A seamless steel pipe geometry parameter automatic detection method and system

Abstract

The present application relates to the field of industrial measurement technology, in particular to a seamless steel pipe geometric parameter automatic detection method and system, the frequency domain phase characteristics of the initial section radius sequence are analyzed to determine the interference signal tendency of each frequency component, and the energy proportion is combined to quantify the frequency domain radius discreteness of each section, so that the surface high-frequency transient interference noise can be accurately stripped; the energy relative change and the frequency domain radius discreteness deviation between adjacent sections are extracted to quantify the section axial instability, and then the overall radius discreteness is dynamically fused, and the accidental error caused by the industrial field steel pipe operation jitter is overcome; finally, the overall radius discreteness is taken as a physical constraint condition conforming to the actual geometric distortion law, and is integrated into the compressed sensing target function to guide the inversion reconstruction, so that the accuracy of reconstructing the seamless steel pipe geometric parameter through sparse data acquisition is higher.

Description

Technical Field

[0001] This invention relates to the field of industrial measurement technology, specifically to an automatic detection method and system for the geometric parameters of seamless steel pipes. Background Technology

[0002] The geometric parameters of seamless steel pipes directly affect the assembly performance and safe service life of the product. Laser scanning technology is commonly used in online inspection to obtain cross-sectional radius data. However, due to physical space limitations such as frame layout and conveyor tracks in actual production lines, laser sensors often cannot be evenly and densely arranged around the circumference of the steel pipe. This results in a large number of sampling blind spots in the acquired measured radius data, exhibiting a highly sparse and incomplete state. To evaluate the overall geometry of the steel pipe based on sparse measured data, existing technologies typically use mathematical geometric fitting algorithms to smooth and complete the incomplete discrete radius data collected by laser sensors, attempting to reconstruct the full cross-sectional geometric profile of the seamless steel pipe.

[0003] However, simple mathematical curve smoothing only focuses on the mathematical continuity of the data surface, ignoring the geometric distortion laws of seamless steel pipes as rigid bodies determined by physical and mechanical mechanisms during the rolling process. It also fails to consider the susceptibility of detection signals to local surface abrupt noise and axial movement vibrations in complex industrial environments. This makes it difficult for reconstruction algorithms to effectively distinguish between true low-frequency physical deformation characteristics and high-frequency transient interference noise in the absence of sufficient sampling points. Consequently, the reconstructed contour curve is easily skewed by local artifacts and lacks true physical morphological support. Ultimately, this results in low accuracy and poor reliability of existing technologies for reconstructing the geometric parameters of seamless steel pipes using sparsely acquired data. Summary of the Invention

[0004] To address the low accuracy and poor reliability of existing technologies that reconstruct the geometric parameters of seamless steel pipes using sparsely acquired data, this invention aims to provide an automatic detection method and system for the geometric parameters of seamless steel pipes. The specific technical solution adopted is as follows: The first aspect of this invention provides an automatic detection method for the geometric parameters of seamless steel pipes, comprising: During the axial movement of the seamless steel pipe, all laser sensors arranged on the same circumferential cross section are synchronously sampled to obtain the cross-sectional angle position of each laser sensor on each cross section and the corresponding measured radius data. Based on the changing trend of the measured radius data of each laser sensor corresponding to the cross-sectional angle position on each cross-section, an initial cross-sectional radius sequence is determined; based on the frequency domain phase characteristics of the initial cross-sectional radius sequence, the interference signal tendency of each frequency component is determined; based on the interference signal tendency and energy proportion characteristics of each frequency component, the frequency domain radius discreteness of each cross-section is determined. Based on the relative energy changes of frequency components between adjacent cross sections and the frequency domain radius dispersion deviation, the axial instability of each cross section is determined; based on the axial instability and frequency domain radius dispersion of each cross section, the overall radius dispersion is determined. Using the overall radius discreteness as a constraint, a compressed sensing objective function is constructed for inversion reconstruction to determine the full cross-sectional geometric parameters of the seamless steel pipe.

[0005] Furthermore, the process of acquiring the interference signal tendency includes: The initial cross-sectional radius sequence is filtered and decomposed by a bandpass filter to determine the local energy and local phase of each frequency component at each cross-sectional angular position. Based on the local phase of all frequency components at each cross-sectional angular position, the corresponding vector average phase is calculated; based on the deviation between the local phase of each frequency component at each cross-sectional angular position and the vector average phase, a cosine function mapping is performed to determine the degree of component deviation; based on the local energy and the degree of component deviation, the effective coherent energy of each frequency component at each cross-sectional angular position is determined; both the local energy and the degree of component deviation are positively correlated with the effective coherent energy; The effective coherent energy of each frequency component at all cross-sectional angular positions is superimposed to determine the total coherent energy; the local energy of each frequency component at all cross-sectional angular positions is superimposed to determine the total component energy; based on the relative magnitude between the total coherent energy and the total component energy, the interference signal tendency of each frequency component is determined; the total coherent energy is positively correlated with the interference signal tendency, and the total component energy is negatively correlated with the interference signal tendency.

[0006] Furthermore, the process of obtaining the frequency domain radius discreteness includes: For each section: The total energy of the cross section is determined by superimposing the total energy of all frequency components; the reference energy ratio of each frequency component is determined based on the total energy of the components and the total energy of the cross section; the tendency of the interference signal is negatively correlated and mapped to determine the tendency weight. The tendency weight of each frequency component is introduced as a weight to weight the process of calculating the reference energy proportion information entropy of each frequency component, thereby determining the frequency domain radius dispersion of each cross section; the frequency domain radius dispersion is normalized to determine the frequency domain radius dispersion of each cross section.

[0007] Furthermore, the process of obtaining the axial instability of the cross-section includes: The total energy of the second-order frequency components corresponding to each cross section is taken as the corresponding principal harmonic energy. Based on the axial sampling sequence of the seamless steel pipe, the corresponding main harmonic variation rate is determined according to the relative deviation between the main harmonic energy of each section and the main harmonic energy of the previous section; an exponential positive correlation mapping is performed on the difference between the frequency domain radius discreteness of each section and the frequency domain radius discreteness of the previous section to determine the discrete fluctuation penalty factor; based on the main harmonic variation rate and the discrete fluctuation penalty factor, the axial instability of each section is determined; both the main harmonic variation rate and the discrete fluctuation penalty factor are positively correlated with the axial instability of the section.

[0008] Furthermore, the process of obtaining the overall radius discreteness includes: A negative correlation mapping is performed on the axial instability of each cross section to determine the corresponding contribution adjustment factor; based on the sum of the contribution adjustment factors of all cross sections, the contribution adjustment factor of each cross section is normalized to determine the comprehensive weight of each cross section. The frequency domain radius discreteness is weighted based on the cross-sectional comprehensive weight to determine the weighted discrete component of each cross-section; the weighted discrete components of all cross-sections are superimposed to determine the overall radius discreteness.

[0009] Furthermore, the process of obtaining the full cross-section geometric parameters includes: Construct a random observation matrix; the rows of the random observation matrix represent each laser sensor, and the columns represent each cross-sectional angle position; the element value of the corresponding element position of each laser sensor in the random observation matrix is ​​1, and the element value of other elements is 0; For each cross section, a data fidelity term is constructed based on the measured radius data of that cross section and the random observation matrix, with the overall radius discreteness as a constraint condition; a compressed sensing objective function is constructed by combining the data fidelity term and the constraint condition; the compressed sensing objective function is iteratively solved to determine the continuous radius vector after convergence for each cross section; the set of continuous radius vectors of each cross section is used as the full cross-sectional geometric parameters of the seamless steel pipe.

[0010] Furthermore, the bandpass filter is a Log-Gabor filter.

[0011] Furthermore, the method for normalizing the frequency domain radius dispersion employs minimum-maximum normalization.

[0012] Furthermore, the process of obtaining the initial cross-sectional radius sequence includes: Based on the cross-sectional angle position of each laser sensor on the current cross-section and the corresponding measured radius data, cubic spline interpolation fitting is performed to determine the initial cross-sectional radius sequence of the corresponding cross-section.

[0013] Secondly, the present invention provides an automatic detection system for the geometric parameters of seamless steel pipes, the system comprising: The data acquisition module is used to synchronously sample all laser sensors arranged on the same circumferential section during the axial movement of the seamless steel pipe, and to acquire the cross-sectional angle position of each laser sensor on each section and the corresponding measured radius data. The first determining module is used to determine an initial cross-sectional radius sequence based on the changing trend of the measured radius data of the cross-sectional angle positions corresponding to each laser sensor on each cross-section; determine the interference signal tendency of each frequency component based on the frequency domain phase characteristics of the initial cross-sectional radius sequence; and determine the frequency domain radius discreteness of each cross-section based on the interference signal tendency and energy proportion characteristics of each frequency component. The second determining module is used to determine the axial instability of each cross section based on the relative energy change of frequency components between adjacent cross sections and the frequency domain radius dispersion deviation; and to determine the overall radius dispersion based on the axial instability and frequency domain radius dispersion of each cross section. The geometric parameter detection module is used to construct a compressed sensing objective function with the overall radius discreteness as a constraint condition for inversion and reconstruction, and to determine the full cross-sectional geometric parameters of the seamless steel pipe.

[0014] Thirdly, the present invention provides a computer device including a memory and a processor. The memory is used to store computer program code, and the processor is used to call and run the computer program code from the memory to perform the method as described in the first aspect or any embodiment of the first aspect of the present invention.

[0015] The present invention has the following beneficial effects: This invention determines the interference signal tendency of each frequency component by analyzing the frequency domain phase characteristics of the initial cross-sectional radius sequence, and quantifies the frequency domain radius discreteness of each cross-section by combining energy proportion characteristics. This enables precise removal of high-frequency transient interference noise from the surface, effectively avoiding the misleading influence of local feature artifacts on the reconstruction process. Utilizing the axial rigidity continuity of the steel pipe, the axial instability of the cross-section is quantified by extracting the relative energy change of frequency components between adjacent cross-sections and the frequency domain radius discreteness deviation. This dynamically fuses the overall radius discreteness, overcoming the random errors caused by the vibration of the steel pipe during operation in industrial settings. Finally, this overall radius discreteness is used as a physical constraint condition conforming to the actual geometric distortion law, integrated into the compressed sensing objective function to guide the inversion reconstruction. This breaks the situation where the objective function easily falls into non-physical local optima under sparse data, ensuring that the reconstruction process strictly conforms to the true morphological mechanism of the seamless steel pipe even with incomplete observation data. This effectively overcomes the shortcomings of existing technologies that lack physical morphological support and have poor reliability in reconstructing contours, resulting in higher accuracy in reconstructing the geometric parameters of seamless steel pipes using sparsely acquired data. Attached Figure Description

[0016] Figure 1 This is a flowchart of an automatic detection method for geometric parameters of seamless steel pipes provided in one embodiment of the present invention; Figure 2 This is a structural diagram of an automatic detection system for the geometric parameters of a seamless steel pipe, provided in one embodiment of the present invention. Detailed Implementation

[0017] The following description, in conjunction with the accompanying drawings, details the specific scheme of the automatic detection method and system for the geometric parameters of seamless steel pipes provided by this invention.

[0018] This invention provides an automatic detection method for the geometric parameters of seamless steel pipes. Please refer to [link / reference]. Figure 1 The diagram illustrates a flowchart of an automatic detection method for the geometric parameters of seamless steel pipes according to an embodiment of the present invention. The method includes: Step S101: During the axial movement of the seamless steel pipe, all laser sensors arranged on the same circumferential cross section are synchronously sampled to obtain the cross-sectional angle position of each laser sensor on each cross section and the corresponding measured radius data.

[0019] Downstream of the sizing mill outlet in the seamless steel pipe production line, a detection area is set up along the axial conveying channel of the seamless steel pipe. Within this detection area, an annular detection bracket, perpendicular to the axial direction of the seamless steel pipe and spanning the conveying rollers, is installed, allowing the seamless steel pipe to pass through its center. To obtain the outer contour data of the seamless steel pipe's cross-section, a set of laser sensors is arranged around the pipe's travel channel on the annular detection bracket. During installation, the optical paths of each sensor are adjusted to ensure that the measuring beams of all laser sensors are strictly within the same physical plane perpendicular to the seamless steel pipe's axis, thus guaranteeing that all laser sensors are targeting the same circumferential cross-section of the pipe. In actual industrial settings, due to spatial constraints imposed by surrounding mechanical support structures and auxiliary equipment, laser sensors typically cannot be arranged in an absolutely uniform and high-density manner on the same circumferential cross-section. To provide the richest possible observational information for the subsequent inversion and reconstruction process, thereby improving the detection accuracy of the full-section geometric parameters, as many laser sensors as possible are added within all available installation space gaps, avoiding interference from on-site mechanical structures. During deployment, non-uniformly spaced, staggered installations were directly utilized using available space gaps on-site, breaking the traditional uniform distribution pattern of sensors. This avoided periodic coupling between fixed measurement points and the regular deformation areas of the steel pipe caused by the rolling process. After all hardware equipment was fixed, the actual physical installation orientation of each laser sensor was precisely calibrated and recorded using on-site calibration instruments. This information was then used to determine the angular position of each laser sensor on various cross-sections using the constructed polar coordinate system and store it in the system.

[0020] During the axial translation of the seamless steel pipe along the bottom conveyor rollers and its passage through the annular detection bracket, strict synchronous sampling control is required to ensure that laser sensors at different angles can accurately scan the same physical cross-section on the pipe. Specifically, an incremental rotary encoder, coaxially connected to the conveyor roller drive motor, acquires the current axial translation speed of the seamless steel pipe in real time. The detection system automatically calculates the required real-time sampling frequency based on the preset axial resolution requirements and this real-time translation speed. Subsequently, the system's synchronous controller continuously generates external hardware trigger pulse signals based on this sampling frequency, synchronously sending them to all laser sensors on the annular detection bracket. Since all sensors are physically constrained to the same circumferential cross-sectional plane, when each laser sensor receives the trigger pulse at the same moment, it synchronously emits a measuring laser beam towards the surface of the seamless steel pipe and receives the reflected echo, thereby calculating the instantaneous physical distance from the sensor's transmitting end to the outer surface of the steel pipe. The central processing unit reads the distance data of the same trigger batch in real time and, combined with the pre-calibrated installation distance reference of each sensor, converts it into the actual radius value of the seamless steel pipe at the corresponding angle at that measurement moment. This completes the synchronous sampling of the current circumferential cross-section and obtains the measured radius data of each laser sensor on that cross-section. As the seamless steel pipe continues to move axially, the system continuously performs cross-section-level synchronous trigger sampling at the above frequency, thereby continuously acquiring the measured radius data corresponding to each cross-section distributed along the axial direction of the steel pipe.

[0021] In one specific implementation of this invention, the sampling frequency of all laser sensors is calculated based on the real-time axial translation speed of the seamless steel pipe and a preset detection axial resolution. Specifically, the real-time axial translation speed is divided by the detection axial resolution, and the quotient is used as the sampling frequency of all laser sensors. The cross-sectional angle position is obtained based on the polar coordinate system of the circumferential cross-section of the seamless steel pipe. Specifically, the geometric center of the cross-section of the seamless steel pipe is used as the origin of the polar coordinate system, and the physical vertical upward direction of the annular detection bracket is used as the polar axis of the polar coordinate system and set as the starting zero-degree angle, thereby establishing a cross-sectional polar coordinate system. Under this polar coordinate system, the entire cross-section is uniformly divided into a preset number of discrete angle points along the circumferential direction, and the fixed angle interval between adjacent discrete angle points is calculated. Subsequently, the actual installation polar angle of each laser sensor under this polar coordinate system is measured, the actual installation polar angle is divided by the fixed angle interval and rounded to the nearest integer, thereby matching the discrete angle point closest to the actual installation polar angle, and the discrete angle point is used as the cross-sectional angle position of the laser sensor in the circumferential direction. In one specific implementation of this invention, the preset detection axial resolution is set to 0.5 mm, which can be adjusted according to the accuracy requirements for detecting the geometric parameters of seamless steel pipes in the specific implementation environment. In another specific implementation of this invention, the preset quantity is set to 360, which can be adjusted according to the required level of detail in the contour reconstruction of the parameters to be measured in the specific implementation environment.

[0022] Step S102: Determine the initial cross-sectional radius sequence based on the changing trend of the measured radius data of the cross-sectional angle position corresponding to each laser sensor on each cross-section; determine the interference signal tendency of each frequency component based on the frequency domain phase characteristics of the initial cross-sectional radius sequence; determine the frequency domain radius discreteness of each cross-section based on the interference signal tendency and energy ratio characteristics of each frequency component.

[0023] Considering that the measured radius data obtained by sparsely arranged laser sensors is highly discrete and incomplete in the circumferential space, and cannot be directly used for subsequent frequency domain decomposition and continuous contour analysis, it is necessary to perform basic data completion and trend transition on these scattered observation points based on the physical continuity characteristics of the seamless steel pipe cross-section. Therefore, the initial cross-sectional radius sequence is determined according to the changing trend of the measured radius data of each laser sensor at the corresponding cross-sectional angle position on each cross-section. By initially transforming the local discrete measured points into smooth transition data covering the entire circumference, a complete basic data carrier is provided for subsequent accurate extraction of frequency domain features and construction of reconstruction constraints.

[0024] In one specific implementation of this invention, based on the cross-sectional angle positions of each laser sensor on the current cross-section and the corresponding measured radius data, cubic spline interpolation fitting is performed to determine the initial cross-sectional radius sequence of the corresponding cross-section. That is, all discrete angle and radius data pairs acquired by the laser sensors within the current cross-section are extracted, and the polynomial coefficients between adjacent observation points are calculated using a cubic spline mathematical model. This smoothly extrapolates the estimated radius values ​​corresponding to all unmeasured angle positions on the entire circumference, forming a continuous initial contour curve. Cubic spline interpolation can match the smoothness of geometric deformation and physical inertia exhibited by seamless steel pipes as rigid structures under the pressure of rolling mills. It effectively avoids the non-physical abrupt changes in the polygonal shape caused by simple linear interpolation, thus maximally matching the true macroscopic physical deformation trend of the steel pipe in the preliminary preprocessing stage, preventing distorted initial fitting artifacts from affecting the subsequent frequency domain decomposition and reconstruction process.

[0025] Due to the presence of local surface abrupt noises such as oxide scale burrs or sensor electrical transients in complex industrial environments, and the inability of simple mathematical geometric fitting algorithms to distinguish these high-frequency transient interferences from the actual physical deformation profile of the steel pipe, this invention aims to accurately identify and remove non-physical interference artifacts from incomplete observation data. Based on the frequency domain phase characteristics of the initial cross-sectional radius sequence, the invention determines the interference signal tendency of each frequency component. By analyzing the phase consistency of different frequency components in the circumferential space, the system can effectively determine the contribution of each frequency component to the abrupt interference signal, thereby avoiding the misleading influence of local feature artifacts on the reconstruction process.

[0026] Preferably, in some possible implementations of the embodiments of the present invention, the process of obtaining the tendency of the interference signal includes: The initial cross-sectional radius sequence is decomposed by a bandpass filter to determine the local energy and local phase of each frequency component at each cross-sectional angle. In other words, the smoothed initial cross-sectional radius sequence is transformed from the spatial domain to the frequency domain using bandpass filtering, extracting the amplitude and phase information of each specific frequency component at discrete positions on the circumference. In one specific implementation of this invention, a Log-Gabor filter is used because it has a Gaussian transfer function on the logarithmic frequency scale and strictly lacks a DC component. This not only overcomes the defect of traditional Gabor filters having non-zero DC components when the bandwidth is too large, but also its frequency response characteristics better match the wideband distribution characteristics of the real physical contour. It can capture multi-scale physical deformation and abrupt noise on the pipe wall surface with extremely high local spatial and frequency resolution, providing more accurate basic characteristic data for subsequent phase coherence analysis.

[0027] The mean value of the local phase of each frequency component at each cross-sectional angular position is calculated to obtain the corresponding vector average phase. A cosine function mapping is then performed based on the deviation between the local phase of each frequency component at each cross-sectional angular position and the vector average phase to determine the degree of component deviation. In this embodiment of the invention, the cosine value of the difference between the local phase of each frequency component at each cross-sectional angular position and the vector average phase is input into the ReLU function, and the corresponding degree of component deviation is output.

[0028] The vector average phase here represents the overall phase concentration trend and dominant structural direction of all frequency components at the angular position of the cross section. Therefore, for each angular position of the cross section, the larger the difference between the local phase of each frequency component at each angular position and the corresponding vector average phase, the further the phase change of that frequency component at that position deviates from the dominant structural trend of the overall signal. That is, the weaker the contribution of that component to the significant structural features (such as interference spikes) at that position. Therefore, by calculating the cosine value of this difference, when the phase is highly consistent (the difference approaches 0 degrees), the output is close to 1 with a high weight, while when the phase deviation is large, the output is a small or even negative value. At this time, by inputting this cosine value into the ReLU function for non-negative truncation, all negative numbers can be forced to 0, thereby effectively avoiding the energy cancellation problem caused by phase inversion and ensuring the monotonicity and physical rationality of the subsequent calculation evaluation index.

[0029] Furthermore, based on the product between local energy and component deviation, the effective coherent energy of each frequency component at each cross-sectional angular position is determined. Since the component deviation characterizes how closely the phase of the frequency component is aligned with the dominant trend of the overall signal at a specific angular position, it is used as a weight to weight the local energy. This ensures that only when the phase of the frequency component is highly consistent with the overall dominant phase is its corresponding local energy considered as effectively superimposed coherent energy. This effectively filters out unstructured background white noise energy that has high local amplitude but disordered phase, and accurately extracts the characteristic energy used to construct abrupt signal.

[0030] The total coherent energy is determined by superimposing the effective coherent energy of each frequency component at all cross-sectional angular positions; the total component energy is determined by superimposing the local energy of each frequency component at all cross-sectional angular positions; and the interference signal tendency of each frequency component is determined based on the relative magnitude between the total coherent energy and the total component energy.

[0031] In one specific implementation of this invention, the process of obtaining the interference signal tendency is expressed by the following formula: ;in, cross section The corresponding number Interference signal tendency of first-order frequency components; The total number of angular positions for each cross-section; cross section The corresponding number The first frequency component is at the 1st order. Effective coherent energy at each cross-sectional angular position; cross section The corresponding number The first frequency component is at the 1st order. Local energy at each cross-sectional angular position; cross section The corresponding number The sum of coherent energy of the first-order frequency components; cross section The corresponding number The total energy of the first-order frequency components. It should be noted that if... If the value is 0, it means that the frequency component of that order has no energy distribution on the current cross section. In this case, the characteristic data of the component is considered abnormal or has no analytical value, and the interference signal tendency of the frequency component of that order is directly set to 0.

[0032] When acquiring the tendency of interference signals, the ratio is obtained by dividing the total effective coherent energy of a certain frequency component over the entire circumference by the total energy of its original component. According to the laws of physical signal synthesis, to form a sharp and interfering pulse signal (such as the spikes generated by sensor electrical transients), multiple frequency components must exhibit high phase alignment at the same position. Therefore, the larger the calculated ratio (approaching 1), the more the total energy of that frequency component is consumed in the highly phase-consistent structured superposition, indicating that the component has a high potential to construct non-physical local abrupt artifacts, and its tendency as an interference signal is higher. Conversely, if the calculated ratio is small (approaching 0), it indicates that the energy of that component is mostly phase-displaced on the circumference, which closely matches the inherent smooth and gradual physical property of seamless steel pipe as a rigid body, and its tendency as an interference signal is extremely low. By calculating the relative ratio of these two energy parameters, the contribution of different frequency components to the interference signal can be objectively measured in a dimensionless form, thereby achieving scientific quantification and precise targeting of non-physical interference characteristics.

[0033] While the spectral data of the initial cross-sectional radius sequence obtained previously can initially reflect the energy distribution of each specific frequency component of the cross-section, direct analysis based on the original frequency domain energy is easily biased by high-frequency artifact energy, leading to the failure of reconstruction constraints. Therefore, this embodiment of the invention takes into account that the deformation of the cross-section of a real seamless steel pipe naturally possesses extremely strong low-frequency sparsity characteristics in the frequency domain, and that the contribution of each frequency component to the actual interference artifacts is different. Therefore, based on the interference signal tendency and energy proportion characteristics of each frequency component, the frequency domain radius discreteness of each cross-section is determined. By constructing a weighted information entropy that introduces anti-interference tendency weights, the energy influence of interference characteristics can be accurately suppressed when quantifying the frequency concentration of cross-sectional deformation, thereby accurately characterizing the cross-sectional morphology distribution that conforms to the real physical properties of the steel pipe.

[0034] Preferably, in some possible implementations of the embodiments of the present invention, the process of obtaining the frequency domain radius discreteness includes: For each section: The total energy of the cross section is determined by superimposing the total energy of all frequency components. Based on the total component energy and the total cross section energy, the reference energy percentage of each frequency component is determined. In this embodiment, the cumulative value of the total component energy of all frequency components is used as the corresponding total cross section energy. Based on the ratio between the total component energy and the total cross section energy, the reference energy percentage of each frequency component is determined. By calculating the proportion of the absolute energy of each frequency component in the overall cross section energy, the absolute energy differences between different test batches and different steel pipe specifications are effectively eliminated. The strength distribution of each frequency component is uniformly transformed into a dimensionless probability space, providing a standardized data foundation for subsequent accurate quantification of the energy dispersion of cross section deformation based on information entropy theory.

[0035] The tendency of the interference signal is negatively correlated and mapped to determine the tendency weight. The tendency weight of each frequency component is introduced as a weight to weight the process of calculating the information entropy of the reference energy proportion of each frequency component, thereby determining the frequency domain radius dispersion of each cross section.

[0036] In one specific implementation of this invention, the process of obtaining the frequency domain radius discreteness is expressed by the formula: ;in, cross section The degree of dispersion of the frequency domain radius; cross section The number of corresponding frequency components; cross section The corresponding number Interference signal tendency of first-order frequency components; cross section The corresponding number The reference energy percentage of the first frequency component; It is a logarithmic function with base 2; cross section The corresponding number The tendency weight of the first frequency component.

[0037] During the rolling process of seamless steel pipes, due to the physical constraints of rigid body mechanics, the macroscopic physical deformation of its cross-section is necessarily low-frequency and highly smooth. This deformation naturally manifests in the frequency domain as energy being extremely concentrated on a very small number of low-frequency orders. According to the mathematical properties of information entropy, when the cross-sectional energy is highly concentrated in a few frequency components, the calculated entropy value (i.e., the degree of dispersion) is extremely small; while when the energy is randomly distributed across a large number of frequency points, the degree of dispersion is extremely large. However, the frequency distribution measured in actual measurements often mixes surface roughness with high-frequency noise energy introduced by the detection environment. If the information entropy is calculated directly, these randomly distributed high-frequency noises will cause the entropy value to be artificially high, thus severely masking the true low-frequency sparsity characteristics of the steel pipe. This invention penalizes the information entropy calculation process by using a bias weight, derived from a negative correlation mapping of the interference signal's bias, to forcibly suppress the contribution of local artifact features (bias approaching 1) formed by highly coherent phase stacking in the entropy calculation (weight approaching 0). This precisely removes non-physical noise interference terms from the complex original frequency domain data, ensuring that the final quantized frequency domain radius dispersion purely and objectively reflects the sparse characteristics of the steel pipe's true physical contour, providing physically meaningful constraint boundaries for reconstruction and inversion. It should be noted that if the reference energy percentage is 0, it indicates that this frequency component does not have any effective deformation energy characteristics on the current cross-section. In this case, the information entropy benchmark term corresponding to this frequency component is directly set to 0, preventing it from participating in the final accumulation calculation of the frequency domain radius dispersion, thus effectively avoiding system computational crashes caused by logarithmic operation failures.

[0038] Finally, to eliminate the dimensional and order-of-magnitude deviations caused by differences in the number of sensors, sampling density fluctuations, and energy scales between seamless steel pipes of different specifications or different testing batches, and to ensure that the discreteness index of the entire cross-section can participate in subsequent global fusion and inversion reconstruction under a unified benchmark, this embodiment of the invention further normalizes the frequency domain radius discreteness to determine the frequency domain radius discreteness of each cross-section. This embodiment of the invention uses a minimum-maximum normalization method to normalize the frequency domain radius discreteness. The minimum and maximum values ​​are determined by sorting the frequency domain radius discreteness calculated from all synchronously scanned cross-sections along the axial direction in a single seamless steel pipe testing experiment and extracting the minimum and maximum values ​​respectively.

[0039] Step S103: Based on the relative energy change of frequency components between adjacent cross sections and the frequency domain radius dispersion deviation, determine the cross section axial instability of each cross section; based on the cross section axial instability and frequency domain radius dispersion of each cross section, determine the overall radius dispersion.

[0040] Considering that in actual industrial settings, seamless steel pipes are prone to random physical vibrations in the lateral or radial direction during transport and inspection due to their own weight, mechanical collisions, or uneven running tracks, such mechanical vibrations can cause severe irregular positional distortions in the measured radius data acquired by the sensors. This leads to significant random errors and distortion risks when relying solely on data from a single cross-section for discreteness assessment. Therefore, this invention determines the axial instability of each cross-section based on the relative energy changes of frequency components between adjacent cross-sections and the frequency domain radius discreteness deviation. By introducing a comprehensive evaluation mechanism for physical continuity and stability along the axial dimension of the steel pipe, the system can effectively identify and quantify the degree of data distortion caused by local operational vibrations, thereby providing a reliable basis for weight adjustment for subsequent high-quality dynamic fusion of global features.

[0041] Preferably, in some possible implementations of the embodiments of the present invention, the process of obtaining the axial instability of the cross section includes: The total energy of the second-order frequency components corresponding to each cross section is taken as the corresponding principal harmonic energy. Based on the axial sampling sequence of the seamless steel pipe, the corresponding principal harmonic variation rate is determined according to the relative deviation between the principal harmonic energy of each cross section and the principal harmonic energy of the previous cross section. In this embodiment of the invention, the corresponding harmonic energy deviation is determined based on the absolute value of the difference between the principal harmonic energy of each cross section and the principal harmonic energy of the previous cross section. The corresponding principal harmonic variation rate is determined based on the ratio between the harmonic energy deviation and the principal harmonic energy of each cross section.

[0042] The most typical process deformation of seamless steel pipes is usually manifested as elliptic deformation caused by the extrusion of rolls or gravity. The second-order frequency component physically corresponds precisely to this elliptical deformation characteristic of the pipe, and its energy often dominates the cross-sectional deformation. Therefore, extracting the second-order frequency component as the main harmonic energy can most sensitively and intuitively capture the fluctuation of the macroscopic physical contour of the steel pipe. By calculating the rate of change in the form of a ratio, the interference of the absolute size of the pipe diameter on the deviation is effectively eliminated, and a standardized measurement of the change amplitude of the main contour of adjacent cross-sections is achieved.

[0043] An exponential positive correlation mapping is performed between the frequency domain radius discreteness of each cross section and the frequency domain radius discreteness of the previous cross section to determine the discrete fluctuation penalty factor. Based on the main harmonic variation rate and the discrete fluctuation penalty factor, the axial instability of each cross section is determined. It should be noted that since the first cross section does not have a previous cross section, this embodiment of the invention directly sets the main harmonic variation rate of the first cross section to 0 and the discrete fluctuation penalty factor of the first cross section to 1. This assigns a default basic stationary state to the first scanning cross section without affecting the logical coherence of the overall axial stability evaluation and the effectiveness of subsequent calculations. Furthermore, for each cross section, if its corresponding main harmonic energy is 0, the corresponding main harmonic variation rate is directly set to 0 to avoid the situation where dividing by 0 makes the formula meaningless.

[0044] In one specific implementation of this invention, the process of obtaining the axial instability of the cross-section is expressed by the following formula: ;in, cross section Axial instability of the cross section; cross section The rate of change of the principal harmonic; The preset control coefficient; cross section The frequency domain radius discreteness; cross section The frequency domain radius discreteness of the previous cross section; It is an exponential function with the natural constant as its base.

[0045] In one specific implementation of this invention, the preferred range of the preset control coefficient is set as follows: The preset control coefficient can be adjusted automatically according to the expected vibration intensity of the conveyor rollers in the specific implementation environment and the system's tolerance to noise. In this embodiment, it is set to 2.0. The preset control coefficient is used to characterize the sensitivity of the exponential penalty for the rate of change of the main harmonic using the difference in frequency domain radius dispersion. In the specific implementation environment, when the flatness of the conveyor rollers is poor and the mechanical vibration is severe, causing the system to easily collect distorted jitter data, in order to identify and eliminate these unreliable cross-sectional data, it is necessary to increase the value of the preset control coefficient (e.g., 3.0), so that even if there is only a small fluctuation in frequency domain radius dispersion between adjacent cross-sections, a drastically amplified discrete fluctuation penalty factor will be generated. Conversely, if the test environment is extremely stable or the system has a high tolerance for small physical changes, it is necessary to decrease the value of the preset control coefficient (e.g., 1.5) to prevent excessive penalty for normal axial physical gradients.

[0046] This invention places the discrete deviation, reflecting microscopic statistical differences, into the exponential term to determine the discrete fluctuation penalty factor used to measure the degree of signal mutation. This penalty factor is then directly multiplied by the rate of change of the main harmonic, reflecting changes in the macroscopic structure. Based on the combined product of these two factors, the axial instability of each cross-section is determined. According to the conservation of rigid body mechanics and the law of axial continuity distribution, if the seamless steel pipe remains stable during transport, the main shape of its cross-sectional profile and the frequency domain sparsity characteristics within it will not undergo instantaneous mutations. Once mechanical vibration occurs, the sensor scanning surface will tilt and relatively misalign, which will not only cause abnormal jumps in the main frequency energy reflecting the main profile but also cause discontinuities in the overall discrete characteristics of the cross-section. This calculation mechanism allows abnormal fluctuations in discrete characteristics to amplify macroscopic changes in a nonlinear manner. When the main frequency energy transitions smoothly and the discrete characteristics are consistent, both the discrete fluctuation penalty factor and the rate of change of the main harmonic are small, and the system determines that the cross-section is in a stable state and outputs a low instability. When there is a coordinated mutation between the two, the instability value after multiplication increases exponentially. Through this design, the present invention achieves the capture and locking of distorted data caused by random mechanical jitter, avoiding interference from the cross-sectional characteristics of accidental errors in the subsequent analysis process.

[0047] Considering the impact of local jitter, the frequency domain radius dispersion calculated from each independent cross-section has varying degrees of reliability differences. If the arithmetic mean or the characteristics of a single cross-section are directly used as global reconstruction constraints, it is easy to be biased by individual distorted data. Therefore, it is necessary to combine the characteristics of multiple cross-sections distributed along the axial direction for comprehensive screening and dynamic evaluation. Thus, this embodiment of the invention determines the overall radius dispersion based on the axial instability and frequency domain radius dispersion of each cross-section. By combining the instability to perform weighted fusion of the dispersion characteristics of each cross-section, the final obtained feature parameters can eliminate the deviation caused by occasional jitter and truly reflect the physical expectation of the axial geometry of the seamless steel pipe.

[0048] Preferably, in some possible implementations of the embodiments of the present invention, the process of obtaining the overall radius discreteness includes: A negative correlation mapping is performed on the axial instability of each cross section to determine the corresponding contribution adjustment factor. Specifically, a reference product is determined based on the product of a preset nonlinear penalty factor and the axial instability of each cross section. The negative of the reference product is used as the exponent of an exponential function with the natural constant as its base, and the output value of this exponential function is calculated and used as the corresponding contribution adjustment factor. This negative exponential mapping method ensures that the calculated contribution adjustment factor decreases nonlinearly for cross sections with higher axial instability. When the instability of a cross section increases due to operational jitter, the system can forcibly reduce the contribution ratio of that cross section in the global evaluation. For cross sections with stable operation and instability close to zero, the contribution adjustment factor is made close to 1, thus retaining its higher weight.

[0049] In one specific implementation of this invention, the preferred range of the preset nonlinear penalty factor is set as follows: The value can be adjusted according to the specific implementation environment's rejection of local jitter data and the smoothing requirements of feature fusion. In this embodiment, it is set to 2.0. The preset nonlinear penalty factor is used to characterize the penalty strength when the cross-sectional contribution is attenuated using a negative exponential function. When there is a lot of interference in the specific implementation environment and the system needs to exclude cross-sectional data with slight jitter distortion from participating in global fusion, the value of the preset nonlinear penalty factor needs to be increased (e.g., to 3.0), so that even if the axial instability of the cross-section increases slightly, the contribution adjustment factor of the cross-section will be significantly attenuated, thereby forcibly reducing the weight of the cross-section. Conversely, if the test environment is relatively stable, or the system needs to show a high tolerance for cross-sectional data with slight instability to ensure that enough cross-sectional features participate in fusion, the value of the preset nonlinear penalty factor needs to be reduced (e.g., to 1.5), so that the exponential decay is smoother and the weight of cross-sectional features with normal physical gradual changes is not excessively reduced.

[0050] The contribution adjustment factor for each cross-section is normalized by summing the contribution adjustment factors of all cross-sections to determine the comprehensive weight of each cross-section. In other words, the comprehensive weight of the cross-section is determined by the ratio between the contribution adjustment factor of each cross-section and the sum of the contribution adjustment factors of all cross-sections. This ensures that the weight ratio of each cross-section is uniformly mapped to between 0 and 1, and the sum of the comprehensive weights of all cross-sections is strictly constrained to 1, thereby eliminating fluctuations in the evaluation benchmark caused by differences in the total number of cross-section samples under different steel pipe lengths or sampling densities. It should be noted that if the sum of the contribution adjustment factors of all cross-sections is 0, it indicates that all synchronously scanned cross-sections in the current testing experiment are in a state of severe physical jitter and data failure. In this case, the current batch of testing and reconstruction process is directly terminated, and the system issues a data acquisition anomaly warning.

[0051] The frequency domain radius dispersion is weighted based on the cross-sectional comprehensive weight to determine the weighted discrete component of each cross-section; the weighted discrete components of all cross-sections are superimposed to determine the overall radius dispersion. In this embodiment, the weighted discrete component of each cross-section is determined based on the product between the cross-sectional comprehensive weight and the frequency domain radius dispersion; the overall radius dispersion is determined based on the cumulative value of the weighted discrete components of all cross-sections. By multiplying and summing the characteristics of each cross-section with the comprehensive weight reflecting its reliability, the random deviations caused by local distorted cross-sections are smoothly filtered out, making the final output overall radius dispersion a weighted statistical expectation reflecting the structural characteristics of the entire steel pipe section. This provides a stable and physically supported reliable constraint for the subsequent construction of the objective function.

[0052] Step S104: Construct a compressed sensing objective function with the overall radius discreteness as a constraint condition for inversion and reconstruction to determine the full cross-sectional geometric parameters of the seamless steel pipe.

[0053] Finally, considering that the measured radius data directly obtained by the laser sensor is highly sparse on the circumference and has a large number of measurement blind spots, simple interpolation or mathematical geometric fitting cannot truly restore the physical deformation of the steel pipe caused by the extrusion of the rolls. Therefore, traditional curve fitting is prone to falling into non-physical local artifacts and distortions. Therefore, this embodiment of the invention uses the overall radius discreteness as a constraint to construct a compressed sensing objective function for inversion reconstruction to determine the full cross-sectional geometric parameters of the seamless steel pipe. By incorporating the overall features obtained based on the all-axial stability assessment into the mathematical solution model, the contour restoration under the condition of incomplete data can strictly fit and be subject to the real physical deformation mechanism of the steel pipe, and finally output a high-precision continuous geometric contour.

[0054] Preferably, in some possible implementations of the embodiments of the present invention, the process of obtaining the full cross-section geometric parameters includes: Construct a random observation matrix; the rows of the random observation matrix represent each laser sensor, and the columns represent each cross-sectional angle position; the element position corresponding to the cross-sectional angle position of each laser sensor in the random observation matrix has a value of 1, and the element position of other elements has a value of 0; the row arrangement order of the random observation matrix corresponds one-to-one with the actual installation polar angle increasing order corresponding to the cross-sectional angle positions of all laser sensors on the current cross-section, and the column arrangement order of the random observation matrix corresponds one-to-one with the spatial increasing order of all discrete angle points divided along the circumferential direction in the polar coordinate system.

[0055] For each cross section, a data fidelity term is constructed based on the measured radius data and random observation matrix of that cross section, with the overall radius discreteness as a constraint condition; a compressed sensing objective function is constructed by combining the data fidelity term and the constraint condition; the compressed sensing objective function is iteratively solved to determine the continuous radius vector after convergence for each cross section; the set of continuous radius vectors of each cross section is used as the full cross section geometric parameters of the seamless steel pipe.

[0056] In one specific implementation of this invention, the compressed sensing objective function is expressed by the formula: ;in, This indicates the search for the variable to minimize the objective function within the parentheses; Let be the continuous radius vector of the k-th cross section to be solved; The measured data vector is formed by arranging the measured radius data obtained by each laser sensor on the k-th cross section in ascending order according to the corresponding actual installation polar angle. The matrix represents random observations. L2 norm operator; L1 norm operator; For data fidelity items; The overall radius is discrete; It is a discrete Fourier transform basis; The preset regularization weight coefficients; This is a frequency domain sparsity penalty term.

[0057] In one specific implementation of this invention, the preferred range of values ​​for the preset regularization weight coefficient is set as follows: The value can be adjusted automatically based on the sparsity of measured data caused by the difference in the number of laser sensors in the specific implementation environment. In this embodiment, it is set to 0.1. The preset regularization weight coefficient is used to balance the ratio of the data fidelity term and the dynamic frequency domain sparsity penalty term in the objective function. In a specific implementation environment, when the laser sensors are arranged extremely sparsely or the on-site interference noise is extremely high, and the system needs to rely more on the physical prior frequency domain sparsity characteristics to forcibly extrapolate the data of the measurement blind zone and filter out high-frequency noise, the value of the preset regularization weight coefficient needs to be increased (e.g., 0.2) to increase the weight of the dynamic frequency domain sparsity penalty term. Conversely, when the laser sensors are arranged relatively densely and the measured data quality is high, and the system needs to retain as much as possible the real local geometric gradient details of the seamless steel pipe surface to prevent the reconstructed full-section geometric parameters from being too smooth and distorted, the value of the preset regularization weight coefficient needs to be decreased (e.g., 0.05) to make the model more dependent on the objective measured data fidelity term.

[0058] The Discrete Fourier Transform (DFT) basis is a standard complex matrix in the field of digital signal processing, which can be directly obtained using existing DFT matrix construction techniques. The specific acquisition process is as follows: The number of rows and columns of the matrix is ​​equal to the total number of discrete angle points along the circumference in polar coordinates. For any element in the matrix, its corresponding row and column indices, starting from 0, are obtained. The product of these indices and the value of pi (pi) is multiplied by the total number of discrete angle points, and a negative imaginary unit is substituted. This result is used as the exponent of an exponential function with the natural constant as its base. The calculated complex value is the element value at the corresponding position in the matrix. By traversing all rows and columns of the matrix, a complete DFT basis can be generated. In practical implementation, those skilled in the art can directly construct the matrix according to the definition principle of the Fourier transform of the above standard, or directly call the Fourier transform basis generating function library in conventional numerical calculation software to obtain the constant matrix; its main function is to provide a fixed mathematical projection space, and in each solution iteration, the continuous radius vector of the space to be solved is losslessly transformed to the frequency domain, so that the algorithm can accurately extract the sparse features of the corresponding frequency coefficients and apply dynamic constraints, thereby guiding the objective function to converge quickly to the optimal solution that conforms to the physical law of low-frequency deformation of rigid bodies.

[0059] When constructing and solving the compressed sensing objective function, the amount of continuous data across the entire cross-section to be solved is far greater than the actual number of deployed sensors, making it a typical mathematically underdetermined problem. To pinpoint a unique physical true solution from an infinite number of mathematical solutions, this invention designs a solution architecture driven by a data fidelity term and a frequency-domain sparsity penalty term that incorporates the overall radius discreteness. The data fidelity term forces constraints on the solution model, ensuring that the reconstructed cross-sectional profile approximates the actual measured values ​​at the angular positions of the existing sensor placements, thus serving as a fundamental premise for guaranteeing the objectivity and accuracy of the reconstructed parameters. In this embodiment, the square of the second norm is used instead of the simple second norm. This makes the objective function a smooth convex function, significantly reducing the computational complexity of partial derivatives and the iteration convergence time when solving the model. Furthermore, it aligns with the physical statistical law that laser sensor measurement errors follow a Gaussian distribution, allowing for a quadratic penalty to be applied to large deviations at the measured points.

[0060] The frequency domain sparsity penalty term aims to introduce prior knowledge of the rigid deformation of seamless steel pipes. Since the actual deformation of steel pipes is mainly composed of a few low-frequency dominant harmonics, its frequency domain performance is naturally sparsity-oriented. By minimizing this term, the iterative algorithm is forced to automatically discard high-frequency random spikes and prioritize using the fewest low-frequency smooth waveforms to fit the incomplete data, thereby reasonably deducing the transition contour within the measurement blind zone. The overall radius dispersion obtained from the previous assessment based on all-axial physical stability is directly introduced into the frequency domain sparsity penalty term as a dynamic adjustment weight. When the overall radius dispersion is high, it indicates that the overall deformation of the current batch of steel pipes is irregular or there is serious physical jitter. This weight is used to automatically amplify the penalty intensity of frequency domain sparsity, using a more forceful method to eliminate high-frequency variation coefficients and suppress the generation of non-physical sawtooth transition artifacts. Conversely, when the overall radius dispersion is extremely low and exhibits obvious low-frequency steady-state characteristics, the penalty intensity is automatically reduced, thereby allowing the model to retain more realistic local geometric gradient details while ensuring the smoothness of the overall contour. By directly elevating the global evaluation index, which reflects the real physical mechanism, to the dynamic boundary of the objective function, the traditional reconstruction algorithm is prone to getting trapped in local mathematical fitting. This ensures that the reconstruction trend strictly conforms to the rigid mechanical properties of the seamless steel pipe, and achieves physical-level fidelity restoration of the geometric parameters of the entire cross section.

[0061] After clarifying the constraints of the objective function, in order to transform the mathematical model into specific contour data, this embodiment of the invention employs the alternating direction multiplier method to iteratively solve the compressed sensing objective function. In the specific iterative solution process, the alternating direction multiplier method introduces auxiliary variables and Lagrange multipliers to decompose the complex optimization problem, which originally contained non-differentiable terms of the first norm, into multiple easily solvable local subproblems for alternating updates. During each iteration, dynamic weights incorporating the discreteness of the overall radius are used to intervene in the soft threshold shrinkage operation during the update process in real time. After multiple alternating iterations, the iteration stops when the calculation result of the objective function meets the preset convergence condition, and the output variable at this point is used as the continuous radius vector after the current cross-section converges. Finally, the continuous radius vectors of each independent cross-section obtained along the steel pipe axis are assembled and spliced ​​according to their physical order to obtain complete and high-precision geometric parameters of the entire cross-section of the seamless steel pipe. It should be noted that the alternating direction multiplier method is a well-known technique in the art and will not be further limited or elaborated upon here.

[0062] The preset convergence condition in this embodiment of the invention is that the change in the solution vector between two consecutive iterations is less than a preset judgment threshold or the number of iterations reaches a preset iteration limit. In this embodiment of the invention, the preset judgment threshold is set as follows: This value can be adjusted automatically based on the required level of detail in the reconstruction of the geometric parameters under test in the specific implementation environment and the computation time limitations of the computing platform. The smaller the preset judgment threshold, the higher the accuracy of the reconstructed cross-sectional geometric parameters, but the required number of iterations and convergence time will also increase accordingly. Conversely, the larger the preset judgment threshold, the faster the model solves, but may sacrifice some local contour reconstruction accuracy. The preset iteration limit is set to 1000 times. This limit can be adjusted automatically according to the real-time cycle requirements of the detection pipeline. It is mainly used to prevent the solution algorithm from getting stuck in a computational dead loop that cannot converge when encountering extremely distorted data.

[0063] In summary, an automatic detection method for the geometric parameters of seamless steel pipes determines the interference signal tendency of each frequency component by analyzing the frequency domain phase characteristics of the initial cross-sectional radius sequence, and quantifies the frequency domain radius discreteness of each cross-section by combining energy proportion characteristics. This enables precise removal of high-frequency transient interference noise from the surface, effectively avoiding the misleading influence of local feature artifacts on the reconstruction process. Utilizing the axial rigidity continuity of the steel pipe, the axial instability of the cross-section is quantified by extracting the relative energy change of frequency components between adjacent cross-sections and the frequency domain radius discreteness deviation, thereby dynamically fusing the overall radius discreteness and overcoming the random errors caused by the vibration of the steel pipe during operation in industrial settings. Finally, this overall radius discreteness is used as a physical constraint condition conforming to the actual geometric distortion law, integrated into the compressed sensing objective function to guide the inversion reconstruction. This breaks the situation where the objective function easily falls into non-physical local optima under sparse data, ensuring that the reconstruction process strictly conforms to the true morphological mechanism of the seamless steel pipe even with incomplete observation data. This effectively overcomes the shortcomings of existing technologies that lack physical morphological support and have poor reliability in reconstructing contours, resulting in higher accuracy in reconstructing the geometric parameters of seamless steel pipes using sparsely acquired data.

[0064] This invention also provides an automatic detection system for the geometric parameters of seamless steel pipes; please refer to [link / reference]. Figure 2 The diagram shows a structural diagram of an automatic detection system for geometric parameters of seamless steel pipes according to an embodiment of the present invention. The system includes: a data acquisition module 201, a first determination module 202, a second determination module 203, and a geometric parameter detection module 204.

[0065] The data acquisition module 201 is used to synchronously sample all laser sensors arranged on the same circumferential cross section during the axial movement of the seamless steel pipe, and acquire the cross section angle position and corresponding measured radius data of each laser sensor on each cross section. The first determining module 202 is used to determine the initial cross-sectional radius sequence based on the changing trend of the measured radius data of the cross-sectional angle position corresponding to each laser sensor on each cross-section; determine the interference signal tendency of each frequency component based on the frequency domain phase characteristics of the initial cross-sectional radius sequence; and determine the frequency domain radius discreteness of each cross-section based on the interference signal tendency and energy ratio characteristics of each frequency component. The second determining module 203 is used to determine the axial instability of each cross section based on the relative energy change of frequency components between adjacent cross sections and the frequency domain radius dispersion deviation; and to determine the overall radius dispersion based on the axial instability and frequency domain radius dispersion of each cross section. The geometric parameter detection module 204 is used to construct a compressed sensing objective function with the overall radius discreteness as a constraint condition for inversion and reconstruction, and to determine the full cross-sectional geometric parameters of the seamless steel pipe.

[0066] This invention also provides a computer device, which includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, the computer device can perform any of the aforementioned automatic detection methods for the geometric parameters of seamless steel pipes.

Claims

1. An automatic detection method for the geometric parameters of seamless steel pipes, characterized in that, The method includes: During the axial movement of the seamless steel pipe, all laser sensors arranged on the same circumferential cross section are synchronously sampled to obtain the cross-sectional angle position of each laser sensor on each cross section and the corresponding measured radius data. Based on the changing trend of the measured radius data of each laser sensor corresponding to the cross-sectional angle position on each cross-section, an initial cross-sectional radius sequence is determined; based on the frequency domain phase characteristics of the initial cross-sectional radius sequence, the interference signal tendency of each frequency component is determined; based on the interference signal tendency and energy proportion characteristics of each frequency component, the frequency domain radius discreteness of each cross-section is determined. Based on the relative energy changes of frequency components between adjacent cross sections and the frequency domain radius dispersion deviation, the axial instability of each cross section is determined; based on the axial instability and frequency domain radius dispersion of each cross section, the overall radius dispersion is determined. Using the overall radius discreteness as a constraint, a compressed sensing objective function is constructed for inversion reconstruction to determine the full cross-sectional geometric parameters of the seamless steel pipe.

2. The automatic detection method for geometric parameters of seamless steel pipes according to claim 1, characterized in that, The process of acquiring the interference signal tendency includes: The initial cross-sectional radius sequence is filtered and decomposed by a bandpass filter to determine the local energy and local phase of each frequency component at each cross-sectional angular position. Based on the local phase of all frequency components at each cross-sectional angular position, the corresponding vector average phase is calculated; based on the deviation between the local phase of each frequency component at each cross-sectional angular position and the vector average phase, a cosine function mapping is performed to determine the degree of component deviation; based on the local energy and the degree of component deviation, the effective coherent energy of each frequency component at each cross-sectional angular position is determined; both the local energy and the degree of component deviation are positively correlated with the effective coherent energy; The effective coherent energy of each frequency component at all cross-sectional angular positions is superimposed to determine the total coherent energy; the local energy of each frequency component at all cross-sectional angular positions is superimposed to determine the total component energy; based on the relative magnitude between the total coherent energy and the total component energy, the interference signal tendency of each frequency component is determined; the total coherent energy is positively correlated with the interference signal tendency, and the total component energy is negatively correlated with the interference signal tendency.

3. The automatic detection method for geometric parameters of seamless steel pipes according to claim 2, characterized in that, The process of obtaining the frequency domain radius discreteness includes: For each section: The total energy of the cross section is determined by superimposing the total energy of all frequency components; the reference energy ratio of each frequency component is determined based on the total energy of the components and the total energy of the cross section; the tendency of the interference signal is negatively correlated and mapped to determine the tendency weight. The tendency weight of each frequency component is introduced as a weight to weight the process of calculating the reference energy proportion information entropy of each frequency component, thereby determining the frequency domain radius dispersion of each cross section; the frequency domain radius dispersion is normalized to determine the frequency domain radius dispersion of each cross section.

4. The automatic detection method for geometric parameters of seamless steel pipes according to claim 2, characterized in that, The process of obtaining the axial instability of the cross section includes: The total energy of the second-order frequency components corresponding to each cross section is taken as the corresponding principal harmonic energy. Based on the axial sampling sequence of the seamless steel pipe, the corresponding main harmonic variation rate is determined according to the relative deviation between the main harmonic energy of each section and the main harmonic energy of the previous section; an exponential positive correlation mapping is performed on the difference between the frequency domain radius discreteness of each section and the frequency domain radius discreteness of the previous section to determine the discrete fluctuation penalty factor; based on the main harmonic variation rate and the discrete fluctuation penalty factor, the axial instability of each section is determined; both the main harmonic variation rate and the discrete fluctuation penalty factor are positively correlated with the axial instability of the section.

5. The automatic detection method for geometric parameters of seamless steel pipes according to claim 1, characterized in that, The process for obtaining the overall radius discreteness includes: A negative correlation mapping is performed on the axial instability of each cross section to determine the corresponding contribution adjustment factor; based on the sum of the contribution adjustment factors of all cross sections, the contribution adjustment factor of each cross section is normalized to determine the comprehensive weight of each cross section. The frequency domain radius discreteness is weighted based on the cross-sectional comprehensive weight to determine the weighted discrete component of each cross-section; the weighted discrete components of all cross-sections are superimposed to determine the overall radius discreteness.

6. The automatic detection method for geometric parameters of seamless steel pipes according to claim 1, characterized in that, The process of obtaining the full cross-section geometric parameters includes: Construct a random observation matrix; the rows of the random observation matrix represent each laser sensor, and the columns represent each cross-sectional angle position; the element value of the corresponding element position of each laser sensor in the random observation matrix is ​​1, and the element value of other elements is 0; For each cross section, a data fidelity term is constructed based on the measured radius data of that cross section and the random observation matrix, with the overall radius discreteness as a constraint condition; a compressed sensing objective function is constructed by combining the data fidelity term and the constraint condition; the compressed sensing objective function is iteratively solved to determine the continuous radius vector after convergence for each cross section; the set of continuous radius vectors of each cross section is used as the full cross-sectional geometric parameters of the seamless steel pipe.

7. The automatic detection method for geometric parameters of seamless steel pipes according to claim 2, characterized in that, The bandpass filter is a Log-Gabor filter.

8. The automatic detection method for geometric parameters of seamless steel pipes according to claim 3, characterized in that, The method for normalizing the frequency domain radius dispersion is min-max normalization.

9. The automatic detection method for geometric parameters of seamless steel pipes according to claim 1, characterized in that, The process of obtaining the initial cross-sectional radius sequence includes: Based on the cross-sectional angle position of each laser sensor on the current cross-section and the corresponding measured radius data, cubic spline interpolation fitting is performed to determine the initial cross-sectional radius sequence of the corresponding cross-section.

10. An automatic detection system for the geometric parameters of seamless steel pipes, characterized in that, The system includes: The data acquisition module is used to synchronously sample all laser sensors arranged on the same circumferential section during the axial movement of the seamless steel pipe, and to acquire the cross-sectional angle position of each laser sensor on each section and the corresponding measured radius data. The first determining module is used to determine an initial cross-sectional radius sequence based on the changing trend of the measured radius data of the cross-sectional angle positions corresponding to each laser sensor on each cross-section; determine the interference signal tendency of each frequency component based on the frequency domain phase characteristics of the initial cross-sectional radius sequence; and determine the frequency domain radius discreteness of each cross-section based on the interference signal tendency and energy proportion characteristics of each frequency component. The second determining module is used to determine the axial instability of each cross section based on the relative energy change of frequency components between adjacent cross sections and the frequency domain radius dispersion deviation; and to determine the overall radius dispersion based on the axial instability and frequency domain radius dispersion of each cross section. The geometric parameter detection module is used to construct a compressed sensing objective function with the overall radius discreteness as a constraint condition for inversion and reconstruction, and to determine the full cross-sectional geometric parameters of the seamless steel pipe.

Images