physical standard test blocks, such as IIW test blocks, and manual operations, which have the following significant limitations:

[0004] Insufficient calibration efficiency and consistency:

[0005] Traditional methods require frequent replacement of standard test blocks of varying specifications and rely on operator experience to adjust probe position and instrument parameters. The calibration process is time-consuming, typically taking over 30 minutes, and suffers from significant repeatability errors, with baseline linearity errors generally exceeding ±1%. Furthermore, factors such as physical test block wear and temperature drift introduce further uncertainty.

[0006] Poor adaptability of dynamic parameters:

[0007] Existing technologies often use a fixed ultrasonic velocity model for time-baseline calibration. However, in actual testing, the ultrasonic velocity of a material can deviate due to composition fluctuations or ambient temperature changes, with a typical temperature drift of approximately 0.5% / °C. For unknown materials or non-standard test pieces, manual ultrasonic velocity calculation is required, which is inefficient and prone to human error.

[0008] ‌Sensitivity calibration accuracy is limited‌:

[0009] Conventional sensitivity calibration uses a mechanical attenuator to adjust signal strength, with a typical step accuracy of ±0.5dB. This makes it difficult to meet the calibration requirements of high-precision digital flaw detectors, such as those requiring a sensitivity margin error of ≤±0.1dB. Furthermore, the subjectivity of manual interpretation of echo amplitude can lead to high dispersion in calibration results.

[0010] Lack of ability to simulate complex working conditions:

[0011] As industrial inspection scenarios diversify, flaw detectors must operate in complex environments with strong noise and multiple reflection paths. Traditional calibration devices only generate a single standard signal and cannot simulate the interference signal superposition or material structural heterogeneity found in actual working conditions, resulting in deviations between calibration results and actual performance.

[0012] Therefore, there is an urgent need to improve the calibration method of ultrasonic flaw detector calibration devices in the existing technology to overcome the technical problems of insufficient calibration efficiency and consistency, poor adaptability of dynamic parameters, limited sensitivity calibration accuracy and lack of ability to simulate complex working conditions. Summary of the Invention

[0013] The purpose of the present invention is to provide a calibration method for an ultrasonic flaw detector calibration device to overcome the technical problems existing in the prior art, such as insufficient calibration efficiency and consistency, poor adaptability of dynamic parameters, limited sensitivity calibration accuracy, and lack of complex working condition simulation capabilities.

[0014] In order to solve the above technical problems, the technical solutions adopted by the present invention are as follows:

[0015] A calibration method for an ultrasonic flaw detector verification device comprises the following steps:

[0016] S1: Detect the specified material parameters of the material to be tested, call the preset material parameter-ultrasonic database, and obtain the ultrasonic wave with the theoretical frequency matching the material to be tested;

[0017] S2: transmitting the ultrasonic wave of the theoretical frequency to the material to be detected, and obtaining the actual propagation path of the ultrasonic wave of the theoretical frequency in the material to be detected;

[0018] S3: Performing real-time dynamic correction on the actual propagation path based on a preset algorithm to generate calibration parameters of the ultrasonic velocity-delay compensation coefficient;

[0019] S4: compensating the transmitted ultrasonic wave based on the calibration parameters of the ultrasonic wave velocity-delay compensation coefficient;

[0020] S5: After the compensated ultrasonic wave enters the material to be inspected, a time baseline signal output by the flaw detector is collected, the time baseline signal is input into the trained deviation calculation model, the deviation calculation model calculates and outputs the time axis scale deviation, and the time baseline linearity is adjusted based on the time axis scale deviation;

[0021] S6: Generate a stepped ultrasonic attenuation signal, collect the amplitude value of the flaw detector echo signal, compare the echo signal amplitude value with a preset amplitude threshold to obtain a dynamic difference between the two, and adjust the attenuation amount in real time based on the dynamic difference.

[0022] Preferably, the specific process of transmitting the ultrasonic wave of the theoretical frequency and obtaining the actual propagation path of the ultrasonic wave of the theoretical frequency in the material to be detected in step S2 is as follows:

[0023] S21: Based on the theoretical frequency, applying a high-frequency alternating voltage through the piezoelectric crystal to convert electrical energy into mechanical vibration to generate an ultrasonic pulse sequence;

[0024] S22: Use acoustic lens or phased array technology to adjust the beam diffusion angle to ensure that the ultrasonic energy is concentrated on the surface of the material to be tested and suppress stray wave interference;

[0025] S23: constructing a theoretical propagation path model according to the ultrasonic wave velocity corresponding to the material to be tested, including direct incidence, refraction, and reflection path branches;

[0026] S24: using the array probe group to synchronously receive the reflected wave, scattered wave, and transmitted wave, and record the time domain signals and amplitude information of each channel of the reflected wave, scattered wave, and transmitted wave;

[0027] S25: Perform short-time Fourier transform on the received reflected waves, scattered waves, and transmitted waves to extract the arrival time difference Δt of the reflected waves on each path, and calculate the actual propagation distance based on the speed of sound;

[0028] S26: Based on the preset nonlinear regression algorithm, optimize the residual between the theoretical path length and the actual measurement value, and iteratively solve the path correction coefficient e ;

[0029] S27: Map the corrected path data to a three-dimensional coordinate system to generate an actual propagation cloud map including the main sound beam, side lobes, and scattering paths, and mark key reflection points.

[0030] Preferably, in step S3, the specific process of performing real-time dynamic correction on the actual propagation path based on a preset algorithm to generate calibration parameters of the ultrasonic velocity-delay compensation coefficient is as follows:

[0031] S31: Use the dual probes to move at a constant speed along the surface of the material to be tested, collect multiple sets of ultrasonic transmission time data, synchronously record the ambient temperature, and compensate for the effect of temperature drift on ultrasonic velocity. The specific compensation formula is as follows:

[0032] Δ v = α ( T - T 0) v 0;

[0033] in, α is the material temperature drift coefficient, Δ v is the ultrasonic velocity compensation, v 0 is the theoretical ultrasonic speed, T is the recorded ambient temperature, T 0 is the standard ambient temperature;

[0034] S32: Construct the propagation time-path length function based on the least squares method. The specific formula is as follows:

[0035] t i= ( L i +e i ) / v real +d ;

[0036] in, t i is the propagation time, L i is the theoretical path length, e i is the path deviation, v real is the actual ultrasonic speed, d System delay;

[0037] S33: Generate a dynamically corrected ultrasonic velocity compensation coefficient based on the propagation time-path length function, and generate a compensation coefficient matrix based on different detection directions.

[0038] Preferably, the training data used when training the deviation calculation model includes:

[0039] The mapping relationship between the baseline waveform and the error label of the specified group in the historical calibration database;

[0040] Real-time acquisition of flaw detector noise spectrum characteristics and ambient temperature-humidity interference parameters;

[0041] The output of the deviation calculation model is the time baseline scale correction amount and the confidence assessment value;

[0042] When the confidence level of the deviation calculation model output is less than the preset value, the manual review mechanism is triggered.

[0043] Preferably, the specific process of step S5 is as follows:

[0044] S51: Capture the time base signal output by the flaw detector in real time, and record the probe delay, ambient temperature and material sound velocity parameters to form a composite data stream of time-amplitude-temperature;

[0045] S52: performing band-pass filtering and amplitude normalization on the original time baseline signal to eliminate probe contact noise and power frequency interference, and generating a standardized input vector for processing by the deviation calculation model;

[0046] S53: The deviation calculation model calculates the time axis scale deviation, and dynamically adjusts the time base linearity according to the time axis scale deviation. The integrated PID controller adjusts the DSP clock division coefficient in real time to ensure that the time base linearity error converges to a preset range.

[0047] Preferably, the deviation calculation model is a convolutional neural network including an input layer, a convolutional layer stacking layer, a feature compression layer and an output layer;

[0048] The input layer receives a 1×1000-dimensional time baseline signal, performs amplitude normalization processing based on the maximum amplitude of the time baseline signal, and eliminates high-frequency interference through sliding window mean filtering. The time baseline signal is a time series signal;

[0049] The convolutional layer stacking layer includes a first convolutional layer, a second convolutional layer, a third convolutional layer, a fourth convolutional layer, and a fifth convolutional layer. The step size of each convolutional layer is 1, and convolution filling is performed by adding a preset value to the edge of the input data;

[0050] Among them, the first convolution layer is used to extract local temporal features, the second convolution layer is used to enhance the sensitivity of high-frequency components, the third convolution layer is used to capture long-range temporal dependencies, the fourth convolution layer is used to establish multi-scale feature representation, and the fifth convolution layer is used to map features to deviation physical quantity dimensions;

[0051] The feature compression layer includes an activation function and a global average pooling layer. The activation function is used to enhance the nonlinear expression capability of the model. The global average pooling layer is used to add a 1×990 global average pooling layer after the fifth convolutional layer to compress the features into a 1×1×1 output to achieve temporal feature aggregation;

[0052] The output layer includes a regression mapping layer and a range adaptation layer. The regression mapping layer outputs a deviation value ΔT and uses a linear activation function for unconstrained numerical mapping. The range adaptation layer normalizes the label value during retraining and maps it to a specified range, corresponding to a ±10% deviation range.

[0053] Preferably, the specific process in step S6 is as follows:

[0054] S61: Generate an ultrasonic excitation pulse and apply a stepped attenuation through the attenuator;

[0055] S62: The flaw detector echo signal is collected through the ADC module, with a sampling rate greater than or equal to 100MS / s and a resolution greater than or equal to 12 bits;

[0056] S63: Capture the echo signal waveform through an oscilloscope and measure the peak amplitude;

[0057] S64: Based on the preset amplitude threshold of the material characteristics, a dynamic difference ΔA between the actual echo amplitude measurement value and the threshold is calculated, where ΔA = |actual echo amplitude - echo amplitude threshold|;

[0058] S65: When ΔA is greater than the tolerance, the adjustment mechanism is triggered. When ΔA ≤ 1dB, the fine-tuning step is 0.1dB. When ΔA > 3dB, the adjustment is switched to a larger step size for rapid convergence.

[0059] Preferably, after step S6, step S7 is further provided, wherein the flaw detector is scanned within a preset dynamic range, and horizontal linear error, vertical linear deviation and resolution threshold data are synchronously collected to generate a calibration report including a three-dimensional error cloud map and a correction suggestion matrix.

[0060] Preferably, the specific process of step S7 is as follows:

[0061] S71: Generate a pulse sequence covering a dynamic range greater than 80dB through the preset DDS signal generator, triggering the flaw detector transmitter to output a high-voltage pulse signal;

[0062] S72: Use the combination of standard attenuator and fixed attenuator to perform attenuation control, and simultaneously record the signal amplitude changes at the flaw detector receiving end to verify whether the dynamic range lower limit threshold meets the preset signal-to-noise ratio requirement;

[0063] S73: Output six groups of equally spaced modulation waves through the synchronous signal generator, adjust the flaw detector scanning range to the specified gear, collect the deviation value between each pulse leading edge and the theoretical scale, and calculate the maximum offset Δ L max , calculate the horizontal linear error. The specific calculation formula of the horizontal linear error is as follows:

[0064] Δ L = (Δ L max / full scale)×100%;

[0065] S74: Adjust the signal amplitude in a specified step within the dynamic range, collect the peak height on the flaw detector display, fit the deviation curve between the measured amplitude and the theoretical value, and calculate the vertical linear error;

[0066] S75: Generate a double-reflector simulation signal with a specified spacing, gradually reduce the signal amplitude until the flaw detector cannot distinguish the double peaks, and record the minimum calculable amplitude at this time as the resolution threshold;

[0067] S76: Map the three parameters of dynamic range, horizontal position, and vertical linear deviation to a three-dimensional coordinate system, generate a continuous error distribution cloud map through an interpolation algorithm, locate the abnormal parameter interval based on the error cloud map, call the preset compensation algorithm library, and output a targeted correction parameter matrix and a calibration report of the correction suggestion matrix.

[0068] The beneficial effects of the present invention include:

[0069] The present invention provides a calibration method for an ultrasonic flaw detector calibration device. The method obtains and transmits an ultrasonic wave with a theoretical frequency matching the material to be inspected, obtains the actual propagation path of the ultrasonic wave and performs real-time dynamic correction, generates calibration parameters for the ultrasonic velocity-delay compensation coefficient, compensates the emitted ultrasonic wave based on the calibration parameters, and after the compensated ultrasonic wave enters the material to be inspected, collects a time baseline signal and inputs it into a trained deviation calculation model to calculate the time axis scale deviation. The time baseline linearity is adjusted based on the time axis scale deviation to generate a stepped ultrasonic attenuation signal. The amplitude value of the flaw detector echo signal is collected, and the echo signal amplitude value is compared with a preset amplitude threshold to obtain a dynamic difference between the two. The attenuation amount is then adjusted in real time based on the dynamic difference. This method solves the problems of insufficient calibration efficiency and consistency, poor dynamic parameter adaptability, limited sensitivity calibration accuracy, and the ability to simulate complex working conditions in the prior art.

[0070] First, the theoretical ultrasonic frequency is matched through the material parameter database, and stray waves are suppressed by combining acoustic lens or phased array technology. A propagation path model is constructed to achieve precise matching of acoustic parameters and material properties, reducing the sound speed error caused by differences in material properties.

[0071] Secondly, through the dynamic real-time compensation mechanism, the temperature drift compensation formula is set to dynamically correct the sound speed, and the propagation time-path function is iteratively solved to solve the system delay, eliminating the interference of ambient temperature and humidity and probe contact noise, improving the real-time performance of the time base calibration, and making it suitable for high-speed industrial detection scenarios.

[0072] Thirdly, through intelligent time baseline deviation correction, a convolutional neural network is used as the deviation calculation model. Through a five-layer convolution structure, including local features → long-range dependencies → multi-scale fusion, the deep features of the time baseline signal are extracted, the global average pooling layer compresses the timing features, and the deviation value is output, so that the time baseline linearity error converges to the required range. The confidence assessment mechanism triggers manual review to avoid the risk of miscalibration.

[0073] Finally, dynamic differential adjustment is achieved through stepped attenuation closed-loop control. Combined with ADC acquisition and oscilloscope peak detection, adaptive attenuation compensation is implemented, effectively improving the echo amplitude control accuracy and vertical linearity indicators. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 The figure is a flow chart of the calibration method of the ultrasonic flaw detector verification device of the present invention.

[0075] Figure 2 Schematic diagram of the deviation calculation model of the present invention. DETAILED DESCRIPTION

[0076] The following is combined with Figure 1~Figure 2 The present invention is described in further detail:

[0077] Example 1

[0078] See attached Figure 1 As shown, a calibration method for an ultrasonic flaw detector verification device includes the following steps:

[0079] S1: Detect the specified material parameters of the material to be tested, call the preset material parameter-ultrasonic database, and obtain the ultrasonic wave with the theoretical frequency matching the material to be tested;

[0080] S2: transmitting the ultrasonic wave of the theoretical frequency to the material to be detected, and obtaining the actual propagation path of the ultrasonic wave of the theoretical frequency in the material to be detected;

[0081] S3: Performing real-time dynamic correction on the actual propagation path based on a preset algorithm to generate calibration parameters of the ultrasonic velocity-delay compensation coefficient;

[0082] S4: compensating the transmitted ultrasonic wave based on the calibration parameters of the ultrasonic wave velocity-delay compensation coefficient;

[0083] S5: After the compensated ultrasonic wave enters the material to be inspected, a time baseline signal output by the flaw detector is collected, the time baseline signal is input into the trained deviation calculation model, the deviation calculation model calculates and outputs the time axis scale deviation, and the time baseline linearity is adjusted based on the time axis scale deviation;

[0084] S6: Generate a stepped ultrasonic attenuation signal, collect the amplitude value of the flaw detector echo signal, compare the echo signal amplitude value with a preset amplitude threshold to obtain a dynamic difference between the two, and adjust the attenuation amount in real time based on the dynamic difference.

[0085] In this example, the material to be tested is 45# steel. Parameter testing of 45# steel revealed a density of 7.85 g / cm³, an elastic modulus of 210 GPa, and a Poisson's ratio of 0.29. The Material Parameter-Ultrasonic Database, which stores over 100,000 material acoustic properties, was used to match the theoretical ultrasonic frequency to 5 MHz with an error of ±0.1 MHz. Using the Material Parameter-Ultrasonic Database, the sound velocity range, optimal ultrasonic frequency for detection, and attenuation coefficient were determined based on the material type 45# steel.

[0086] A piezoelectric crystal applies a high-frequency alternating voltage to generate an ultrasonic pulse sequence of a specified pulse width. An acoustic lens with a curvature radius of 50 mm and made of epoxy resin is used to control the ultrasonic beam spread angle within ±5°, focusing it on the material surface and suppressing stray waves, achieving a signal-to-noise ratio of ≥40 dB. Based on the ultrasonic velocity of 5920 m / s for 45# steel, a model of direct incidence, refraction, and reflection paths was constructed, with an incidence angle of 45° and a refraction angle of 24°. A 64-element phased array probe was deployed to simultaneously acquire time domain signals of the main beam reflection, defect scattering, and backwall transmission, with an amplitude resolution of 0.1 mV. A short-time Fourier transform was performed on the reflected waves to extract the time difference Δt = 0.2 μs between adjacent channels and calculate the actual propagation distance. A Levenberg-Marquardt nonlinear regression algorithm was used to optimize the residual error between the theoretical path and the actual measurement. A path correction coefficient ε = 0.15 was iteratively calculated with a 95% confidence interval. Map the corrected path to a three-dimensional coordinate system and mark the defect location, such as a Ø2mm flat-bottom hole, and the sidelobe energy distribution, with a proportion of ≤5%.

[0087] Use a dual probe with separate transmitters and receivers to move along the surface of the steel at a constant speed of 0.5m / s to collect 100 sets of propagation time data with a standard deviation of ≤5ns. When the measured temperature T = 28℃ and the standard temperature T0 = 20℃, the steel sound velocity compensation amount is:

[0088] Δv= α (T-T0) v 0=0.0005×(28-20)×5920=23.68m / s;

[0089] Actual speed of sound v real =5920+23.68=5943.68m / s.

[0090] A propagation time-path function was established and fitted using the least squares method to generate a compensation coefficient matrix. The time axis scale deviation was calculated using the deviation calculation model. The model output a time axis scale deviation of ΔT = 0.12 μs with a confidence level of 98%. This triggered a PID controller to adjust the DSP clock division coefficient in 0.01% steps, reducing the time base linearity error from 0.5% to 0.1%. A signal generator outputted a 5 MHz pulse, which was then passed through an attenuator to apply stepped attenuation with 1 dB steps and a range of 0-80 dB. For example, the initial attenuation was 20 dB, the echo amplitude was 2.5 V, the threshold was 2.8 V, the dynamic difference was 0.3 dB, and a 0.1 dB fine adjustment was triggered.

[0091] Example 2

[0092] On the basis of Example 1, the specific process of transmitting the ultrasonic wave of the theoretical frequency and obtaining the actual propagation path of the ultrasonic wave of the theoretical frequency in the material to be detected in step S2 is as follows:

[0093] S21: Based on the theoretical frequency, applying a high-frequency alternating voltage through the piezoelectric crystal to convert electrical energy into mechanical vibration to generate an ultrasonic pulse sequence;

[0094] S22: Use acoustic lens or phased array technology to adjust the beam diffusion angle to ensure that the ultrasonic energy is concentrated on the surface of the material to be tested and suppress stray wave interference;

[0095] S23: constructing a theoretical propagation path model according to the ultrasonic wave velocity corresponding to the material to be tested, including direct incidence, refraction, and reflection path branches;

[0096] S24: using the array probe group to synchronously receive the reflected wave, scattered wave, and transmitted wave, and record the time domain signals and amplitude information of each channel of the reflected wave, scattered wave, and transmitted wave;

[0097] S25: Perform short-time Fourier transform on the received reflected waves, scattered waves, and transmitted waves to extract the arrival time difference Δt of the reflected waves on each path, and calculate the actual propagation distance based on the speed of sound;

[0098] S26: Based on the preset nonlinear regression algorithm, optimize the residual between the theoretical path length and the actual measurement value, and iteratively solve the path correction coefficient e ;

[0099] S27: Map the corrected path data to a three-dimensional coordinate system to generate an actual propagation cloud map including the main sound beam, side lobes, and scattering paths, and mark key reflection points.

[0100] In this embodiment, the piezoelectric crystal material is PZT-5H, with a Curie temperature of 350°C and an electromechanical coupling coefficient of 0.51. The excitation device is a pulse generator that outputs a high-frequency alternating voltage with a peak value of 400V, a frequency of 5MHz, a pulse width of 100ns, and a repetition rate of 1kHz. When the voltage is applied, the piezoelectric crystal vibrates longitudinally, generating an ultrasonic pulse train with a center frequency of 5±0.2MHz and a bandwidth of 3.8-6.2MHz. The waveform parameters are a Gaussian-windowed sine wave with a sidelobe suppression ratio greater than or equal to 30dB.

[0101] The lens is made of epoxy resin and its acoustic impedance is 3.0×10 6 Rayl, curvature radius 40mm3. Beam control: The diffusion angle is compressed from ±15° to ±5°, the focal depth below the workpiece surface is 5mm, and a 32-element linear array probe is configured with an element spacing of 0.5mm and a delay resolution of 10ns. The transmission delay calculation formula is:

[0102] τ=( R / c ) ×[( R ²+( n ×d )²) 1 / 2 - R ];

[0103] focal length R 50mm, the speed of sound in steel c 5920m / s, array element spacing d is 0.5mm, τ is the launch delay, n is the number of relay nodes in the transmission path.

[0104] Input material parameter values ​​to the path model to calculate the path branch. The theoretical path length L is 10 mm thick at vertical incidence. Snell's law is used to calculate the angle of incidence. i Refraction angle when 1=30° i 2, sinth 1 / c 1 = sinθ 2 / c 2 →θ 2 =arcsin ( c 2 ·sinθ 1 / c 1) = 16.2 ° .in c 1=6320m / s, c 2 is 1480m / s. F The reflection path length of a 2mm pore defect is L=2×(5²+1²) 1 / 2 =10.2mm.

[0105] The 64-element annular array probe has an outer diameter of 20mm, an inner diameter of 5mm, and a bandwidth of 2-10MHz. The synchronous sampling rate is 100MS / s, the quantization accuracy is 14bit, the inter-channel delay error is ≤2ns, and the main beam echo amplitude is v 1 = 2.3v, corresponding to the bottom echo. Defect scattering signal amplitude v 2=0.15V, SNR=21dB, penetration signal amplitude v 3=1.8 v , used for sound attenuation rate calculation.

[0106] The window function of the short-time Fourier transform is a Hamming window with a length of 256 points, an overlap rate of 75%, and a frequency resolution of 19.5 kHz. The time difference between the direct wave and the defect echo is extracted, Δt = 3.2 μs, and the actual propagation distance is calculated:

[0107] L real = c ×Δt+ L0=6320m / s×3.2μs+10mm=20.22mm+10mm= 30.22mm.

[0108] in, L 0 is the system calibration reference value.

[0109] The measured temperature is T = 25°C, and the sound velocity correction calculation is as follows:

[0110] Δc= α (T-T0) c 0=0.0005×5×6320=15.8m / s;

[0111] Path correction factor e =0.18mm, 95% confidence interval ±0.03mm. After correction, the deviation between the theoretical path and the measured value is reduced from 0.25mm to 0.05mm.

[0112] Coordinate system mapping: Create a 3D mesh of the workpiece with a resolution of 0.1mm×0.1mm×0.1mm, and interpolate the path data to the mesh nodes.

[0113] Cloud Rendering: The main beam path can be highlighted in red, with an energy contribution of 80%. The sidelobe path can be set to a semi-transparent blue, with an energy contribution of 15% and an angle of ±7°. The scattering path can be set to a yellow point cloud marker, corresponding to the defect coordinates x=5.2mm, y=3.1mm, and z=10.0mm. Key Point Annotation: The defect reflection point is marked with a 2mm diameter sphere and an amplitude-depth curve is attached, with the peak occurring at z=10mm.

[0114] In step S3, the specific process of performing real-time dynamic correction on the actual propagation path based on a preset algorithm to generate calibration parameters of the ultrasonic velocity-delay compensation coefficient is as follows:

[0115] S31: Use the dual probes to move at a constant speed along the surface of the material to be tested, collect multiple sets of ultrasonic transmission time data, synchronously record the ambient temperature, and compensate for the effect of temperature drift on ultrasonic velocity. The specific compensation formula is as follows:

[0116] Δ v = α ( T - T 0) v 0;

[0117] in, α is the material temperature drift coefficient, Δ v is the ultrasonic velocity compensation, v 0 is the theoretical ultrasonic speed, T is the recorded ambient temperature, T0 is the standard ambient temperature;

[0118] S32: Construct the propagation time-path length function based on the least squares method. The specific formula is as follows:

[0119] t i = ( L i +e i ) / v real +d ;

[0120] in, t i is the propagation time, L i is the theoretical path length, e i is the path deviation, v real is the actual ultrasonic speed, d System delay;

[0121] S33: Generate a dynamically corrected ultrasonic velocity compensation coefficient based on the propagation time-path length function, and generate a compensation coefficient matrix based on different detection directions.

[0122] Example 3

[0123] Based on Example 1 or Example 2, the training data used when training the deviation calculation model includes:

[0124] The mapping relationship between the baseline waveform and the error label of the specified group in the historical calibration database;

[0125] Real-time acquisition of flaw detector noise spectrum characteristics and ambient temperature-humidity interference parameters;

[0126] The output of the deviation calculation model is the time baseline scale correction amount and the confidence assessment value;

[0127] When the confidence level of the deviation calculation model output is less than the preset value, the manual review mechanism is triggered.

[0128] The specific process of step S5 is as follows:

[0129] The S51 uses the NI-PXIe-5162 PCIe high-speed data acquisition card with a sampling rate of 1 GS / s and a bandwidth of 200 MHz. It captures the time-baseline signal output by the flaw detector in real time. It also connects to a temperature sensor and a sound velocity measurement module to record probe delay, ambient temperature, and material sound velocity parameters, forming a composite time-amplitude-temperature data stream. It records the time-baseline signal timestamp, simultaneously overlays the probe delay compensation value, quantifies the time-baseline voltage peak, samples the ambient temperature once per second, and writes metadata into the data stream header.

[0130] S52: Bandpass filter and amplitude normalize the original time baseline signal, normalize the time baseline amplitude to the range [0,1], and use a 4th-order Butterworth filter to suppress low-frequency power frequency interference and high-frequency noise to eliminate probe contact noise and power frequency interference. The window length is 1024 points, corresponding to a time of 10.24 μs, and the step size is 512 points. Continuous time series segments are generated. The input vector dimension is: 1×1024 (time series) + 3×1 (temperature, sound velocity, probe delay) = 1027 dimensions. Generate a standardized input vector for processing by the deviation calculation model.

[0131] S53: The deviation calculation model receives a 1027-dimensional standardized vector, including the preprocessed time baseline signal and environmental parameters. The deviation calculation model calculates the time axis scale deviation, and dynamically adjusts the time baseline linearity according to the time axis scale deviation. The integrated PID controller adjusts the DSP clock division coefficient in real time to ensure that the time baseline linearity error converges to within a preset range.

[0132] See also Figure 2 As shown, the deviation calculation model is a convolutional neural network including an input layer, a convolutional layer stacking layer, a feature compression layer and an output layer;

[0133] The input layer receives a 1×1000-dimensional time baseline signal, performs amplitude normalization processing based on the maximum amplitude of the time baseline signal, and eliminates high-frequency interference through sliding window mean filtering. The time baseline signal is a time series signal;

[0134] The convolutional layer stacking layer includes a first convolutional layer, a second convolutional layer, a third convolutional layer, a fourth convolutional layer, and a fifth convolutional layer. The step size of each convolutional layer is 1, and convolution filling is performed by adding a preset value to the edge of the input data;

[0135] Among them, the first convolution layer consists of 32 5×1 kernels for extracting local temporal features, the second convolution layer consists of 64 3×1 kernels for enhancing the sensitivity of high-frequency components, the third convolution layer consists of 128 7×1 kernels for capturing long-range temporal dependencies, the fourth convolution layer consists of 64 multi-scale kernels, 1×1, 3×1, and 5×1 in parallel, for establishing multi-scale feature representation, and the fifth convolution layer consists of 1 1×1 kernel for mapping features to the deviation physical quantity dimension, i.e., linear activation.

[0136] The feature compression layer includes an activation function and a global average pooling layer. The activation function is used to enhance the nonlinear expression capability of the model. The global average pooling layer is used to add a 1×990 global average pooling layer after the fifth convolutional layer to compress the features into a 1×1×1 output to achieve temporal feature aggregation;

[0137] The output layer includes a regression mapping layer and a range adaptation layer. The regression mapping layer outputs a deviation value, ΔT, using an unconstrained numerical mapping with a linear activation function. During retraining, the range adaptation layer normalizes the label values ​​and maps them to a specified range, which can be set to [-1, 1], corresponding to a ±10% deviation range. The deviation calculation model outputs ΔT = 0.12 μs with a 98% confidence level, triggering a PID controller to adjust the DSP clock division factor in 0.01% steps, reducing the timebase linearity error from 0.5% to 0.1%.

[0138] Example 4

[0139] Based on Example 1, Example 2, or Example 3, the specific process in step S6 is as follows:

[0140] S61: Generates an ultrasonic excitation pulse and applies stepped attenuation through the attenuator. The initial attenuation step is set to 0dB, and subsequent attenuation steps are set to 10dB, 20dB, 30dB, 40dB, 50dB, 60dB, 70dB, and 80dB, with each step lasting 50ms. Dynamic attenuation step adjustment mechanism: When ΔA ≤ 1dB, the step size is 0.1dB; when ΔA > 3dB, the step size is 3dB. Pulse width is 100ns, repetition frequency is 1kHz, and energy density is 0.5mJ / mm².

[0141] S62: The flaw detector echo signal is acquired through the ADC module, with a sampling rate greater than or equal to 100MS / s and a resolution greater than or equal to 12 bits. The ADC module trigger delay is calibrated to ±2ns, capturing the complete rising edge of the echo signal. The storage depth per channel is 32k points, covering the entire echo time window.

[0142] S63: Use an oscilloscope to capture the echo signal waveform and measure the peak amplitude. Perform an extreme value search using a sliding window with a 50ns window width, eliminating ringing noise interference and sub-waves with amplitudes < 5% of the main peak.

[0143] S64: Dynamically calculate the threshold based on the material attenuation coefficient μ = 0.02dB / mm, preset the amplitude threshold, and calculate the dynamic difference ΔA between the measured echo amplitude and the threshold, ΔA = |measured echo amplitude - echo amplitude threshold|.

[0144] S65: When ΔA is greater than the tolerance, the adjustment mechanism is triggered. When ΔA≤1dB, the fine-tuning step is 0.1dB; when ΔA>3dB, the adjustment is switched to a large step size to achieve rapid convergence.

[0145] After step S6, step S7 is also set to scan the flaw detector within the preset dynamic range, synchronously collect horizontal linear error, vertical linear deviation and resolution threshold data, and generate a calibration report including a 3D error cloud map and a correction suggestion matrix.

[0146] The specific process of step S7 is as follows:

[0147] S71: Generate a pulse sequence covering a dynamic range greater than 80dB through the preset DDS signal generator, triggering the flaw detector transmitter to output a high-voltage pulse signal;

[0148] S72: Use the combination of standard attenuator and fixed attenuator to perform attenuation control, and simultaneously record the signal amplitude changes at the flaw detector receiving end to verify whether the dynamic range lower limit threshold meets the preset signal-to-noise ratio requirement;

[0149] S73: Output six groups of equally spaced modulation waves through the synchronous signal generator, adjust the flaw detector scanning range to the specified gear, collect the deviation value between each pulse leading edge and the theoretical scale, and calculate the maximum offset Δ L max , calculate the horizontal linear error. The specific calculation formula of the horizontal linear error is as follows:

[0150] Δ L = (Δ L max / full scale)×100%;

[0151] S74: Adjust the signal amplitude in a specified step within the dynamic range, collect the peak height on the flaw detector display, fit the deviation curve between the measured amplitude and the theoretical value, and calculate the vertical linear error;

[0152] S75: Generate a double-reflector simulation signal with a specified spacing, gradually reduce the signal amplitude until the flaw detector cannot distinguish the double peaks, and record the minimum calculable amplitude at this time as the resolution threshold;

[0153] S76: Map the three parameters of dynamic range, horizontal position, and vertical linear deviation to a three-dimensional coordinate system, generate a continuous error distribution cloud map through an interpolation algorithm, locate the abnormal parameter interval based on the error cloud map, call the preset compensation algorithm library, and output a targeted correction parameter matrix and a calibration report of the correction suggestion matrix.

[0154] Example 5

[0155] In another implementation of Example 1, the flaw detector calibration device is preheated for more than 15 minutes after being turned on, the signal generator in the device is monitored, the output end is measured with an oscilloscope, the frequency is effectively displayed on the oscilloscope, and the frequency knob of the device is adjusted from the lowest phase to the highest phase continuously, paying attention to its continuity, and recording the frequencies displayed at the lowest and highest ends, which is the frequency range.

[0156] Frequency stability is calculated as follows: ,

[0157] Where: f w is the frequency stability, f 1 is the receiver display frequency measured after 15 minutes, in MHz; f 0 is the display frequency of the measurement receiver, in MHz;

[0158] Power on the device and allow it to warm up for at least 15 minutes. Connect its output port to the input port of a measuring receiver. Select a frequency and record the frequency displayed by the measuring receiver and the device. Calculate the frequency accuracy. Connect the device input port to the output port of an ultrasonic flaw detector. Then, connect the device output signal to an oscilloscope. Adjust the oscilloscope's voltage and time measurement ranges to the appropriate ranges. Use the oscilloscope to measure the number of sine waves within each envelope. Observe the attenuator settings for each setting. The sum of the attenuation for each setting is the attenuation range, expressed in dB. Observe and record the stepping of each setting for the device, expressed in dB.

[0159] Turn on the device and preheat for more than 15 minutes. Set the frequency and voltage values ​​on the device interface, set the corresponding frequency on the measurement receiver, connect the device output port to the measurement receiver input port, and after the receiver reading stabilizes, calculate the signal voltage using the following formula:

[0160] ;

[0161] in, a is the power value measured by the receiver, in dBm, and U is the peak value of the signal voltage, in V.

[0162] Turn on the device and preheat for more than 15 minutes, set a sine wave output amplitude point, then set the reference frequency point, set the corresponding frequency on the measuring receiver, connect the device output port to the measuring receiver input port, read the power value A0, evenly select the specified frequency point within the device output frequency range, change the output frequency, and read the power value A0 at each frequency point. i , the sine wave flatness is calculated by the following formula:

[0163] ;

[0164] Among them, A0 is the reference frequency power value, unit is dBm, A i is the power value at each frequency point, in dBm. is the flatness of the sine wave within this frequency range.

[0165] After the device is turned on and preheated for more than 15 minutes, connect the output port of the device to the input port of the spectrum analyzer, set the signal amplitude to the maximum, and set the output frequency f 1. The spectrum analyzer displays the peak level of the output frequency fundamental wave. Read the measured value and record it in the corresponding table.

[0166] Change the spectrum analyzer center frequency to f 2, f 3. f 4. f 5 ( f 2, f 3. f 4. f 5 means 2 times f 1,3 times f 1,4 times f 1.5 times f 1) Read the measured values ​​of harmonic levels in sequence.

[0167] The total harmonic distortion (%) of a sine wave is calculated using the following formula:

[0168] ;

[0169] in, L 1 is the fundamental peak level, in dBm, L 2, L 3. L 4. L 5 are the 2nd, 3rd, 4th and 5th harmonic peak levels, in dBm; THD It is the total harmonic distortion, unit is %.

[0170] In summary, the calibration method for an ultrasonic flaw detector provided by the present invention obtains and emits an ultrasonic wave of a theoretical frequency matching the material to be detected, obtains the actual propagation path of the ultrasonic wave and performs real-time dynamic correction, generates calibration parameters for the ultrasonic velocity-delay compensation coefficient, compensates the emitted ultrasonic wave based on the calibration parameters, and after the compensated ultrasonic wave enters the material to be detected, collects a time baseline signal and inputs it into a trained deviation calculation model to calculate the time axis scale deviation. The time baseline linearity is adjusted based on the time axis scale deviation to generate a stepped ultrasonic attenuation signal, collects the amplitude value of the flaw detector echo signal, compares the echo signal amplitude value with a preset amplitude threshold to obtain a dynamic difference between the two, and adjusts the attenuation amount in real time based on the dynamic difference. This method solves the problems of insufficient calibration efficiency and consistency, poor dynamic parameter adaptability, limited sensitivity calibration accuracy, and complex working condition simulation capabilities in the prior art.

Claims

1. A calibration method for an ultrasonic flaw detector verification device, characterized in that: The following steps are involved: S1: Detect the specified material parameters of the material to be tested, call the preset material parameter-ultrasonic database, and obtain the ultrasonic wave with the theoretical frequency matching the material to be tested; S2: transmitting the ultrasonic wave of the theoretical frequency to the material to be detected, and obtaining the actual propagation path of the ultrasonic wave of the theoretical frequency in the material to be detected; S3: Performing real-time dynamic correction on the actual propagation path based on a preset algorithm to generate calibration parameters of the ultrasonic velocity-delay compensation coefficient; S4: compensating the transmitted ultrasonic wave based on the calibration parameters of the ultrasonic wave velocity-delay compensation coefficient; S5: After the compensated ultrasonic wave enters the material to be inspected, a time baseline signal output by the flaw detector is collected, the time baseline signal is input into the trained deviation calculation model, the deviation calculation model calculates and outputs the time axis scale deviation, and the time baseline linearity is adjusted based on the time axis scale deviation; S6: generating a stepped ultrasonic attenuation signal, collecting the amplitude value of the flaw detector echo signal, comparing the amplitude value of the echo signal with a preset amplitude threshold to obtain a dynamic difference between the two, and adjusting the attenuation amount in real time based on the dynamic difference; In step S3, the specific process of performing real-time dynamic correction on the actual propagation path based on a preset algorithm to generate calibration parameters of the ultrasonic velocity-delay compensation coefficient is as follows: S31: Use the dual probes to move at a constant speed along the surface of the material to be tested, collect multiple sets of ultrasonic transmission time data, synchronously record the ambient temperature, and compensate for the effect of temperature drift on ultrasonic velocity. The specific compensation formula is as follows: D v = α ( T - T 0) v 0; in, α is the material temperature drift coefficient, Δ v is the ultrasonic velocity compensation, v 0 is the theoretical ultrasonic speed, T is the recorded ambient temperature, T 0 is the standard ambient temperature; S32: Construct the propagation time-path length function based on the least squares method. The specific formula is as follows: t i = ( L i +ε i ) / v real +δ ； in, t i is the propagation time, L i is the theoretical path length, ε i is the path deviation, v real is the actual ultrasonic speed, δ System delay; S33: generating a dynamically corrected ultrasonic velocity compensation coefficient based on the propagation time-path length function, and generating a compensation coefficient matrix based on different detection directions; After step S6, step S7 is also set to scan the flaw detector within the preset dynamic range, synchronously collect horizontal linear error, vertical linear deviation, and resolution threshold data, and generate a calibration report containing a 3D error cloud map and a correction suggestion matrix. The specific process is as follows: S71: Generate a pulse sequence covering a dynamic range greater than 80dB through the preset DDS signal generator, triggering the flaw detector transmitter to output a high-voltage pulse signal; S72: Use the combination of standard attenuator and fixed attenuator to perform attenuation control, and simultaneously record the signal amplitude changes at the flaw detector receiving end to verify whether the dynamic range lower limit threshold meets the preset signal-to-noise ratio requirement; S73: Output six groups of equally spaced modulation waves through the synchronous signal generator, adjust the flaw detector scanning range to the specified gear, collect the deviation value between each pulse leading edge and the theoretical scale, and calculate the maximum offset Δ L max , calculate the horizontal linear error. The specific calculation formula of the horizontal linear error is as follows: Δ L =(Δ L max / full scale) × 100%; S74: Adjust the signal amplitude in a specified step within the dynamic range, collect the peak height on the flaw detector display, fit the deviation curve between the measured amplitude and the theoretical value, and calculate the vertical linear error; S75: Generate a double-reflector simulation signal with a specified spacing, gradually reduce the signal amplitude until the flaw detector cannot distinguish the double peaks, and record the minimum calculable amplitude at this time as the resolution threshold; S76: Map the three parameters of dynamic range, horizontal position, and vertical linear deviation to a three-dimensional coordinate system, generate a continuous error distribution cloud map through an interpolation algorithm, locate the abnormal parameter interval based on the error cloud map, call the preset compensation algorithm library, and output a targeted correction parameter matrix and a calibration report of the correction suggestion matrix.

2. The calibration method of an ultrasonic flaw detector verification device according to claim 1, characterized in that: The specific process of transmitting the ultrasonic wave of the theoretical frequency and obtaining the actual propagation path of the ultrasonic wave of the theoretical frequency in the material to be detected in step S2 is as follows: S21: Based on the theoretical frequency, applying a high-frequency alternating voltage through the piezoelectric crystal to convert electrical energy into mechanical vibration to generate an ultrasonic pulse sequence; S22: Use acoustic lens or phased array technology to adjust the beam diffusion angle to ensure that the ultrasonic energy is concentrated on the surface of the material to be tested and suppress stray wave interference; S23: constructing a theoretical propagation path model according to the ultrasonic wave velocity corresponding to the material to be tested, including direct incidence, refraction, and reflection path branches; S24: using the array probe group to synchronously receive the reflected wave, scattered wave, and transmitted wave, and record the time domain signals and amplitude information of each channel of the reflected wave, scattered wave, and transmitted wave; S25: Perform short-time Fourier transform on the received reflected waves, scattered waves, and transmitted waves to extract the arrival time difference Δt of the reflected waves on each path, and calculate the actual propagation distance based on the speed of sound; S26: Based on the preset nonlinear regression algorithm, optimize the residual between the theoretical path length and the actual measurement value, and iteratively solve the path correction coefficient ε ; S27: Map the corrected path data to a three-dimensional coordinate system to generate an actual propagation cloud map including the main sound beam, side lobes, and scattering paths, and mark key reflection points.

3. The calibration method of an ultrasonic flaw detector verification device according to claim 1, characterized in that: The training data used when training the deviation calculation model includes: The mapping relationship between the baseline waveform and the error label of the specified group in the historical calibration database; Real-time acquisition of flaw detector noise spectrum characteristics and ambient temperature-humidity interference parameters; The output of the deviation calculation model is the time baseline scale correction amount and the confidence assessment value; When the confidence level of the deviation calculation model output is less than the preset value, the manual review mechanism is triggered.

4. The calibration method of an ultrasonic flaw detector verification device according to claim 3, characterized in that: The specific process of step S5 is as follows: S51: Capture the time base signal output by the flaw detector in real time, and record the probe delay, ambient temperature and material sound velocity parameters to form a composite data stream of time-amplitude-temperature; S52: performing band-pass filtering and amplitude normalization on the original time baseline signal to eliminate probe contact noise and power frequency interference, and generating a standardized input vector for processing by the deviation calculation model; S53: The deviation calculation model calculates the time axis scale deviation, and dynamically adjusts the time base linearity according to the time axis scale deviation. The integrated PID controller adjusts the DSP clock division coefficient in real time to ensure that the time base linearity error converges to a preset range.

5. The calibration method of an ultrasonic flaw detector verification device according to claim 4, characterized in that: The deviation calculation model is a convolutional neural network including an input layer, a convolutional layer stacking layer, a feature compression layer and an output layer; The input layer receives a 1×1000-dimensional time baseline signal, performs amplitude normalization processing based on the maximum amplitude of the time baseline signal, and eliminates high-frequency interference through sliding window mean filtering. The time baseline signal is a time series signal; The convolutional layer stacking layer includes a first convolutional layer, a second convolutional layer, a third convolutional layer, a fourth convolutional layer, and a fifth convolutional layer. The step size of each convolutional layer is 1, and convolution filling is performed by adding a preset value to the edge of the input data; Among them, the first convolution layer is used to extract local temporal features, the second convolution layer is used to enhance the sensitivity of high-frequency components, the third convolution layer is used to capture long-range temporal dependencies, the fourth convolution layer is used to establish multi-scale feature representation, and the fifth convolution layer is used to map features to deviation physical quantity dimensions; The feature compression layer includes an activation function and a global average pooling layer. The activation function is used to enhance the nonlinear expression capability of the model. The global average pooling layer is used to add a 1×990 global average pooling layer after the fifth convolutional layer to compress the features into a 1×1×1 output to achieve temporal feature aggregation; The output layer includes a regression mapping layer and a range adaptation layer. The regression mapping layer outputs a deviation value ΔT and uses a linear activation function for unconstrained numerical mapping. The range adaptation layer standardizes the label value during retraining and maps it to a specified range.

6. The calibration method of an ultrasonic flaw detector verification device according to claim 1, characterized in that: The specific process in step S6 is as follows: S61: Generate an ultrasonic excitation pulse and apply a stepped attenuation through the attenuator; S62: Collecting the flaw detector echo signal through the ADC module; S63: Capture the echo signal waveform through an oscilloscope and measure the peak amplitude; S64: Based on the preset amplitude threshold of the material characteristics, a dynamic difference ΔA between the actual echo amplitude measurement value and the threshold is calculated, where ΔA = |actual echo amplitude - echo amplitude threshold|; S65: When ΔA is greater than the tolerance, the adjustment mechanism is triggered. When ΔA ≤ 1dB, the fine-tuning step is 0.1dB. When ΔA > 3dB, the adjustment is switched to a larger step size for rapid convergence.

Images