Real-time analysis and defect identification method for inner wall ultrasonic testing data of seamless steel pipe

Abstract

The application provides a seamless steel pipe inner wall ultrasonic detection data real-time analysis and defect identification method, relates to the seamless steel pipe detection technical field, and comprises the following steps: acquiring an ultrasonic echo signal sequence, performing adaptive mode decomposition to extract a characteristic vector set, establishing an adaptive reference model of a normal inner wall state, calculating a deviation metric value and performing correlation clustering to identify a defect region, and further determining a defect type and position. The application realizes high-precision automatic identification of the seamless steel pipe inner wall defect, and improves the detection efficiency and accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of seamless steel pipe detection technology, and particularly relates to a seamless steel pipe inner wall ultrasonic detection data real-time analysis and defect identification method. BACKGROUND

[0002] As an important industrial base material, seamless steel pipes are widely used in the fields of petroleum, natural gas, chemical industry, power and aerospace. The inner wall quality of seamless steel pipes directly affects their performance and safety, so effective defect detection of the inner wall of seamless steel pipes is of great significance. Ultrasonic detection technology has become one of the main methods for detecting the inner wall of seamless steel pipes due to its non-destructive and high-precision characteristics.

[0003] Traditional ultrasonic detection of the inner wall of seamless steel pipes usually uses fixed threshold judgment or simple signal processing methods for defect identification. With the increasing quality requirements of industrial production for seamless steel pipes, the traditional detection method is facing more and more challenges and deficiencies. SUMMARY

[0004] The embodiments of the present application provide a seamless steel pipe inner wall ultrasonic detection data real-time analysis and defect identification method, which can solve the problems in the prior art.

[0005] In a first aspect, the embodiments of the present application provide a seamless steel pipe inner wall ultrasonic detection data real-time analysis and defect identification method, comprising:

[0006] Obtaining an ultrasonic echo signal sequence of the inner wall of the seamless steel pipe;

[0007] Adaptive mode decomposition is performed on each ultrasonic echo signal to obtain a plurality of intrinsic mode functions and a residual trend component, and a differentiated feature extraction is performed to obtain a feature vector set;

[0008] The feature vector set is divided into a plurality of spatial grid units, baseline values of each feature dimension are extracted, and a reference feature vector of each spatial grid unit is calculated in combination with a neighborhood unit and a historical reference characteristic vector to establish an adaptive reference model of a normal inner wall state;

[0009] According to the adaptive reference model, a deviation measure value of each feature vector of the inner wall of the seamless steel pipe is calculated, and the deviation measure value is associated and clustered in the spatial dimension and the time dimension to identify a continuous region whose deviation measure value exceeds an adaptive threshold value as a defect region;

[0010] For each defect region, a waveform feature parameter of the corresponding ultrasonic echo signal sequence is extracted, and is matched with a preset defect type discrimination criterion to determine the defect type and the defect position, and a defect identification result is obtained.

[0011] For each ultrasonic echo signal, adaptive mode decomposition is performed to obtain multiple intrinsic mode functions and a residual trend component. Differential feature extraction is then performed to obtain a set of feature vectors, including:

[0012] Adaptive mode decomposition is performed on each ultrasonic echo signal. By iteratively extracting oscillation components at different frequency scales, the ultrasonic echo signal is decomposed into multiple intrinsic mode functions and a residual trend component. The number of intrinsic mode functions is adaptively determined according to the energy distribution of the ultrasonic echo signal.

[0013] The analytic signal is obtained by performing a complex domain transformation on each intrinsic mode function, its instantaneous phase sequence is extracted, and the instantaneous frequency change trajectory is obtained by performing time differentiation on the instantaneous phase sequence. The instantaneous amplitude envelope curve is obtained by calculating the magnitude sequence of the analytic signal.

[0014] The residual trend components are subjected to multi-order difference operations on the time axis to extract the gradient sequence and its corresponding statistical features as gradient change features.

[0015] The instantaneous frequency change trajectory and the instantaneous amplitude envelope curve are time-aligned and combined with the gradient change characteristics to obtain the feature vector set of each ultrasonic echo signal.

[0016] The set of feature vectors is divided into multiple spatial grid cells. Baseline values ​​for each feature dimension are extracted, and combined with neighboring cells and historical baseline characteristic vectors, baseline feature vectors for each spatial grid cell are calculated to establish an adaptive baseline model for the normal inner wall state, including:

[0017] The feature vector set is divided into spatial domains according to the circumferential position index and the axial position index to form multiple spatial grid units, each spatial grid unit corresponding to a subset of feature vectors;

[0018] For each subset of feature vectors, the baseline value of each feature dimension is calculated to obtain the initial reference feature vector. The initial reference feature vectors of each spatial grid cell and its neighboring cells are then subjected to spatial adaptive smoothing to obtain the smoothed reference feature vector.

[0019] From historical detection data, extract the historical baseline feature vector of each spatial grid cell, calculate the time decay weight based on the time interval between the historical detection period and the current detection period, and calculate the cumulative weight based on the cumulative number of samples.

[0020] Based on the time decay weight and the cumulative weight, the smoothed baseline feature vector and the historical baseline feature vector are weighted and summed to obtain the baseline feature vector of each spatial grid cell, and an adaptive baseline model of the normal inner wall state is constructed.

[0021] For each feature vector subset, calculate the baseline value of each feature dimension, obtain the initial reference feature vector, and perform spatial adaptive smoothing processing on the initial reference feature vector of each spatial grid cell and its neighborhood cell to obtain a smoothed reference feature vector, comprising:

[0022] For each feature vector subset corresponding to each spatial grid cell, extract the numerical distribution of all feature vectors in each feature dimension, perform robust statistical analysis on the numerical distribution in each feature dimension, calculate the center position estimate resistant to outlier interference as the baseline value of each feature dimension, and combine to obtain the initial reference feature vector of each spatial grid cell;

[0023] For each spatial grid cell, calculate its neighborhood cell with spatial continuity in the circumferential direction and axial direction, and extract the corresponding initial reference feature vector to form a neighborhood reference feature vector group;

[0024] According to the spatial distance between each neighborhood cell and the spatial grid cell, a spatial decay function is constructed, and a sample reliability function is constructed according to the sample number of each neighborhood cell, and based on the spatial decay function and the sample reliability function, the neighborhood initial baseline feature vector group is filtered and smoothed to obtain the smoothed reference feature vector of each spatial grid cell.

[0025] According to the adaptive reference model, the deviation measure value of each feature vector of the inner wall of the seamless steel pipe is calculated, and it is associated and clustered in spatial dimension and time dimension, and the continuous region whose deviation measure value exceeds the adaptive threshold is identified as a defect region, comprising:

[0026] According to the reference feature vector of each spatial grid cell in the adaptive reference model, the deviation measure value of each feature vector of the detection point of the inner wall of the seamless steel pipe is calculated;

[0027] The deviation measure value is constructed into a two-dimensional spatial distribution map according to the circumferential index and the axial index, and a space-time evolution matrix is constructed combined with the detection time sequence, wherein the space-time evolution matrix records the deviation measure value change trajectory of each detection point at multiple detection time;

[0028] The space-time evolution matrix is subjected to two-dimensional correlation clustering, and by calculating the connectivity index of each detection point in the spatial dimension and the consistency index in the time dimension, a plurality of deviation measure value aggregation regions are identified;

[0029] From the deviation measure value aggregation region, a continuous region whose deviation measure value exceeds the adaptive threshold is extracted, and the continuous region is marked as a suspected defect region.

[0030] performing two-dimensional correlation clustering on the spatio-temporal evolution matrix, identifying a plurality of deviation metric value aggregation regions by calculating a connectivity index of each detection point in the spatial dimension and a consistency index in the time dimension, including:

[0031] Based on the spatio-temporal evolution matrix, the spatial correlation degree between the deviation metric value of each detection point and the deviation metric value of its adjacent detection point is quantified to obtain the connectivity index, and the stability of the deviation metric value of each detection point in the continuous time sequence is calculated to obtain the consistency index.

[0032] Performing two-dimensional coupling analysis on the connectivity index and the consistency index, constructing a two-dimensional correlation degree evaluation function, and calculating the two-dimensional correlation degree value of each detection point in spatial connectivity and time consistency.

[0033] Based on the two-dimensional correlation degree value, the detection points meeting the preset correlation condition are merged into the same clustering cluster, and the clustering boundary is determined according to the two-dimensional correlation degree value distribution characteristics of the detection points in each clustering cluster, and a plurality of deviation metric value aggregation regions that present connected distribution in space and consistent evolution in time are identified.

[0034] For each defect region, the waveform feature parameters of the corresponding ultrasonic echo signal sequence are extracted, and matched with the preset defect type discrimination criterion to determine the defect type and defect position, and the defect recognition result is obtained, including:

[0035] For each defect region, the multi-dimensional waveform feature parameters of the corresponding ultrasonic echo signal sequence are extracted, including the amplitude feature parameters reflecting the geometric shape of the defect and the frequency spectrum feature parameters reflecting the internal structure of the defect.

[0036] Based on the multi-dimensional waveform feature parameters and the preset defect type discrimination criterion, multi-level matching is performed, including preliminary classification of the amplitude feature parameters to determine the macro category of the defect, and refined identification of the frequency spectrum feature parameters to determine the microscopic characteristics of the defect.

[0037] According to the matching results of the macro category and microscopic characteristics, combined with the spatial position information of the ultrasonic echo signal sequence, a defect recognition result containing defect type identification and defect position coordinates is generated.

[0038] The second aspect of the embodiment of the application provides a seamless steel pipe inner wall ultrasonic detection data real-time analysis and defect identification system, including:

[0039] The first unit is used for acquiring the ultrasonic echo signal sequence of the inner wall of the seamless steel pipe.

[0040] The second unit is configured to perform adaptive mode decomposition on each ultrasonic echo signal to obtain a plurality of intrinsic mode functions and a residual trend component, and perform differential feature extraction to obtain a feature vector set;

[0041] The third unit is configured to divide the feature vector set into a plurality of spatial grid units, extract baseline values of each feature dimension, and calculate reference feature vectors of each spatial grid unit in combination with neighbor units and historical reference feature vectors to establish an adaptive reference model of a normal inner wall state.

[0042] The fourth unit is configured to calculate deviation metric values of each feature vector of the inner wall of the seamless steel pipe according to the adaptive reference model, and perform spatial dimension and time dimension correlation clustering on the deviation metric values to identify a continuous area with a deviation metric value exceeding an adaptive threshold as a defect area.

[0043] The fifth unit is configured to extract waveform feature parameters of a corresponding ultrasonic echo signal sequence of each defect area, and match the waveform feature parameters with preset defect type discrimination criteria to determine a defect type and a defect position of the defect area, and obtain a defect recognition result.

[0044] A third aspect of the embodiment of the application,

[0045] An electronic device is provided, comprising:

[0046] A processor;

[0047] A memory for storing processor-executable instructions;

[0048] The processor is configured to invoke the instructions stored in the memory to perform the method described above.

[0049] A fourth aspect of the embodiment of the application,

[0050] A computer-readable storage medium is provided, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the method described above.

[0051] The beneficial effects of the present application are as follows:

[0052] The adaptive mode decomposition technology is used to process the ultrasonic echo signal, which can effectively separate different frequency components, extract differential features, enhance the defect features in the signal, and improve the accuracy and efficiency of signal processing, thereby laying a good foundation for subsequent defect recognition.

[0053] The adaptive reference model based on the spatial grid unit is established, and neighbor units and historical reference feature vectors are combined for calculation, so that the model can adapt to the changes of the normal state of the steel pipe inner wall, improve the robustness and adaptability of the model, and reduce the misjudgment rate.

[0054] The present application realizes accurate positioning and identification of the defect area through the correlation clustering and adaptive threshold judgment in the spatial dimension and the time dimension, and can accurately determine the defect type and position through the waveform characteristic parameter and the preset discrimination criterion matching, thereby improving the reliability and efficiency of the detection and effectively guaranteeing the quality safety of the seamless steel pipe. BRIEF DESCRIPTION OF DRAWINGS

[0055] Figure 1 A flowchart of a seamless steel pipe inner wall ultrasonic detection data real-time analysis and defect identification method according to an embodiment of the present application is shown in

[0056] Figure 2 A flowchart of an ultrasonic echo signal adaptive modal decomposition and feature extraction method is shown in DETAILED DESCRIPTION

[0057] In order to make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the present application.

[0058] The technical solutions of the present application will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes can not be described in some embodiments.

[0059] Figure 1 A flowchart of a seamless steel pipe inner wall ultrasonic detection data real-time analysis and defect identification method according to an embodiment of the present application is shown in Figure 1 The method comprises the following steps.

[0060] An ultrasonic echo signal sequence of the inner wall of the seamless steel pipe is acquired.

[0061] Each ultrasonic echo signal is subjected to adaptive modal decomposition to obtain a plurality of intrinsic modal functions and a residual trend component, and is subjected to differential feature extraction to obtain a feature vector set.

[0062] The feature vector set is divided into a plurality of spatial grid units, the baseline value of each feature dimension is extracted, and the baseline feature vector of each spatial grid unit is calculated in combination with the neighborhood unit and the historical reference characteristic vector to establish an adaptive reference model of the normal inner wall state.

[0063] According to the adaptive benchmark model, a deviation metric value of each feature vector of the inner wall of the seamless steel pipe is calculated, and is associated and clustered in a spatial dimension and a time dimension, a continuous area with a deviation metric value exceeding an adaptive threshold is identified as a defect area;

[0064] For each defect area, a waveform feature parameter of a corresponding ultrasonic echo signal sequence is extracted, and is matched with a preset defect type discrimination criterion to determine a defect type and a defect position, and a defect recognition result is obtained.

[0065] Figure 2 An adaptive modal decomposition and feature extraction method flowchart for an ultrasonic echo signal is shown in FIG. 1. In an optional implementation, for each ultrasonic echo signal, adaptive modal decomposition is performed to obtain a plurality of intrinsic mode functions and a residual trend component, and differential feature extraction is performed to obtain a feature vector set, including:

[0066] The adaptive modal decomposition is performed on each ultrasonic echo signal to extract oscillation components of different frequency scales through iteration, and the ultrasonic echo signal is decomposed into a plurality of intrinsic mode functions and a residual trend component, wherein the number of intrinsic mode functions is adaptively determined according to the energy distribution of the ultrasonic echo signal;

[0067] An analytic signal is obtained by performing complex domain transformation on each intrinsic mode function, an instantaneous phase sequence is extracted, and a time differential operation is performed on the instantaneous phase sequence to obtain an instantaneous frequency variation trajectory, and an instantaneous amplitude envelope curve is obtained by calculating a modulus sequence of the analytic signal;

[0068] A gradient sequence and a corresponding statistical feature are extracted by performing a multi-order difference operation on the residual trend component along the time axis as a gradient variation feature;

[0069] The instantaneous frequency variation trajectory and the instantaneous amplitude envelope curve are time-aligned, and the gradient variation feature is combined to obtain a feature vector set of each ultrasonic echo signal.

[0070] In actual application, the collected ultrasonic echo signal needs to be preprocessed, including denoising and normalization. Taking an ultrasonic detection as an example, the sampling rate of the collected echo signal is 10 MHz, the signal length is 1024 points, and the preprocessed signal is used as the input of the adaptive modal decomposition.

[0071] Adaptive mode decomposition is an analysis method for non-linear and non-stationary signals, which extracts different frequency components in the signal through an iterative screening process. In this method, the adaptive mode decomposition algorithm is applied to the pre-processed ultrasonic echo signal. In the specific implementation process, the extreme points of the signal are detected, the upper and lower envelope lines are constructed, the mean envelope line is calculated, and the mean envelope line is subtracted from the original signal to obtain the candidate intrinsic mode function. It is judged whether the candidate function meets the definition condition of intrinsic mode function: the number of oscillations is close to the number of extreme points and the mean envelope is approximately zero. If the condition is met, the function is confirmed as an intrinsic mode function, and the function is subtracted from the original signal. The above process is repeated for the remaining signal. When the remaining signal no longer contains effective oscillation components or the number of iterations reaches the preset threshold, the decomposition process is terminated, and the remaining signal is regarded as a residual trend component.

[0072] Taking the above sampling data as an example, after adaptive mode decomposition, 6 intrinsic mode functions and 1 residual trend component are obtained. The first two intrinsic mode functions contain the main frequency components of the signal, with frequency ranges of 3-5 MHz and 1-2 MHz, respectively, corresponding to the main reflection characteristics of the detected material.

[0073] Each intrinsic mode function is subjected to complex domain transformation to construct an analytical signal. In actual implementation, the Hilbert transform is used to convert the real-valued intrinsic mode function into a complex analytical signal in the complex domain. For the first intrinsic mode function in the example data, its analytical signal can be represented as a complex number formed by the original signal and its Hilbert transform. From the analytical signal, the instantaneous phase sequence is extracted by calculating the argument of the complex number. The instantaneous frequency trajectory is obtained by time-differentiating the instantaneous phase sequence. The analysis of the first intrinsic mode function shows that its instantaneous frequency is relatively stable at the first 300 points of the signal, about 4 MHz, and then decreases significantly to about 2.8 MHz, which reflects the frequency dispersion effect of ultrasonic waves in the medium.

[0074] At the same time, the instantaneous amplitude envelope curve is obtained by calculating the modulus of the analytical signal. The analysis of the example data shows that the amplitude envelope curve of the first intrinsic mode function reaches a peak value of about 0.85 (after normalization) in the interval of 100-300 points, and then gradually decays, which reflects the absorption characteristics of the material to ultrasonic waves.

[0075] For the residual trend component, a multi-order difference operation is performed to extract gradient information. In implementation, first-order, second-order and third-order differences are used, corresponding to the physical concepts of velocity, acceleration and jerk respectively. The statistical properties of each order difference sequence are calculated, including mean, variance, kurtosis and skewness. The residual trend component of the example data is analyzed. The mean of the first-order difference is 0.03, and the variance is 0.002, indicating that the trend change is relatively gentle. The kurtosis of the second-order difference is 4.5, and the skewness is 0.8, reflecting the existence of nonlinear characteristics in the signal.

[0076] The instantaneous frequency change trajectory and the instantaneous amplitude envelope curve are time-aligned. Since different intrinsic mode functions have different time characteristics, it is necessary to ensure the consistency of the features in the time dimension. In actual implementation, the instantaneous frequency and amplitude envelope curve are sampled at fixed time intervals (such as 10 points for one time window). The average value, maximum value and minimum value of each window are extracted as features. For the example data, about 100 time window feature values are extracted.

[0077] In combination with the gradient change characteristics, the final feature vector set is constructed. Taking the example data as an example, the feature vector generated by each ultrasonic echo signal has a dimension of 650, including the instantaneous frequency features (300 dimensions) of each intrinsic mode function, the amplitude envelope features (300 dimensions), and the gradient features (50 dimensions) of the residual trend component. These features comprehensively reflect the frequency, energy and structural information of the ultrasonic wave propagating in the material, which can be used for subsequent defect classification or material property evaluation.

[0078] By implementing the above method, the subtle feature differences in the ultrasonic echo signal can be effectively captured, and the accuracy and reliability of defect detection can be improved. Experimental verification shows that compared with traditional Fourier analysis or wavelet analysis, this method has obvious advantages in identifying small defects in materials.

[0079] In an optional implementation, the feature vector set is divided into multiple spatial grid cells, the baseline values of each feature dimension are extracted, and the baseline feature vectors of each spatial grid cell are calculated in combination with the neighborhood cells and historical reference characteristic vectors, to establish an adaptive reference model of the normal inner wall state, including:

[0080] The feature vector set is divided into multiple spatial grid cells according to the circumferential position index and the axial position index, and each spatial grid cell corresponds to a group of feature vector subsets;

[0081] For each feature vector subset, the baseline value of each feature dimension is calculated to obtain an initial baseline feature vector. The initial baseline feature vectors of each spatial grid cell and its neighborhood cells are subjected to spatial adaptive smoothing processing to obtain a smoothed baseline feature vector.

[0082] From the historical detection data, the historical reference feature vectors of each spatial grid unit are extracted, the time decay weight is calculated according to the time interval between the historical detection period and the current detection period, and the cumulative weight is calculated according to the cumulative sample quantity;

[0083] Based on the time decay weight and the cumulative weight, the smoothed reference feature vector is obtained by weighted summation of the smoothed reference feature vector and the historical reference feature vector, and the adaptive reference model of the normal inner wall state is constructed.

[0084] In the specific embodiment, the obtained feature vector set is divided in the spatial domain according to the circumferential position index and the axial position index to form a plurality of spatial grid units. For example, for the inner wall of a cylindrical steel pipe, the circumferential direction can be divided into 36 uniform regions (each 10 degrees as a region), and the axial direction can be divided into 20 uniform regions, thereby forming 36x20=720 spatial grid units, each spatial grid unit corresponding to a feature vector subset of a specific position region on the inner wall of the steel pipe. Taking the spatial grid unit G(12, 5) as an example, it represents the 12th circumferential region and the 5th axial region, and the unit contains the feature vectors of all acquisition points in the corresponding region.

[0085] For each feature vector subset, the baseline value of each feature dimension is calculated to obtain an initial reference feature vector. Assuming that each feature vector contains four feature dimensions of temperature, pressure, vibration frequency and surface roughness, for all feature vectors in the spatial grid unit G(12, 5), the median of the four feature dimensions is calculated respectively as the initial reference feature vector of the unit. For example, the initial reference feature vector of G(12, 5) is [72.3℃, 3.5MPa, 112.8Hz, 25.6μm].

[0086] In order to reduce the influence of spatial noise, spatial adaptive smoothing processing is performed on the initial reference feature vectors of each spatial grid unit and its neighborhood units. For G(12, 5), its neighborhood units include G(11, 5), G(13, 5), G(12, 4), G(12, 6), G(11, 4), G(11, 6), G(13, 4), G(13, 6) a total of 8 units. According to the spatial correlation, the center unit G(12, 5) can be set to a weight of 0.4, the four adjacent units each set to a weight of 0.1, and the four diagonal adjacent units each set to a weight of 0.05. The initial reference feature vectors of the nine units are weighted and averaged to obtain the smoothed reference feature vector of G(12, 5), such as [73.1℃, 3.6MPa, 110.5Hz, 26.2μm].

[0087] From the historical detection data, the historical reference feature vectors of each spatial grid unit are extracted. Assuming that the historical reference feature vectors of G(12, 5) in the previous three detections are L1 = [71.2℃, 3.3MPa, 108.7Hz, 22.1μm], L2 = [72.5℃, 3.5MPa, 110.2Hz, 24.3μm] and L3 = [72.9℃, 3.6MPa, 109.8Hz, 25.7μm], and the corresponding detection times are 6 months ago, 3 months ago and 1 month ago.

[0088] According to the time interval between the historical detection period and the current detection period, the time decay weight is calculated. The longer the time interval, the smaller the weight. For the historical data 1 month ago, 3 months ago and 6 months ago, the time decay weights can be set to 0.8, 0.5 and 0.3 respectively. At the same time, the cumulative weight is calculated according to the cumulative sample number, reflecting the statistical reliability of the historical data. Assuming that the cumulative sample numbers of the previous three detections are 120, 105 and 95 respectively, and the current detection sample number is 30, then the cumulative weights are 0.8, 0.75 and 0.7 respectively.

[0089] Based on the time decay weight and the cumulative weight, the smoothed reference feature vector and the historical reference feature vector are weighted and summed to obtain the final reference feature vector. The weight of the current smoothed reference feature vector can be set to 0.6, and the combined weight of the three historical reference feature vectors is 0.4, which is distributed according to the product of the time decay weight and the cumulative weight. The specific calculation is as follows: the weight of the current smoothed reference feature vector is 0.6, the weights of the historical vectors L1, L2 and L3 are 0.4×(0.8×0.8) / (0.8×0.8+0.5×0.75+0.3×0.7)=0.174, 0.4×(0.5×0.75) / (0.8×0.8+0.5×0.75+0.3×0.7)=0.127 and 0.4×(0.3×0.7) / (0.8×0.8+0.5×0.75+0.3×0.7)=0.099 respectively.

[0090] The vectors are weighted and summed to obtain the final reference feature vector of G(12, 5) [72.85℃, 3.57MPa, 110.18Hz, 25.82μm]. The reference feature vectors of all 720 spatial grid units are calculated by this method to form a complete adaptive reference model. This adaptive reference model has three adaptabilities: spatial adaptability, which reduces local noise through neighborhood smoothing; time adaptability, which tracks the slow changes of the inner wall state through time decay weight; sample adaptability, which balances the reliability of new and old data through cumulative weight. In practical application, this model can effectively reduce the false alarm rate caused by normal aging, environmental changes, etc., while maintaining high sensitivity to real anomalies.

[0091] After the adaptive reference model is completed, the current detected feature vector and the reference feature vector of the corresponding spatial grid cell are analyzed for deviation, the deviation degree is calculated and a threshold is set, and the abnormal inner wall area is identified to provide accurate decision basis for equipment maintenance.

[0092] In an optional embodiment, for each feature vector subset, the baseline value of each feature dimension is calculated to obtain an initial reference feature vector, and the initial reference feature vectors of each spatial grid cell and its neighborhood cells are subjected to spatial adaptive smoothing processing to obtain a smoothed reference feature vector, including:

[0093] For each feature vector subset corresponding to each spatial grid cell, the numerical distribution of all feature vectors in each feature dimension is extracted, and robust statistical analysis is performed on the numerical distribution in each feature dimension to calculate the center position estimate resistant to outlier interference as the baseline value of each feature dimension, and the initial reference feature vector of each spatial grid cell is obtained by combination;

[0094] For each spatial grid cell, its neighborhood cells with spatial continuity in the circumferential direction and the axial direction are calculated, and the corresponding initial reference feature vectors are extracted to form a neighborhood reference feature vector group;

[0095] According to the spatial distance between each neighborhood cell and the spatial grid cell, a spatial decay function is constructed, and a sample reliability function is constructed according to the sample number of each neighborhood cell, and the neighborhood initial baseline feature vector group is subjected to filtering and smoothing operation based on the spatial decay function and the sample reliability function to obtain the smoothed reference feature vector of each spatial grid cell.

[0096] In the specific embodiment, after the spatial grid cell division is completed, for each grid cell G(i, j), the feature vectors of all data points in the cell are collected to form a feature vector subset V(i, j)={v1, v2,..., v k}, k is the number of detection points included in each grid cell, and for each feature dimension m, the values of all feature vectors in the dimension are extracted to form a numerical distribution set V m (i, j)={v 1m , v 2m ,..., v km}.

[0097] For each numerical distribution set V m(i,j), and performing a robust statistical analysis, which refers to a statistical method that can obtain reliable estimates even in the presence of outliers or outliers, in the present embodiment, the median is used as an outlier-resistant central position estimator, the median is the value located in the middle position after sorting the data, and is not sensitive to outliers, the specific calculation steps are as follows: for the numerical value distribution set V m (i,j), the values in it are sorted in ascending order, and the sorted set V m '(i,j)={v' 1m , v' 2m ,..., v' km} is obtained, if k is odd, the median is the value located at the (k+1) / 2 position after sorting, that is, med(V m (i,j))=v' ((k+1) / 2)m ; if k is even, the median is the average of the values located at the k / 2 and k / 2+1 positions after sorting, that is, med(V m (i,j))=(v' (k / 2) m+v' (k / 2+1) m) / 2.

[0098] The median calculated on each feature dimension is taken as the baseline value b m (i,j)=med(V m (i,j) of the dimension, and the initial baseline feature vector b(i,j) of the grid cell G(i,j) is obtained by combination, that is, b(i,j)=[b1(i,j), b2(i,j),..., b n (i,j)].

[0099] In a specific data case, a certain area is divided into 4x4 grid cells, each grid cell contains 10 data points, and each data point is represented by a 5-dimensional feature vector, taking grid cell G(2,3) as an example, the feature vectors of the 10 data points contained in the cell are as follows: v1 = [3.2, 1.5, 4.8, 2.1, 0.9], v2 = [3.5, 1.8, 4.5, 2.3, 1.1], v3 = [3.1, 1.4, 4.7, 2.0, 0.8], v4 = [3.4, 1.7, 4.6, 2.2, 1.0], v5 = [3.3, 1.6, 4.9, 2.4, 0.7], v6 = [3.6, 1.9, 4.4, 2.5, 1.2], v7 = [3.0, 9.0, 4.3, 2.6, 0.6], v8 = [3.7, 2.0, 4.2, 2.7, 1.3], v9 = [2.9, 1.3, 4.1, 2.8, 0.5], v 10= [3.8, 2.1, 15.0, 2.9, 1.4].

[0100] For the first feature dimension, the values of all feature vectors in this dimension are extracted to obtain a numerical distribution set V1(2, 3) = {3.2, 3.5, 3.1, 3.4, 3.3, 3.6, 3.0, 3.7, 2.9, 3.8}, which is sorted to obtain V1'(2, 3) = {2.9, 3.0, 3.1, 3.2, 3.3, 3.4, 3.5, 3.6, 3.7, 3.8}, since the number of data points k = 10 is even, the median is the average of the values at the 5th and 6th positions after sorting, i.e. b1(2, 3) = (3.3 + 3.4) / 2 = 3.35.

[0101] For the second feature dimension, the values of all feature vectors in this dimension are extracted to obtain a numerical distribution set V2(2, 3) = {1.5, 1.8, 1.4, 1.7, 1.6, 1.9, 9.0, 2.0, 1.3, 2.1}, it can be seen that the value 9.0 is obviously an outlier, which is sorted to obtain V2'(2, 3) = {1.3, 1.4, 1.5, 1.6, 1.7, 1.8, 1.9, 2.0, 2.1, 9.0}, the median is the average of the values at the 5th and 6th positions after sorting, i.e. b2(2, 3) = (1.7 + 1.8) / 2 = 1.75.

[0102] Similarly, for the third, fourth and fifth feature dimensions, b3(2, 3) = 4.55, b4(2, 3) = 2.45, b5(2, 3) = 0.95 are calculated, and the baseline values of all feature dimensions are combined to obtain the initial baseline feature vector b(2, 3) = [3.35, 1.75, 4.55, 2.45, 0.95] of the grid cell G(2, 3).

[0103] The above process is repeated for all spatial grid cells to obtain the initial baseline feature vector of each spatial grid cell, which directly reflects the concentration trend of data in each spatial region, but is affected by uneven sample size and random noise. To improve the reliability of the baseline feature vector, further spatial adaptive smoothing processing is performed.

[0104] The spatial adaptive smoothing process determines the neighborhood units of each spatial grid cell. In a cylindrical device monitoring scene, the neighborhood units include the grid cells adjacent in the circumferential direction and the axial direction. Taking the spatial grid cell G(3, 7) as an example, its circumferential adjacent cells are G(3, 6) and G(3, 8), and its axial adjacent cells are G(2, 7) and G(4, 7), which together constitute the neighborhood unit set of G(3, 7).

[0105] In order to consider the influence of spatial distance on the smoothing process, a spatial attenuation function is constructed. In actual implementation, a distance inverse ratio function is used as the spatial attenuation function. For example, if the center point distance between the spatial grid cell G(3, 7) and its neighborhood unit G(3, 6) is 5 units, then the corresponding spatial attenuation weight is 1 / 5 = 0.2. Similarly, the distances between G(3, 7) and G(3, 8), G(2, 7), and G(4, 7) are 5, 10, and 10 units respectively, and the corresponding spatial attenuation weights are 0.2, 0.1, and 0.1 respectively.

[0106] Meanwhile, considering the influence of the number of samples in each neighborhood unit, a sample reliability function is constructed. In this embodiment, the sample reliability function is calculated based on the logarithm of the number of samples. For example, if the spatial grid cell G(3, 6) contains 80 sample points, then its sample reliability weight is calculated as log(80) / log(200) = 0.9, where 200 is a preset reference sample number. Similarly, the sample numbers of G(3, 8), G(2, 7), and G(4, 7) are 90, 60, and 50 respectively, and the corresponding sample reliability weights are 0.92, 0.86, and 0.84 respectively.

[0107] Based on the spatial attenuation weight and the sample reliability weight, a weighted average of the neighborhood reference feature vector is performed to realize the filtering and smoothing operation. Taking a certain feature dimension as an example, the initial baseline values of G(3, 7) and its neighborhood units G(3, 6), G(3, 8), G(2, 7), and G(4, 7) are 82.3, 81.5, 82.8, 80.9, and 83.1 respectively. Through weighted average calculation, the smoothed baseline value of G(3, 7) in this feature dimension is 81.9. Similarly, the smoothed baseline values of all feature dimensions are calculated to obtain the smoothed reference feature vector of the spatial grid cell G(3, 7).

[0108] The above smoothing process is repeated for all spatial grid cells to obtain the smoothed reference feature vector of each spatial grid cell. Compared with the initial reference feature vector, the smoothed reference feature vector can better reflect the continuous variation of the feature in space, reduce the fluctuations caused by uneven sample quantity and random noise, and improve the reliability and stability of the reference feature.

[0109] In practical applications, the smoothed reference feature vector can be used as an important reference benchmark to determine whether the device is in a normal operating state. When the deviation between the feature vector of the newly collected data point and the smoothed reference feature vector of the corresponding spatial position exceeds the preset threshold, it can be determined that the device state reflected by the data point is abnormal, and further analysis or alarm processing is required.

[0110] In an optional embodiment, according to the adaptive reference model, the deviation metric value of each feature vector of the inner wall of the seamless steel pipe is calculated, and is associated and clustered in the spatial dimension and the time dimension, and a continuous region with a deviation metric value exceeding the adaptive threshold is identified as a defect region, including:

[0111] According to the reference feature vector of each spatial grid element in the adaptive reference model, the deviation metric value of each detection point feature vector of the inner wall of the seamless steel pipe is calculated;

[0112] The deviation metric value is constructed into a two-dimensional spatial distribution map according to the circumferential index and the axial index, and a space-time evolution matrix is constructed in combination with the detection time sequence, wherein the space-time evolution matrix records the deviation metric value change trajectory of each detection point at multiple detection times;

[0113] The space-time evolution matrix is subjected to two-dimensional associated clustering, and by calculating the connectivity index of each detection point in the spatial dimension and the consistency index in the time dimension, a plurality of deviation metric value aggregation regions are identified;

[0114] From the deviation metric value aggregation region, a continuous region with a deviation metric value exceeding the adaptive threshold is extracted, and the continuous region is marked as a suspected defect region.

[0115] In the specific embodiment, when calculating the deviation metric value based on the adaptive reference model, the original data of the inner wall of the seamless steel pipe is obtained, including multi-dimensional feature information obtained from ultrasonic scanning or electromagnetic detection, and for the feature vector of each detection point, the reference feature vector in the corresponding spatial grid element is searched, for example, for a detection point located at the circumferential index 45 and the axial index 128, the feature vector is [0.85, 0.56, 0.32, 0.91], and the reference feature vector at this position is [0.82, 0.53, 0.30, 0.89], and the deviation metric value is 0.058 by calculating the Euclidean distance. In order to improve the calculation efficiency, a pre-allocated memory is used to create an array with the same size as the number of detection points, and the deviation metric values of all detection points are batch calculated.

[0116] In the process of constructing the two-dimensional spatial distribution map and the spatio-temporal evolution matrix, the calculated deviation metric values are arranged according to the circumferential index and the axial index to form a two-dimensional matrix with a size of MxN, wherein M represents the number of circumferential grids (such as 360 units), and N represents the number of axial grids (such as 1024 units).

[0117] For a one-time detection of a seamless steel pipe, part of the data in the deviation metric value distribution matrix is as follows: the values of the region with circumferential indexes 40 to 45 and axial indexes 125 to 130 are [[0.025, 0.031, 0.028, 0.029, 0.030, 0.027], [0.032, 0.045, 0.062, 0.058, 0.051, 0.036], [0.029, 0.057, 0.086, 0.093, 0.065, 0.042], [0.031, 0.049, 0.072, 0.079, 0.063, 0.038], [0.028, 0.036, 0.043, 0.047, 0.042, 0.033], [0.026, 0.029, 0.030, 0.031, 0.029, 0.027]] at multiple time points t1, t2,..., t p The data is collected, and these two-dimensional matrices are stacked in time sequence to form a three-dimensional spatio-temporal evolution matrix with a dimension of PxMxN, wherein P is the number of detection time points.

[0118] When performing two-dimensional correlation clustering on the spatio-temporal evolution matrix, the connectivity index in the spatial dimension is calculated, and for each detection point with a deviation metric value exceeding an initial threshold (such as 0.05), it is checked whether the points in its eight-neighborhood also exceed the threshold. When the deviation metric value of a detection point such as (43, 127) is 0.093, and there are multiple points (such as (43, 126) with a value of 0.086, (42, 127) with a value of 0.079, etc.) around it that also exceed the threshold, these points are grouped into the same spatially connected region. The region growing algorithm is used to start from the point with the highest deviation metric value, and recursively add adjacent points that meet the threshold condition to the current region until it cannot be expanded any further.

[0119] In addition, the consistency index in the time dimension is calculated, and the changes in the deviation metric values of the same detection point at consecutive time points are compared. For example, the deviation metric values of the detection point (43, 127) at three consecutive time points are [0.089, 0.093, 0.091], which have a small fluctuation, indicating that the anomaly is stable and has a high temporal consistency. The deviation metric values of another detection point (50, 200) at three time points are [0.062, 0.035, 0.028], which have a large fluctuation and are considered to be noise interference. Regions with high spatial connectivity and high temporal consistency are preferentially identified as defect regions.

[0120] When extracting continuous regions that exceed a preset deviation threshold from the clustered area of ​​deviation metric values, an appropriate deviation threshold, such as 0.075, is set based on historical detection data and expert experience. This threshold can be dynamically adjusted based on the material characteristics and production process of seamless steel pipes to identify all detection points whose deviation metric values ​​exceed this threshold and to verify the spatial continuity of these points.

[0121] For example, in the above example, a continuous region containing detection points such as (43, 127) and (43, 126) is identified. Its average deviation metric value is 0.085, which exceeds the preset threshold. The area, shape features, and boundary characteristics of the continuous region are calculated. When the area of ​​the region exceeds the preset minimum size (such as 12 continuous detection points) and the shape is relatively regular (the ratio of perimeter to area is less than the preset threshold), the region is marked as a suspected defect region. The outline, location, area, and severity of the suspected defect region are saved for subsequent analysis and verification. At the same time, a visualization result is generated, which displays the distribution of suspected defects on the seamless steel pipe unfolding diagram in a color-coded manner, making it easier for technicians to quickly locate and evaluate defects.

[0122] In one optional implementation, a two-dimensional correlation clustering is performed on the spatiotemporal evolution matrix. By calculating the connectivity index of each detection point in the spatial dimension and the consistency index in the temporal dimension, multiple regions with deviations from the metric values ​​are identified, including:

[0123] Based on the spatiotemporal evolution matrix, the deviation metric value of each detection point is quantified, and the spatial correlation between the deviation metric values ​​of each detection point and its neighboring detection points is obtained to obtain the connectivity index. The stability of the deviation metric value of each detection point on the continuous time series is calculated to obtain the consistency index.

[0124] The connectivity index and the consistency index are coupled in two dimensions to construct a two-dimensional correlation evaluation function, and the two-dimensional correlation values ​​of each detection point in terms of spatial connectivity and temporal consistency are calculated.

[0125] Based on the dual-dimensional correlation degree values, iterative clustering is performed on each detection point. Detection points that meet the preset correlation conditions are grouped into the same cluster. The cluster boundaries are determined according to the distribution characteristics of the dual-dimensional correlation degree values ​​of the detection points in each cluster. Multiple deviation value clustering regions that are spatially connected and temporally consistent are identified.

[0126] In the embodiment, based on the spatiotemporal evolution matrix, the spatial correlation degree between the deviation metric value of each detection point and the deviation metric value of the adjacent detection point is quantified to obtain the connectivity index. Specifically, the deviation metric values of all detection points at the t-th moment are extracted from the spatiotemporal evolution matrix to form a spatial distribution sequence. For the detection point a, the difference in the deviation metric value between the detection point a and the adjacent detection point b is investigated. Let the deviation metric value of the detection point a at the t-th moment be P(a, t), and the deviation metric value of the detection point b at the t-th moment be P(b, t). The relative difference value D(a, b, t) = |P(a, t) - P(b, t)| / max(P(a, t), P(b, t)) is calculated. When the deviation metric values of the adjacent detection points are close, the value of D(a, b, t) is smaller, indicating that these points have higher correlation in space. A spatial weight function W(a, b) is introduced, which decreases with the increase of the physical distance between the detection points a and b. The connectivity index S(a) of the detection point a is calculated as S(a) = Σ(W(a, b) x (1 - D(a, b, t))), b ∈ N(a), where N(a) represents the neighborhood set of the detection point a. In actual application, the detection points within a radius of 50 meters around the detection point a can be selected as its neighborhood points.

[0127] The stability degree of the deviation metric value of each detection point in the continuous time sequence is calculated to obtain the consistency index. For the detection point a, the variation trend of the deviation metric value in the time dimension is analyzed. The deviation metric value sequence {V(a, t-k), V(a, t-k+1),..., V(a, t)} of the detection point a in the time window [t-k, t] is investigated. The discrete degree of the sequence is calculated. In actual application, the time window can be set to 24 hours, and the sampling interval is 1 hour. The time fluctuation index F(a, t) is introduced to represent the fluctuation of the deviation metric value of the detection point a in the specified time window. F(a, t) can be obtained by calculating the coefficient of variation of the deviation metric value sequence. The smaller the coefficient of variation, the higher the consistency of the detection point in the time dimension. The consistency index T(a) of the detection point a is defined as T(a) = 1 - F(a, t), and the value range is [0, 1]. The larger the value, the stronger the time consistency.

[0128] The connectivity index and the consistency index are coupled in two dimensions, a two-dimensional correlation degree evaluation function is constructed, the two-dimensional correlation degree values of each detection point in spatial connectivity and time consistency are calculated, a two-dimensional evaluation space is constructed, the horizontal coordinate represents the spatial connectivity index S(a), the vertical coordinate represents the time consistency index T(a), the detection point a can be represented as the coordinate point (S(a), T(a)), the two-dimensional correlation degree evaluation function C(a) is defined as C(a) = a x S(a) + b x T(a), wherein a and b are weight coefficients, a+b = 1, reflecting the relative importance of spatial connectivity and time consistency in anomaly identification, in actual application, the weight can be adjusted according to the specific scene, and the typical value is a = 0.6, b = 0.4, indicating that the spatial connectivity is slightly heavier than the time consistency.

[0129] Based on the two-dimensional correlation degree values, the iteration clustering processing is performed on each detection point, the density clustering idea is adopted, the correlation degree threshold C th and the distance threshold D th are set, initially, all detection points are not classified, the highest unclassified detection point in the two-dimensional correlation degree is selected as a seed point, a new clustering cluster is started, the seed point is marked as part of the current clustering cluster and added to the processing queue, a detection point is taken out from the queue, the surrounding detection points are checked, if the two-dimensional correlation degree C(b) of the adjacent detection point is greater than the threshold C th , and the physical distance from the current detection point is less than the distance threshold D th , the adjacent detection point is added to the current clustering cluster and the processing queue, the above process is repeated until the processing queue is empty, the formation of a clustering cluster is completed, the highest point in the remaining unclassified points is selected as a new seed point, the formation of the next clustering cluster is started, and all points with a correlation degree greater than the threshold C th are classified into a certain clustering cluster.

[0130] By judging the clustering boundary, multiple deviation metric value aggregation areas which are distributed in space and evolve in time are identified, for each formed clustering cluster, the boundary characteristics are analyzed. The average two-dimensional correlation degree C avg of the detection points in the clustering cluster and the correlation degree gradient between the boundary detection points and the non-clustering area detection points are calculated, the boundary detection points are defined as the points adjacent to at least one non-clustering area detection point in the clustering cluster, the boundary definition index B is defined as the average difference between the correlation degrees of the boundary detection points and the adjacent non-clustering area detection points, when the B value is greater than a preset threshold B th , it is considered that the clustering cluster boundary is clear, and it is a clear deviation metric value aggregation area.

[0131] In an alternative embodiment, for each defect region, a waveform feature parameter of the corresponding ultrasonic echo signal sequence is extracted, and matched with a preset defect type discrimination criterion to determine the defect type and defect position, and obtain a defect recognition result, including:

[0132] For each defect region, a multi-dimensional waveform feature parameter of the corresponding ultrasonic echo signal sequence is extracted, the multi-dimensional waveform feature parameter including an amplitude type feature parameter reflecting the geometric shape of the defect, and a frequency spectrum type feature parameter reflecting the internal structure of the defect.

[0133] Based on the multi-dimensional waveform feature parameter, a multi-level matching is performed with a preset defect type discrimination criterion, the multi-level matching including a preliminary classification of the amplitude type feature parameter to determine a macroscopic category of the defect, and a refined identification of the frequency spectrum type feature parameter to determine a microscopic characteristic of the defect.

[0134] According to the matching result of the macroscopic category and microscopic characteristic, in combination with spatial position information of the ultrasonic echo signal sequence, a defect recognition result containing a defect type identification and a defect position coordinate is generated.

[0135] In the specific embodiment, for each identified defect region, a corresponding ultrasonic echo signal sequence is extracted from original ultrasonic detection data, and according to the coordinate range of the defect region, an A-scan signal set of the relevant region is intercepted from a complete three-dimensional ultrasonic data body (X-Y-Z), and to ensure signal integrity, a boundary is expanded outward by 5 mm as a buffer zone. For each defect region, usually 50-100 groups of A-scan signals are extracted for feature analysis.

[0136] The multi-dimensional waveform feature parameter of the ultrasonic echo signal sequence includes: a peak amplitude (PA) calculated as a ratio of the maximum amplitude of the defect echo signal to the echo amplitude of the reference reflecting surface, usually expressed in dB; an echo duration (ED) measured as a length of a duration period in which the echo signal amplitude exceeds a background noise threshold, in microseconds; a waveform symmetry degree (WS) calculated as a left-right symmetry coefficient of the defect signal waveform, with a value range of 0-1, and the closer to 1, the more symmetrical; and an amplitude fluctuation degree (AV) calculated as a standard deviation of the amplitude in the duration period divided by the average amplitude. For a typical under-thickness defect, PA=0.85 (indicating 1.4 dB weaker than the reference surface), ED=0.65 microseconds, WS=0.78, and AV=0.12 are obtained.

[0137] The spectral feature parameters include: main frequency component (MF), the frequency spectrum of the defect echo signal is calculated by fast Fourier transform, and the frequency corresponding to the highest energy point is extracted; band width (BW), the full width at half maximum in the frequency spectrum is calculated; frequency concentration (FC), the concentration degree of signal energy around the main frequency within the range of ±1 MHz is measured; high frequency attenuation ratio (HAR), the ratio of high frequency band (such as more than 5 MHz) energy to total energy is calculated. For a defect sample containing a crack, the feature parameters are MF = 3.2 MHz, BW = 2.1 MHz, FC = 0.68, and HAR = 0.42.

[0138] Based on the extracted multi-dimensional waveform feature parameters, multi-level matching is performed with the preset defect type discrimination criteria. The amplitude feature parameters are used for macro classification, and the defects are preliminarily divided into three categories: volume type defects (such as pores, inclusions), surface type defects (such as cracks, delaminations) and geometric type defects (such as thickness changes, steps). The discrimination criteria include: volume type defects usually have higher waveform symmetry (WS>0.75) and shorter echo duration (ED<0.5 microseconds); surface type defects are characterized by lower waveform symmetry (WS<0.6) and longer echo duration (ED>0.8 microseconds); geometric type defects exhibit moderate waveform symmetry (0.6<WS<0.75) and lower amplitude fluctuation (AV<0.15).

[0139] After macro classification, spectral feature parameters are used for refined identification to determine the micro characteristics of the defects. For volume type defects, the main frequency component and the band width can be used to distinguish between pores (MF close to the center frequency of the probe, BW narrow) and inclusions (MF deviates from the center frequency of the probe, BW wide); for surface type defects, the high frequency attenuation ratio and the frequency concentration can be used to distinguish between cracks (HAR>0.4, FC<0.7) and delaminations (HAR<0.3, FC>0.8); for geometric type defects, the comprehensive analysis of amplitude features and spectral features can distinguish between thickness changes and step structures.

[0140] In practical application, a feature parameter database is established in advance, which contains the feature parameter ranges of various typical defects. For example, the parameter ranges of pore type defects are set as: PA = 0.6-0.9, ED = 0.3-0.5 microseconds, WS = 0.8-0.95, AV = 0.1-0.25, MF close to the center frequency of the probe ±0.5 MHz, BW = 1.5-2.5 MHz, FC = 0.7-0.9, HAR = 0.2-0.3. By calculating the matching degree of the feature parameters of the defect to be identified with the parameter ranges of each defect type in the database, the defect type with the highest probability is determined.

[0141] According to the matching result of the macro category and the micro characteristic, a defect recognition result is generated in combination with spatial position information of the ultrasonic echo signal sequence, the spatial position information including plane coordinates (X-Y coordinates) and depth coordinates (Z coordinates) of the defect, the plane coordinates being determined by a position of the defect region on a C-scan image, and the depth coordinates being calculated by calculating a time delay from a transmission pulse to a defect echo signal and converted in combination with a material sound speed, wherein the material sound speed can be obtained by measuring a standard test block.

[0142] The defect recognition result is output in a data table form, each record containing a defect ID, a defect type, defect position coordinates (X, Y, Z), defect size (length, width, estimated height), and a confidence score, for example, for a defect identified as a pore, the record is: ID=12, type="pore", position=(125.3mm, 78.2mm, 4.5mm), size=(2.1mm, 1.8mm, 1.5mm), confidence=0.92, and these results can be further used for defect visualization display and defect severity evaluation.

[0143] The seamless steel pipe inner wall ultrasonic testing data real-time analysis and defect recognition system of the embodiment of the present application comprises:

[0144] The first unit is configured to acquire an ultrasonic echo signal sequence of the inner wall of the seamless steel pipe.

[0145] The second unit is configured to perform adaptive modal decomposition on each ultrasonic echo signal to obtain a plurality of intrinsic mode functions and a residual trend component, perform differential feature extraction, and obtain a feature vector set.

[0146] The third unit is configured to divide the feature vector set into a plurality of spatial grid units, extract baseline values of each feature dimension, calculate reference feature vectors of each spatial grid unit in combination with neighbor units and historical reference characteristic vectors, and establish an adaptive reference model of a normal inner wall state.

[0147] The fourth unit is configured to calculate deviation metric values of each feature vector of the inner wall of the seamless steel pipe according to the adaptive reference model, and perform correlation clustering of the deviation metric values in spatial and time dimensions, and identify a continuous region with a deviation metric value exceeding an adaptive threshold as a defect region.

[0148] The fifth unit is configured to extract waveform feature parameters of the ultrasonic echo signal sequence corresponding to each defect region, and match the waveform feature parameters with preset defect type discrimination criteria to determine a defect type and a defect position of the defect region, and obtain a defect recognition result.

[0149] The third aspect of the embodiment of the present application provides an electronic device comprising:

[0150] a processor;

[0151] a memory for storing processor-executable instructions;

[0152] wherein the processor is configured to invoke the instructions stored by the memory to perform the method as described above.

[0153] In a fourth aspect, the present application provides a computer readable storage medium having stored thereon computer program instructions which, when executed by a processor, implement the method as described above.

[0154] The present application can be a method, an apparatus, a system, and / or a computer program product. The computer program product can include a computer readable storage medium having computer readable program instructions stored therein, which, when executed by a processor, perform various aspects of the present application.

[0155] It should be noted that the above-mentioned embodiments are merely used to illustrate the technical solutions of the present application, rather than limiting the present application; although the present application has been described in detail with reference to the above-mentioned embodiments, those skilled in the art should understand that the technical solutions recorded in the above-mentioned embodiments can be modified or equivalent replacements can be made to some or all of the technical features; and the modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for real-time analysis of ultrasonic testing data and defect recognition of the inner wall of a seamless steel pipe, characterized in that, The method comprises the following steps: obtaining an ultrasonic echo signal sequence of the inner wall of a seamless steel pipe; performing adaptive modal decomposition on each ultrasonic echo signal to obtain a plurality of intrinsic mode functions and a residual trend component, and performing differential feature extraction to obtain a feature vector set, comprising: performing adaptive modal decomposition on each ultrasonic echo signal to extract oscillation components of different frequency scales through iteration, and decomposing the ultrasonic echo signal into a plurality of intrinsic mode functions and a residual trend component, wherein the number of intrinsic mode functions is adaptively determined according to the energy distribution of the ultrasonic echo signal; performing complex domain transformation on each intrinsic mode function to obtain an analytical signal, extracting its instantaneous phase sequence, and performing time differentiation operation on the instantaneous phase sequence to obtain an instantaneous frequency variation trajectory, and obtaining an instantaneous amplitude envelope curve by calculating the modulus sequence of the analytical signal; performing multi-order difference operation on the residual trend component in the time axis to extract a gradient sequence and its corresponding statistical features as gradient variation features; aligning the instantaneous frequency variation trajectory and the instantaneous amplitude envelope curve in time, combining the gradient variation features to obtain a feature vector set of each ultrasonic echo signal; dividing the feature vector set into a plurality of spatial grid units, extracting the baseline value of each feature dimension, and combining the neighborhood units and historical reference characteristic vectors to calculate the reference feature vector of each spatial grid unit, and establishing an adaptive reference model of the normal inner wall state; According to the adaptive reference model, calculate the deviation measure value of each feature vector of the inner wall of the seamless steel pipe, and associate and cluster it in the spatial dimension and the time dimension, identify the continuous area whose deviation measure value exceeds the adaptive threshold as the defect area; For each defect area, extract the waveform feature parameters of the corresponding ultrasonic echo signal sequence, and match them with the preset defect type discrimination criterion to determine the defect type and defect position, and obtain the defect recognition result.

2. The method of claim 1, wherein, The feature vector set is divided into a plurality of spatial grid units, the baseline value of each feature dimension is extracted, and the neighborhood units and historical reference characteristic vectors are combined to calculate the reference feature vector of each spatial grid unit, and an adaptive reference model of the normal inner wall state is established, comprising: The feature vector set is divided into a plurality of spatial grid units according to the circumferential position index and the axial position index, forming a plurality of spatial grid units, each spatial grid unit corresponding to a group of feature vector subsets; For each feature vector subset, calculate the baseline value of each feature dimension to obtain an initial reference feature vector, and perform spatial adaptive smoothing processing on the initial reference feature vectors of each spatial grid unit and its neighborhood units to obtain a smoothed reference feature vector; From the historical detection data, extract the historical reference feature vector of each spatial grid unit, calculate the time decay weight according to the time interval between the historical detection period and the current detection period, and calculate the cumulative weight according to the cumulative sample number; Based on the time decay weight and the cumulative weight, the smoothed reference feature vector is weighted and summed with the historical reference feature vector to obtain a reference feature vector of each spatial grid cell, and an adaptive reference model of a normal inner wall state is constructed.

3. The method of claim 2, wherein, For each feature vector subset, baseline values of each feature dimension are calculated to obtain an initial reference feature vector. The initial reference feature vectors of each spatial grid cell and its neighboring cells are subjected to spatial adaptive smoothing processing to obtain a smoothed reference feature vector, including: For each feature vector subset corresponding to each spatial grid cell, the numerical distribution of all feature vectors in each feature dimension is extracted, and a robust statistical analysis is performed on the numerical distribution in each feature dimension to calculate a center position estimator resistant to outlier interference as the baseline value of each feature dimension, and the initial reference feature vector of each spatial grid cell is obtained by combination; For each spatial grid cell, its neighborhood cells with spatial continuity in the circumferential direction and the axial direction are calculated, and the corresponding initial reference feature vectors are extracted to form a neighborhood reference feature vector group; According to the spatial distance between each neighborhood cell and the spatial grid cell, a spatial decay function is constructed, and a sample reliability function is constructed according to the number of samples of each neighborhood cell. Based on the spatial decay function and the sample reliability function, the neighborhood initial baseline feature vector group is subjected to filtering and smoothing operation to obtain the smoothed reference feature vector of each spatial grid cell.

4. The method of claim 1, wherein, According to the adaptive reference model, the deviation measure value of each feature vector of the inner wall of the seamless steel pipe is calculated, and it is associated and clustered in spatial dimension and time dimension to identify a continuous region with deviation measure value exceeding an adaptive threshold as a defect region, including: According to the reference feature vector of each spatial grid cell in the adaptive reference model, the deviation measure value of each detection point feature vector of the inner wall of the seamless steel pipe is calculated; The deviation measure value is constructed into a two-dimensional spatial distribution map according to the circumferential index and the axial index, and a space-time evolution matrix is constructed in combination with the detection time sequence, wherein the space-time evolution matrix records the deviation measure value change trajectory of each detection point at multiple detection time points; The space-time evolution matrix is subjected to two-dimensional correlation clustering, and by calculating the connectivity index of each detection point in the spatial dimension and the consistency index in the time dimension, a plurality of deviation measure value aggregation regions are identified; From the deviation measure value aggregation region, a continuous region with deviation measure value exceeding an adaptive threshold is extracted, and the continuous region is marked as a suspected defect region.

5. The method of claim 4, wherein, The space-time evolution matrix is subjected to two-dimensional correlation clustering, and by calculating the connectivity index of each detection point in the spatial dimension and the consistency index in the time dimension, a plurality of deviation measure value aggregation regions are identified, including: Based on the space-time evolution matrix, the spatial correlation degree between the deviation measure value of each detection point and the deviation measure value of its adjacent detection point is quantified to obtain a connectivity index, and the stability degree of the deviation measure value of each detection point in a continuous time sequence is calculated to obtain a consistency index; The connectivity index and the consistency index are analyzed in two dimensions, a two-dimensional correlation degree evaluation function is constructed, and a two-dimensional correlation degree value of each detection point in spatial connectivity and temporal consistency is calculated; Based on the two-dimensional correlation degree value, each detection point is iteratively clustered, detection points meeting a preset correlation condition are merged into the same cluster, and the two-dimensional correlation degree value distribution characteristics of the detection points in each cluster are used to determine the cluster boundary, and multiple deviation metric value aggregation areas that are continuously distributed in space and consistently evolved in time are identified.

6. The method of claim 1, wherein, For each defect area, the waveform feature parameters of the corresponding ultrasonic echo signal sequence are extracted, and matched with a preset defect type discrimination criterion to determine the defect type and defect position, and a defect recognition result is obtained, including: For each defect area, the multi-dimensional waveform feature parameters of the corresponding ultrasonic echo signal sequence are extracted, the multi-dimensional waveform feature parameters including amplitude type feature parameters reflecting the geometric shape of the defect and spectrum type feature parameters reflecting the internal structure of the defect; Based on the multi-dimensional waveform feature parameters, a multi-level matching is performed with a preset defect type discrimination criterion, the multi-level matching including a preliminary classification of the amplitude type feature parameters to determine the macro category of the defect, and a refined identification of the spectrum type feature parameters to determine the microscopic characteristics of the defect; According to the matching results of the macro category and the microscopic characteristics, and in combination with the spatial position information of the ultrasonic echo signal sequence, a defect recognition result containing a defect type identifier and a defect position coordinate is generated.

7. A system for real-time analysis of ultrasonic testing data and defect recognition of the inner wall of seamless steel pipes, for implementing the method according to any one of claims 1 to 6, characterized in that, Including: A first unit configured to acquire an ultrasonic echo signal sequence of an inner wall of a seamless steel pipe; A second unit configured to perform adaptive modal decomposition on each ultrasonic echo signal to obtain a plurality of intrinsic mode functions and a residual trend component, and perform differential feature extraction to obtain a feature vector set; A third unit configured to divide the feature vector set into a plurality of spatial grid units, extract baseline values of each feature dimension, calculate baseline feature vectors of each spatial grid unit in combination with neighborhood units and historical reference characteristic vectors, and establish an adaptive reference model of a normal inner wall state; A fourth unit configured to calculate deviation metric values of each feature vector of the inner wall of the seamless steel pipe based on the adaptive reference model, and perform correlation clustering in spatial and temporal dimensions to identify continuous regions with deviation metric values exceeding an adaptive threshold as defect regions; A fifth unit configured to extract waveform feature parameters of the corresponding ultrasonic echo signal sequence of each defect area, and match the waveform feature parameters with a preset defect type discrimination criterion to determine the defect type and defect position, and obtain a defect recognition result.

8. An electronic device, comprising: Including: A processor; A memory for storing processor-executable instructions; The processor is configured to invoke the instructions stored in the memory to execute the method of any one of claims 1 to 6.

9. A computer-readable storage medium having stored thereon computer program instructions, wherein, The computer program instructions are executed by the processor to implement the method of any one of claims 1 to 6. The computer program instructions are executed by the processor to implement the method of any one of claims 1 to 6.

Images