Steel seamless gas cylinder inner surface defect detection method and system based on digital twinning

Abstract

The present application discloses a seamless steel gas cylinder inner surface defect detection method and system based on digital twinning, relates to the technical field of defect detection, and realizes full coverage, high precision and self-adaptive online detection of small defects on the inner surface of a thin-walled seamless steel gas cylinder under real service conditions for the first time. By deeply integrating high-frequency guided wave physical perception and digital twinning virtual model and introducing a multi-stage signal quality closed-loop control mechanism, the present application not only breaks through the dependence of traditional nondestructive testing on shutdown, surface treatment and fixed period, but also significantly improves the reliability and engineering practicability of defect identification, thereby providing a popularized technical paradigm for the safe and intelligent operation and maintenance of pressure-bearing special equipment.

Description

Technical Field

[0001] This invention relates to the field of defect detection technology, and in particular to a method and system for detecting defects on the inner surface of seamless steel gas cylinders based on digital twins. Background Technology

[0002] Seamless steel gas cylinders are widely used in industrial and medical fields, and their safety is paramount. During long-term use, the inner wall of gas cylinders is prone to defects such as corrosion pits and microcracks due to residual moisture or corrosive media. These defects are difficult to detect but develop rapidly and are a major cause of cylinder failure or even explosion. Currently, safety inspections of gas cylinders mainly rely on periodic inspections every few years. This method is time-consuming and cannot provide real-time monitoring of the cylinder's health. Moreover, conventional hydrostatic tests only reflect the overall strength of the cylinder and are not sensitive to small, localized defects; wall thickness measurements can only detect a limited number of points and cannot cover the entire inner surface.

[0003] Existing nondestructive testing methods also have their limitations. Magnetic particle and penetrant testing can only examine the outer surface and cannot reach the inner wall. Conventional ultrasonic testing on thin-walled gas cylinders is difficult to distinguish between defect waves and bottom waves due to near-field interference, and it requires surface grinding before testing, which is inconvenient. X-ray testing carries radiation risks and is not suitable for daily use. Acoustic emission testing requires additional pressure on the gas cylinder and cannot be performed under normal operating conditions. Although low-frequency ultrasonic guided waves can detect over longer distances, their long wavelengths are insufficient to distinguish minute corrosion or cracks. While high-frequency guided wave technology is more sensitive to minute defects, under actual working conditions of the gas cylinder (with media inside and valves not disassembled), the received signal is very complex and mixed with various interferences, making it difficult for traditional signal processing methods to effectively extract useful defect information.

[0004] Digital twin technology offers a new approach to equipment inspection, but existing research mostly remains at the level of simulation or data visualization, and a complete inspection method that can closely integrate guided wave detection data with the actual state of the gas cylinder has not yet been developed. In particular, there is a lack of a systematic solution that can perform multi-level processing and quality assessment of the raw signal under complex operating conditions, and ultimately achieve accurate imaging on the digital model.

[0005] Therefore, there is an urgent need for a technical solution based on digital twins for detecting defects on the inner surface of seamless steel gas cylinders. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a method for detecting defects on the inner surface of seamless steel gas cylinders based on digital twins, specifically including the following steps: S1. Under the condition that the thin-walled steel seamless gas cylinder is in online use, the cylinder valve is not disassembled, and the cylinder is filled with liquefied gas or compressed gas medium, a high-frequency guided wave sensor array installed on the outer wall of the thin-walled steel seamless gas cylinder emits modal guided waves to the cylinder body of the thin-walled steel seamless gas cylinder, and after the modal guided waves are reflected or scattered by the inner surface of the thin-walled steel seamless gas cylinder, the time domain waveform data and spatial coordinates of each high-frequency guided wave sensor are recorded. S2. Obtain the geometric parameters and real-time service parameters of the current thin-walled steel seamless gas cylinder, and construct a three-dimensional geometric solid model of the current thin-walled steel seamless gas cylinder based on the geometric parameters. Register the spatial coordinates of each high-frequency guided wave sensor to the outer surface of the three-dimensional geometric solid model of the current thin-walled steel seamless gas cylinder, and generate a digital twin of the current thin-walled steel seamless gas cylinder by combining the real-time service parameters of the current thin-walled steel seamless gas cylinder. S3. Perform phased correction on the time-domain waveform data of each high-frequency guided wave sensor, and determine the correction result of each phase. Based on the determination result, obtain the updated time-domain waveform data of each high-frequency guided wave sensor. S31: Based on multi-scale time-frequency joint characterization, the time-domain waveform data of each high-frequency guided wave sensor is corrected in the first stage to generate the first corrected waveform data of each high-frequency guided wave sensor. S311: Perform continuous wavelet transform on the time-domain waveform data of the current high-frequency guided wave sensor to obtain a complex coefficient matrix at several scales; S312: Based on the complex coefficient matrix, calculate the instantaneous amplitude sequence and instantaneous phase sequence for each scale, forming several data pairs consisting of the instantaneous amplitude sequence and the instantaneous phase sequence; S313: Based on the preset guided wave main mode frequency range, select an effective scale subset from several data pairs, and perform fusion processing on the instantaneous amplitude sequence in the effective scale subset. At the same time, perform fusion processing on the instantaneous phase sequence in the effective scale subset to obtain the fused instantaneous amplitude sequence and the fused instantaneous phase sequence, respectively. S314: The time-domain waveform data of the current high-frequency guided wave sensor is reconstructed using the fused instantaneous amplitude sequence and the fused instantaneous phase sequence to generate the first corrected waveform data of the current high-frequency guided wave sensor; S32: Obtain the guided wave mode energy concentration range of the first corrected waveform data of the current high-frequency guided wave sensor, and obtain the amplitude stability index based on the guided wave mode energy concentration range. Determine whether the first corrected waveform data meets the preset modal purity threshold. If it meets the threshold, use the first corrected waveform data of the current high-frequency guided wave sensor as the updated time domain waveform data. If it does not meet the threshold, mark the first corrected waveform data of the current high-frequency guided wave sensor as the first waveform data to be corrected, and proceed to S33. S321: Perform a Fourier transform on the first corrected waveform data of the current high-frequency guided wave sensor to obtain the amplitude spectrum of the first corrected waveform data; S322: Obtain the wall thickness parameters based on the geometric parameters of the current thin-walled seamless steel gas cylinder, and at the same time obtain the modal theory dispersion relationship. Combine the wall thickness parameters and the modal theory dispersion relationship to calculate the center frequency of the guided wave main mode, and set the initial bandwidth with the center frequency of the guided wave main mode as the center to form a candidate energy concentration interval. S323: Search for the local maximum point of the amplitude spectrum of the first corrected waveform data within the candidate energy concentration interval, and expand the bandwidth by a preset ratio with the local maximum point as the center to obtain the guided wave mode energy concentration interval; S324: Extract the time-domain envelope sequence of the first corrected waveform data within the energy concentration range of the guided wave mode, and calculate the ratio of the amplitude standard deviation to the mean of the time-domain envelope sequence as an amplitude stability index; S325: Obtain the preset modal purity threshold; S326: Compare the amplitude stability index with the modal purity threshold. If the amplitude stability index is less than or equal to the modal purity threshold, the first corrected waveform data is determined to meet the modal purity requirements, and the current first corrected waveform data of the high-frequency guided wave sensor is used as the updated time-domain waveform data. If the amplitude stability index is greater than the modal purity threshold, the first corrected waveform data is determined to not meet the modal purity requirements, the current first corrected waveform data of the high-frequency guided wave sensor is marked as the first waveform data to be corrected, and proceed to S33. S33: Based on the spatial-phase consistency constraint, perform a second-stage correction on the first waveform data to be corrected of the current high-frequency guided wave sensor to generate the second corrected waveform data of the current high-frequency guided wave sensor. S331: Calculate the geometric distance between the current high-frequency guided wave sensor and the adjacent high-frequency guided wave sensor based on the spatial coordinates of the current high-frequency guided wave sensor and the spatial coordinates of the adjacent high-frequency guided wave sensor. S332: Calculate the theoretical time delay difference based on geometric distance and the propagation speed of guided waves in thin-walled seamless steel cylinders; S333: Perform Hilbert transform on the first waveform data to be corrected of the current high-frequency guided wave sensor and the first waveform data to be corrected of the adjacent high-frequency guided wave sensor respectively to obtain their respective analytical signals, and extract the instantaneous phase sequence of the current high-frequency guided wave sensor and the instantaneous phase sequence of the adjacent high-frequency guided wave sensor from their respective analytical signals. S334: Perform phase compensation on the instantaneous phase sequences of adjacent high-frequency guided wave sensors based on the theoretical time delay difference to obtain the compensated adjacent instantaneous phase sequences; S335: Calculate the phase residual between the current instantaneous phase sequence of the high-frequency guided wave sensor and the compensated adjacent instantaneous phase sequence, and average the phase residual to obtain the spatial-phase consistency error; S336: Perform phase correction on the first waveform data to be corrected of the current high-frequency guided wave sensor based on the spatial-phase consistency error, and generate the second corrected waveform data of the current high-frequency guided wave sensor; S34: Obtain the cross-correlation peak offset between the second corrected waveform data of the current high-frequency guided wave sensor and the second corrected waveform data of the adjacent high-frequency guided wave sensor, and determine whether the second corrected waveform data meets the preset channel synchronization tolerance threshold. If it meets the threshold, the second corrected waveform data of the current high-frequency guided wave sensor is used as the updated time domain waveform data. If it does not meet the threshold, the second corrected waveform data of the current high-frequency guided wave sensor is marked as the second waveform data to be corrected, and proceed to S35. S341: Perform cross-correlation calculation on the second corrected waveform data of the current high-frequency guided wave sensor and the second corrected waveform data of the adjacent high-frequency guided wave sensor to obtain the cross-correlation function; S342: Determine the time index of the sampling point with the largest amplitude in the cross-correlation function, and use the difference between the time index of the sampling point with the largest amplitude and the time index of the zero-delay position as the cross-correlation peak offset; S343: Calculate the main period of the signal based on the center frequency of the main mode of the guided wave, and use the preset proportion of the main period of the signal as the theoretical allowable synchronization offset; S344: Obtain the preset channel synchronization tolerance threshold; S345: Compare the absolute values ​​of the cross-correlation peak offset and the channel synchronization tolerance threshold. If the absolute value of the cross-correlation peak offset is less than or equal to the channel synchronization tolerance threshold, the second corrected waveform data is determined to meet the channel synchronization requirements, and the current second corrected waveform data of the high-frequency guided wave sensor is used as the updated time-domain waveform data. If the absolute value of the cross-correlation peak offset is greater than the channel synchronization tolerance threshold, the second corrected waveform data is determined to not meet the channel synchronization requirements, the current second corrected waveform data of the high-frequency guided wave sensor is marked as the second waveform data to be corrected, and proceed to S35. S35: Based on sparse dictionary learning and reconstruction, perform a third-stage correction on the second waveform data to be corrected of the current high-frequency guided wave sensor to generate the third corrected waveform data of the current high-frequency guided wave sensor. S351: The second waveform data to be corrected from the current high-frequency guided wave sensor is segmented to obtain several time-domain signal blocks; S352: Construct an overcomplete dictionary, wherein the overcomplete dictionary contains several atoms, each of which is a waveguide basis function with different frequencies and attenuation characteristics; S353: For each time-domain signal block, solve for the sparse coefficient vector in an overcomplete dictionary; S354: Reconstruct each time-domain signal block by linear combination of sparse coefficient vector and overcomplete dictionary to obtain each reconstructed signal block; S355: Concatenate all reconstructed signal blocks according to the original timing sequence to generate the third corrected waveform data of the current high-frequency guided wave sensor; S36: Obtain the residual noise entropy value of the third corrected waveform data of the current high-frequency guided wave sensor in the defect sensitive frequency band, and determine whether the third corrected waveform data meets the preset imaging availability criterion threshold. If it meets the threshold, the third corrected waveform data of the current high-frequency guided wave sensor is used as the updated time-domain waveform data. If it does not meet the threshold, the third corrected waveform data of the current high-frequency guided wave sensor is marked as the third waveform data to be corrected, and the iteration termination judgment condition is executed. The third waveform data to be corrected is used as input to re-enter S31 until the updated time-domain waveform data of the current high-frequency guided wave sensor is obtained. S361: Set the defect-sensitive frequency band according to the center frequency of the main waveguide mode; S362: Bandpass filtering is performed on the third corrected waveform data of the current high-frequency guided wave sensor to retain the signal components within the defect-sensitive frequency band; S363: Perform wavelet packet decomposition on the signal components within the defect-sensitive frequency band to obtain several sub-band coefficients; S364: Calculate the Shannon entropy of each sub-band coefficient and average the Shannon entropy of all sub-band coefficients to obtain the residual noise entropy value; S365: Obtain the preset imaging availability criterion threshold; S366: Compare the residual noise entropy value with the imaging availability criterion threshold. If the residual noise entropy value is less than or equal to the imaging availability criterion threshold, the third corrected waveform data is determined to meet the imaging availability requirements, and the current third corrected waveform data of the high-frequency guided wave sensor is used as the updated time-domain waveform data. If the residual noise entropy value is greater than the imaging availability criterion threshold, the third corrected waveform data is determined to not meet the imaging availability requirements, the current third corrected waveform data of the high-frequency guided wave sensor is marked as the third waveform data to be corrected, and the current iteration number is checked to see if it is less than the preset maximum iteration number. If so, the third waveform data to be corrected is used as input to re-enter S31 until the updated time-domain waveform data of the current high-frequency guided wave sensor is obtained. If not, the current third corrected waveform data is used as the updated time-domain waveform data.

[0007] S4. Map the updated time-domain waveform data of each high-frequency guided wave sensor to the inner surface grid nodes of the digital twin of the thin-walled steel seamless gas cylinder. Calculate the scattering energy distribution of the inner surface grid nodes using the inverse time-shift imaging algorithm to obtain the current set of candidate regions for inner surface defects of the thin-walled steel seamless gas cylinder. S5. Based on the candidate region set of internal surface defects, identify the location and geometric contour of corrosion pits or cracks on the internal surface of the current thin-walled steel seamless gas cylinder.

[0008] This invention also discloses a digital twin-based system for detecting defects on the inner surface of seamless steel gas cylinders, which includes the following modules: Transmission module: Used to transmit modal guided waves to the cylinder of thin-walled seamless steel gas cylinder when the cylinder is in online use, the cylinder valve is not disassembled, and the cylinder is filled with liquefied gas or compressed gas medium. The module records the time-domain waveform data and spatial coordinates of each high-frequency guided wave sensor after the modal guided waves are reflected or scattered by the inner surface of the thin-walled seamless steel gas cylinder. Data acquisition module: used to acquire the geometric parameters and real-time service parameters of the current thin-walled steel seamless gas cylinder, and construct a three-dimensional geometric solid model of the current thin-walled steel seamless gas cylinder based on the geometric parameters. The spatial coordinates of each high-frequency guided wave sensor are registered to the outer surface of the three-dimensional geometric solid model of the current thin-walled steel seamless gas cylinder. Combined with the real-time service parameters of the current thin-walled steel seamless gas cylinder, a digital twin of the current thin-walled steel seamless gas cylinder is generated. Correction module: used to perform phased correction on the time-domain waveform data of each high-frequency guided wave sensor, and to judge the correction result of each phase, and to obtain the updated time-domain waveform data of each high-frequency guided wave sensor based on the judgment result. Region partitioning module: used to map the updated time-domain waveform data of each high-frequency guided wave sensor to the inner surface grid nodes of the digital twin of the thin-walled steel seamless gas cylinder, and calculate the scattering energy distribution of the inner surface grid nodes through the inverse time-off imaging algorithm to obtain the current inner surface defect candidate region set of the thin-walled steel seamless gas cylinder. Identification module: Based on the candidate region set of internal surface defects, it identifies the location and geometric contour of corrosion pits or cracks on the internal surface of the current thin-walled steel seamless gas cylinder.

[0009] The embodiments of the present invention have the following technical effects: This invention effectively solves the problems encountered by existing detection methods in practical applications through a technical process of high-frequency guided wave detection, multi-level signal correction, and digital model imaging. Addressing the issue that traditional periodic inspections cannot monitor in real time, this invention can directly detect under normal gas cylinder operating conditions without stopping the machine or disassembling valves. To address the complexity and interference of high-frequency guided wave signals, this invention designs a three-stage, progressively advancing signal correction and judgment process. First, multi-scale time-frequency analysis is performed on the original signal to screen out signal components related to the target guided wave mode and fuse and reconstruct them to obtain the first corrected waveform. Then, its amplitude stability index is calculated and compared with a preset modal purity threshold. If the signal is sufficiently pure, it is directly used for imaging; otherwise, it proceeds to the second stage. The second stage utilizes the spatial relationship between sensors to calculate the theoretical time delay difference and performs phase compensation and correction on signals from adjacent channels to solve the signal asynchrony problem. After correction, the cross-correlation peak offset is calculated and compared with a preset channel synchronization tolerance threshold. If the synchronization requirement is met, it is used as a usable signal; otherwise, it proceeds to the third stage. The third stage employs a sparse dictionary learning method to deeply reconstruct the signal, suppressing residual noise. After reconstruction, the residual noise entropy value of the signal within the defect-sensitive frequency band is calculated and compared with a preset imaging usability criterion threshold. If the signal quality meets the standard, it is used for imaging; if not, the signal is sent back to the first stage to begin a new round of correction loops until a usable signal is obtained. Finally, all signals that pass the quality judgment are mapped onto the digital twin model of the gas cylinder, and the defect location and contour on the inner surface are accurately calculated using a reverse time-shift imaging algorithm.

[0010] In summary, this invention achieves for the first time a comprehensive, high-precision, and adaptive online detection of minute defects on the inner surface of thin-walled seamless steel gas cylinders under real-world service conditions. By deeply integrating high-frequency guided wave physical sensing with a digital twin virtual model and introducing a multi-stage signal quality closed-loop control mechanism, it not only overcomes the dependence of traditional non-destructive testing on downtime, surface treatment, and fixed cycles, but also significantly improves the reliability and engineering practicality of defect identification, providing a scalable technical paradigm for the safe and intelligent operation and maintenance of pressure-bearing special equipment. Attached Figure Description

[0011] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0012] Figure 1This is a flowchart of a method for detecting defects on the inner surface of a seamless steel gas cylinder based on digital twins, provided in an embodiment of the present invention. Figure 2 This is a structural block diagram of the steel seamless gas cylinder inner surface defect detection system based on digital twin provided in an embodiment of the present invention; Figure 3 This is a cross-sectional schematic diagram of the installation position of one of the high-frequency guided wave sensors provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the original time-domain waveform data of one of the high-frequency guided wave sensors provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the updated time-domain waveform data of one of the high-frequency guided wave sensors provided in an embodiment of the present invention; Figure 6 This is a scattering energy distribution map presented using the original time-domain waveform data, provided by an embodiment of the present invention; Figure 7 This is a scattering energy distribution map presented by an embodiment of the present invention when using updated time-domain waveform data. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0014] Example 1: As Figure 1 As shown, the present invention provides a method for detecting defects on the inner surface of seamless steel gas cylinders based on digital twins, comprising the following steps: S1. Under the condition that the thin-walled steel seamless gas cylinder is in online use, the cylinder valve is not disassembled, and the cylinder is filled with liquefied gas or compressed gas medium, a high-frequency guided wave sensor array installed on the outer wall of the thin-walled steel seamless gas cylinder emits modal guided waves to the cylinder body of the thin-walled steel seamless gas cylinder, and after the modal guided waves are reflected or scattered by the inner surface of the thin-walled steel seamless gas cylinder, the time domain waveform data and spatial coordinates of each high-frequency guided wave sensor are recorded. It is worth noting that in this embodiment, the high-frequency guided wave sensor array is first fixed to the outer wall of a thin-walled seamless steel gas cylinder, such as... Figure 3As shown. To obtain the spatial coordinates of each sensor, a laser tracker or 3D scanner is used to scan the outer surface of the gas cylinder before installation to create a point cloud model of the outer surface of the gas cylinder. Then, the position of the center of each sensor is located on the point cloud model, and their height relative to the reference plane at the bottom of the gas cylinder and their angular position around the axis of the cylinder are recorded, thereby determining the precise three-dimensional spatial coordinates of each sensor; For acquiring time-domain waveform data, a signal generator sends pulsed electrical signals of specific frequency and amplitude to the excitation sensor, exciting modal guided waves to propagate within the gas cylinder. When the guided waves encounter defects or boundaries on the inner surface of the gas cylinder and reflect back, the receiving sensor converts the mechanical vibration into electrical signals. These analog electrical signals are amplified by a preamplifier and then digitized by a data acquisition card at a high sampling rate to obtain discrete voltage-time series data, such as... Figure 4 As shown. Finally, the series of voltage values ​​recorded by each sensor over time are saved as a time-domain waveform data file, and stored in correspondence with the previously acquired spatial coordinates for subsequent processing.

[0015] S2. Obtain the geometric parameters and real-time service parameters of the current thin-walled steel seamless gas cylinder, and construct a three-dimensional geometric solid model of the current thin-walled steel seamless gas cylinder based on the geometric parameters. Register the spatial coordinates of each high-frequency guided wave sensor to the outer surface of the three-dimensional geometric solid model of the current thin-walled steel seamless gas cylinder, and generate a digital twin of the current thin-walled steel seamless gas cylinder by combining the real-time service parameters of the current thin-walled steel seamless gas cylinder. It is worth noting that, in this embodiment, the process of obtaining the geometric parameters and real-time service parameters of the current thin-walled seamless steel gas cylinder is as follows: First, by consulting the gas cylinder nameplate or factory certificate of conformity, the basic geometric dimensions of the gas cylinder, such as the standard outer diameter, nominal wall thickness, and total length, are recorded. Simultaneously, an ultrasonic thickness gauge is used to randomly select several measuring points at different locations on the gas cylinder for actual measurement to correct for actual wall thickness deviations caused by manufacturing tolerances or long-term service, thereby obtaining accurate actual geometric parameters. For real-time service parameters, the pressure and temperature sensors installed at the gas cylinder valve are used to directly read the current medium pressure and gas temperature inside the gas cylinder. These parameters reflect the current operating status of the gas cylinder. Subsequently, a three-dimensional geometric solid model of the current thin-walled steel seamless gas cylinder is constructed based on the determined actual geometric parameters. Specifically, using CAD software, the actual outer diameter, actual wall thickness, and total length of the gas cylinder are input to create a standard cylindrical shell three-dimensional model. If the gas cylinder has end caps at both ends, corresponding end structures are added according to the standard end cap type, such as hemispherical or elliptical, to ensure that the geometry of the model is completely consistent with the actual gas cylinder. Next, the spatial coordinates of each high-frequency guided wave sensor are registered to the outer surface of the three-dimensional geometric model. Since the height of each sensor relative to the reference plane at the bottom of the gas cylinder and its angle around the cylinder's axis were recorded in step S1, the same coordinate system is established on the three-dimensional model: with the center of the gas cylinder's bottom surface as the origin and the cylinder's axis as the Z-axis. The position of each sensor on the Z-axis is determined based on its height value, and its rotational position around the Z-axis is determined based on its angle value. This allows for precise positioning of the mounting point of each sensor on the outer surface of the model, and the sensor data is then associated with that point. Finally, a digital twin of the current thin-walled seamless steel gas cylinder is generated by combining its real-time service parameters. Specifically, based on the three-dimensional geometric solid model, material properties are assigned to the model, including the elastic modulus and density of the steel. Real-time pressure and temperature parameters are then bound to the model as boundary conditions or state variables. For example, the current pressure value is mapped to the inner surface of the model to simulate the effect of internal pressure on minute deformations of the wall thickness. In this way, a virtual model that contains accurate geometry and reflects the current physical state and sensor layout—the digital twin of the current thin-walled seamless steel gas cylinder—is constructed.

[0016] S3. Perform staged correction on the time-domain waveform data of each high-frequency guided wave sensor, and determine the correction result of each stage. Based on the determination result, obtain the updated time-domain waveform data of each high-frequency guided wave sensor, such as... Figure 5 As shown; S31: Based on multi-scale time-frequency joint characterization, the time-domain waveform data of each high-frequency guided wave sensor is corrected in the first stage to generate the first corrected waveform data of each high-frequency guided wave sensor. S311: Perform continuous wavelet transform on the time-domain waveform data of the current high-frequency guided wave sensor to obtain a complex coefficient matrix at several scales; S312: Based on the complex coefficient matrix, calculate the instantaneous amplitude sequence and instantaneous phase sequence for each scale, forming several data pairs consisting of the instantaneous amplitude sequence and the instantaneous phase sequence; S313: Based on the preset guided wave main mode frequency range, select an effective scale subset from several data pairs, and perform fusion processing on the instantaneous amplitude sequence in the effective scale subset. At the same time, perform fusion processing on the instantaneous phase sequence in the effective scale subset to obtain the fused instantaneous amplitude sequence and the fused instantaneous phase sequence, respectively. S314: The time-domain waveform data of the current high-frequency guided wave sensor is reconstructed using the fused instantaneous amplitude sequence and the fused instantaneous phase sequence to generate the first corrected waveform data of the current high-frequency guided wave sensor; It is worth noting that in this embodiment, the time-domain waveform data of the current high-frequency guided wave sensor is first subjected to continuous wavelet transform to obtain a complex coefficient matrix at several scales. Specifically, a suitable mother wavelet for thin-walled guided wave analysis is selected, such as the Morlet wavelet, which has good time-frequency localization characteristics. Using the time-domain waveform data as input, a set of discrete scale parameters from fine to coarse is set, such as an integer sequence from 1 to 64, and the wavelet coefficients at each scale are calculated sequentially. For each time point, different scales correspond to different frequency components, ultimately resulting in a two-dimensional matrix where rows represent time and columns represent scale. Each element in the matrix is ​​a complex number, containing the amplitude and phase information at that time and scale. Next, based on the obtained complex coefficient matrix, instantaneous amplitude and instantaneous phase sequences are calculated for each scale. For any given scale column, its complex coefficients can be represented as real and imaginary parts, or in polar coordinates. The instantaneous amplitude sequence is the magnitude (absolute value) of all complex coefficients at that scale, forming an amplitude curve that varies with time; the instantaneous phase sequence is the argument of all complex coefficients at that scale, forming a phase curve that varies with time. In this way, each scale corresponds to a pair of data: an instantaneous amplitude sequence and an instantaneous phase sequence. Several such data pairs constitute a detailed description of the original signal at different frequency components. Subsequently, based on the preset guided wavelet dominant mode frequency range, an effective scale subset is selected from several data pairs. Determining the preset guided wavelet dominant mode frequency range is an objective process. First, the material properties of the steel used in the gas cylinder and the actual geometric parameters obtained through step S2 are consulted. Using these parameters, the phase velocity and group velocity of the T(0,1) mode within the current wall thickness and frequency range are calculated using guided wave theory formulas, which can be Lamb wave or rod wave dispersion equations. The T(0,1) mode is a commonly used longitudinal stretching mode. Combined with the center frequency of the excitation sensor used, for example, 50kHz, the dominant propagating mode at that frequency and its corresponding effective frequency band are inferred. Typically, this frequency band is a certain bandwidth range above and below the center frequency, for example, ±20% of the center frequency. Then, each scale of the continuous wavelet transform is mapped back to its corresponding center frequency. Scales whose center frequencies fall within the aforementioned effective frequency band are identified, and the data pairs of these scales are retained to form the effective scale subset. Data pairs of other scales are considered noise or non-target mode interference and are discarded. The instantaneous amplitude sequences from the effective scale subset are fused, and the instantaneous phase sequences are fused simultaneously, yielding fused instantaneous amplitude sequences and fused instantaneous phase sequences, respectively. The specific fusion method is as follows: for each time point, the average instantaneous amplitude of all effective scales at that moment is calculated as the fused instantaneous amplitude; similarly, the average instantaneous phase of all effective scales at that moment is calculated as the fused instantaneous phase. This fusion method can smooth out random fluctuations caused by a single scale while preserving the energy characteristics and phase continuity of the main modes. Finally, the time-domain waveform data of the current high-frequency guided wave sensor is reconstructed using the fused instantaneous amplitude sequence and the fused instantaneous phase sequence to generate the first corrected waveform data of the current high-frequency guided wave sensor. The reconstruction process is an approximate implementation of the inverse transform: the fused instantaneous amplitude is used as the modulus of the new complex coefficients, and the fused instantaneous phase is used as the argument of the new complex coefficients to reconstruct an ideal complex coefficient sequence. Although the strictly defined inverse continuous wavelet transform requires integration across all scales, in this embodiment, the empirical mode decomposition idea related to the Hilbert-Huang transform or the method of directly constructing an analytic signal can be used to recombine this set of fused amplitudes and phases into an analytic signal, which is then restored to a time-domain signal through inverse Fourier transform or simple numerical integration. This generated new signal is the first corrected waveform data.

[0017] The main benefits of reconstruction lie in noise reduction and mode separation. The original time-domain waveform data is mixed with various modes of guided wave, electronic noise, and environmental interference. After decomposing the signal into the time-frequency domain through continuous wavelet transform, only the effective scale component related to the target guided wave mode T(0,1) is retained, while other irrelevant frequency components are eliminated. Then, through fusion and reconstruction, these purified components are recombined into a time-domain signal. The resulting first corrected waveform data has significantly reduced background noise and a clearer, more prominent target mode signal compared to the original data, providing high-quality basic data for subsequent judgment of signal purity and more refined phase correction.

[0018] S32: Obtain the guided wave mode energy concentration range of the first corrected waveform data of the current high-frequency guided wave sensor, and obtain the amplitude stability index based on the guided wave mode energy concentration range. Determine whether the first corrected waveform data meets the preset modal purity threshold. If it meets the threshold, use the first corrected waveform data of the current high-frequency guided wave sensor as the updated time domain waveform data. If it does not meet the threshold, mark the first corrected waveform data of the current high-frequency guided wave sensor as the first waveform data to be corrected, and proceed to S33. S321: Perform a Fourier transform on the first corrected waveform data of the current high-frequency guided wave sensor to obtain the amplitude spectrum of the first corrected waveform data; S322: Obtain the wall thickness parameters based on the geometric parameters of the current thin-walled steel seamless gas cylinder, and at the same time obtain the T(0,1) mode theoretical dispersion relationship. Combine the wall thickness parameters and the T(0,1) mode theoretical dispersion relationship to calculate the center frequency of the guided wave main mode, and set the initial bandwidth with the center frequency of the guided wave main mode as the center to form a candidate energy concentration interval. S323: Search for the local maximum point of the amplitude spectrum of the first corrected waveform data within the candidate energy concentration interval, and expand the bandwidth by a preset ratio with the local maximum point as the center to obtain the guided wave mode energy concentration interval; S324: Extract the time-domain envelope sequence of the first corrected waveform data within the energy concentration range of the guided wave mode, and calculate the ratio of the amplitude standard deviation to the mean of the time-domain envelope sequence as an amplitude stability index; S325: Obtain the preset modal purity threshold; S326: Compare the amplitude stability index with the modal purity threshold. If the amplitude stability index is less than or equal to the modal purity threshold, the first corrected waveform data is determined to meet the modal purity requirements, and the current first corrected waveform data of the high-frequency guided wave sensor is used as the updated time-domain waveform data. If the amplitude stability index is greater than the modal purity threshold, the first corrected waveform data is determined to not meet the modal purity requirements, the current first corrected waveform data of the high-frequency guided wave sensor is marked as the first waveform data to be corrected, and proceed to S33. In this embodiment, the first corrected waveform data of the current high-frequency guided wave sensor is subjected to a Fourier transform to obtain its amplitude spectrum. Specifically, a valid time-domain signal segment is selected from the first corrected waveform data, typically the stable time period after the guided wave arrives. A Fast Fourier Transform (FFT) algorithm is applied to convert this segment from the time domain to the frequency domain. The modulus of the complex spectral value corresponding to each frequency point is calculated to obtain the energy level at that frequency, thus forming an amplitude spectrum curve that varies with frequency. This curve visually illustrates the energy distribution of the signal at different frequencies.

[0019] Next, to determine the candidate energy concentration range, it is necessary to obtain the wall thickness parameters by combining the geometric parameters of the current thin-walled seamless steel cylinder, and to obtain the T(0,1) modal theoretical dispersion relation. Regarding the wall thickness parameters, the average or representative wall thickness values ​​of the actual wall thickness at various locations of the cylinder obtained from measurements in step S2 are used. The T(0,1) modal theoretical dispersion relation can be obtained by consulting the standard formula in the acoustics handbook, or by using finite element simulation software to establish a simplified plate model for calculation. To accurately describe this relationship, a wave equation for a cylindrical shell is first established using elastic dynamics theory. This equation is related to the angular frequency... The transcendental equation for wave number k: Where h is the wall thickness, E is the elastic modulus, and v is Poisson's ratio. Let c be the density. Solving this equation yields a series of discrete modal curves. For the T(0,1) mode, i.e., the torsional zeroth-order mode, its core characteristic is that it exhibits non-dispersion characteristics in the low-frequency range. This means that within this frequency range, the wave phase velocity c... p Group speed c g The velocity remains almost constant and does not change with frequency. As the frequency increases, the curve begins to bend, entering the dispersion region, where the velocity becomes frequency-dependent. Substituting the specific material parameters and wall thickness of the gas cylinder, the above characteristic equation is numerically solved, and f is plotted. k-curve graph. Group velocity V g This represents the propagation speed of the guided wave energy envelope and is a core parameter for subsequent signal processing and time correction. In the obtained T(0,1) modal dispersion curve, with frequency f as the horizontal axis and wavenumber k as the vertical axis, the following steps are performed to determine V. g : Locating the flat segment: Observe the curve branch corresponding to the T(0,1) mode. In the low-frequency region, the curve appears as a horizontal straight line or an approximately horizontal line segment. This corresponds to non-dispersion characteristics.

[0020] Read the speed value: Directly read the speed value corresponding to the flat segment. Because in the non-dispersion region c p ≈c g This constant velocity value is directly used as the group velocity V under the current wall thickness. g .

[0021] By combining wall thickness parameters and the dispersion relation of the T(0,1) mode theory, the center frequency of the dominant guided wave mode is calculated. In a thin-walled cylinder, the T(0,1) mode is a predominant longitudinal stretching wave mode, exhibiting non-dispersion characteristics in the low-frequency range, meaning its velocity is essentially constant. The calculation of the center frequency aims to find the frequency point where the energy is most concentrated. The specific calculation formula is as follows: ; Among them, f c V represents the center frequency of the dominant mode of the guided wave. g The group velocity of mode T(0,1) at the current wall thickness is obtained by querying the dispersion curve and finding the velocity value corresponding to the flat segment in the low-frequency region at the current wall thickness; L eff This represents the effective path length for guided wave propagation. For gas cylinders with a straight section, it is usually the straight-line distance between the excitation sensor and the receiving sensor. If it is a surround detection, the average propagation distance needs to be calculated based on the sensor layout.

[0022] An initial bandwidth is set centered on the center frequency of the dominant waveguide mode, forming a candidate energy concentration range. The initial bandwidth can be set to ±10% or ±20% of the center frequency, creating a frequency range centered on the center frequency. , It is ±10% or ±20%. This range is the candidate energy concentration interval.

[0023] Subsequently, the local maximum points of the amplitude spectrum of the first corrected waveform data are searched within the candidate energy concentration interval. This means finding the peak points on the spectral curve within the aforementioned frequency range. Assume the frequency corresponding to the highest peak found is f. peak Taking this local maximum point f as an example peak Centered on the signal, the bandwidth is expanded by a preset ratio, for example, by doubling the initial bandwidth, to obtain the final energy concentration range of the guided wave mode. This more accurately covers the frequency range where the actual signal energy is most concentrated, making it more adaptable than the theoretical center frequency range alone.

[0024] Next, the time-domain envelope sequence of the first corrected waveform data is extracted within the energy concentration region of the guided wave mode, and the amplitude stability index is calculated. First, the first corrected waveform data is processed using the Hilbert transform to obtain its analytic signal, and then the instantaneous envelope of the signal, i.e., the time-domain envelope sequence, is calculated. This sequence reflects the trend of signal amplitude variation over time. Then, the statistical characteristics of this envelope sequence are calculated. The specific formula for calculating the amplitude stability index is as follows: ; Among them, S stab Indicates the amplitude stability index, The standard deviation of the time-domain envelope sequence is represented. It represents the mean of the time-domain envelope sequence, that is, the average size of all envelope amplitudes.

[0025] The calculated amplitude stability index measures the degree of amplitude fluctuation of a signal within the region of concentrated energy of the dominant mode. If the signal is pure and stable, its envelope should be relatively smooth, with a small standard deviation and a large mean, resulting in a small ratio. If the signal contains a large amount of noise, multimodal interference, or reflection reverberation, the envelope will fluctuate drastically, the standard deviation will increase, and the ratio will increase. Therefore, the smaller the index, the more stable and pure the signal.

[0026] The preset modal purity threshold is based on statistical results derived from a large amount of experimental data or simulation. Multiple repeated tests are conducted in a laboratory environment using a known defect-free standard gas cylinder, and the amplitude stability index under normal conditions is recorded. The maximum value plus three times the standard deviation is taken as the lower limit of the threshold.

[0027] Finally, the amplitude stability index is compared with the modal purity threshold. If the amplitude stability index is less than or equal to the modal purity threshold, it indicates that the amplitude fluctuation of the current signal is within an acceptable range and there is little interference. The first corrected waveform data is determined to meet the modal purity requirements and is directly used as the updated time-domain waveform data for subsequent processing or directly for imaging. If the amplitude stability index is greater than the modal purity threshold, it indicates that there are many unstable components or interference in the signal. It is determined that the requirements are not met, and it is marked as the first waveform data to be corrected, and proceeds to S33 for further phase correction and reconstruction.

[0028] The advantage of this approach is that it provides a quantitative and automated signal quality screening mechanism. Through the implementation processes S321 to S326, this invention can intelligently distinguish which sensor data is reliable and which is unreliable. Reliable data can be used directly, saving computational resources; unreliable data triggers subsequent fine-tuning processes to prevent inferior data from contaminating the final imaging results. This hierarchical processing strategy not only improves detection efficiency but also significantly enhances the system's robustness and accuracy in complex operating conditions, ensuring high reliability of the final digital twin imaging.

[0029] S33: Based on the spatial-phase consistency constraint, perform a second-stage correction on the first waveform data to be corrected of the current high-frequency guided wave sensor to generate the second corrected waveform data of the current high-frequency guided wave sensor. S331: Calculate the geometric distance between the current high-frequency guided wave sensor and the adjacent high-frequency guided wave sensor based on the spatial coordinates of the current high-frequency guided wave sensor and the spatial coordinates of the adjacent high-frequency guided wave sensor. S332: Calculate the theoretical time delay difference based on geometric distance and the propagation speed of guided waves in thin-walled seamless steel cylinders; S333: Perform Hilbert transform on the first waveform data to be corrected of the current high-frequency guided wave sensor and the first waveform data to be corrected of the adjacent high-frequency guided wave sensor respectively to obtain their respective analytical signals, and extract the instantaneous phase sequence of the current high-frequency guided wave sensor and the instantaneous phase sequence of the adjacent high-frequency guided wave sensor from their respective analytical signals. S334: Perform phase compensation on the instantaneous phase sequences of adjacent high-frequency guided wave sensors based on the theoretical time delay difference to obtain the compensated adjacent instantaneous phase sequences; S335: Calculate the phase residual between the current instantaneous phase sequence of the high-frequency guided wave sensor and the compensated adjacent instantaneous phase sequence, and average the phase residual to obtain the spatial-phase consistency error; S336: Perform phase correction on the first waveform data to be corrected of the current high-frequency guided wave sensor based on the spatial-phase consistency error, and generate the second corrected waveform data of the current high-frequency guided wave sensor; In this embodiment, the geometric distance between the current high-frequency guided wave sensor and its adjacent high-frequency guided wave sensors is first calculated. Since the spatial coordinates of each sensor have been obtained in step S2, the distance between them can be calculated using spherical geometry or the planar geometry of an unfolded cylindrical surface. Assume the coordinates of the current high-frequency guided wave sensor are... The coordinates of its adjacent high-frequency guided wave sensor are Where R is the outer radius of the gas cylinder, it can be approximated that the radius of the sensor installation positions is the same. Let be the angle about the axis, and z be the axial height. The linear geometric distance D between them is... geo The distance between two points in three-dimensional space can be calculated using the formula for the distance between two points, as follows: ; D geo This indicates the geometric distance between the current high-frequency guided wave sensor and its adjacent high-frequency guided wave sensor. This represents the angular difference between the two sensors in the circumferential direction: z1 represents the axial height coordinate of the current high-frequency guided wave sensor, and z2 represents the axial height coordinate of the adjacent high-frequency guided wave sensor.

[0030] The theoretical time delay difference is calculated based on the geometric distance and the propagation speed of guided waves in a thin-walled seamless steel gas cylinder. The specific calculation formula is as follows: ; This represents the theoretical time delay difference.

[0031] The physical meaning of the calculated theoretical time delay difference is: if the guided wave travels at a constant group velocity V... group The time interval required for the signal to propagate along the shortest path from the current sensor to an adjacent sensor, or vice versa. It represents a time difference benchmark for two sensors receiving the same wavefront signal in an ideally homogeneous medium without interference.

[0032] Subsequently, Hilbert transforms are performed on the first waveform data to be corrected from the current high-frequency guided wave sensor and the first waveform data to be corrected from the adjacent high-frequency guided wave sensor to obtain their respective analytic signals, and the instantaneous phase sequence is extracted. The detailed process of the Hilbert transform is as follows: for any real-valued time-domain waveform signal x(t), a Hilbert transform operation is performed on it to generate an imaginary part signal. This makes the original signal an analytic signal in complex form. , where j is the imaginary unit. This analytic signal contains the amplitude and phase information of the original signal. The method for extracting the instantaneous phase sequence from the analytic signal is: for each time point t i Calculate the complex number z(t) i The argument of is the arctangent function, and the formula is: The instantaneous phase sequence that changes over time is obtained. This operation is performed on the first waveform data to be corrected for the current sensor and the adjacent sensor respectively to obtain the instantaneous phase sequence of the current high-frequency guided wave sensor and the instantaneous phase sequence of the adjacent high-frequency guided wave sensor.

[0033] Phase compensation is performed on the instantaneous phase sequences of adjacent high-frequency guided wave sensors based on the theoretical time delay difference, resulting in compensated adjacent instantaneous phase sequences. Because the guided wave travels from one sensor to another with a time delay, alignment on the time axis or compensation in the phase domain are necessary. Specifically, the instantaneous phase sequences of adjacent sensors are... Translate on the timeline Each time unit constitutes a new time variable. Assuming the wave propagates from the current point to the adjacent point, if the adjacent reception is later, it needs to be pushed back to align; conversely, if the wave propagates in the opposite direction, the direction of propagation needs to be clearly defined. Here, it is assumed that the phase difference between two points arriving before the same wavefront needs to be compared. Typically, the later-arriving signal is advanced or the earlier-arriving signal is delayed to simulate the state at the same moment. More precisely, a compensated adjacent instantaneous phase sequence is constructed. The purpose of this step is to eliminate the inherent phase lag caused by the different geometric propagation distances, so that the two phase sequences should theoretically coincide.

[0034] The phase residual between the current instantaneous phase sequence of the high-frequency guided wave sensor and the compensated adjacent instantaneous phase sequences is calculated, and the phase residual is averaged to obtain the spatial-phase consistency error. The formula for calculating the phase residual is as follows: ; This represents the phase residual at time t. This represents the value of the instantaneous phase sequence of the current high-frequency guided wave sensor at time t. This represents the value of the instantaneous phase sequence of adjacent high-frequency guided wave sensors at time t after compensation.

[0035] To obtain a single spatial-phase consistency error value The root mean square error or the average absolute value of the phase residuals within the effective time period is taken. The root mean square error is used as the final indicator.

[0036] The physical meaning of the calculated phase residual is the degree of phase inconsistency between two adjacent sensors after eliminating the effects of theoretical propagation delay. This inconsistency can be caused by various factors, such as changes in local material properties, differences in sensor coupling states, electronic channel deviations, or scattering caused by early micro-defects. Therefore, it is a key indicator for measuring the spatial consistency of signals.

[0037] Finally, phase correction is performed on the first waveform data to be corrected from the current high-frequency guided wave sensor based on the spatial-phase consistency error, generating the second corrected waveform data. Specifically, since we know that neighboring sensors provide more stable or reference phase information, or that a phase field can be constructed using multiple neighboring sensors, the spatial-phase consistency error value is used to evaluate the reliability of the current sensor's phase. If the spatial-phase consistency error value is large, it indicates that the current sensor's phase may have a systematic deviation. The phase information from neighboring sensors is then used to correct the current sensor's phase. For example, a correction factor is constructed based on the average phase residual of all neighboring sensors to adjust the current sensor's phase sequence. Specifically, the instantaneous phase of the current sensor can be replaced with a smoothed phase value, or data from time points with excessively large phase residuals can be simply removed. More commonly, this error estimate is used to assign a lower weight to the current sensor's data in subsequent back-projection imaging algorithms, or during time-domain reconstruction, a model based on neighboring sensor phase predictions is used to correct the phase component of the current signal. Finally, by combining the phase information corrected from S31 with the original amplitude information, a new time-domain signal is reconstructed, which is the second corrected waveform data of the current high-frequency guided wave sensor. The specific advantages of implementing processes S331 to S336 for this invention are as follows: Improving signal quality: By introducing spatial-phase consistency constraints, random noise or systematic phase drift in individual sensor data can be identified and corrected, significantly improving data reliability.

[0038] Enhanced anti-interference capability: By utilizing the spatial correlation between adjacent sensors, incoherent noise can be effectively suppressed, because noise is usually uncorrelated between different sensors, while real guided wave signals have spatial continuity.

[0039] Improving imaging accuracy: Defect imaging in digital twins relies on precise wavefield data. Phase-corrected second-correction waveform data can more accurately reflect the true wave propagation behavior, thereby reducing artifacts and improving the accuracy of defect localization and quantification in subsequent inversion imaging.

[0040] Automated quality control: This process automatically quantifies the consistency of sensor data, providing the system with an objective quality assessment standard, and can screen and process high-quality data without human intervention.

[0041] S34: Obtain the cross-correlation peak offset between the second corrected waveform data of the current high-frequency guided wave sensor and the second corrected waveform data of the adjacent high-frequency guided wave sensor, and determine whether the second corrected waveform data meets the preset channel synchronization tolerance threshold. If it meets the threshold, the second corrected waveform data of the current high-frequency guided wave sensor is used as the updated time domain waveform data. If it does not meet the threshold, the second corrected waveform data of the current high-frequency guided wave sensor is marked as the second waveform data to be corrected, and proceed to S35. S341: Perform cross-correlation calculation on the second corrected waveform data of the current high-frequency guided wave sensor and the second corrected waveform data of the adjacent high-frequency guided wave sensor to obtain the cross-correlation function; S342: Determine the time index of the sampling point with the largest amplitude in the cross-correlation function, and use the difference between the time index of the sampling point with the largest amplitude and the time index of the zero-delay position as the cross-correlation peak offset; S343: Calculate the main period of the signal based on the center frequency of the main mode of the guided wave, and use the preset proportion of the main period of the signal as the theoretical allowable synchronization offset; S344: Obtain the preset channel synchronization tolerance threshold; S345: Compare the absolute values ​​of the cross-correlation peak offset and the channel synchronization tolerance threshold. If the absolute value of the cross-correlation peak offset is less than or equal to the channel synchronization tolerance threshold, the second corrected waveform data is determined to meet the channel synchronization requirements, and the current second corrected waveform data of the high-frequency guided wave sensor is used as the updated time-domain waveform data. If the absolute value of the cross-correlation peak offset is greater than the channel synchronization tolerance threshold, the second corrected waveform data is determined to not meet the channel synchronization requirements, the current second corrected waveform data of the high-frequency guided wave sensor is marked as the second waveform data to be corrected, and proceed to S35. In this embodiment, the second corrected waveform data of the current high-frequency guided wave sensor is first cross-correlated with the second corrected waveform data of adjacent high-frequency guided wave sensors to obtain the cross-correlation function. The specific calculation formula of the cross-correlation function is as follows: ; This represents the cross-correlation function, which is a time lag. The function reflects the similarity of two signals under different relative displacements. x(t) represents the amplitude of the second corrected waveform data of the current high-frequency guided wave sensor at time t; y(t) represents the amplitude of the second corrected waveform data of the adjacent high-frequency guided wave sensor at time t; N represents the total number of sampling points involved in the calculation.

[0042] The calculated cross-correlation function measures the similarity of the waveforms of the guided wave signals received by two sensors and their time synchronization. If the two signals are perfectly synchronized and have no time delay, the cross-correlation function will peak at zero delay; if there is a time delay, the peak will deviate from the zero-delay position. By observing the peak position and height of the cross-correlation function, the strength of the signal correlation and the specific time deviation can be determined.

[0043] In the cross-correlation function, determine the time index of the sampling point with the largest amplitude and the time index of the zero-delay location. Specifically, iterate through all the calculation results of the cross-correlation function, find the point with the largest value, and record the corresponding time lag index, denoted as . The time index of the zero-latency location is... The position, that is, the state when the two signals are perfectly aligned and there is no relative movement. Cross-correlation peak offset. The calculation formula is: ; The physical meaning of the calculated cross-correlation peak offset is: the residual time deviation between the current sensor signal and the adjacent sensor signal in actual measurement. This deviation may be caused by differences in electronic channel delay, small changes in the propagation path, or residual errors in the signal processing. It directly reflects the degree of asynchrony between the two channels on the time axis.

[0044] Next, the signal principal period is calculated based on the center frequency of the guided wave master mode, and a preset proportion of the signal principal period is used as the theoretical allowable synchronization offset. The formula for calculating the signal principal period is as follows: T main =1 / f c .

[0045] In engineering practice, 1 / 4 or 1 / 8 of the main cycle is often chosen as the boundary. This is because exceeding half a cycle can lead to phase ambiguity (i.e., inability to distinguish between leading and lagging), while exceeding 1 / 4 of a cycle may introduce significant phase errors. Therefore, a preset ratio of 1 / 4 is used here. Theoretically, the allowable synchronization offset is 0.25 times the signal's main cycle.

[0046] The theoretically permissible synchronization offset is the maximum time deviation allowed between two channels without affecting subsequent imaging accuracy and phase analysis. If the actual deviation exceeds this value, it indicates that the signal asynchrony has become severe enough to potentially mislead defect localization or cause interference artifacts.

[0047] Subsequently, a preset channel synchronization tolerance threshold is obtained. The process of setting this threshold is reasonable and objective. For example, in a laboratory environment, multiple sets of normal gas cylinder data are tested, the distribution of their cross-correlation peak offsets is statistically analyzed, the maximum offset within the 99% confidence interval is taken, and rounded up to a fixed value that is easy to calculate, such as 1 microsecond or the corresponding number of sampling points. This fixed value is the preset channel synchronization tolerance threshold, which is a constant.

[0048] Finally, the absolute values ​​of the cross-correlation peak offset and the channel synchronization tolerance threshold are compared. If the absolute value of the cross-correlation peak offset is less than or equal to the channel synchronization tolerance threshold, the second corrected waveform data is determined to meet the channel synchronization requirements, and the current second corrected waveform data of the high-frequency guided wave sensor is used as the updated time-domain waveform data. This is because the time deviation between the two channels is within an acceptable range, the signal has good consistency, and it can be directly used for subsequent digital twin construction or imaging. If the absolute value of the cross-correlation peak offset is greater than the channel synchronization tolerance threshold, the second corrected waveform data is determined to not meet the channel synchronization requirements, the current second corrected waveform data of the high-frequency guided wave sensor is marked as the second waveform data to be corrected, and the process proceeds to S35. This is because an excessively large time deviation indicates a serious synchronization problem in the data of that channel, and direct use would lead to imaging distortion or misjudgment.

[0049] The advantages of this judgment are as follows: by quantifying the time synchronization between channels, it ensures that all sensor data involved in the processing are consistent in time reference, avoiding image misalignment caused by channel differences; it automatically identifies channels with poor synchronization, preventing them from contaminating the overall detection results and improving the robustness of the system; good channel synchronization is the foundation of high-precision tomographic imaging, and this step provides high-quality data support for the subsequent generation of accurate digital twin images.

[0050] The specific advantages of the present invention through the implementation process of S341 to S345 are as follows: It not only focuses on the amplitude and phase purity of the signal, but also pays attention to the time synchronization in the multi-sensor array, forming a multi-level quality control system. The whole process is based on objective mathematical indicators and preset thresholds, eliminating the need for manual judgment and realizing full automation of the detection process. By eliminating channels with poor synchronization, the risk of false positives or false negatives caused by hardware differences or environmental interference is significantly reduced, making the final generated digital twin of thin-walled steel seamless gas cylinder more realistic and reliable.

[0051] S35: Based on sparse dictionary learning and reconstruction, perform a third-stage correction on the second waveform data to be corrected of the current high-frequency guided wave sensor to generate the third corrected waveform data of the current high-frequency guided wave sensor. S351: The second waveform data to be corrected from the current high-frequency guided wave sensor is segmented to obtain several time-domain signal blocks; S352: Construct an overcomplete dictionary, wherein the overcomplete dictionary contains several atoms, each of which is a waveguide basis function with different frequencies and attenuation characteristics; S353: For each time-domain signal block, solve for the sparse coefficient vector in an overcomplete dictionary; S354: Reconstruct each time-domain signal block by linear combination of sparse coefficient vector and overcomplete dictionary to obtain each reconstructed signal block; S355: Concatenate all reconstructed signal blocks according to the original timing sequence to generate the third corrected waveform data of the current high-frequency guided wave sensor; In this embodiment, the second waveform data to be corrected from the current high-frequency guided wave sensor is first segmented to obtain several time-domain signal blocks. Specifically, the longer time-domain waveform data is cut into multiple small, non-overlapping or slightly overlapping segments according to a fixed time length, such as a fixed duration containing the main guided wave pulse. This is done to reduce computational complexity and allow for localized processing based on the characteristics of the signal at different times.

[0052] Next, an overcomplete dictionary is constructed, containing several atoms, each of which is a waveguide basis function with different frequencies and attenuation characteristics. The construction process is as follows: First, based on the material and geometric parameters of the thin-walled seamless steel gas cylinder, the phase velocity and group velocity of the T(0,1) mode within the frequency band of interest are calculated using waveguide theory formulas. Then, a set of discrete frequency points is set, for example, from -20% to +20% of the center frequency, with a step size of 1 kHz. For each frequency point, a corresponding sinusoidal oscillation function is constructed as the basic waveform. Considering the energy attenuation during actual propagation, the above sinusoidal function is multiplied by an exponential attenuation term. , is the attenuation coefficient, which is related to frequency and material loss; x is the normalized time or distance variable. Furthermore, to cover different phase start points and envelope shapes, Gaussian envelopes or other window functions can be introduced to modulate the basic waveform. Arranging these waveform vectors, processed with different frequencies, attenuation rates, and phase shifts, column-wise forms a matrix, which is called an overcomplete dictionary. It is called overcomplete because the number of atoms in the dictionary is far greater than the dimension of the signal. This means that the same signal can be approximated by many different combinations of atoms, thus providing a larger selection space to capture the subtle features of the signal.

[0053] Subsequently, for each time-domain signal block, a sparse coefficient vector is solved in an overcomplete dictionary. This is an optimization problem aiming to find a set of coefficients such that dictionary atoms weighted with these coefficients can best reconstruct the original signal block, while requiring that this set of coefficients be as sparse as possible, i.e., most coefficients are zero or close to zero. Specific solution methods typically employ matching pursuit algorithms or basis pursuit denoising. Taking orthogonal matching pursuit as an example, the steps are: initialize the residual to the current time-domain signal block; in each iteration, calculate the absolute value of the inner product of the similarity between all atoms in the dictionary and the current residual, and select the atom with the highest similarity; record the position of this atom and its corresponding weight coefficient; update the residual, i.e., subtract the contribution of the selected atom from the original signal; repeat the above process until the preset sparsity is reached, i.e., the number of selected atoms or the residual is sufficiently small. The final non-zero coefficients and their corresponding position indices together form the sparse coefficient vector. This process essentially filters out a few typical waveforms and their intensities that best represent the true characteristics of the guided wave from a chaotic background of noise.

[0054] Next, each time-domain signal block is reconstructed using a linear combination of the sparse coefficient vector and an overcomplete dictionary, resulting in a reconstructed signal block. Specifically, all non-zero coefficient values ​​in the sparse coefficient vector obtained in the previous step are extracted, and their corresponding dictionary atoms—the selected basis functions—are identified. These atoms are then multiplied by their corresponding coefficient values, and all results are summed. Mathematically, this can be expressed as: , where c i The non-zero sparsity coefficients, d i is the corresponding dictionary atom, and E is the reconstructed signal. Since only the most representative atoms are retained and the parts with extremely small coefficients are discarded, which usually correspond to random noise or irrelevant interference, the reconstructed signal block is smoother and purer than the original signal block. It removes high-frequency noise and irrelevant frequency components, but retains the main shape and energy distribution of the guided wave.

[0055] Finally, all reconstructed signal blocks are spliced ​​together according to their original timing sequence to generate the third corrected waveform data of the current high-frequency guided wave sensor. This involves sequentially connecting the reconstructed time-domain signal blocks to restore a complete long-term time-domain waveform. If there is overlap in the segmentation, averaging or weighted fusion needs to be performed in the overlapping areas to ensure continuity at the splicing points and avoid abrupt changes.

[0056] Through the implementation process of S351 to S355, the present invention has the following specific advantages: Sparse dictionary learning can extract the main physical patterns from complex mixed signals while filtering out non-sparse components, which are considered noise. This is particularly effective in detection environments with low signal-to-noise ratios, significantly improving signal quality. Unlike traditional filtering methods that may blur signal edges, dictionary-based learning adaptively selects the most suitable basis function for the current signal, thus better preserving key information such as the shape, amplitude, and arrival time of the guided wave pulse, which is beneficial for subsequent accurate positioning. Guided waves may undergo mode conversion or be affected by complex boundary reflections during actual propagation, leading to non-stationary signals. Sparse representation can flexibly adapt to such changes, fitting complex waveform structures by combining different atoms, improving the ability to process data under complex conditions. The entire process is based on mathematical optimization algorithms, eliminating the need for manually setting complex filtering parameters, reducing the influence of subjective factors on the detection results, and making the entire data processing flow more standardized and reliable.

[0057] S36: Obtain the residual noise entropy value of the third corrected waveform data of the current high-frequency guided wave sensor in the defect sensitive frequency band, and determine whether the third corrected waveform data meets the preset imaging availability criterion threshold. If it meets the threshold, the third corrected waveform data of the current high-frequency guided wave sensor is used as the updated time-domain waveform data. If it does not meet the threshold, the third corrected waveform data of the current high-frequency guided wave sensor is marked as the third waveform data to be corrected, and the iteration termination judgment condition is executed. The third waveform data to be corrected is used as input to re-enter S31 until the updated time-domain waveform data of the current high-frequency guided wave sensor is obtained. S361: Set the defect-sensitive frequency band according to the center frequency of the main waveguide mode; S362: Bandpass filtering is performed on the third corrected waveform data of the current high-frequency guided wave sensor to retain the signal components within the defect-sensitive frequency band; S363: Perform wavelet packet decomposition on the signal components within the defect-sensitive frequency band to obtain several sub-band coefficients; S364: Calculate the Shannon entropy of each sub-band coefficient and average the Shannon entropy of all sub-band coefficients to obtain the residual noise entropy value; S365: Obtain the preset imaging availability criterion threshold; S366: Compare the residual noise entropy value with the imaging availability criterion threshold. If the residual noise entropy value is less than or equal to the imaging availability criterion threshold, the third corrected waveform data is determined to meet the imaging availability requirements, and the current third corrected waveform data of the high-frequency guided wave sensor is used as the updated time-domain waveform data. If the residual noise entropy value is greater than the imaging availability criterion threshold, the third corrected waveform data is determined to not meet the imaging availability requirements, the current third corrected waveform data of the high-frequency guided wave sensor is marked as the third waveform data to be corrected, and the current iteration number is checked to see if it is less than the preset maximum iteration number. If so, the third waveform data to be corrected is used as input to re-enter S31 until the updated time-domain waveform data of the current high-frequency guided wave sensor is obtained. If not, the current third corrected waveform data is used as the updated time-domain waveform data.

[0058] In this embodiment, the defect-sensitive frequency band is first set based on the center frequency of the waveguide master mode. Specifically, the center frequency of the waveguide master mode calculated in step S32 is selected as the benchmark. Considering that defects such as cracks and corrosion exhibit the most significant scattering and reflection within a specific frequency range in actual detection, a relatively narrow frequency band centered on the center frequency is typically chosen. For example, the lower limit frequency is set to 0.8 times the center frequency of the waveguide master mode, and the upper limit frequency is set to 1.2 times the center frequency of the waveguide master mode. This range is the defect-sensitive frequency band because within this band, the waveguide wavelength has a high degree of matching with common defect sizes, and the signal-to-noise ratio is relatively good, thus most effectively reflecting defect information.

[0059] Next, bandpass filtering is applied to the third-corrected waveform data from the current high-frequency guided wave sensor to retain signal components within the defect-sensitive frequency band. Specifically, a digital filter is used, with its passband range set to the aforementioned [lower limit frequency, upper limit frequency]. The third-corrected waveform data is input into this filter, removing frequency components below the lower limit frequency and above the upper limit frequency. After filtering, the resulting signal contains only energy within the defect-sensitive frequency band, which helps to further remove out-of-band noise and interference from other unrelated modes, allowing subsequent analysis to focus more on signal characteristics related to the defect.

[0060] Subsequently, wavelet packet decomposition is performed on the signal components within the defect-sensitive frequency band to obtain several sub-band coefficients. Wavelet packet decomposition is a more refined signal analysis method than traditional wavelet transform; it can decompose not only the low-frequency components but also the high-frequency components, thus achieving a uniform division of the signal frequency band. The specific process is as follows: the filtered signal is used as input, a suitable wavelet basis function is selected, and the number of decomposition levels is set. Through the cascading operation of multiple low-pass and high-pass filters, the signal frequency band is continuously subdivided into smaller sub-bands. Each decomposition level generates a set of coefficients, ultimately yielding a series of sub-band coefficient vectors representing different minute frequency ranges in the final level. These coefficients reflect the energy distribution of the signal in each minute frequency band.

[0061] Next, the Shannon entropy of each sub-band coefficient is calculated, and the Shannon entropy of all sub-band coefficients is averaged to obtain the residual noise entropy value. First, for each sub-band coefficient vector, its corresponding probability distribution is calculated. Assume the... The coefficient vector of the k sub-bands is c k If its length is M, then the energy percentage p of each sample point is... k,i This can be approximated as the proportion of the square of the absolute value of the coefficient vector at that point to the sum of the squares of the absolute values ​​of all coefficients in the vector. The entropy value of this sub-band is calculated using the Shannon entropy formula. Shannon entropy measures the uncertainty or randomness of the signal within that sub-band. If the signal is a pure periodic waveform, its energy distribution is relatively concentrated, resulting in a lower entropy value; if the signal is full of random noise, its energy distribution tends to be uniform, resulting in a higher entropy value. After calculating the Shannon entropy of all sub-bands, the arithmetic mean of the entropy values ​​of all sub-bands is taken to obtain the final residual noise entropy value.

[0062] The physical meaning of the calculated residual noise entropy value is: the complexity or disorder of the random noise component in the remaining signal after preliminary processing and frequency band filtering. The lower the entropy value, the more ordered the signal is, the closer it is to the ideal deterministic waveform, and the less noise there is; the higher the entropy value, the more random interference is mixed in the signal, and the poorer the quality.

[0063] In a laboratory environment, multiple tests were conducted using a known defect-free standard specimen, collecting a large amount of waveform data under normal conditions. For each set of normal data, the residual noise entropy value was calculated as described above, and the entropy distribution of these normal data was statistically analyzed. The maximum value of these normal data entropy values ​​was taken, and an appropriate safety margin, such as 5%-10% of the maximum value, was added as a fixed limit. This fixed value is the preset imaging usability criterion threshold. It is a constant that does not change with the object being tested and represents the upper limit of the highest noise level that may occur under ideal defect-free conditions.

[0064] Finally, before entering the S36 procedure, an iteration counter is independently initialized for each high-frequency guided wave sensor, with an initial value of 0. Each time S366 is entered, the current iteration counter value is read first. The residual noise entropy value is compared with the imaging availability criterion threshold. If the residual noise entropy value is less than or equal to the imaging availability criterion threshold, the third corrected waveform data is determined to meet the imaging availability requirements, and the current third corrected waveform data of the high-frequency guided wave sensor is used as the updated time-domain waveform data. This is because the noise level in the signal is within an acceptable range, and the signal has sufficient clarity and consistency to generate a high-quality digital twin image. If the residual noise entropy value is greater than the imaging availability criterion threshold, the third corrected waveform data is determined not to meet the imaging availability requirements, and the current iteration count is checked to see if it is less than the preset maximum iteration count. If the current iteration count is less than the preset maximum iteration count, the current third corrected waveform data of the high-frequency guided wave sensor is marked as the third waveform data to be corrected, and the third waveform data to be corrected is used as input to re-enter S31 until the updated time-domain waveform data of the current high-frequency guided wave sensor is obtained. This is because excessively high noise entropy indicates poor signal quality, and directly using it for imaging will produce severe artifacts or mask real defects. It is necessary to return to the first stage for more thorough preprocessing or correction. If the residual noise entropy value is greater than the imaging usability criterion threshold, and the current iteration number is greater than or equal to the preset maximum iteration number, the current third-corrected waveform data is output as the updated time-domain waveform data.

[0065] Based on extensive experimental data, most normal signal correction processes typically reduce the noise entropy below the threshold within 3-5 iterations. Data that fails to converge after more than 5 iterations is usually considered anomalous. Therefore, setting the maximum number of iterations to 5 is sufficient to cover the vast majority of effective correction scenarios.

[0066] The advantage of this judgment is that: By using Shannon entropy, an information theory metric, the noise content of the signal is objectively quantified, avoiding errors caused by subjective judgment. Only when the signal noise is low enough is it allowed to enter the imaging stage, ensuring the clarity and reliability of the digital twin image from the source. If the condition is not met, it returns to S31 for reprocessing, forming an adaptive iterative optimization closed loop until high-quality data is obtained, which greatly improves the robustness and success rate of the system.

[0067] Through the implementation process of S361 to S366, the present invention has the following specific advantages: By combining wavelet packet decomposition and Shannon entropy, noise characteristics can be evaluated simultaneously in the frequency and time domains, providing a more comprehensive and accurate assessment than a single time or frequency domain indicator. A complete chain from preprocessing to final usability determination has been established, achieving full automation and quality control of the detection process and reducing the need for manual intervention. By rigorously screening signals with high signal-to-noise ratios, the risk of missed detections and false detections has been effectively reduced, ensuring that the final digital twin of the thin-walled seamless steel gas cylinder can realistically and accurately reflect its internal state.

[0068] S4. Map the updated time-domain waveform data of each high-frequency guided wave sensor to the inner surface grid nodes of the digital twin of the thin-walled steel seamless gas cylinder. Calculate the scattering energy distribution of the inner surface grid nodes using the inverse time-shift imaging algorithm to obtain the current set of candidate regions for inner surface defects of the thin-walled steel seamless gas cylinder. In this embodiment, the updated time-domain waveform data of each high-frequency guided wave sensor is first mapped to the inner surface mesh nodes of the digital twin of the thin-walled seamless steel gas cylinder. Specifically, the digital twin model of the thin-walled seamless steel gas cylinder constructed in step S2 is used. This model already includes the three-dimensional geometry of the inner surface of the cylinder and discretized mesh node information. For each sensor installation location, the propagation path between the sensor and each inner surface mesh node is typically calculated. Since guided waves mainly propagate axially and circumferentially and may be reflected or scattered on the inner surface, it is assumed that the signal originates from the sensor, travels a certain path to a mesh node on the inner surface, and then returns or is received by other sensors. In single-sided detection mode, direct waves or single-reflection waves are typically considered; in double-sided or multi-sided modes, transmission or multiple reflections are considered. To simplify the mapping process, ray tracing or distance-weighted averaging methods can be used. Specifically, for each grid node on the inner surface, its geometric distance *d* relative to the current sensor and the theoretical propagation time are calculated: A three-dimensional digital model of the inner surface of the thin-walled seamless steel gas cylinder to be detected is created and discretized into a high-density set of grid nodes. For each grid node on the inner surface, Dijkstra's algorithm from graph theory is used to calculate its geometric distance relative to the currently active sensor on the triangular mesh model of the gas cylinder's inner surface, t = d / V. gThen, a signal segment corresponding to a time window of the propagation time is extracted from the updated time-domain waveform data. This segment represents the signal characteristics that the sensor should receive if a scattering source is present at that grid node. This extracted signal segment is assigned to the corresponding grid node as the excitation or response input for that node. If multiple sensors cover the same grid node, signals from different sensors can be superimposed or averaged to enhance signal strength. In this way, the original one-dimensional time-domain waveform data is transformed into a spatiotemporal data field distributed on a two-dimensional inner surface grid, meaning that each grid node possesses time-domain signal information related to its location.

[0069] Subsequently, the scattered energy distribution of the inner surface grid nodes is calculated using a reverse time migration imaging algorithm, thereby obtaining the candidate region set of inner surface defects for the current thin-walled seamless steel gas cylinder. The core idea of ​​reverse time migration imaging is to simulate the wave propagation process, but in the opposite direction. The specific steps are as follows: First, the time-domain signal mapped onto the inner surface grid nodes is regarded as a virtual source or received record. For each grid node, its recorded time-domain signal is used as the source term, substituted into the wave equation, and inversely integrated on the time axis, i.e., from the final time T back to the initial time 0. In this process, a zero-frequency background field or reference field is simultaneously propagated forward, or more commonly, cross-correlation imaging conditions are used. Specifically, for each grid node (X,Y), two wave fields are defined: one is the actual recorded wave field from the sensor. The other is a virtual wave field generated through numerical simulation that radiates outward from the grid node. The reverse time migration imaging formula is usually expressed as follows: the imaging profile I(X,Y) is equal to the sum of the cross-correlation of the two wavefields over time. ; Where I(X,Y) represents the scattered energy distribution value at the grid node (X,Y) on the inner surface of the thin-walled seamless steel gas cylinder, X s This indicates the location of the sensor.

[0070] In practical calculations, for efficiency, a two-way wavefield extension method is typically used: the data recorded by the sensor is used as boundary conditions to extend inward to obtain the source wavefield; simultaneously, assuming each grid node is a potential scattering source, the receiving wavefield is extended outward. When the source and receiving wavefields coincide and are in phase at a certain time and spatial location, their product or cross-correlation reaches its maximum value. After integrating over all time points, the resulting scalar field I(X,Y) is the scattering energy distribution map, as shown below. Figure 7As shown in the figure, if the energy value at a certain grid node is significantly higher than the surrounding background noise level, it indicates the presence of a strong scattering center at that location, which is highly likely to be the location of a defect. Therefore, by grouping together all grid nodes whose energy values ​​exceed a preset energy threshold (e.g., three standard deviations of the background mean), a candidate set of defects on the inner surface of the thin-walled seamless steel gas cylinder is formed. These candidate regions are the initially identified suspected defect locations, which can be further confirmed through manual verification or more refined analysis.

[0071] Through this process, the present invention achieves a direct conversion from raw sensor data to intuitive defect images. The reverse time migration algorithm effectively handles wavefield interference and diffraction phenomena in complex media. Compared to traditional tomography or Time-of-Flight (TOF) localization methods, it provides higher resolution defect imaging results, especially for irregularly shaped or edge-located defects, offering better localization accuracy. Simultaneously, mapping the data to the surface mesh of the digital twin allows the imaging results to be directly visualized on the 3D model, enabling operators to intuitively view the location, size, and morphology of defects, greatly improving detection efficiency and the usability of the results.

[0072] S5. Based on the candidate region set of internal surface defects, identify the location and geometric contour of corrosion pits or cracks on the internal surface of the current thin-walled steel seamless gas cylinder. In this embodiment, the location and geometric contour of corrosion pits or cracks on the inner surface of the current thin-walled steel seamless gas cylinder are identified based on the candidate region set of inner surface defects. The specific operation begins with spatial clustering analysis of the candidate region set. Since the scattered energy distribution map obtained from reverse time-shift imaging often shows a connected, high-energy region around the same real defect rather than an isolated point, a connected component-based labeling algorithm is used to process these high-energy grid nodes. The labeling algorithm could be, for example, seed filling or a disjoint-set data structure. By setting a neighborhood rule, such as 8-neighborhood or 26-neighborhood, all interconnected high-energy grid nodes with energy values ​​exceeding a threshold are grouped into an independent connected component. Each connected component represents a potential independent defect target.

[0073] Next, for each connected component, its specific geometric features are extracted to distinguish between corrosion pits and crack-like defects. For location determination, the average coordinates of all mesh nodes in the connected component, or the coordinates of the mesh node with the highest energy value, are taken as the center coordinates of the defect. For obtaining the geometric profile, we need to analyze the shape characteristics of the connected component on the 3D mesh. First, the area of ​​the connected component (i.e., the number of mesh nodes it contains) is calculated by multiplying it by the area of ​​a single mesh and the perimeter (i.e., the number of mesh nodes on the boundary of the connected component). Then, its aspect ratio or ellipse fitting parameters are calculated. If the shape of the connected component is close to a circle or ellipse, and the aspect ratio is close to 1, it is identified as a corrosion pit; because corrosion is usually a uniform or non-uniform loss of material, the resulting pits are often relatively rounded in planar projection. Conversely, if the aspect ratio of the connected component is significantly greater than a preset value, such as 2 or 3, or its shape exhibits obvious linear extension characteristics, it is identified as a crack-like defect; because cracks have sharp edges and slender shapes.

[0074] To more accurately depict the geometric contour, we connect the boundary points of all mesh nodes belonging to the defect in the connected component. For corrosion pits, their contours can be fitted to an ellipse or polygon, and their maximum and minimum diameters recorded. For cracks, their centerlines can be extracted using a skeletonization algorithm to obtain the crack length and orientation angle, while also recording their width distribution. Furthermore, by combining the scattered energy intensity information obtained in step S4, the depth or severity of the defect can be further estimated. For example, stronger energy usually means more intense scattering, which may correspond to a deeper pit or a wider crack opening. Finally, the center coordinates, type, and geometric contour data defined by the boundary mesh nodes of each connected component are recorded to form a complete defect identification result.

[0075] In this way, the present invention can automatically extract specific defect information from complex imaging data. This method not only identifies the location of the defect but also describes its appearance and can distinguish between different types of defects. This is crucial for subsequent gas cylinder safety assessments because corrosion pits and cracks affect gas cylinder strength differently, requiring different assessment standards and treatment strategies. Through this meticulous classification and contour extraction, precise data support can be provided for the lifecycle management of gas cylinders, ensuring the practicality and reliability of the detection results.

[0076] Example 2: Figure 2 As shown, this embodiment also discloses a digital twin-based system for detecting defects on the inner surface of seamless steel gas cylinders, including the following modules: Transmission module: Used to transmit modal guided waves to the cylinder of thin-walled seamless steel gas cylinder when the cylinder is in online use, the cylinder valve is not disassembled, and the cylinder is filled with liquefied gas or compressed gas medium. The module records the time-domain waveform data and spatial coordinates of each high-frequency guided wave sensor after the modal guided waves are reflected or scattered by the inner surface of the thin-walled seamless steel gas cylinder. Data acquisition module: used to acquire the geometric parameters and real-time service parameters of the current thin-walled steel seamless gas cylinder, and construct a three-dimensional geometric solid model of the current thin-walled steel seamless gas cylinder based on the geometric parameters. The spatial coordinates of each high-frequency guided wave sensor are registered to the outer surface of the three-dimensional geometric solid model of the current thin-walled steel seamless gas cylinder. Combined with the real-time service parameters of the current thin-walled steel seamless gas cylinder, a digital twin of the current thin-walled steel seamless gas cylinder is generated. Correction module: used to perform phased correction on the time-domain waveform data of each high-frequency guided wave sensor, and to judge the correction result of each phase, and to obtain the updated time-domain waveform data of each high-frequency guided wave sensor based on the judgment result. Region partitioning module: used to map the updated time-domain waveform data of each high-frequency guided wave sensor to the inner surface grid nodes of the digital twin of the thin-walled steel seamless gas cylinder, and calculate the scattering energy distribution of the inner surface grid nodes through the inverse time-off imaging algorithm to obtain the current inner surface defect candidate region set of the thin-walled steel seamless gas cylinder. Identification module: Based on the candidate region set of internal surface defects, it identifies the location and geometric contour of corrosion pits or cracks on the internal surface of the current thin-walled steel seamless gas cylinder.

[0077] Example 3: To verify the effectiveness and robustness of the three-stage progressive signal correction and judgment process proposed in this invention in the detection of defects on the inner surface of seamless steel gas cylinders, a certain model of seamless steel gas cylinder was selected as the experimental object. A high-frequency guided wave sensor array was arranged in a specific area on its outer surface for actual detection experiments. Keeping the gas cylinder stationary, a control signal generator sent a pulse electrical signal to the excitation sensor, exciting modal guided waves to propagate within the gas cylinder. The original time-domain waveform data of a single high-frequency guided wave sensor located at the same installation position was collected. The gas cylinder cross-sectional structure corresponding to this position is shown below. Figure 3 As shown, Figure 3 A schematic cross-sectional view of the gas cylinder at the sensor installation location is shown, clearly defining the geometric features and boundary conditions of the detection area to ensure a clear and consistent physical background for the experimental data. The raw data is then processed using the full workflow proposed in this invention. First, the raw time-domain waveform data without any corrections is acquired, and its waveform characteristics are as follows... Figure 4 As shown. From Figure 4As can be seen, the original waveform is affected by environmental noise interference, mode aliasing, and phase asynchrony between sensors, resulting in a cluttered signal and severely submerged effective defect echoes. Directly using it for inverse time migration imaging would produce a large number of artifacts. Subsequently, the above-mentioned original data is input into the three-stage correction module of this invention: In the first stage, multi-scale time-frequency analysis is performed on the original signal to screen out signal components related to the target guided wave mode and fuse and reconstruct them to obtain the first corrected waveform. Then, its amplitude stability index is calculated and compared with a preset mode purity threshold. If the signal is sufficiently pure, it represents the updated time-domain waveform data obtained after processing; if it does not meet the requirements, it proceeds to the second stage. In the second stage, the spatial positional relationship between sensors is used to calculate the theoretical time delay difference, and phase compensation and correction are performed on the signals of adjacent channels to solve the signal asynchrony problem. After correction, the cross-correlation peak offset is calculated and compared with a preset channel synchronization tolerance threshold. If the synchronization requirement is met, it represents the updated time-domain waveform data obtained after processing; if it does not meet the requirements, it proceeds to the third stage. In the third stage, a sparse dictionary learning method is used to deeply reconstruct the signal to suppress residual noise. After reconstruction, the residual noise entropy value of the signal within the defect-sensitive frequency band is calculated and compared with a preset imaging availability criterion threshold. If the signal quality meets the standard, it represents the updated time-domain waveform data obtained after processing; if it does not meet the standard, the signal is sent back to the first stage to start a new round of correction loops until the corresponding threshold of any stage is met, i.e., the final updated time-domain waveform data is output, such as... Figure 5 As shown. It should be stated that, as Figure 5 The updated time-domain waveform data of one of the high-frequency guided wave sensors shown should be regarded as the original time-domain waveform data of one of the high-frequency guided wave sensors, which is the updated time-domain waveform data obtained after any stage of correction, or the updated time-domain waveform data obtained after a new round of correction cycle.

[0078] By comparison Figure 4 and Figure 5 It can be clearly observed that the high-frequency random noise on the left side of the updated time-domain waveform is significantly suppressed, and the peak value on the right side is significantly higher and more contracted than before the correction. This demonstrates more accurate peak identification, making the previously obscured effective reflection signal outline clearer and sharper. This is helpful for subsequently constructing a scattering energy distribution map with lower noise and other interference, such as... Figure 7 As shown, this can effectively improve the accuracy of subsequent defect identification. Conversely, if an uncorrected method is used, such as... Figure 4 If the original time-domain waveform data shown is used to construct a scattering energy distribution map, it will directly result in the constructed scattering energy distribution map as follows: Figure 6As shown, noise and other interference are extremely severe, which is very detrimental to subsequent defect identification operations. This proves that the present invention, through a three-level progressive strategy of time-frequency analysis, phase compensation, and sparse reconstruction, can effectively separate the target mode from interference noise, and the signal consistency is significantly enhanced. Figure 5 This indicates that for signals with poor initial quality, when a single round of three-stage correction is insufficient to eliminate complex noise, multiple rounds of iteration can be used to gradually remove noise components, demonstrating the robustness of this mechanism in handling non-stationary noise and channel synchronization errors in the field. In summary, through comparison... Figure 4 , Figure 5 and comparison Figure 6 , Figure 7 The three-stage progressive correction and iterative judgment method proposed in this invention can effectively overcome complex problems such as large noise interference, mixed modes and asynchronous channels in on-site detection, and provide a reliable data foundation for accurate defect identification.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting defects on the inner surface of seamless steel gas cylinders based on digital twins, characterized in that, Includes the following steps: S1. Under the condition that the thin-walled steel seamless gas cylinder is in online use, the cylinder valve is not disassembled, and the cylinder is filled with liquefied gas or compressed gas medium, a high-frequency guided wave sensor array installed on the outer wall of the thin-walled steel seamless gas cylinder emits modal guided waves to the cylinder body of the thin-walled steel seamless gas cylinder, and after the modal guided waves are reflected or scattered by the inner surface of the thin-walled steel seamless gas cylinder, the time domain waveform data and spatial coordinates of each high-frequency guided wave sensor are recorded. S2. Obtain the geometric parameters and real-time service parameters of the current thin-walled steel seamless gas cylinder, and construct a three-dimensional geometric solid model of the current thin-walled steel seamless gas cylinder based on the geometric parameters. Register the spatial coordinates of each high-frequency guided wave sensor to the outer surface of the three-dimensional geometric solid model of the current thin-walled steel seamless gas cylinder, and generate a digital twin of the current thin-walled steel seamless gas cylinder by combining the real-time service parameters of the current thin-walled steel seamless gas cylinder. S3. Perform staged correction on the time-domain waveform data of each high-frequency guided wave sensor, and determine the correction result of each stage separately. Based on the determination result, obtain the updated time-domain waveform data of each high-frequency guided wave sensor; including: S31: Based on multi-scale time-frequency joint characterization, the time-domain waveform data of each high-frequency guided wave sensor is corrected in the first stage to generate the first corrected waveform data of each high-frequency guided wave sensor. S32: Obtain the guided wave mode energy concentration range of the first corrected waveform data of the current high-frequency guided wave sensor, and obtain the amplitude stability index based on the guided wave mode energy concentration range. Determine whether the first corrected waveform data meets the preset modal purity threshold. If it meets the threshold, use the first corrected waveform data of the current high-frequency guided wave sensor as the updated time domain waveform data. If it does not meet the threshold, mark the first corrected waveform data of the current high-frequency guided wave sensor as the first waveform data to be corrected, and proceed to S33. S33: Based on the spatial-phase consistency constraint, perform a second-stage correction on the first waveform data to be corrected of the current high-frequency guided wave sensor to generate the second corrected waveform data of the current high-frequency guided wave sensor. S34: Obtain the cross-correlation peak offset between the second corrected waveform data of the current high-frequency guided wave sensor and the second corrected waveform data of the adjacent high-frequency guided wave sensor, and determine whether the second corrected waveform data meets the preset channel synchronization tolerance threshold. If it meets the threshold, the second corrected waveform data of the current high-frequency guided wave sensor is used as the updated time domain waveform data. If it does not meet the threshold, the second corrected waveform data of the current high-frequency guided wave sensor is marked as the second waveform data to be corrected, and proceed to S35. S35: Based on sparse dictionary learning and reconstruction, perform a third-stage correction on the second waveform data to be corrected of the current high-frequency guided wave sensor to generate the third corrected waveform data of the current high-frequency guided wave sensor. S36: Obtain the residual noise entropy value of the third corrected waveform data of the current high-frequency guided wave sensor in the defect sensitive frequency band, and determine whether the third corrected waveform data meets the preset imaging availability criterion threshold. If it meets the threshold, the third corrected waveform data of the current high-frequency guided wave sensor is used as the updated time-domain waveform data. If it does not meet the threshold, the third corrected waveform data of the current high-frequency guided wave sensor is marked as the third waveform data to be corrected, and the iteration termination judgment condition is executed. The third waveform data to be corrected is used as input to re-enter S31 until the updated time-domain waveform data of the current high-frequency guided wave sensor is obtained. S4. Map the updated time-domain waveform data of each high-frequency guided wave sensor to the inner surface grid nodes of the digital twin of the thin-walled steel seamless gas cylinder. Calculate the scattering energy distribution of the inner surface grid nodes using the inverse time-shift imaging algorithm to obtain the current set of candidate regions for inner surface defects of the thin-walled steel seamless gas cylinder. S5. Based on the candidate region set of internal surface defects, identify the location and geometric contour of corrosion pits or cracks on the internal surface of the current thin-walled steel seamless gas cylinder.

2. The method for detecting defects on the inner surface of seamless steel gas cylinders based on digital twins according to claim 1, characterized in that, The first-stage correction of the time-domain waveform data of each high-frequency guided wave sensor based on multi-scale time-frequency joint characterization generates the first corrected waveform data of each high-frequency guided wave sensor, including: S311: Perform continuous wavelet transform on the time-domain waveform data of the current high-frequency guided wave sensor to obtain a complex coefficient matrix at several scales; S312: Based on the complex coefficient matrix, calculate the instantaneous amplitude sequence and instantaneous phase sequence for each scale, forming several data pairs consisting of the instantaneous amplitude sequence and the instantaneous phase sequence; S313: Based on the preset guided wave main mode frequency range, select an effective scale subset from several data pairs, and perform fusion processing on the instantaneous amplitude sequence in the effective scale subset. At the same time, perform fusion processing on the instantaneous phase sequence in the effective scale subset to obtain the fused instantaneous amplitude sequence and the fused instantaneous phase sequence, respectively. S314: The time-domain waveform data of the current high-frequency guided wave sensor is reconstructed using the fused instantaneous amplitude sequence and the fused instantaneous phase sequence to generate the first corrected waveform data of the current high-frequency guided wave sensor.

3. The method for detecting defects on the inner surface of seamless steel gas cylinders based on digital twins according to claim 2, characterized in that, The process involves acquiring the guided wave mode energy concentration range of the first corrected waveform data of the current high-frequency guided wave sensor, obtaining an amplitude stability index based on the guided wave mode energy concentration range, calculating a modal purity threshold based on the amplitude stability index, and determining whether the first corrected waveform data meets the modal purity threshold. If it does, the first corrected waveform data of the current high-frequency guided wave sensor is used as the updated time-domain waveform data; otherwise, the first corrected waveform data of the current high-frequency guided wave sensor is marked as the first waveform data to be corrected, and the process proceeds to S33, including: S321: Perform a Fourier transform on the first corrected waveform data of the current high-frequency guided wave sensor to obtain the amplitude spectrum of the first corrected waveform data; S322: Obtain the wall thickness parameters based on the geometric parameters of the current thin-walled steel seamless gas cylinder, and at the same time obtain the modal theory dispersion relationship. Combine the wall thickness parameters and the modal theory dispersion relationship to calculate the center frequency of the guided wave main mode, and set the initial bandwidth with the center frequency of the guided wave main mode as the center to form a candidate energy concentration interval. S323: Search for the local maximum point of the amplitude spectrum of the first corrected waveform data within the candidate energy concentration interval, and expand the bandwidth by a preset ratio with the local maximum point as the center to obtain the guided wave mode energy concentration interval; S324: Extract the time-domain envelope sequence of the first corrected waveform data within the energy concentration range of the guided wave mode, and calculate the ratio of the amplitude standard deviation to the mean of the time-domain envelope sequence as an amplitude stability index; S325: Obtain the preset modal purity threshold; S326: Compare the amplitude stability index with the modal purity threshold. If the amplitude stability index is less than or equal to the modal purity threshold, the first corrected waveform data is determined to meet the modal purity requirements, and the first corrected waveform data of the current high-frequency guided wave sensor is used as the updated time-domain waveform data. If the amplitude stability index is greater than the modal purity threshold, the first corrected waveform data is determined to not meet the modal purity requirements, the first corrected waveform data of the current high-frequency guided wave sensor is marked as the first waveform data to be corrected, and proceed to S33.

4. The method for detecting defects on the inner surface of seamless steel gas cylinders based on digital twins according to claim 3, characterized in that, The second-stage correction of the first waveform data to be corrected of the current high-frequency guided wave sensor based on spatial-phase consistency constraints, generating the second corrected waveform data of the current high-frequency guided wave sensor, includes: S331: Calculate the geometric distance between the current high-frequency guided wave sensor and the adjacent high-frequency guided wave sensor based on the spatial coordinates of the current high-frequency guided wave sensor and the spatial coordinates of the adjacent high-frequency guided wave sensor. S332: Calculate the theoretical time delay difference based on geometric distance and the propagation speed of guided waves in thin-walled seamless steel cylinders; S333: Perform Hilbert transform on the first waveform data to be corrected of the current high-frequency guided wave sensor and the first waveform data to be corrected of the adjacent high-frequency guided wave sensor respectively to obtain their respective analytical signals, and extract the instantaneous phase sequence of the current high-frequency guided wave sensor and the instantaneous phase sequence of the adjacent high-frequency guided wave sensor from their respective analytical signals. S334: Perform phase compensation on the instantaneous phase sequences of adjacent high-frequency guided wave sensors based on the theoretical time delay difference to obtain the compensated adjacent instantaneous phase sequences; S335: Calculate the phase residual between the current instantaneous phase sequence of the high-frequency guided wave sensor and the compensated adjacent instantaneous phase sequence, and average the phase residual to obtain the spatial-phase consistency error; S336: Perform phase correction on the first waveform data to be corrected of the current high-frequency guided wave sensor based on the spatial-phase consistency error, and generate the second corrected waveform data of the current high-frequency guided wave sensor.

5. The method for detecting defects on the inner surface of seamless steel gas cylinders based on digital twins according to claim 4, characterized in that, The process involves acquiring the cross-correlation peak offset between the current high-frequency guided wave sensor's second corrected waveform data and the second corrected waveform data of adjacent high-frequency guided wave sensors, calculating the channel synchronization tolerance threshold based on the cross-correlation peak offset, determining whether the second corrected waveform data meets the channel synchronization tolerance threshold, and if it does, using the current high-frequency guided wave sensor's second corrected waveform data as the updated time-domain waveform data; otherwise, marking the current high-frequency guided wave sensor's second corrected waveform data as the second waveform data to be corrected, and proceeding to step S35, which includes: S341: Perform cross-correlation calculation on the second corrected waveform data of the current high-frequency guided wave sensor and the second corrected waveform data of the adjacent high-frequency guided wave sensor to obtain the cross-correlation function; S342: Determine the time index of the sampling point with the largest amplitude in the cross-correlation function, and use the difference between the time index of the sampling point with the largest amplitude and the time index of the zero-delay position as the cross-correlation peak offset; S343: Calculate the main period of the signal based on the center frequency of the main mode of the guided wave, and use the preset proportion of the main period of the signal as the theoretical allowable synchronization offset; S344: Obtain the preset channel synchronization tolerance threshold; S345: Compare the absolute values ​​of the cross-correlation peak offset and the channel synchronization tolerance threshold. If the absolute value of the cross-correlation peak offset is less than or equal to the channel synchronization tolerance threshold, the second corrected waveform data is determined to meet the channel synchronization requirements, and the current second corrected waveform data of the high-frequency guided wave sensor is used as the updated time-domain waveform data. If the absolute value of the cross-correlation peak offset is greater than the channel synchronization tolerance threshold, the second corrected waveform data is determined to not meet the channel synchronization requirements, the current second corrected waveform data of the high-frequency guided wave sensor is marked as the second waveform data to be corrected, and proceed to S35.

6. The method for detecting defects on the inner surface of seamless steel gas cylinders based on digital twins according to claim 5, characterized in that, The third-stage correction of the second waveform data to be corrected of the current high-frequency guided wave sensor based on sparse dictionary learning and reconstruction, generating the third corrected waveform data of the current high-frequency guided wave sensor, includes: S351: The second waveform data to be corrected from the current high-frequency guided wave sensor is segmented to obtain several time-domain signal blocks; S352: Construct an overcomplete dictionary, wherein the overcomplete dictionary contains several atoms, each of which is a waveguide basis function with different frequencies and attenuation characteristics; S353: For each time-domain signal block, solve for the sparse coefficient vector in an overcomplete dictionary; S354: Reconstruct each time-domain signal block by linear combination of sparse coefficient vector and overcomplete dictionary to obtain each reconstructed signal block; S355: Concatenate all reconstructed signal blocks according to the original timing sequence to generate the third corrected waveform data of the current high-frequency guided wave sensor.

7. The method for detecting defects on the inner surface of seamless steel gas cylinders based on digital twins according to claim 6, characterized in that, The process of acquiring the residual noise entropy value of the third corrected waveform data of the current high-frequency guided wave sensor within the defect-sensitive frequency band, calculating the imaging availability criterion threshold based on the residual noise entropy value, and determining whether the third corrected waveform data meets the imaging availability criterion threshold, includes: if it meets the threshold, the third corrected waveform data of the current high-frequency guided wave sensor is used as the updated time-domain waveform data; if it does not meet the threshold, the third corrected waveform data of the current high-frequency guided wave sensor is marked as the third waveform data to be corrected, and the third waveform data to be corrected is used as input to re-enter S31 until the updated time-domain waveform data of the current high-frequency guided wave sensor is obtained. S361: Set the defect-sensitive frequency band according to the center frequency of the main waveguide mode; S362: Bandpass filtering is performed on the third corrected waveform data of the current high-frequency guided wave sensor to retain the signal components within the defect-sensitive frequency band; S363: Perform wavelet packet decomposition on the signal components within the defect-sensitive frequency band to obtain several sub-band coefficients; S364: Calculate the Shannon entropy of each sub-band coefficient and average the Shannon entropy of all sub-band coefficients to obtain the residual noise entropy value; S365: Obtain the preset imaging availability criterion threshold; S366: Compare the residual noise entropy value with the imaging availability criterion threshold. If the residual noise entropy value is less than or equal to the imaging availability criterion threshold, the third corrected waveform data is determined to meet the imaging availability requirements, and the current third corrected waveform data of the high-frequency guided wave sensor is used as the updated time-domain waveform data. If the residual noise entropy value is greater than the imaging availability criterion threshold, the third corrected waveform data is determined to not meet the imaging availability requirements, the current third corrected waveform data of the high-frequency guided wave sensor is marked as the third waveform data to be corrected, and the current iteration number is checked to see if it is less than the preset maximum iteration number. If so, the third waveform data to be corrected is used as input to re-enter S31 until the updated time-domain waveform data of the current high-frequency guided wave sensor is obtained. If not, the current third corrected waveform data is used as the updated time-domain waveform data.

8. A digital twin-based system for detecting defects on the inner surface of seamless steel gas cylinders, used to perform the digital twin-based method for detecting defects on the inner surface of seamless steel gas cylinders as described in any one of claims 1-7, characterized in that, Includes the following modules: Transmission module: Used to transmit modal guided waves to the cylinder of thin-walled seamless steel gas cylinder when the cylinder is in online use, the cylinder valve is not disassembled, and the cylinder is filled with liquefied gas or compressed gas medium. The module records the time-domain waveform data and spatial coordinates of each high-frequency guided wave sensor after the modal guided waves are reflected or scattered by the inner surface of the thin-walled seamless steel gas cylinder. Data acquisition module: used to acquire the geometric parameters and real-time service parameters of the current thin-walled steel seamless gas cylinder, and construct a three-dimensional geometric solid model of the current thin-walled steel seamless gas cylinder based on the geometric parameters. The spatial coordinates of each high-frequency guided wave sensor are registered to the outer surface of the three-dimensional geometric solid model of the current thin-walled steel seamless gas cylinder. Combined with the real-time service parameters of the current thin-walled steel seamless gas cylinder, a digital twin of the current thin-walled steel seamless gas cylinder is generated. Correction module: used to perform phased correction on the time-domain waveform data of each high-frequency guided wave sensor, and to judge the correction result of each phase, and to obtain the updated time-domain waveform data of each high-frequency guided wave sensor based on the judgment result. Region partitioning module: used to map the updated time-domain waveform data of each high-frequency guided wave sensor to the inner surface grid nodes of the digital twin of the thin-walled steel seamless gas cylinder, and calculate the scattering energy distribution of the inner surface grid nodes through the inverse time-off imaging algorithm to obtain the current inner surface defect candidate region set of the thin-walled steel seamless gas cylinder. Identification module: Based on the candidate region set of internal surface defects, it identifies the location and geometric contour of corrosion pits or cracks on the internal surface of the current thin-walled steel seamless gas cylinder.

Images